How Sustained Low Interest Rates Reshape Pension Strategic Asset Allocation: Cross-Country Evidence from Mature Pension Markets

1. Literature and Institutional Background Review

This section proceeds from verifiable public sources to lay out the institutional facts and the literature coordinates on which this paper's argument rests, and thereby situates the conventional analytical framework that this paper seeks to reassess. In this paper, the conventional analytical framework refers to a research tradition that renders the mechanism by which persistently low interest rates reshape pension asset allocation as a direct causal arrow — market interest rates decline, the present value of liabilities rises, the duration gap widens, rational sponsors turn to liability-driven investment (LDI) — hence selling equities and buying long-duration bonds — and treats "the interest rate" as a clean scalar that enters the allocation decision function directly, adjudicating whether "reaching for yield dominates or ALM pressure dominates" by whether a regression coefficient is significant. The entire endeavor of this paper is a systematic reassessment of the identification validity and mechanistic completeness of this conventional framework.

At the level of stylized facts concerning asset scale and allocation patterns, mature pension markets carry the overwhelming bulk of global pension assets. According to the Thinking Ahead Institute's Global Pension Assets Study 2026, the group of markets this paper focuses on — the United States, the United Kingdom, the Netherlands, Canada, Australia, Japan, and Switzerland (hereafter defined as P7) — accounts for approximately 91% of global pension assets (WTW 2026). In terms of allocation patterns, over a window of roughly twenty years, the average strategic allocation of P7 has undergone an observable reordering — equities at approximately 48%, bonds at approximately 31%, other assets at approximately 19%, and cash at approximately 3%, with the equity share down by about 9 percentage points from twenty years earlier, while the share of "other" assets (alternatives, private equity, infrastructure, etc.) has risen by about 6 percentage points (WTW 2026). In terms of institutional form, the benefit obligations of this group of markets are migrating from defined benefit (DB) to defined contribution (DC), with the DC asset share having reached approximately 63% — though this average masks substantial internal divergence (72% in the United States, 90% in Australia, 44% in Canada) (WTW 2026). It is precisely the still-substantial DB stock that constitutes the arena in which interest-rate shocks genuinely bite on the liability side and trigger the strategic allocation responses with which this paper is concerned.

On the specific transmission channel from "interest rates → overseas allocation," one widely cited relationship in the literature is that each 1-percentage-point decline in the domestic-currency government bond yield is associated with an approximately 3.4-percentage-point increase in the overseas asset share, attributed to reaching-for-yield motives; its finer decomposition shows that a 1-percentage-point yield decline corresponds to roughly a 1-percentage-point decline in the bond share, a roughly 0.5-percentage-point increase in the equity share (approximately 1.5 percentage points for emerging-market equities), and an approximately 3.2-percentage-point increase in the overseas share for DB plans (BIS 2024). Chapter 3 of this paper will argue that this relationship is a naive-correlation-plus-reaching-for-yield narrative — precisely the object that needs to be reassessed. A related estimate in the same direction holds that, after stripping out valuation effects, each 1-percentage-point decline in the risk-free rate corresponds to an approximately 0.66-percentage-point increase in risk-asset exposure (Konradt 2023).

On the relationship between demographics and the equilibrium interest rate, the literature is by now quite mature. Gagnon, Johannsen, and López-Salido show that demographic factors alone can explain approximately 1.25 percentage points of the decline in the equilibrium real interest rate since 1980, with the original text explicitly characterizing this contribution as "much, if not all" of the permanent decline (Gagnon, Johannsen, and López-Salido 2016). This fact directly supports the premise in Chapter 1 of this paper that "demographics are an important upstream driver of the equilibrium real interest rate." Another upstream channel related to the pension liability side is plan maturity: as plan membership ages and cash flows turn from net inflows to net outflows, liability duration is independently lengthened, and demand for long-dated bonds and LDI is independently pulled up; research from the Organisation for Economic Co-operation and Development (OECD) and the Global Financial Stability Report of the International Monetary Fund (IMF) document a consistent set of facts pointing toward maturity driving bond allocation, net cash flow turning negative, and duration mismatch (OECD 2024; IMF 2025).

At the level of accounting standards, International Accounting Standard 19 (as revised in 2011, hereafter IAS19R), which took effect in 2013, constitutes the core institutional fact of Chapter 2 of this paper. This standard, effective January 1, 2013, eliminated the corridor method (removing deferred smoothing of actuarial gains or losses), eliminated the discretionary recognition of expected return on plan assets (ERR), replaced it with the net interest method (computed as the discount rate multiplied by the net defined benefit liability or asset), and required that all remeasurement differences be recognized in other comprehensive income (OCI) (IFRS Foundation 2011). On the published-evidence side, Anantharaman and Chuk, using a Canadian sample, demonstrate that following the adoption of IAS19R, firms shifted pension assets from equities to bonds, attributing this primarily to the elimination of the ERR — that is, replacing the ERR with the discount rate reduced the risk firms took on in pension investment, as reflected in their pension asset allocation (Anantharaman and Chuk 2018). This finding supports the core claim of Chapter 2 of this paper.

At the level of taxation, the classic framework of Black and Tepper shows that fully funding a pension, holding bonds within the pension account, and shifting equity risk to shareholders' own portfolios is tax-efficient (Black 1980; Tepper 1981). The Tax Cuts and Jobs Act (TCJA) of 2017 reduced the corporate tax rate from 35% in 2017 to 21% in 2018, and firms rushed to make 2017 DB contributions ahead of the tax cut in order to deduct them at the higher rate; related research documents an average increase of approximately 27% in DB contributions in 2017 (Gaertner, Lynch, and Vernon 2018), and other research documents that the tax incentive induced approximately an additional $3 billion in contributions among medium and large plans, with a sharp reversal in 2018 (BIS 2021).

At the level of settlement and collateral, the European Market Infrastructure Regulation (EMIR) requires all counterparties trading in the European Union to post variation margin (VM) in cash rather than in bonds, while pension funds do not naturally hold large amounts of cash (Macfarlanes 2019). The timeline of the pension clearing exemption is also publicly verifiable: on June 9, 2022, the European Commission adopted a delegated regulation extending the exemption by one year, with the EU exemption ultimately expiring on June 18, 2023, while the UK Treasury extended the exemption for UK and European Economic Area pensions by two years, to June 18, 2025 (Norton Rose Fulbright 2023). As a manifest event exposing this institutional constraint, following the UK "mini-budget" of September 23, 2022, gilt yields spiked sharply, triggering margin and collateral calls exceeding £70 billion for LDI funds and pension schemes, prompting the Bank of England to conduct emergency targeted gilt purchases to restore order (Federal Reserve Bank of Chicago 2023; Bank of England 2022). The Bank of England has also flagged margin procyclicality as an international policy concern, documenting the vicious spiral of collateral calls and forced gilt sales (Bank of England 2022). These two sets of institutional facts constitute the public evidentiary basis for Chapter 5 of this paper.

Finally, at the level of behavioral fingerprints during the decumulation (payout) phase, the pension risk transfer market has set records. US pension risk transfer (PRT) recorded 568 transactions and $51.8 billion in premiums in 2022, a historical record (Aon 2023); UK buyout and buy-in transactions reached approximately £49–50 billion in 2023, also a historical record (LCP 2024). Chapter 4 of this paper will use this fingerprint to substantiate the corresponding time-series claim.

In sum, the conventional analytical framework compresses the institutional heterogeneity described above into a single exogenous interest-rate scalar, summarizing pension behavior with a single "interest rate → allocation" slope; this paper instead argues that it is precisely these compressed-away institutional layers — accounting conventions, hedging costs and regulation, behavioral reference points, and settlement physics — that carry the true mechanism by which the allocation function is rewritten.

2. Introduction: Question, Judgment, and Structure

(1) Research Background: A Receding Era of Low Interest Rates, and a Class of Long-Term Liabilities That Cannot Exit

Over the past four decades, real interest rates in the world's major advanced economies have undergone a long-cycle, seemingly structural decline. By around 2020, both nominal and real long-end sovereign rates had simultaneously hit historic lows, and "secular stagnation" moved from a fringe hypothesis to a shared operating premise across mainstream monetary policy and asset management circles. After 2022, however, as inflation returned and monetary policy tightened rapidly, this era began to recede at an equally steep pace. For the vast majority of financial institutions, this rise and fall in the interest-rate environment amounts merely to a single balance-sheet remeasurement; but for pension funds — a peculiar class of long-horizon institutional investors — it touches the very foundation of institutional survival, because pension liabilities do not exit alongside interest rates: they are long-term obligations spanning decades, jointly locked in by demographic structure and benefit promises, and virtually incapable of being actively wound down.

The object of study in this paper focuses on mature pension markets. This paper uses P7 to denote the set of markets — the United States, the United Kingdom, the Netherlands, Canada, Australia, Japan, and Switzerland — that feature relatively complete disclosure and relatively mature institutions. The focus on this group is because these markets not only carry the overwhelming bulk of global pension assets, but their internal institutional heterogeneity also happens to provide a natural basis for comparison in identifying mechanisms. In terms of asset scale, P7 accounts for approximately 91% of global pension assets, making it an unavoidable core sample for understanding global pension behavior (WTW 2026). In terms of allocation patterns, over a window of some twenty years, the average strategic allocation of P7 has undergone an observable reordering — equities at approximately 48%, bonds at approximately 31%, other assets at approximately 19%, and cash at approximately 3% — with the equity share having shifted down significantly from twenty years earlier, and the "other" share having risen significantly (WTW 2026). In terms of institutional form, the benefit obligations of this group of markets are migrating from DB to DC, with the DC asset share having reached approximately 63% — though this average masks substantial internal divergence (72% in the United States, 90% in Australia, 44% in Canada) — and it is precisely the still-substantial DB stock that constitutes the arena in which interest-rate shocks genuinely bite on the liability side and trigger all of the strategic allocation responses with which this paper is concerned (WTW 2026).

This background is especially acute for pensions because of two unavoidable features of their liabilities. First, liabilities are long in duration and highly sensitive to interest rates: under discounting arithmetic, the present value of liabilities is monotonically decreasing in the discount rate, and each downward step in interest rates correspondingly inflates the present value of liabilities — a point beyond dispute, and the common accounting starting point shared by all four competing explanations. Second, liabilities cannot exit: unlike banks, which can shrink their balance sheets, or insurers, which can stop selling new policies, the benefit promises of a mature DB plan are already locked in for existing beneficiaries, and a plan whose cash flow has turned deeply negative must still hold duration-matched assets to avoid forced fire sales, even in the face of elevated interest rates. It is the superposition of these two features that means the question of "how the interest-rate environment reshapes pension funds' strategic asset allocation" cannot be reduced to an ordinary asset remeasurement, but must instead be understood as a question about the behavioral function of a long-term liability-holding entity under a changing interest-rate regime.

(2) The Core Question: Not "What Was Bought," but "Why the Allocation Function Was Systematically Rewritten"

The core question of this paper is: through what mechanism did persistently low interest rates systematically alter pension funds' strategic asset allocation function?

It must first be clarified that this is not a question of describing a phenomenon. The claim that "pension funds bought more equities and alternatives, and increased their exposure to long-dated bonds and LDI, during the low-rate period" is broadly true in the data, but it is not itself the question this paper seeks to answer — at most it is the starting point this paper seeks to explain. What this paper seeks to answer is why low interest rates systematically altered the allocation function — that is, under what mechanism an interest-rate movement exogenous to pension funds is transmitted, refracted, and amplified, ultimately rewriting these institutions' strategic decision rules about what to hold, how much can be held, and what reference point to adjust against. Describing the phenomenon answers "where did the allocation move"; explaining the mechanism answers "how was the allocation's decision function itself rewritten." This distinction is precisely where this paper departs from the large body of literature in the field that merely restates regression coefficients.

Among the many possible mechanisms, this paper is organized throughout around a single focal judgment — the core of the entire paper: does reaching for yield dominate, or does ALM pressure dominate?

These two mechanisms can, on the surface, both explain the same set of phenomena — pension funds increasing their allocation to risk assets and long-dated bonds during periods of low interest rates — but the behavioral nature underlying them is entirely different, and their policy implications are even diametrically opposed. If reaching for yield dominates — that is, pension funds, facing insufficient fixed-income returns under low rates, actively raise their risk appetite to meet target return objectives and rush into equities and alternatives — then the policy focus should be on managing their risk appetite and procyclical behavior. If ALM pressure dominates — that is, pension funds passively adjust their asset side because the present value of liabilities inflates as rates decline, in order to match duration and control the asset-liability gap — then the policy focus should be on managing their discount-rate rules and liability measurement conventions. The former calls for managing risk appetite; the latter calls for managing discount-rate rules — the two point toward entirely different regulatory toolkits. The dispute between "reaching for yield and ALM" is thus, in essence, a judgment about the fundamental nature of pension fund behavior, not a technical matter that can be settled merely by estimating a larger regression and checking whether a coefficient is significant. This judgment stance constitutes the anchor for all subsequent analysis in this paper.

(3) The Paper's Overall Judgment: The Observed "Interest Rate → Allocation" Coefficient Is Unreliable; the True Mechanism Is Hidden in Four Neglected Layers

The overall judgment of this paper can be summarized in one sentence, which is also the paper's most central contribution relative to the existing literature: the observed "interest rate → allocation" coefficient is unreliable owing to a confounding demographic common cause and demand-side reflexivity; the true mechanism does not lie in this contaminated coefficient, but is hidden in four neglected layers — accounting conventions, hedging, behavior, and settlement.

This overall judgment contains two progressively layered parts.

The first part is a negative, foundational identification warning, which constitutes the entire task of Chapter 1 of this paper. The most common approach in the literature is to write strategic asset allocation as a dependent variable of interest rates, regress "interest rate → allocation" into a single slope, and default to treating interest rates as exogenous to pension funds. This paper contends: this default is precisely least valid during a prolonged period of low interest rates. The observed "interest rate → allocation" slope is contaminated or reversed by two forces simultaneously — first, population aging as an omitted common cause, which both depresses the equilibrium real interest rate and independently raises pension funds' liability duration and demand for long-dated bonds, causing the interest-rate coefficient to absorb effects that properly belong to demographics; second, the reflexive production of the low-interest-rate environment by pension funds' rigid demand, whereby price-insensitive buying created by institutional design continuously depresses the term premium and prolongs the low-rate environment itself, thereby reversing the causal arrow from "interest rate → allocation" to "demand → environment." Consequently, before any structural causal interpretation is undertaken, this coefficient is already unreliable; the dispute between "reaching for yield dominance and ALM dominance" is, at this identification layer, judged to be a misdiagnosis of one and the same contaminated coefficient. This is the foundational identification warning that the entire empirical exercise of this paper must raise before proceeding.

The second part is a constructive mechanism reconstruction, which is the task of Chapters 2 through 5 of this paper; here this paper only previews its place within the overall judgment, without unpacking its internal identification. Given that the observed coefficient is unreliable, where does the true mechanism lie? This paper contends that the true mechanism driving the rewriting of the allocation function has been systematically neglected by the mainstream narrative within four layers:

• The accounting-convention layer: what enters the allocation function is never "the interest rate" itself, but the "shadow rates" refracted by four mutually misaligned institutional accounting regimes (the IAS19R and US Financial Accounting Standards Board Statement No. 715 profit-and-loss recognition conventions, the funding convention under Section 430 of the Internal Revenue Code, the tax-deductibility convention, and the nominal discounting convention); allocation drift is driven by the opening and closing of the wedge between these conventions, and can decouple from the path of the equilibrium real interest rate.
• The hedging layer: the true constraint on pension funds' overseas allocation is not the monotonic decline in the domestic-currency interest rate, but rather the post-hedge carry and the corner-solution-style regulatory penalty on currency mismatch; treating "a 1-percentage-point decline in the domestic-currency rate raises overseas assets" as an established causal relationship is a shallow misreading of correlation.
• The behavioral layer: the true trigger for de-risking on the liability side is often a behavioral reference point (such as whether a prior peak has been crossed, or whether a round-number threshold has been touched), rather than the outcome of first-order optimization; allocation adjustments consequently display the behavioral fingerprint of a one-directional ratchet and psychological clustering around round numbers.
• The settlement layer: what determines what pension funds can hold, and how much leverage they can add, during periods of stress is the settlement-physics constraint of collateral and margin (the species mismatch of cash variation margin, the reflexive loop of collateral valuation), rather than liability duration itself.

It must be stated with the utmost care: the introduction and Chapter 1 of this paper bear full evidentiary responsibility only for the first part above — the identification-layer warning — and within that, the only claim that is genuinely identified is the negative conclusion of "debunking the demographic common cause." The mechanism claims for the four layers in the second part above will be argued individually in the subsequent chapters; their identification strength varies, and some are downgraded to design-plus-literature synthesis owing to the unavailability of micro-level data. This paper will explicitly flag each such instance, chapter by chapter, and will never pass it off here as an already-identified conclusion.

(4) Preview of the Five-Chapter Structure

This paper unfolds progressively across five layers, forming a complete chain from identification warning to mechanism reconstruction:

• Chapter 1, Identification-Layer Warning: This is the foundation of the entire paper. It does not compete with any mechanism chapter for the same pool of residual variance, and answers only one question — why the observed "interest rate → allocation" coefficient cannot be read as structural causation. It accomplishes this through three legs: debunking the naive regression using a predetermined demographic instrument (the sole load-bearing leg), demand-side reflexivity as a theoretical proposition supplemented by decumulation-phase checks, and distributional politics as an explicitly unidentified qualitative case. This chapter is the identification gate through which all subsequent empirical claims must first pass.
• Chapter 2, The Accounting-Convention Layer: Argues that what enters the allocation function is the shadow rate rather than the interest rate, with its core identifiable evidence being the IAS19R standard change as a "quasi-exogenous interest-rate shock."
• Chapter 3, The True Constraint on Overseas Allocation: Replaces the shallow causality of "domestic-currency rate → monotonic overseas allocation" with "post-hedge carry" and "regulatory corner solutions."
• Chapter 4, The Reference-Point Ratchet: Argues for the behavioral trigger of liability-side de-risking and its systematic deviation from first-order optimization.
• Chapter 5, The Boundary of Settlement Physics: Argues that settlement constraints, rather than liability duration, determine the holding boundary during periods of stress.

(5) Methodological Stance: Judgment Precedes Data, with Identification Design at the Core

This paper self-consciously adopts an explicit methodological stance that runs throughout, and that directly determines the discipline with which this paper draws on evidence.

First, judgment precedes data. This paper's organizing principle is not "answer whatever question the available data permits," but rather "first establish a judgment about the fundamental nature of pension fund behavior, and then test it with an identification design." This means that when the micro-level identification data required for a key judgment are unavailable under the data conditions on which this study relies, this paper's choice is not to avoid that judgment and retreat to a question that the data can answer but that is of no consequence — rather, it is to honestly present the identification design, explicitly flag the data gap, and use the support or refutation that published literature offers for the direction predicted by that design as downgraded evidence. In all such instances, this paper explicitly labels the item as downgraded under identification-strategy and data-availability constraints, and never disguises a downgraded item as an already-identified empirical result.

Second, identification design is at the core, and logic is made explicit. This paper's primary criterion for judging whether a mechanism claim holds is not whether a regression coefficient is significant, but whether the identification design can genuinely separate that mechanism from its competing hypotheses. Accordingly, this paper deliberately makes logical relationships explicit in its argumentation — for every key inference, it strives to state, in the structure of premise, conclusion, and falsifiable condition, the boundary within which it holds, rather than using rhetoric to paper over gaps in identification.

Third, the evidentiary status of three types of propositions is honestly distinguished. This paper strictly distinguishes and separately labels: aggregate facts directly traceable to public sources; micro-level identification data that a load-bearing leg requires but that are unavailable owing to micro-data availability constraints; and pre-registered expected directions fixed in writing before the data were examined. What this paper calls "pre-registration" refers to an internal research commitment that fixes hypotheses, expected directions, and falsification thresholds in writing before the data are accessed, and is not affiliated with any external registry. Owing to micro-data availability constraints, this paper reports no regression point estimates, significance levels, or sample sizes that were not genuinely obtained; every quantitative figure appearing in the text is either a sourced public fact, an estimate reported by published literature by other authors (with attribution), or an explicitly labeled pre-registered direction awaiting estimation or a predicted sign. This discipline will be most concentratedly reflected in the treatment of the identification-layer warning in Chapter 1 below — because that chapter is precisely a specimen in which the identification design is exceptionally strong, yet the micro-level identification data are precisely unavailable.

3. Chapter 1 The Identification Warning: Why the Observed "Interest Rate → Allocation" Coefficient Cannot Be Read as Causal

3.1 The Positioning of This Chapter: A Foundational Identification Checkpoint That Only Falsifies, Never Asserts

This chapter is the foundation of the entire empirical exercise, but its posture is defensive and negative. It advances no positive claim about allocation, does not claim to have estimated the "true elasticity of allocation with respect to the interest rate," and does not compete with any subsequent mechanism chapter for the right to explain the same residual variance. It does exactly one thing: it demonstrates why the observed "interest rate → allocation" regression coefficient cannot be read as structural causation.

This positioning must be stated unambiguously at the outset, because it directly determines the evidentiary boundary of every claim in this chapter. The chapter rests on three legs with strictly divided labor, and the evidentiary status of the three legs is entirely different — they must never be conflated:

• Leg One, demographic-confounder debunking — this is the only load-bearing leg of the chapter, and the only place where a genuinely identified claim is made. It does nothing but debunk: it demonstrates that the naive "interest rate → allocation" coefficient is contaminated by demographics as an omitted common cause, and, where data permit, quantifies a lower bound on this contamination bias. It only falsifies; it never asserts, and under no circumstances does it flip into any positive allocation claim.
• Leg Two, demand-side reflexivity — this is a theoretical proposition, supplemented by a side test from the decumulation (payout) phase. It argues that the causal arrow may reverse during periods of low interest rates (demand → environment, rather than interest rate → allocation), but it explicitly does not claim to have identified causality; its evidentiary status is that of a sharply argued judgment supported by a side test, not an identified result.
• Leg Three, distributive-politics non-recognition — this is an explicitly unidentified qualitative case. It advances a third perspective distinct from both reaching for yield and asset-liability management (ALM) — the political refusal to acknowledge low rates, forcing increased equity allocation to prop up a high discount rate — but because of an accounting identity, it cannot be purely econometrically identified, and is therefore presented only through process tracing and narrative case study, entering no econometric identification surface whatsoever.

The core conclusion of this chapter claims only the negative result that the first leg genuinely earns through identification — namely, that the naive coefficient is contaminated by the demographic common cause and is therefore untrustworthy. It claims no positive allocation proposition, and does not claim that either the reflexivity leg or the distributive-politics leg has identified causality. This graded treatment of the identification burden — clearly distinguishing what can be identified (the debunking leg), what is merely a theoretical proposition (the reflexivity leg), and what simply cannot be identified and can only serve as a case study (the distributive-politics leg) — is itself the core judgment increment of this chapter relative to research that merely recites coefficients.

3.2 The Debunking Leg: How Population Aging, as an Omitted Common Cause, Contaminates the Naive Coefficient

The entire load-bearing weight of this chapter rests on the diagnosis of one causal chain. This chain's logic can be characterized by two parallel legs and one unclosed backdoor, presented below link by link, each annotated with its evidentiary status.

3.2.1 Leg A: Population Aging to the Decline in the Equilibrium Real Interest Rate (An Established, Exogenous Upstream Driver)

The first leg is the depressive effect of population aging on the equilibrium real interest rate. The mechanism is life-cycle saving logic: rising life expectancy, a lengthening retirement period, and increased precautionary saving, compounded by the relative scarcity of effective labor supply, jointly depress the equilibrium real interest rate. The literature on this chain is extremely mature, and this paper treats it as an established, exogenous common-cause upstream driver.

This link is supported by directly verifiable public evidence. Demographic factors alone can explain roughly a 1.25 percentage point decline in the equilibrium real interest rate since 1980, and the original source explicitly states that this contribution is "much, if not all" of the permanent decline (Gagnon, Johannsen, and López-Salido 2016); the Federal Reserve Bank of San Francisco's series of studies on the neutral rate trend point in the same direction (Holston, Laubach, and Williams 2016). This fact is critical to this chapter because it substantiates the premise that "demographics are an important upstream driver of the equilibrium real interest rate" — and it is precisely this premise that qualifies demographics as a candidate common cause contaminating the "interest rate → allocation" coefficient.

3.2.2 Leg B: Population Aging to Pension Liability Duration and Demand for Long Bonds or LDI (An Established Mechanistic Entity)

The second leg is that the same population aging, through an entirely different channel, independently raises pension funds' demand for long-duration bonds. This channel is plan maturity: as plan membership ages, cash flows shift from net inflow to net outflow, the investment horizon shortens, and risk tolerance mechanically declines, liability duration is independently raised, and the allocation demand for long bonds and liability-driven investment (LDI) is independently pulled up.

The critical property of this link is that every step of it is demographically driven and independent of the level of interest rates. This point must be made clear with a counterexample: even if interest rates were high, a mature plan whose cash flow has already turned deeply negative must still hold duration-matched bonds to avoid being forced into fire sales of assets at the point of payout. In other words, the force by which Leg B pulls long-bond demand does not need low interest rates to be triggered — it is the mechanical consequence of plan maturity itself. This paper treats it as an established mechanistic entity, and its direction is supported by public evidence: relevant OECD research and the IMF's Global Financial Stability assessments document a consistent set of facts — maturity pointing toward bonds, net cash flow turning negative, and duration mismatch (OECD 2024; IMF 2025).

3.2.3 The Unclosed Backdoor and Coefficient Contamination (the Sole Identified Load-Bearing Conclusion of This Chapter)

Now, taking the two legs together, the debunking logic emerges. Leg A and Leg B share the same exogenous shock — population aging. The typical baseline practice in the literature is to regress allocation on interest rates, obtaining an "interest rate → allocation" slope. The problem is that, in this regression, the interest rate is in fact a downstream proxy for Leg A (demographics affect the equilibrium real interest rate, which in turn affects interest rates), while demographics' independent effect on allocation via Leg B has not been separately controlled for. As a result, the interest rate coefficient absorbs the effect that should properly be attributed to Leg B (demographics affecting duration demand) — a classic case of omitted variable bias, or, in the language of causal diagrams, a backdoor opened by the confounder "demographics" that has not been closed.

Because this backdoor remains unclosed, the observed "interest rate → allocation" coefficient systematically overstates the independent effect of interest rates — it mistakenly credits to interest rates the portion of long-bond allocation actually driven by demographics through duration demand. This is the sole conclusion in this chapter that is truly identified and that serves as the core load-bearing result: what is identified is the debunking judgment that "the naive coefficient is contaminated," not any positive allocation claim.

Here a clarification of orthogonality with another type of bias is needed, to avoid confusion. Accounting-convention pre-screening (the measurement and sample-selection problem addressed in Chapter 2) is orthogonal to the structural common-cause bias addressed here: even if accounting conventions were perfect and the sample suffered no attrition whatsoever, as long as the demographic common cause exists, the backdoor bias identified here would still exist. Therefore, the debunking exercise in this chapter is not a repetition of Chapter 2 — it is an independent, more upstream identification checkpoint.

3.2.4 Identification Design: Predetermined Demographic Instruments, Cross-Country Quasi-Experiments, and Coefficient Stability

To bring the above debunking logic down to the empirical level requires an identification design capable of filtering out Leg A and exposing the absorbed Leg B effect. The primary strategy of the debunking leg of this chapter consists of three instruments:

First, a Bartik shift-share predetermined demographic instrument (Bartik 1991; Goldsmith-Pinkham, Sorkin, and Swift 2020). Its construction idea is to multiply base-period (e.g., 1990, or plan-founding-year) age-group population shares by the national-level shift in aging, yielding a "predetermined" component of population aging. This predetermined component is added as a control variable in the regression — its role is to filter out the endogenous demographic portion embedded in Leg A, thereby exposing the Leg B effect absorbed by the interest rate coefficient.

Second, a cross-country quasi-experiment based on "common equilibrium interest rate, heterogeneous demographics." The eurozone and multiple East Asian countries face a highly common global equilibrium real interest rate, yet exhibit starkly different population aging trajectories. This institutional fact provides a natural comparison: if the response slope of allocation is stratified across countries by the speed of population aging (rather than by the level of interest rates) — for example, if Japan, which is aging fastest, has the steepest slope even though it faces the same global equilibrium rate as other countries — then this is direct evidence that the demographic common cause is at work.

Third, a coefficient stability test. By comparing the relative influence of observable and unobservable selection on the coefficient (Oster 2019), this assesses whether the movement in the interest rate coefficient after adding the predetermined demographic component is sufficient to support a judgment of common-cause dominance.

It must be emphasized that the responsibility of this debunking leg is strictly bounded: it only debunks, only quantifies a lower bound on bias, and never overreaches. Specifically, it never turns around to positively predict the sign of "demographics affecting the allocation slope" (doing so would open a new backdoor via the channel "demographics affect the equilibrium interest rate, which in turn affects the term premium," thereby invalidating the instrument's exclusion restriction); it never supplies an instrumental variable for the reflexivity leg; it never claims any positive variance component; and it never serves as the engine separating the accounting-convention layer from the identification layer. This bounding of responsibility is not a technical detail — it is precisely the premise on which this chapter's identification strategy can hold: it reduces the exclusion-restriction requirement of the debunking leg to a minimum — all that is needed is that the base-period predetermined share not feed back, through plan-specific shocks to contemporaneous allocation, into the contemporaneous interest rate coefficient's residual; there is no need to claim, and this chapter does not claim, that demographics positively drive the sign of allocation.

3.2.5 Pre-Registered Expectations and Falsification Thresholds

In keeping with the discipline that judgment precedes data and expectations are fixed before data, the core expectation of the debunking leg was fixed before any data were examined, and is deliberately designed to be refutable at both ends:

• Primary expectation: after adding the predetermined demographic component, the interest rate coefficient shrinks significantly but does not go to zero. Here, "shrinks" foreshadows the direction of contamination, while "does not go to zero" faithfully preserves an independent valuation or discounting channel for interest rates — this is never a claim that interest rates are irrelevant, but rather that the interest rate effect has been overstated because a demographic effect has been absorbed into the interest rate coefficient. The magnitude of the shrinkage is the lower bound on the bias.
• Cross-sectional expectation: within a cross-section for a single year, where interest rates are differenced out, plans with greater maturity or deeper aging exhibit a higher share of long bonds (the net effect of Leg B, with a positive sign).
• Cross-country stratification expectation: the baseline slope is stratified by the speed of population aging rather than by the level of interest rates, with faster-aging economies (Japan) exhibiting a steeper slope.
• Falsification threshold: if, after adding the Bartik predetermined component, the interest rate coefficient remains stable and does not shrink (i.e., the coefficient stability test does not support common-cause dominance), then the foundation of this chapter fails, and this paper honestly accepts the falsification. This is a threshold for retreat fixed ex ante, not an ex post defense.

This pre-registration structure directly confronts the most damaging strengthened counterargument against the debunking leg: namely, that "once the predetermined demographic component is controlled for, the interest rate coefficient does not shrink significantly at all — the bias you predicted was imaginary." This paper does not evade this challenge; instead, it accepts the test and fixes the retreat threshold ex ante: if the coefficient truly does not shrink, the foundation of this chapter retreats. At the same time, this paper reduces the risk of "the bias being imaginary" through three mutually independent corroborating pieces of evidence that do not share the same identifying assumption — the cross-section (same year, interest rates differenced out), cross-country stratification (slope varies with the speed of demographic change rather than the level of interest rates), and the decumulation-phase side test. Because these three do not share the same identifying assumption, they will not all collapse together should any single assumption fail.

3.2.6 Honest Disposition: Identification Design in Place, Micro-Level Identification Downgraded

Here the evidentiary discipline of this paper must be strictly observed, and the evidentiary status of the debunking leg must be stated honestly and without ambiguity.

Publicly verified components: both demographic upstream links on which the debunking logic depends have had their direction verified. Leg A (demographics affecting the equilibrium real interest rate, roughly 1.25 percentage points, "much, if not all") is directly verified by Gagnon, Johannsen, and López-Salido (Gagnon, Johannsen, and López-Salido 2016); the direction of Leg B (demographics affecting duration and long-bond demand) is supported by the OECD's and IMF's documentation of maturity pointing toward bonds, net cash flow turning negative, and duration mismatch (OECD 2024; IMF 2025). Moreover, the very target of this debunking exercise has public background as well: the practice in the literature of presenting "a 1 percentage point decline in the domestic-currency interest rate corresponds to a rise in the offshore asset share" as a naive correlation (for example, reporting that a 1 percentage point decline in domestic sovereign bond yields corresponds to roughly a 3.4 percentage point rise in the offshore asset share, attributed to reaching for yield) (BIS 2024) is exactly the object against which this chapter warns — this coefficient absorbs into the interest rate or yield coefficient an effect that should properly belong to the demographic Leg B; during periods of low interest rates, demographics and interest rates are collinear in the same direction via Leg A, which is precisely the publicly verifiable backdrop for the unclosed backdoor.

Downgrade statement (micro data unavailable): the core debunking test of the debunking leg — namely, observing whether the interest rate coefficient shrinks significantly after adding the predetermined demographic component (base-period age-group shares multiplied by the national aging shift), with the magnitude of shrinkage constituting the lower bound on bias, together with the cross-country quasi-experiment and the coefficient stability test — requires a base-period age-group-share panel and an OECD Global Pension Statistics cross-country allocation panel (with duration approximated by country-level weighting). These are unavailable and have not been constructed, owing to constraints on micro-data availability. Consequently, subject to this constraint, this paper does not report an interest rate coefficient, does not report a shrinkage magnitude, does not report point estimates from the coefficient stability test, and does not report a sample size. This paper honestly presents three points: first, the identification design is in place — this paper holds that the identification design of the debunking leg is valid, on the grounds that its exclusion restriction rests on a bounded function (debunking only, never overreaching), and the concern that "the instrument opens its own backdoor" has been closed by the bounding of responsibility (forgoing any positive claim); second, the pre-registered direction is as stated above (the coefficient shrinks significantly but does not go to zero, refutable at both ends by data); third, the literature-based synthesis supports the direction — both demographic upstream links have been verified, so the debunking judgment that "the interest rate coefficient absorbs the omitted demographic Leg B effect" has public support at the mechanism level, but "the specific magnitude of the shrinkage after controlling for predetermined demographics" cannot be given owing to constraints on micro-data availability, and requires a micro-level panel.

An honest annotation regarding the magnitude of the bias lower bound (a robustness caveat): this paper's self-assessment of this chapter's identification design is "passable, but with a reservation." The reservation is this: when the predetermined demographic component is used as a control variable for debunking purposes, if Leg A (demographics affecting the equilibrium real interest rate) contains a genuine discounting channel, then controlling for demographics would simultaneously remove part of the genuine interest rate effect — this would make the magnitude of "the shrinkage" as a lower bound on bias contestable. This paper does not evade this point: the core conclusion of this chapter claims only the debunking warning that "the observed coefficient is untrustworthy" (this warning holds as long as the coefficient moves significantly once the common-cause proxy is introduced), so this concern does not undermine the core conclusion; but precisely because Leg A contains a genuine discounting channel, the point estimate of the magnitude of the bias lower bound cannot be pinned down. This paper therefore honestly labels this as "direction confirmed (at the literature level), magnitude unavailable (owing to the absence of micro-level identification)," and under no circumstances misreports any number for the shrinkage magnitude. It is worth noting that the very fact that Leg A's 1.25 percentage point contribution has been verified in fact heightens the credibility of the contamination concern — the greater demographics' contribution to the equilibrium real interest rate, the more of the demographic component the interest rate coefficient absorbs; but for this very reason, the point estimate of the magnitude becomes harder to cleanly disentangle.

3.3 The Reflexivity Leg: How Demand Produces the Environment — An Explicitly Unidentified Theoretical Proposition

The debunking leg has demonstrated that the observed coefficient is contaminated by the demographic common cause, but this is only half of the identification warning. The other half comes from a force running in the opposite direction: the rigid demand of pension funds is not merely a passive response to interest rates — it also reflexively produces the low interest rate environment itself. The evidentiary status of this leg is fundamentally different from that of the debunking leg — it is a theoretical proposition, supplemented by a side test from the decumulation phase, and it explicitly does not claim to have identified causality. This paper strictly maintains this distinction here.

3.3.1 Theoretical Proposition: The Reversal of the Arrow from "Interest Rate → Allocation" to "Demand → Environment"

The logical chain of the reflexivity loop is as follows: institutional design (mandatory defined benefit (DB) plans, duration-matching requirements, large public pension funds) first creates a body of price-insensitive, rigid long-bond buying demand; this buying demand persistently absorbs the supply of sovereign long bonds, depresses the term premium, and weakens the constraint that "bond vigilantes" impose on fiscal policy; sovereigns are thereby able to operate more durably in a low interest rate environment; low interest rates, in turn, amplify duration-matching buying demand via discounting, forming a positive feedback loop.

The key to this chain lies in the direction of the causal arrow: it runs from "the stock of demand affecting the persistence of the environment," which is exactly the opposite of the mainstream direction of "interest rates affecting allocation." Thus, even setting aside the debunking leg, the observed coefficient has a second source of untrustworthiness — because within this reflexive loop, the interest rate (or, more precisely, the term premium) is downstream of pension demand rather than upstream of it, and regressing allocation on interest rates is tantamount to treating an endogenous variable as an exogenous explanatory variable. This is the second reason this paper judges the core identifying assumption of "interest rate exogeneity" to be "least tenable during periods of low interest rates."

3.3.2 Side-Test Design and the Decumulation-Phase "Semi-Counterfactual"

The positive proposition of the reflexivity leg is supported by two types of side tests, rather than by using demographically driven pension reform legislation as an instrumental variable — because that instrument would itself fail via the very backdoor identified by the debunking leg ("demographics affect the equilibrium interest rate, which in turn affects the term premium"), invalidating its exclusion restriction. This is an identification discipline that must be held firm: this paper would rather let the reflexivity leg remain a theoretical proposition than use an instrument known to open a backdoor to masquerade as identification.

The first type of side test is cross-sectional structural evidence: differencing out pure duration demand using "the spread between domestic sovereign bonds and foreign bonds or swaps of the same duration," leaving only the domestic-specific premium, then interacting this with the steepness of the domestic maturity wall, to test whether the structural share of domestic long bonds held by pension funds negatively predicts the level of the term premium and its sensitivity to global interest rate shocks.

The second type of side test is the decumulation-phase "semi-counterfactual," which is the closing falsifier of this chapter. Its logic is highly elegant: after 2022, interest rates reversed (rose), while the population continued to age monotonically — the two legs, for the first time, took opposite signs. If the interest rate causal channel is real, a rise in interest rates should trigger a rotation back into equities; if the demographic causal channel is real, continued population aging should continue to drive further bond allocation. During the low interest rate period the two are collinear and cannot be separated; the sign reversal after 2022 precisely breaks this collinearity, allowing the policy cycle to serve as a sign instrument, from which the relative weight of the two legs can be back-solved. Formally, letting allocation equal the interest rate coefficient times the interest rate, plus the demographic coefficient times demographics, plus an error term: during the low interest rate period the interest rate coefficient and the demographic coefficient are inseparable due to collinearity, and the sign reversal after 2022 breaks this collinearity, thereby allowing the two coefficients to be separately estimated.

3.3.3 Pre-Registered Expectations and the Retreat Threshold

• Reflexivity-proposition expectation (a theoretical proposition, not identified): the domestic long-bond holding share of pension funds negatively predicts the term premium and its sensitivity to global interest rate shocks.
• Pre-registered falsification threshold for the decumulation-phase closing test: fixed explicitly ex ante — if the rebound in equity share from 2022 to 2025 exceeds more than half of the cumulative upward shift during the low interest rate period, and is stratified along interest rates rather than demographics, then interest-rate-independent causation is judged to dominate, and this chapter's demographic common-cause line of argument retreats. This is an honest pre-registered retreat, not an ex post defense.

3.3.4 Honest Disposition: The Theoretical Proposition Remains Sharp, Without Masquerading as Identification

The evidentiary status of the reflexivity leg must likewise be honestly disclosed. The share-premium measure needed for the cross-sectional identification engine of the reflexivity leg is subject to measurement error, and the decumulation-phase window is only about 3 years — a short sample. This paper therefore explicitly does not claim that the reflexivity leg has identified causality — it remains a sharply argued theoretical proposition, supplemented by a decumulation-phase side test. It should be specifically clarified that the value of the decumulation-phase side test derives from structure (the sign reversal breaking collinearity) rather than from sample size; thus the fact that "the sample is short" does not weaken the logical value of the side-test design, but it does mean this paper cannot offer any point estimate on this basis. In terms of direction during the decumulation phase, verifiable public facts are consistent with the expected direction of the reflexivity leg — for example, Japanese institutional investors (including the unhedged exposure of the Government Pension Investment Fund of Japan) significantly sold off foreign bonds in 2022 (Council on Foreign Relations 2023), consistent with the direction of "the decumulation-phase sign reversal"; but this is only a directional, literature- and fact-level consistency, not micro-level identification. This paper reports no positive allocation elasticity and provides no instrumental variable for the reflexivity leg.

3.4 The Distributive-Politics Leg: An Explicitly Unidentified Third Perspective, Presented Only as a Qualitative Case

The third leg of this chapter is the weakest in evidentiary status among the three legs, yet the sharpest in the perspective it offers. It advances a third mechanism distinct from both reaching for yield and ALM — distributive-politics non-recognition — but it is explicitly unidentified in terms of causality, and is presented only as a qualitative case, entering no econometric identification surface whatsoever. This paper explains thoroughly here why this leg "cannot be identified and can only serve as a case," because this is precisely the most honest component of this chapter's graded treatment of the identification burden.

3.4.1 Mechanism Hypothesis: How Political Non-Recognition Reverses Causality

The logic of distributive politics is as follows: for public DB plans that use the expected-return method, lowering the discount rate would make the funding gap visible, thereby politically demanding tax increases or higher contributions — something the plan sponsor (e.g., a state government) is unwilling to bear. As a result, there is a political tendency to refuse to lower the discount rate; and to maintain an elevated discount rate that is accountingly self-consistent, it must be supported by an elevated assumed rate of return; and to sustain this elevated assumed rate of return, the asset side is forced to increase allocation to equities and private equity.

The causal arrow of this chain is "the discount rate affects allocation," and it completely reverses the usual causal narrative: it is not "interest rates are low, so equities must be bought," but rather "there is a political refusal to acknowledge that interest rates are low, and equities must be bought in order to prop up this judgment." The sharpness of this perspective lies precisely in how it differs from both reaching for yield and ALM — reaching for yield is preference-driven, ALM is liability-driven, while distributive politics is accounting-legitimacy-driven.

3.4.2 Why It Cannot Be Identified: The Accounting-Identity Trap

Despite its sharp perspective, this paper must honestly acknowledge that this leg cannot be purely econometrically identified under the expected-return method. The reason is an accounting-identity trap. Under the accounting identity of the expected-return method, the discount rate is defined as the assumed return, and the assumed return is in turn a function of allocation — this is an arithmetic relationship. This means that the proposition "political stickiness drives increased equity allocation" is observationally equivalent to the competing proposition "the mark-to-market group is merely passively bound by ALM constraints": the same set of data showing "a high discount rate accompanied by high equity allocation" can be read either as distributive-politics non-recognition or as the natural result of ALM constraints — the two cannot be distinguished in the data.

This paper attempted to rescue this into the identification layer using within-group political variables combined with a difference-in-differences stratification, but it can be shown that this attempt does not escape the identity: shifting the identity from the level to the first-difference layer, the second leg — "stickiness affects equity allocation" versus "the mark-to-market group is bound by ALM" — remains observationally equivalent. To genuinely cut this apart would require an ALM-intensity measure independent of allocation, and this is precisely unattainable. This paper therefore downgrades the distributive-politics leg to a qualitative, case-based pillar of distributive politics: preserving its sharpness of judgment (non-recognition is a third perspective distinct from reaching for yield or ALM, with a reversed causal arrow), but explicitly marking it as unidentified in terms of causality, presented through process tracing and narrative case studies (such as the discount-rate battle in a specific state plan), not entering the econometric identification surface of the identification layer, and not serving as an identified-layer testable implication.

This disposition itself responds to the second strengthened counterargument against the overall judgment — namely, "you keep the distributive-politics leg as a third perspective, but under the expected-return method the discount rate and allocation are an accounting identity; 'stickiness drives increased equity allocation' and 'the mark-to-market group is merely bound by ALM constraints' are completely observationally equivalent, you simply cannot cut them apart, and keeping it around is the emperor's new clothes of the identification layer." This paper's response is: this criticism is fully accepted, and it has already been honestly disposed of through downgrading — it is precisely because this paper explicitly distinguishes that the distributive-politics leg cannot be identified and can only serve as a case, rather than forcibly stuffing it into the identification layer to pad it out, that this chapter's graded treatment of the identification burden holds together. This grading of "what can be identified, what is merely a theoretical proposition, and what simply cannot be identified" is precisely the judgment increment of this chapter relative to conventional robustness checks.

3.4.3 Honest Disposition: The Data Support Only a Case Narrative

The case data sources for the distributive-politics leg include the U.S. Public Plans Database (containing long panels of discount rates, allocation, and maturity), classifications of public pension governance, and state legislative records. But it must be honestly noted: these data support only case narrative and correlational description, and cannot be elevated to causal identification owing to the accounting identity plus ALM observational equivalence. This paper therefore lists no econometric identification strategy for this leg, does not fold it into the identified-layer testable implications, and treats it only as a qualitative pillar.

3.5 This Chapter's Increment Relative to the Conventional Framework, and Disposition of Two Strengthened Counterarguments

Before closing this chapter, it is necessary to clarify its judgment increment — because a sharp challenge would be: isn't all of this "demographic common cause plus reverse causality" simply the textbook common sense of omitted variables and reverse causation? Wouldn't adding a demographic control variable and running an instrumental-variable regression settle the matter — on what grounds does this deserve to be called a foundational, book-length warning?

This paper's answer has two layers, corresponding precisely to the two strengthened counterarguments this chapter disposes of.

The first layer: this chapter upgrades the coefficient problem into an identification-validity warning, and this is precisely where the increment relative to the conventional framework lies. The conventional framework treats "reaching for yield versus ALM" as a regression-coefficient problem and defaults to assuming interest rate exogeneity — this falls squarely into the five mechanical argumentation patterns this paper has pre-identified as ones to avoid: first, treating regression significance as equivalent to established conclusion without mechanism discrimination; second, ignoring the endogeneity problem of where cross-country identification could come from, given that interest rates are a globally common shock; third, treating correlation as causation; fourth, "year fixed effects absorbing the common interest rate shock" — in a cross-country panel, since the interest rate as a common shock would be wholly absorbed by year fixed effects, a purely time-series design cannot provide any source of identification for the interest rate effect; fifth, "mistaking mark-to-market-convention ratio changes for behavior" — misreading proportional shifts caused by passive valuation reshuffling under mark-to-market conventions as active reallocation. What this chapter does is far more than simply "adding a control variable": it demonstrates that the core exogeneity assumption of the baseline identification strategy (interest rate exogeneity) systematically fails within the study window, and it reverses the causal arrow (demand affecting the environment). Precisely for this reason it is foundational: every identified claim in Chapters 2 through 5 must first pass through this identification checkpoint, or else what they are interpreting is nothing but a contaminated coefficient.

The second layer: regarding the accounting-identity trap, this paper fully accepts it and has already honestly disposed of it. As described above, this paper acknowledges that, under the accounting identity of the expected-return method combined with ALM constraints, the distributive-politics leg cannot be purely econometrically identified, and has already downgraded it to a qualitative case, removed from the identification debt. This disposition, taken together with the previous one, itself constitutes evidence that this chapter is not merely a restatement of common sense: this chapter clearly distinguishes what can be identified (the debunking leg, the sole load-bearing and identified claim), what is merely a theoretical proposition (the reflexivity leg, explicitly not claiming identification), and what simply cannot be identified and can only serve as a case (the distributive-politics leg, explicitly unidentified) — this graded treatment of the identification burden is precisely what the conventional framework lacks, and is precisely the weight this chapter carries as a foundational identification warning.

3.6 Chapter Summary

As the foundation of the entire empirical exercise, this chapter answers only one question — why the observed "interest rate → allocation" coefficient cannot be read as causal — and offers the following clearly graded conclusions.

First, with respect to the identified, load-bearing conclusion: the observed "interest rate → allocation" coefficient is contaminated by demographics as an omitted common cause, and is therefore untrustworthy. This debunking judgment is genuinely earned through identification by the debunking leg; its two demographic upstream links (Leg A, approximately 1.25 percentage points; Leg B, direction) have both been verified by public sources (Gagnon, Johannsen, and López-Salido 2016; OECD 2024; IMF 2025), and the debunking logic has public support at the mechanism level; this paper holds that its identification design is valid, for reasons already given in the bounding of responsibility in Section 3.2.4. But it must be honestly noted: the debunking leg's Bartik debunking regression itself is downgraded owing to constraints on micro-data availability (the required micro-level panel is unavailable), and the magnitude of the bias lower bound cannot be pinned down — this chapter asserts only direction (the coefficient is untrustworthy), asserts no magnitude, and reports no coefficient that has not actually been obtained.

Second, with respect to the theoretical proposition: the demand-side reflexivity leg argues that the causal arrow reverses during periods of low interest rates (demand affecting the environment), constituting a second source of untrustworthiness for the observed coefficient; it is supported by cross-sectional structural evidence and a decumulation-phase "semi-counterfactual" as side tests, explicitly does not claim to have identified causality, and reports no positive elasticity.

Third, with respect to the qualitative case: distributive-politics non-recognition offers a third perspective distinct from reaching for yield or ALM (accounting-legitimacy-driven, with a reversed causal arrow), but cannot be identified owing to the accounting identity of the expected-return method combined with ALM observational equivalence, and is presented only through process tracing and narrative case study, entering no econometric identification surface.

The overall judgment of this chapter is therefore this: the dispute between "reaching-for-yield dominance and ALM dominance" is, at the identification layer, a misdiagnosis of the same contaminated coefficient. Until this coefficient is debunked and this arrow's possible reversal is identified, any approach that treats the dispute between the two as something that can be settled simply by "estimating a bigger regression and looking at the coefficient" is building on a contaminated foundation. The true mechanism does not lie in this coefficient — it is hidden within the four neglected layers of accounting conventions, hedging, behavior, and settlement, to be argued one by one in the following four chapters. This is the foundational identification warning that must be erected before the empirical work of this entire paper can begin.

4. Chapter 2 The Shadow Rate: What Enters the Allocation Function Is Never "the Interest Rate," but the Shadow Rate Each of Four Misaligned Accounting Regimes Refracts

4.1 Restating the Problem: Why "Rate → Allocation" Is a Mis-specified Causal Channel

The mainstream narrative writes the mechanism by which long-run low interest rates reshape pension strategic asset allocation as a direct causal arrow: market rates fall, the present value of liabilities rises, the duration gap widens, and rational sponsors rationally pivot to liability-driven investment (LDI), selling equities and buying long-duration bonds. This narrative is not wrong at the descriptive level — mature pension markets did indeed experience a systematic wave of de-equitization around 2013 — but it treats "the interest rate" as a clean scalar that enters the allocation decision function directly. The core judgment of this chapter runs counter to this: what enters the pension allocation function is never the market rate itself, but rather the "shadow rate" that each of four mutually misaligned institutional accounting regimes refracts from the market rate; allocation drift is driven by the opening and closing of the wedges between these accounting bases, and can decouple from — or even run counter to — the path of the equilibrium real rate.

This judgment matters because it sits at the same level as the paper's central question — whether reaching for yield or ALM pressure dominates — yet raises a third possibility: a considerable portion of allocation drift is neither reaching for yield nor pure ALM hedging, but endogenous to accounting standards. If so, the policy implications branch again: managing risk appetite is useless, and managing discount-rate rules is only half right; what truly needs monitoring is the accounting wedge torn open between standards, funding law, and tax law.

It must be stated plainly at the outset: the core conclusion of this chapter claims only that this "translation mechanism" exists and has a measurable shape — its identification design is given by the IAS19R standard difference-in-differences (an identified design, with micro-level dose identification downgraded); it does not claim to be able to separate the accounting-basis effect from ALM constraint strength via cross-regime sign comparison (that is observationally equivalent, requiring internationally comparable ALM indices across countries, which are unattainable given data availability constraints — see the counterargument treatment in 4.5). The judgment is falsifiable and non-obvious: the mainstream narrative attributes the wave of de-equitization around 2013 wholesale to the concurrent low-rate environment; this chapter argues that a considerable part of it is endogenous to the accounting standard — if International Financial Reporting Standards (IFRS) sponsors did not exhibit significantly greater de-equitization relative to US-GAAP-only sponsors before versus after IAS19R, or if the magnitude of de-equitization correlates more strongly with sponsor balance-sheet fragility than with plan liability duration (interest-rate sensitivity), then this judgment is falsified.

4.2 Discounting Arithmetic: A Bookkeeping Starting Point, Not Evidence

Let us first settle the least contested — and therefore least judgment-additive — link. That falling market rates raise the present value of liabilities is an identity of discounting arithmetic: the present value of liabilities decreases monotonically in the discount rate. The reaching-for-yield hypothesis, the ALM hypothesis, the distributional-politics hypothesis, and the accounting-wedge hypothesis — all four competing explanations agree on this point, and it therefore cannot discriminate between any pair of competing hypotheses. This link is a bookkeeping starting point, not an evidentiary layer.

Specifically, if one replaces the nominal discount rate with the real (inflation-adjusted) rate, the liability "explosion" under the Dutch Financial Assessment Framework (Financieel Toetsingskader, FTK) disappears — a fact that had once been treated as supporting evidence within the traditional analytical framework. This chapter honestly downgrades it to background: it is merely the arithmetic consequence of swapping the independent variable in the discounting formula, a tautology that carries no identifying function and does not enter the chapter's core conclusion. The genuine judgment increment is compressed into two deeper layers: balance-sheet fragility — the intensity of de-equitization varies with sponsor balance-sheet fragility rather than plan duration; and wedge closure — the true de-risking trigger is the re-convergence and wedge closure across accounting bases during the decumulation (payout) phase of low rates, not low rates themselves.

4.3 The Load-Bearing Leg: IAS19R as a Quasi-Exogenous Rate Shock (the Sole Load-Bearing Identified Design)

4.3.1 Institutional Facts: How the Standard Cut Off the "Accounting Subsidy for Holding Equities" in One Stroke

The sole load-bearing identification engine of this chapter (an identified design) reads IAS19R, effective in 2013, as a "quasi-exogenous rate shock." Its institutional content is fully verifiable: IAS 19 (2011 amendment) took effect on 1 January 2013; it abolished the corridor method (eliminating deferred smoothing of actuarial gains or losses); it abolished the elective expected return on plan assets (the expected return on plan assets is no longer separately recognized), replacing it with the net-interest method — i.e., net interest is calculated by multiplying the discount rate by the net defined-benefit liability or asset; remeasurements are recognized in full in other comprehensive income (IFRS Foundation 2011).

This standard change is functionally equivalent to an "exogenous downward shift in the institutional expected rate of return," unrelated to the level of the equilibrium real rate. The crux is that it cut off, in one stroke, two "accounting subsidies" that equity holdings had enjoyed under the old standard: first, the corridor method smoothed the annual volatility of equities into the income statement, systematically understating the reported cost of holding equities; second, the elective expected-return provision allowed sponsors to assume expected equity returns above 8%, directly used to flatter pension expense. Once these two subsidies were erased, rational sponsors had an incentive to swap equities for bonds sharing the same underlying rate as the discount rate, in order to eliminate net-interest noise — this is the accounting motive for LDI, not the rate-hedging motive. Both phenomenally manifest as equity-to-bond rotation, but the driving variables are entirely different: the former tracks the magnitude of the lost accounting subsidy, the latter tracks interest-rate sensitivity (duration).

4.3.2 Dose-Response: The Differential Prediction That Elevates "Hypothesis" to "Identified" (the Load-Bearing Element of This Chapter)

Institutional facts alone are not sufficient to elevate the above mechanism from hypothesis to identified. The critical differential prediction lies in dose-response: the magnitude of de-equitization rises with "the aggressiveness of the old expected-return assumption multiplied by balance-sheet fragility (pension volatility relative to sponsor equity)," and is only weakly correlated with liability duration. This is the operational cut that elevates the mechanism from hypothesis to identified — the pure rate-hedging hypothesis predicts de-equitization magnitude tracking duration, while the accounting-basis hypothesis predicts it tracking fragility; the two are orthogonal or even opposite. This cross-sectional directional contrast precisely sidesteps the absence of internationally comparable ALM indices: it uses within-group (same standard regime, same country) cross-sectional variation in fragility versus duration to achieve separation, and therefore does not carry the ALM observational-equivalence burden of cross-regime sign comparison (see 4.5).

The main regression takes the form: the coefficient on the triple interaction of "IFRS sponsor times post-reform times aggressiveness of the old expected-return assumption" for the change in equity weight, with standard-regime, year, and industry fixed effects, plus controls for funding status, liability duration, interest rates, and sponsor size. The pre-registered expectation, fixed before seeing the data: this coefficient is significantly negative (the more aggressive the old assumption, the sharper the de-equitization in the IFRS group, post-reform), and the elasticity of de-equitization with respect to balance-sheet fragility is significantly larger than its elasticity with respect to liability duration; the pre-reform pre-trend placebo is insignificant (parallel trends hold); the US-GAAP-only placebo group shows no significant de-equitization (because US Financial Accounting Standards No. 715 did not simultaneously abolish expected return and the corridor). The falsification threshold is likewise fixed in advance and symmetric: if this coefficient is insignificant, or if the elasticity of de-equitization with respect to liability duration instead dominates, the judgment is falsified and reverts to a single-factor or ALM narrative; if the pre-trend diverges significantly, this load-bearing leg is downgraded to correlational evidence, the chapter loses its sole load-bearing identification engine, and the core conclusion must be retracted to the level of a mechanism hypothesis. This downgrade path is fixed in advance and is not revised after seeing the data.

4.3.3 Identification Design in Place, Micro-Dose Identification Downgraded (Honest Boundary)

The layered status of the evidence must be presented honestly. The institutional-fact layer (the content and timing of IAS19R) is fully verifiable (IFRS Foundation 2011); the identification-design layer passes (the 2013 effective date of IAS19R was determined by the International Accounting Standards Board's standards-revision agenda, exogenous to IFRS sponsors and inapplicable to US-GAAP-only sponsors, with the interest-rate path common to both groups, so that the standard-regime-times-time between-group difference absorbs the common rate shock). But the dose triple-difference main regression and the cross-sectional cut of "fragility elasticity versus duration elasticity" are downgraded owing to constraints on micro-data availability: the required cross-regime panel of pension footnote disclosures — firm-by-firm hand-collected values of the old expected-return assumption, corridor-usage status, other-comprehensive-income detail, asset-allocation weights, and funding status, grouped by listing venue or reporting standard into IFRS versus US-GAAP-only — is unattainable. Consequently, subject to this constraint, this paper does not report this coefficient, does not report the elasticity-contrast coefficient, and does not report sample size. The identification design is in place; the micro-data are unattainable.

Published literature supporting this direction is verifiable, and requires one refinement that must not be post-hoc rationalized. Anantharaman and Chuk, using a Canadian sample, show that following adoption of IAS19R, firms shifted pension assets from equities to bonds, attributing this primarily to the elimination of the expected rate of return (ERR) — i.e., replacing the expected rate of return with the discount rate reduced firms' risk-taking in pension investment, as reflected in their pension asset allocation (Anantharaman and Chuk 2018). This direction supports the chapter's core assertion that "losing the accounting subsidy triggers de-equitization" (literature-level support, not micro-identification). An important refinement: the chapter's mechanism was originally stated as "corridor elimination causes equity-to-bond rotation," whereas the published evidence more precisely locates the primary cause in the elimination of the expected rate of return (rather than the corridor itself) — this is consistent with, and even more favorable to, this load-bearing leg's design of treating "the aggressiveness of the old expected-return assumption" as a continuous dose (the dose happens to hang precisely on the expected rate of return), but it should be honestly noted: the literature's primary cause is the expected rate of return, with the corridor as a parallel channel. Furthermore, research on whether the announcement or implementation of IAS19R carried real economic consequences provides strong evidence of an other-comprehensive-income effect of IAS19R on allocation (consistent in direction). The literature's Canadian IFRS sample already embeds a quasi-treatment of "IFRS-adopting countries," but does not perform a standard-regime difference-in-differences, nor does it treat the aggressiveness of the expected rate of return as a continuous dose — this is precisely the incremental contribution of this load-bearing leg's design, and precisely where data constraints prevent execution. This chapter therefore does not pass off directional literature support as identified causation.

4.4 The Timing of the Wedge: Low-Rate Periods Are Not Where Bond-Buying Accelerates — Wedge Closure Is the Trigger (an Identified Design)

A second misalignment arises from the separation of funding-basis and accounting-basis rates. Three rates — the Internal Revenue Code Section 430 segment-smoothed funding rate, the immediate accounting discount rate under Financial Accounting Standards No. 715, and the buyout-basis rate — were torn apart into a wedge during the low-rate period. The relevant smoothing legislation smooths the funding rate using a 25-year historical average, keeping it far above the accounting spot rate, producing a misalignment in which plans appear "adequately funded on a funding basis, requiring no contributions" while being "deeply underfunded on an accounting basis." Sponsors thus enjoyed a contribution holiday, tucking the accounting shortfall into other comprehensive income rather than the income statement, and institutionally delaying allocation migration. This means that in the years of deepest low rates, sponsors were, if anything, in no hurry to buy bonds — precisely the reverse timing implied by the single-factor narrative. By the decumulation phase (from 2022 onward), rising rates caused the three rates to re-converge, closing the wedge and releasing the smoothing corridor that had capped it — and contributions, buyouts, and LDI all jumped in tandem. The true de-risking trigger is wedge closure, not low rates themselves.

The identification strategy for this wedge leg is: wedge width equals the Section 430 smoothed-segment rate minus the Section 715 spot discount rate, constructible year by year, as a panel covariate explaining changes in contributions and bond weight; using the two smoothing statutes of 2012 and 2014 as legislative timing shocks for exogenous expansion of the wedge (the legislative intent was to offset highway-fund or fiscal shortfalls, exogenous to pension allocation), comparing sponsors with high versus low wedge exposure via difference-in-differences; using Canadian or UK plans (not subject to Section 430, with a different funding basis) as placebos, predicting that the pattern of "de-risking dormancy during the low-rate period, an eruption of wedge closure in the decumulation phase" appears only in the United States.

Evidentiary status: the legislative timing is publicly verifiable (historical segment-rate tables, the statutes themselves), but a plan-level dual-basis panel of funding status and allocation migration is unattainable, making it impossible to execute the "high- versus low-wedge-exposure difference-in-differences with Canadian/UK placebos." This wedge leg is therefore downgraded to design-plus-direction: the identification design is in place, the micro-data are unattainable. Its timing predictions (de-risking dormancy in the low-rate period, a jump in contributions, buyouts, or LDI following wedge closure in the decumulation phase, occurring only in the United States) are consistent in direction with verifiable decumulation-phase evidence — the record-setting US and UK buyout wave of 2022–2023 falls precisely in the window of high rates and wedge closure (this behavioral fingerprint is publicly verified in Chapters 4 and 5).

4.5 Two Strengthened Counterarguments and Their Treatment

4.5.1 Counterargument One: The Accounting Wedge Is Observationally Equivalent to ALM Constraint Strength

The strongest opposing explanation is: the so-called accounting-wedge effect is observationally equivalent to differences in ALM constraint strength across countries or accounting regimes. The opposite-signed cross-regime allocation pattern can be fully explained by "expected-return-method countries have loose ALM regulation, while market-value or risk-free-curve countries have strict ALM regulation"; if IFRS-group de-equitization is greater after IAS19R, the strongest counter-explanation is not "loss of the accounting subsidy" but rather "the IFRS or other-comprehensive-income basis lets equity volatility hit reported net worth directly, triggering harder funding or contribution constraints" — a purely mechanical ALM consequence, observationally equivalent to the accounting wedge. On this basis, cross-regime sign comparison cannot discriminate between "accounting basis determines allocation" and "ALM constraint strength determines allocation."

The treatment is to bound the domain of applicability and absorb. First, we concede that this counterargument holds at the level of cross-regime sign comparison — that comparison is therefore honestly downgraded to a corroborating mechanism, not an independent identification; the chapter's core conclusion does not claim to separate the accounting effect from ALM via cross-regime sign comparison. Second, we absorb the identification difficulty this counterargument points to as a source of identifying power for the main design: the IAS19R standard-regime difference-in-differences precisely sidesteps this counterargument — it varies only the timing of the accounting regime (2013 IAS19R exogenous to IFRS sponsors, inapplicable to US-GAAP-only sponsors) within the same ALM strength (same country, same regulatory environment), so that "two groups within the same ALM environment exhibiting different dose responses" does not carry the ALM observational-equivalence burden. The more critical differential evidence lies in the direction of cross-sectional heterogeneity: the ALM hypothesis predicts de-equitization magnitude tracking plan duration or interest-rate sensitivity (hedging need), while the accounting-wedge hypothesis predicts it tracking sponsor balance-sheet fragility (the ratio of pension volatility to equity, the aggressiveness of the old expected-return assumption) — the two give orthogonal or even opposite cross-sectional predictions for the same IAS19R shock. The counterargument is valid against cross-regime sign comparison but invalid against the standard-regime difference-in-differences main design.

4.5.2 Counterargument Two: The Standard Merely "Recognized" a De-risking Direction Already Determined by Rates

The second counterargument questions causal order and exogeneity: IAS19R merely "recognized" on the reported balance sheet a de-risking already occurring because of low rates — the true driver remains rates raising liabilities through ALM, with sponsors already intending to hedge, and the standard change merely accelerated or made explicit an already-existing direction. Moreover, 2013 fell within a period of deepening low rates, so the "quasi-exogenous rate shock" is in fact highly entangled with rates themselves — the standard effect and the rate effect are collinear, and the parallel-trends assumption for the difference-in-differences may not hold around 2013.

The treatment is design-based rebuttal plus a bounded concession. Rebutting exogeneity: the 2013 effective date of IAS19R was determined by the International Accounting Standards Board's standard-revision agenda, exogenous to any individual sponsor's pension-allocation decision, and applicable to IFRS sponsors but not to US-GAAP-only sponsors — this is an allocation at the standard-regime level, not a sponsor's own choice; the rate path is common to both groups, so the between-group difference in the difference-in-differences absorbs the common rate shock. This is precisely the correct answer to the "year dummies absorb common rate shocks" clause within the mechanical argumentation pattern the whole paper guards against: using standard-regime-times-time differencing rather than pure time-series, the common rate is differenced away. Rebutting "recognition versus driving": the dose-response design directly distinguishes the two — if it is merely "recognizing an already-determined direction," de-equitization magnitude should track liability duration or funding shortfall depth (pre-existing hedging need); if it is "standard-driven," magnitude should track the aggressiveness of the old expected-return assumption (how much accounting subsidy was lost). The dose-response of "the more accounting subsidy lost, the more equities cut" cannot be produced by the pure recognition hypothesis (under which the aggressiveness of the old assumption should be unrelated to de-equitization). The bounded concession (parallel trends): we acknowledge that parallel trends require testing — using the pre-2013 pre-trend placebo (the IFRS and US-GAAP-only groups should show no systematic difference in de-equitization slope before the reform) as a pre-check; if the pre-trend diverges significantly, this design retreats and the load-bearing leg is downgraded to correlational evidence. This concession is written into the pre-registration and is not rationalized after the fact; the US-GAAP-only group serves as a separate placebo (Financial Accounting Standards No. 715 did not simultaneously abolish expected return and the corridor).

4.6 Corroborating and Hypothesis-Level Rings (Explicitly Marked as Non-Load-Bearing)

The following rings each refract additional shadow rates, but none are load-bearing; their evidentiary status is honestly labeled.

A third misalignment arises from the decomposition of the nominal discount rate. The nominal discount rate equals the real rate plus expected inflation plus the interest-rate risk premium; the nominal scalar masks the interest-rate-risk-premium component, and the ratio of linked or inflation-swap exposure moves in the same direction as the estimated interest-rate risk premium (rather than the nominal yield). But the decomposition of the interest-rate risk premium is sensitive to the decomposition method used (model-dependent), and so it remains at the hypothesis level and does not enter the core conclusion.

A fourth misalignment arises from the deductibility basis under tax law. As low rates push coupons toward zero, the numerator of the "bond-holding tax spread" in the Black-Tepper tax-arbitrage framework tends to zero, collapsing the tax shield on the allocation dimension, and arbitrage migrates to the timing of contributions. The 2017 Tax Cuts and Jobs Act induced a debt-funded contribution surge, yet the within-fund bond-to-equity ratio was unchanged — living evidence of exactly this. This is timing-layer corroboration or a mechanism hypothesis, not treated as an identified layer. Verifiable directional support: the 2017 Tax Cuts and Jobs Act cut the corporate tax rate from 35% in 2017 to 21% in 2018, and firms rushed to make 2017 DB contributions ahead of the rate cut in order to deduct them at the higher rate; related research records an average increase of about 27% in 2017 DB contributions (Gaertner, Lynch, and Vernon 2018), and other research records roughly $3 billion in additional contributions induced by the tax advantage among mid-to-large plans, with a strong reversal in 2018 (BIS 2021). The Black and Tepper framework — that a fully funded pension plan investing in bonds within the pension fund, shifting equity risk to shareholders, is tax-efficient — is likewise verified (Black 1980; Tepper 1981). These directions support the mechanism hypothesis that "the tax shield collapses on the allocation dimension, with arbitrage migrating to contribution timing."

The opposite-signed cross-sectional allocation pattern across discounting regimes (expected-return method versus market-value spot versus risk-free curve) serves as cross-regime corroboration that the mechanism exists. To be explicitly marked: this ring is observationally equivalent to "differing ALM constraint strength across countries determines the sign," and is therefore treated as mechanism corroboration, not an independent identified proposition; separating it would require internationally comparable ALM indices across countries, which are unattainable given data constraints, and it therefore remains permanently corroborative until such data exist. Its data sources (US public pension discounting assumptions, UK Pensions Regulator annual disclosures, Dutch central bank framework data) remain to be verified, and this paper does not cite them pending verification.

Finally, to guard against contamination from the market-value basis, one must test, after stripping out valuation effects, whether allocation migration remains residually present — passive valuation reshuffling is not the same as active reallocation; this is a robustness safeguard against the "market-value-basis denominator illusion," corresponding to the "market-value-basis ratio treated as behavior" clause in the mechanical argumentation pattern the whole paper guards against.

4.7 Chapter Summary

What enters the allocation function is never "the interest rate," but the shadow rate that each of four misaligned accounting regimes refracts; allocation drift is driven by the opening and closing of the accounting wedge. This chapter's core conclusion — "the translation mechanism exists and has a measurable shape (the standard-regime difference-in-differences identification design)" — presents an evidentiary structure of three layers: the institutional-fact layer (the content and timing of IAS19R) is fully verifiable; the identification-design layer passes; the micro-dose identification is downgraded (the cross-regime footnote panel is unattainable), but published literature (Anantharaman and Chuk 2018) supports the core assertion directionally, with the refinement that the primary cause is the elimination of the expected rate of return honestly noted. The discounting arithmetic is downgraded to background; the judgment increment lies in balance-sheet fragility (de-equitization tracking fragility rather than duration) and wedge closure (the de-risking trigger being wedge closure rather than low rates); cross-regime sign comparison is downgraded to background or corroboration. This chapter does not overstep to claim separation of the accounting effect from ALM via cross-regime sign comparison.

5. Chapter 3 The Real Overseas Constraint: Not a Monotonic Decreasing Function of the Domestic-Currency Rate, but Corner Solutions Created by Hedged Carry and Regulation

5.1 Restating the Problem: Correcting the Monotonic Causality of "1 Percentage Point to 3.4 Percentage Points of Reaching for Yield"

Why do pension funds increase their allocation to overseas assets during periods of low interest rates? The ready-made answer comes from a widely cited correlation: for every 1 percentage point decline in the domestic-currency government bond yield, the overseas asset share rises by approximately 3.4 percentage points, attributed to reaching for yield (BIS 2024). The conventional analytical framework follows this coefficient, treating "a low domestic-currency long-end rate monotonically raises the overseas share" as a settled causal direction, treating overseas allocation as a monotonic channel for reaching for yield, and treating the exchange rate merely as a passive byproduct.

The core judgment of this chapter runs counter to this: the true constraint on the increment of overseas allocation during periods of low interest rates is dual, and neither part is "a low domestic-currency long-end rate monotonically raises the overseas share." The first is a constraint substitution — what enters pension decision-making is not the domestic-currency nominal interest-rate differential, but the hedged carry; the second is an upper-bound corner — the cross-country differences in the ceiling on overseas allocation are determined chiefly by regulatory capital or matching penalties on hedged foreign debt. This judgment corrects the earlier shallow treatment that regarded overseas allocation as a monotonic channel for reaching for yield (a treatment that took "1 percentage point to 3.4 percentage points" as a settled causal direction).

It is necessary at the outset to draw an honest boundary around the core conclusion: the main body of this chapter is supported only by two load-bearing legs with clean identification designs — the hedged-carry leg (within the same fund, the divergence between "hedged foreign debt" and "unhedged foreign debt or equity," an identification design already specified) plus the institutional-contrast leg (a difference-in-differences between pension funds and insurers within the same country, an identification design already specified). Cross-border reverse pricing ("allocation affects price") is an exploratory, secondary leg: its common factor has not been ruled out, the instrumental variable cited is a eurozone-synchronized shock, and there is no within-same-fund internal contrast; it is explicitly flagged as non-load-bearing and, failing to meet the bar, is relegated to a literature discussion. The core conclusion asserts only the portion that the two load-bearing legs can genuinely identify.

Let us first firmly establish the target of correction. The original source can be checked verbatim: "a 1 percentage point decline in the domestic-currency government bond yield is associated with an approximately 3.4 percentage point rise in overseas asset holdings," explicitly attributed to a reaching-for-yield motive; a finer breakdown is: a 1 percentage point decline in yield corresponds to roughly a 1 percentage point decline in the bond share and a roughly 0.5 percentage point rise in the equity share (1.5 percentage points for emerging-market equities), with the overseas share of DB plans rising by about 3.2 percentage points (BIS 2024). This target of correction is fully established — the source itself states that this is a naive correlation plus a reaching-for-yield narrative, precisely the object that the hedged-carry leg is meant to replace with "hedged carry" and that the institutional-contrast leg is meant to explain via regulatory corner solutions accounting for cross-sectional differences.

5.2 Constraint Substitution: Hedged Carry Is the True Driving Variable (Load-Bearing, Identification Design Specified)

5.2.1 Mechanism: How Hedging Costs Decouple Overseas Allocation from the Domestic-Currency Long End

The first load-bearing leg replaces the driving variable of overseas allocation, from the domestic-currency nominal interest-rate differential to the hedged carry. Pension funds holding long-duration foreign-currency bonds typically use rolling 3-month or 6-month FX swaps to hedge, at a cost approximately equal to the short-end interest-rate differential between the two countries plus the cross-currency basis (the deviation from covered interest rate parity). Thus the hedged carry equals the 10-year foreign-currency rate minus the 10-year domestic-currency rate minus the 3-month hedging cost, and its sign is determined by the shape of the yield curve and the basis, decoupling from — or even flipping sign relative to — the domestic-currency long-end rate. The increment in overseas allocation is driven by "hedged carry being positive as a precondition," rather than being pulled by the domestic-currency long end; by the decumulation phase, an inverted short end causes the hedging cost to eat into the interest-rate differential, turning the hedged carry negative, so overseas allocation reverses even while the domestic-currency long end remains low. This timing — the domestic-currency long end still low, while overseas allocation has already turned — is something the monotonic causality of the conventional framework cannot produce.

5.2.2 Identification Source: The Divergence between "Hedged" and "Unhedged" within the Same Fund

The key to identification lies not in level correlation but in the sign divergence, within the same fund, across three types of exposure in response to the same shock. Here one must confront head-on the strongest counter-explanation (the reinforced version of counter-argument one): the so-called "reduction of hedged foreign debt" is actually simply a purely liability-driven, first-order optimal response that requires no new variable such as "hedged carry" — a rise in interest rates lowers the present value of liabilities, improves the funding ratio, and narrows the duration gap, so a rational de-risking reduces overseas or foreign-currency exposure together as risk assets, which is observationally equivalent to hedged carry.

The treatment is to use an internal contrast to force the two hypotheses into opposite sign predictions, rather than relying on level correlation. Liability-driven behavior is "a decline in the total risk budget," predicting that hedged foreign debt and unhedged foreign debt or equity are reduced in the same direction (both are risk assets, cut together); hedging-cost-driven behavior predicts that only hedged foreign debt is reduced, with unhedged foreign debt or equity not necessarily reduced in tandem (what is cut is the portion for which "hedged carry has turned negative"). The divergence between "hedged foreign debt" and "unhedged foreign debt or equity" within the same fund is a fingerprint that liability-driven behavior cannot produce (liability-driven behavior does not distinguish between hedged and unhedged). This paper further argues: if the pure risk budget were asymmetric, rationally one should cut more of the unhedged foreign debt carrying exchange-rate risk — which is precisely the opposite of the observed cut in hedged foreign debt, making the direction of divergence harder to explain away. In terms of specification, if, after controlling for the liability side (changes in the funding ratio, changes in the duration gap), the hedged-return spread still significantly predicts changes in hedged foreign debt exposure, then liability-driven behavior is insufficient to explain it; countries where the cross-currency basis has widened (Japan, the Netherlands) show a higher elasticity to "hedging cost" than to "nominal interest-rate differential," further corroborating the cost channel; exogenous cost changes (public series of the cross-currency basis or the deviation from covered interest rate parity) serve as a quasi-exogenous mover of hedging cost. Honest boundary (pre-committed falsification threshold): if the overseas divestment of 2022–2023 is mainly explained by the narrowing of the parent fund's duration gap, hedged and unhedged foreign debt are reduced in the same direction with no divergence, and the incremental explanatory power of hedged carry after controlling for the liability side is not significant, then this leg is falsified and the judgment is withdrawn.

5.2.3 Identification Design in Place, Micro Data Unavailable

The evidentiary status is stratified as follows. The mechanism-fact layer of this leg is verifiable and consistent in direction: Japanese institutional investors (including the unhedged exposure of the Government Pension Investment Fund of Japan) significantly sold off foreign debt in 2022 (Council on Foreign Relations 2023); the Dutch civil servants' pension fund (ABP) cut its dollar hedge ratio on equities from 50% to 25% (IPE 2023); the relevant BIS study itself states a naive correlation (BIS 2024) — all consistent in direction with "the constraint is hedged carry rather than a monotonic function of the domestic-currency rate." But the regression on the divergence between "hedged foreign debt" and "unhedged foreign debt or equity" within the same fund is downgraded: the fund-level panel of hedged/unhedged sub-flows and month-by-month hedge ratios that it requires is unavailable (the sub-items for the Government Pension Investment Fund of Japan, the Netherlands, and Japanese life insurers are relatively obtainable but require period-by-period manual construction, and disclosure in most countries is coarse). Therefore, subject to this limitation, this paper does not report this coefficient, does not report the incremental explanatory power, and does not report the test statistic for the divergence. The identification design is in place; the micro data are unavailable. "The incremental explanatory power of hedged carry after controlling for the liability side" is precisely the part that cannot be delivered due to data constraints.

The direction and magnitude of the hedge-ratio reduction have a verifiable pair of figures: the Dutch civil servants' pension fund (ABP) lowered the dollar hedge ratio on its equity exposure from 50% to 25% (the object being equity, not foreign debt) (IPE 2023). Likewise, the Government Pension Investment Fund of Japan's adoption of a new policy asset allocation on October 31, 2014, shifting to four asset classes at 25% each (a large reduction in domestic bonds from roughly 60% previously), is a verified background fact regarding the direction of this leg during the decumulation phase (GPIF 2014).

5.3 The Upper-Bound Corner: Regulator-Manufactured Home Bias (Load-Bearing, Identification Design Specified)

5.3.1 Mechanism: How Regulatory Penalties Pin Overseas Allocation at a Corner

The second load-bearing leg attributes the cross-country differences in the ceiling on overseas allocation to regulation, rather than to cross-country differences in the willingness to reach for yield. Regulation imposes a penalty on "hedged foreign debt" — the matching adjustment does not recognize rolling hedges (the position of the UK Prudential Regulation Authority), or high capital is imposed on currency mismatch (under Solvency II), or direct quantitative caps are set — which raises the effective capital cost of hedged foreign debt and lowers the institutional ceiling on overseas allocation. Thus, even when the hedged carry is attractive during periods of low interest rates, full overseas allocation is not possible: this is a corner solution, not an interior optimum within a return budget. Countries with lax institutional regimes (the Netherlands, Canada's "prudent person plus no quantitative cap") show high elasticity, while tightly regulated countries (constrained by the currency mismatch provisions of Solvency II) show near-zero elasticity. Ignoring this term causes a pooled regression to mistakenly attribute "cross-sectional differences in regulatory tightness" to "interest-rate elasticity," biasing the coefficient.

The division of labor between the two legs is complementary: the hedged-carry leg determines "along which variable overseas allocation moves" (hedged carry, rather than the domestic-currency rate), i.e., the substitution of the first-order driving variable; the institutional-contrast leg determines "how far overseas allocation can move" (where the regulatory corner lies), i.e., the endogenization of the boundary of the feasible region.

5.3.2 Identification Source: Pension Funds versus Insurers within the Same Country

Here again one must confront head-on the strongest counter-explanation (the reinforced version of counter-argument two): the cross-country cross-sectional difference in overseas allocation is merely a difference in market depth or duration supply (countries with shallow domestic long-bond markets are forced to allocate overseas), or a cross-country difference in pure reaching-for-yield willingness, rather than regulatory penalty — the regulatory-penalty index is collinear with market depth, so one cannot prove that it is specifically the capital or matching penalty mechanism that is pinning the ceiling.

The treatment is to use the contrast between "pension funds and insurers within the same country" to difference out the common factor of market depth. Pension funds and insurers face the same domestic market depth, the same domestic-currency rate, and the same duration supply within a given country; the only systematic difference is the regulatory constraint (insurers are subject to the currency mismatch provisions of Solvency II, while occupational pension funds are often exempt). If, under the same interest rate and the same market, insurers' hedged foreign debt is suppressed while pension funds' is not, then the market-depth or preference factor is absorbed by the contrast, and the remaining difference is attributed to regulatory penalty — this is something the market-depth hypothesis cannot produce (market depth treats pension funds and insurers alike). Layering on a difference-in-differences around institutional reform events: episodes of regulatory tightening (rulings that the matching adjustment does not recognize rolling hedges, upward revisions to Solvency II currency mismatch capital) suppress the overseas allocation of constrained entities at a point when interest rates and market depth are unchanged, with the placebo being that exempt entities (occupational pension funds) do not move over the same period — the time dimension further rules out cross-sectional collinearity. Honest boundary (pre-committed falsification threshold): if, after controlling for a "hedged foreign debt regulatory-penalty intensity index," the overseas difference is still mainly explained by interest rates or duration gap or market depth, if pension funds and insurers show no systematic difference under the same interest rate, and if episodes of regulatory tightening show no significant suppression effect on constrained entities, then this leg is falsified and the judgment is withdrawn.

5.3.3 Identification Design in Place, Micro Data Unavailable

The evidentiary status is stratified as follows. The existence of the regulatory-penalty mechanism on which this leg relies is verifiable: under the European Market Infrastructure Regulation, all counterparties trading within the EU must post variation margin in cash (rather than bonds); pension funds do not hold large amounts of cash, which is especially onerous when large cash variation margin must be posted during stress periods — this is precisely a "type-mismatch tax," and also one branch of "the regulatory penalty on hedged foreign debt" (Macfarlanes 2019); regulatory texts such as Solvency II currency mismatch capital and the UK Prudential Regulation Authority's position on the matching adjustment for rolling hedges are publicly verifiable. That is, the existence of the mechanism "regulation pins hedged foreign debt at a corner" has public support. But the regression combining the pension-fund-versus-insurer difference-in-differences within the same country plus the institutional reform shock is downgraded: the "hedged foreign debt regulatory-penalty intensity index" it requires (a large body of institutional coding) and the panel aligning pension-fund and insurer allocations within the same country are unavailable. Therefore, subject to this limitation, this paper does not report the elasticity, nor does it report the difference-in-differences coefficient. The identification design is in place; the micro data are unavailable. At the design level: pension funds and insurers face the same market depth, domestic-currency rate, and duration supply; the common market-depth factor is absorbed by the contrast, differences in liability structure are absorbed by institution-type fixed effects, and identification rests on the within-time difference-in-differences around regulatory-tightening events, with the placebo being the exempt entities — this paper holds that this identification design is valid; but the causal magnitude of the cross-sectional elasticity difference is unavailable due to data constraints.

5.4 Exploratory Secondary Leg: Cross-Border Reverse Pricing (Unidentified, Relegated to Literature Discussion)

The third leg examines the reverse channel of "allocation affects price": whether the term premium at the long end in the target country is sensitive to the "net buying share of foreign pension funds plus insurers," and whether this remains significant after controlling for local central bank holdings and net supply. The identifying power of this leg depends entirely on whether the global common factor can be ruled out — this requires a clean cross-border instrumental variable that moves demand for a single country's debt only, does not touch that country's fundamentals, and is not contemporaneous with the global interest-rate environment. The instruments cited, such as Solvency II or the Dutch new contract, are both eurozone-synchronized shocks, unable to separate "Dutch pension funds' demand for German government bonds" from "a shift in the demand of pan-European institutions as a whole," and therefore do not satisfy this requirement. Hence this leg is downgraded to a literature discussion, is non-load-bearing, and reports no causal claim; its qualitative value is retained as a basis for downstream decumulation-phase contagion pricing. Its evidentiary gap is honestly recorded: pension funds and insurers are often disclosed on a combined basis, individual fund holdings are mostly unavailable, and a clean cross-border instrumental variable that can be separated from the global common factor is lacking. This mirrors, in reverse, the very mechanistic argumentation pattern that this paper as a whole cautions against — "correlation as causation" — precisely because the common factor cannot be separated out, this leg must not be elevated to a causal proposition.

5.5 Increment over the Conventional Framework and Chapter Summary

Relative to the conventional framework, this chapter corrects the baseline causal direction: it replaces the core explanatory variable, from the domestic-currency interest-rate differential to hedged carry, and demonstrates that its sign can flip; it reconstructs home bias, from a preference into a regulator-manufactured artifact that can be moved by policy. The conventional framework took "1 percentage point to 3.4 percentage points" as settled causation, treated overseas allocation as a monotonic channel for reaching for yield, and treated the exchange rate merely as a passive byproduct; this chapter demonstrates that the true constraint is "hedged carry being positive as a precondition" plus "a ceiling set by a regulator-manufactured corner," correcting the shallow treatment.

The core conclusion of this chapter — "hedged carry plus the regulatory corner are the true constraint, correcting the shallow causality of 1 percentage point to 3.4 percentage points" — presents, in terms of evidence, three layers: the target of correction (the BIS's 1 percentage point to 3.4 percentage points and the reaching-for-yield narrative) is fully verifiable; the mechanism facts of the two load-bearing legs (the Government Pension Investment Fund of Japan's 2022 sell-off of foreign debt, the Dutch civil servants' pension fund's halving of its hedge ratio, EMIR cash variation margin, Solvency II currency mismatch capital) are verifiable and support the direction; the fund-level or index-level identification regressions for both legs are downgraded (micro data unavailable, identification design in place). Cross-border reverse pricing is explicitly flagged as unidentified and relegated to a literature discussion. The core conclusion covers only the portion that the two load-bearing legs can genuinely identify — it asserts only the two claims that "overseas allocation moves along hedged carry" and "the ceiling is determined by the regulatory corner," and does not assert that the reverse-pricing claim that "allocation affects price" has been identified.

6. Chapter 4: The Reference-Point Ratchet: Liability-Side "De-Risking" as a Unidirectional Lag of Loss Aversion, Not a First-Order Optimization of the Return Gap

6.1 From "More Prudent" to "Non-Optimizing": What This Chapter Falsifies Is Not a Direction but a Label

The first three chapters dealt with how interest-rate signals are refracted by institutional accounting and market structure before reaching decision-makers: the refraction of accounting conventions (Chapter 2), the unreliability of the identification coefficient itself (Chapter 1), and the true post-hedge carry constraint on overseas allocation (Chapter 3). This chapter turns to a deeper layer: even if decision-makers read an undistorted signal, their response function to that signal is itself not the textbook-assumed expected-utility optimization.

The most conspicuous behavior in mature pension markets over the past two decades is the "de-risking glide path" under the liability-driven investment (LDI) framework: as a plan matures and its funding ratio improves, it progressively reduces growth-type assets (mainly equities and some alternatives) and increases matching-type assets (long-duration bonds and interest-rate hedging instruments) along a predetermined path. The industry almost uniformly narrates this as "more prudent, more optimal"; the acceleration of de-risking and buyouts during the sharp rate rise of 2022-2024 is likewise narrated as "hedging is cheaper when rates are high, the duration gap narrows, hence acceleration is rational."

The judgment of this chapter collides head-on with this narrative, but the manner of collision must be defined with great precision, or it will slide into the coefficient-restatement and correlation-as-causation traps warned against in earlier chapters. This chapter does not falsify the direction of "de-risking" itself—plans are indeed de-risking, and this is not in dispute; what this chapter falsifies is the label attached to it: "de-risking is a symmetric, path-independent, first-order optimizing response triggered by the return gap." The chapter's affirmative claim is: the triggering mechanism of liability-side de-risking is the reference points and behavioral biases of trustees and sponsors (the loss-aversion ratchet, the perceptual salience of psychological round numbers, and the transfer of reference-point ownership), which systematically deviate from expected-utility optimization.

This claim is a judgment rather than a matter of regression coefficients precisely because it is observationally equivalent, in direction, to the optimization explanation—both predict "de-risking." To separate them cannot rely on "whether the coefficient is significant"; it can only rely on designing three testable fingerprints that run counter to optimization: an asymmetric and path-dependent dead zone, spikes of perceptual salience at psychological round numbers outside statutory thresholds, and excess de-risking explained by the dominant governance party even after controlling for hedging cost. Each fingerprint corresponds to an independent, separately falsifiable leg; if any one leg is overturned by the data, the corresponding claim is voided without implicating the other two. This is the chapter's core increment relative to a mere descriptive account of "de-risking clustering": it splits a broad direction into three falsifiable claims, each with its own clean discriminating identification.

Before laying out the three legs, a boundary line must first be drawn, one that determines what this chapter can and cannot claim.

6.2 The Boundary Line: This Chapter Claims Only That the Liability-Side Trigger Is Behavioral, Not That All Asset-Side Adjustments Are Behavioral

An easy mistake for an honest behavioral-finance argument to make is to expand "some deviations from optimization are behavioral" into "all observed adjustments are behavioral." This chapter rejects that expansion from the outset.

The specific dividing line occurs at the phenomenon of "increased allocation to alternative assets." During the low-rate period, mature pension funds broadly increased allocations to infrastructure, private equity, private credit, real estate, and other real assets; in the P7 market, the "other" asset category rose to about 19% over two decades (equities about 48%, bonds about 31%, other about 19%, cash about 3%), and this drift is often cited as ironclad evidence of reaching for yield (WTW 2026). But there exists a pure, behavior-free optimization explanation: if the stock-bond correlation flipped from negative to positive during the low-rate period, bonds lost their diversification identity, the portfolio's total risk budget rose passively, and rational funds shifted toward real assets with low correlation to both stocks and bonds. In the Markowitz framework, this is precisely the first-order-condition solution under a new covariance input—the covariance matrix changes, and optimal weights move accordingly. It is optimization, and it is observationally fully equivalent to "a rational fund re-optimizing under updated covariances."

This chapter's treatment of this covariance-repair mechanism is explicit: it is not a member of this chapter, and is mentioned only as an "optimization control boundary." Its role is to falsify the label of reaching for yield (attributing increased allocation to alternatives to greedily chasing returns is wrong; it may simply be a passive reshuffling of the risk budget), but it does not falsify first-order optimization itself—quite the opposite, it is a clean example of first-order optimization. Therefore, hanging covariance repair under the umbrella of "falsifying first-order optimization" would be a logical error, which this chapter refuses to commit. The true resolution of asset-side "increased allocation to alternatives" is partly handled by the accounting-convention mechanism of Chapter 2 (the disappearance of the accounting subsidy behind de-equitization) and partly falls under this optimization control boundary.

This yields the chapter's shared boundary statement (fixed in advance here to prevent overreach): the chapter's core conclusion claims only that "the triggering mechanism of liability-side de-risking is behavioral reference points, systematically deviating from first-order optimization," borne by three legs—the unidirectional-ratchet leg, the round-number-threshold leg, and the reference-point-ownership-transfer leg—each with its own clean discriminating identification. The chapter does not claim that all asset-side allocation adjustments are behavioral; the covariance repair behind increased allocation to alternatives is explicitly labeled as optimization, serves as a control boundary, is observationally equivalent to rational re-optimization, and is not included under this chapter's falsification umbrella.

This boundary is less a gesture of modesty than a precondition for identification: precisely because the optimizing portion (covariance repair) is honestly carved out, the "behavioral deviation" claimed by the remaining three legs can be cleanly identified—otherwise any instance of "de-risking" could be absorbed by an opponent with the single sentence "this is just the covariance changing."

6.3 First Leg: The Unidirectional Ratchet—The Lag Fingerprint of Loss Aversion

6.3.1 Claim and Mechanism

The claim of the first leg is: the de-risking glide path of mature pension funds almost never reverses. When interest rates fall, growth-type assets that have already been reduced are rarely replenished. This phenomenon itself is widely observed—the industry designs de-risking glide paths to be ratchet-like (once a threshold is crossed and allocation is reduced, it does not go back), and narrates this as prudence. The increment of this leg lies in offering a mechanistic explanation for this unidirectionality that runs counter to optimization, and in designing a test that separates the two.

The causal chain of the mechanism is as follows (each link marked with its evidentiary status):

First, low interest rates inflate the present value of liabilities (established: this is pure discounting arithmetic, a consensus across the four hypotheses, see the background in Chapter 1). The funding ratio deteriorates because the denominator inflates, and the trustee's attention is anchored on this ratio, eager to reduce the loss already incurred (hypothesis: the funding ratio is a binding reference point, not merely a monitored indicator).

Second, when the funding ratio crosses a certain trigger threshold, the trustee registers this new level as a reference point (anchoring effect compounded with the endowment effect)—the de-risking that crosses the threshold is encoded as "safety already realized" (to be established: the trigger point is defined relative to a reference point, not relative to the optimal weight).

Third, when falling rates push the ratio back into its earlier range, replenishing growth-type assets requires "using an already-banked surplus to re-assume risk." Under a loss-averse value function, this is encoded as a loss and is therefore avoided. Allocation with respect to the interest-rate path therefore takes on a unidirectional, path-dependent (hysteretic) form (to be established: this is exactly the dead-zone asymmetry that needs to be tested).

The critical divergence is: a pure expected-utility optimizer's adjustments are symmetric—reducing and replenishing face the same first-order condition, with no directional preference. The observed unidirectional ratchet is precisely the fingerprint of loss aversion. If allocation were truly symmetric and bidirectional, this leg would not hold.

6.3.2 The Strongest Counterargument: This Is Just Transaction Costs, Not Behavioral Bias

This leg must confront a powerful optimization explanation head-on—inertial-band policy. Reallocation between growth-type and matching-type assets carries real transaction costs (bid-ask spreads, market impact, re-hedging costs), and LDI structural adjustments also involve operational friction. Any optimization model with friction will produce a "no-adjustment band": within the band, nothing moves; adjustment occurs only when deviation is large enough. This alone is sufficient to produce the lag of "not readily replenishing after reducing," with no need for any loss aversion or reference point. Observing lag is not the same as observing behavioral bias—this is the equivalence this leg most needs to break through.

Dealing with this counterargument relies neither on refutation nor absorption, but on delimiting its domain of applicability and designing discriminating identification. The friction hypothesis and the reference-point hypothesis give opposite predictions for the shape of the dead zone:

• Friction-induced stickiness predicts a symmetric dead zone (adjustment costs do not distinguish direction: reducing and replenishing face the same spread and impact) that is independent of historical path (the width of the no-adjustment band is determined by current cost parameters, unrelated to "whether a previous high was breached in the past").
• The reference-point hypothesis predicts an asymmetric dead zone (easy to reduce, hard to replenish—replenishing, which means "using realized surplus to re-risk," is encoded as a loss) that is conditioned by history (the dead-zone width widens with whether the fund has recently experienced a funding-ratio drawdown, or has previously breached a prior high).

Accordingly, this leg concedes that transaction costs do produce part of a symmetric dead zone (taking this as a baseline and control variable), but maintains that the asymmetric component and the path-dependent component of the dead zone are residuals that the friction model cannot generate and that are specific to reference points. Identification rests on a carefully constructed matched pair: within the same year and same jurisdiction, pair funds with approximately identical growth-type exposure, asset scale, and liquidity profile (hence approximately identical transaction-cost parameters), but with different historical funding-ratio peak paths. With friction held equal, the symmetric dead zone is differenced out by the pairing; any remaining difference in the reduce-versus-replenish threshold can only be explained by path dependence (reference point plus endowment effect). If, after pairing, the probability of replenishment does not vary with differences in the peak path, the friction hypothesis wins and this leg is voided.

6.3.3 Model Design and Pre-Registration (Fixed Prior to Any Data)

• Objective: to separate "loss-aversion reference points causing an asymmetric, path-dependent dead zone" from "transaction costs causing a symmetric, history-independent dead zone." The object of discrimination is the two components—asymmetry and path dependence of the dead zone—that the friction model cannot generate.
• Identification strategy: matching funds of the same scale and liquidity plus differences in peak path. Supplemented by a comparison of growth-type weights on the "upward path versus downward path" at the same interest-rate level (a direct fingerprint of the lag).
• Main specification: with "whether growth-type assets are replenished" (or the magnitude of replenishment) as the dependent variable, the core explanatory variable is the difference between the reduce-threshold and the replenish-threshold in interest rates (dead-zone width) multiplied by whether a prior high was ever breached (path-dependence interaction), with pair fixed effects absorbing friction.
• Robustness checklist: first, the moderation of dead-zone width by "whether a funding-ratio drawdown has recently occurred" (continuous version); second, a sign test on the difference in growth-type weights along the upward versus downward path; third, a placebo—pure DC plans and sovereign reserve funds (with no funding-ratio reference-point constraint) should not display an asymmetric dead zone; fourth, checking the stability of conclusions under varying matching tolerances.
• Pre-registered expectations: the judgment supported (this leg holds) is that, after pair differencing, the group that has breached a prior high shows a significantly higher replenishment threshold than the group that has not, growth-type weights are systematically lower along the upward path than the downward path, and pure-DC or reserve-fund placebos do not display this asymmetry; the judgment falsified (this leg voided) is that, after pairing, the replenishment probability or threshold does not vary with differences in the peak path, or the dead zone shows the same asymmetry in pure-DC or reserve funds (in which case it is not specific to reference points), and the lag is attributed to transaction costs, with this leg reduced to a literature discussion.

6.3.4 Evidentiary Status: Downgraded, With the Weakest Point Honestly Flagged

The fund-level annual allocation and funding-ratio panels needed for the main test—from the UK Pensions Regulator's annual disclosures, De Nederlandsche Bank, the Office of the Superintendent of Financial Institutions of Canada, and relevant US filings—as well as the "same scale and liquidity, different peak path" matched pairs, are unavailable given the constraints on micro-data accessibility. Accordingly, subject to this constraint, this paper does not report dead-zone width, does not report the replenishment-threshold difference, and does not report sample size.

It must be honestly flagged that this leg is the weakest in evidentiary terms in this chapter, and the nature of the weakness is intrinsic to the proposition itself: replenishment events are inherently sparse—the very fact that de-risking almost never reverses means that "replenishment when interest rates fall" is an event that occurs very rarely. This directly limits the statistical power to identify "the replenishment threshold exceeds the reduction threshold." For this reason, the chapter has pre-fixed "the number of replenishment events" as a power precondition: if the number of replenishment events across jurisdictions falls below a preset threshold, the main test is downgraded to directional or one-sided evidence, explicitly flagged as power-limited, and identification is not claimed. Given the data constraints preventing the execution of the matching, this leg stops at "design in place plus power precondition," and does not claim to have achieved identification. At the literature level, the unidirectionality of de-risking glide paths is widely observed in the industry (constituting suggestive support for the direction), but the residual specific to reference points—"replenishment probability varies with peak path after pairing"—is unavailable given the data constraints.

6.4 Second Leg: Clustering at Round-Number Thresholds—The Accumulation Fingerprint of Perceptual Salience

6.4.1 Claim and Mechanism

The claim of the second leg is: the actual incidence density of de-risking shows anomalous clustering and discontinuities at psychologically salient funding-ratio marks, and this clustering cannot be explained by continuous expected-utility optimization or by codified regulatory or accounting thresholds—it arises from the trustee's perceptual salience toward "round-number thresholds" (round-number anchoring).

The mechanism is a problem of cognitive approximation. Trustees face a continuous state space of funding ratios, but humans do not think in continuous variables; instead they approximate it using sparse, salient anchors (round percentages in tens or hundreds). Crossing a salient mark (such as from 99% to 101%, passing through 100%) triggers a discrete review and de-risking action, while continuous movement between two anchors does not trigger one. The result is that action density piles up and hollows out on either side of the anchors—a sharp contrast to the smooth, continuous adjustment density predicted by optimization. This leg is a purely perceptual phenomenon, independent of the other two legs and separately falsifiable.

6.4.2 An Orthogonal Cut Against Rule-Driven Explanations

The confound this leg most needs to guard against is mistaking clustering at psychological round numbers for clustering driven by codified statutory or accounting thresholds. Many funding-ratio marks do carry statutory consequences: the Netherlands' 105% and 119.1%, the US Pension Protection Act's 100%, and so on; crossing them triggers statutory contribution or recovery-plan requirements. Clustering at these points is rule-driven—an institutional response rather than a behavioral bias.

This leg's separation from rule-driven explanations is clean and orthogonal: the rule-driven account concerns the rule effect produced by the statutorily consequential threshold itself, while this leg concerns whether pure psychological round numbers with no statutory consequence (such as 90%, 110%, 120%) still attract clustering. The identification strategy is designed accordingly—first use bunching estimation (drawing on the tax-kink and discontinuity literature) to construct a frictionless counterfactual density and quantify the excess mass at each mark; the key step is to remove all thresholds with statutory consequences from the candidate anchor set, and then test whether significant excess mass remains at the remaining pure psychological round numbers. If clustering disappears after removing statutory thresholds, the rule hypothesis wins; if it persists, perceptual salience is established.

6.4.3 Model Design and Pre-Registration

• Objective: to separate "clustering at psychological round numbers caused by perceptual salience" from "clustering driven by codified regulatory or accounting thresholds." The object of discrimination is whether excess mass remains at psychological round numbers after removing all statutory thresholds.
• Identification strategy: bunching estimation plus removal of statutory thresholds.
• Main specification: bunching estimation of the funding-ratio distribution (or the distribution of de-risking trigger points), with excess mass equal to actual density minus counterfactual density divided by counterfactual density, estimated at psychological round numbers after removing statutory thresholds.
• Robustness checklist: first, a skewness test of clustering above 100% and sparseness below it in the funding-ratio distribution; second, whether clustering marks shift with "the number being watched" across countries with different reporting conventions (the Netherlands' spot coverage ratio versus the 12-month average policy coverage ratio); third, sensitivity to the polynomial order of the counterfactual and the bunching window width; fourth, a placebo—random marks with no behavioral significance should show no spikes.
• Pre-registered expectations: the judgment supported (this leg holds) is that, after removing statutory thresholds, significant positive excess mass remains at pure psychological round numbers (90%, 110%, 120%), the clustering marks shift with reporting convention, and the distribution clusters above and is sparse below 100%; the judgment falsified (this leg voided) is that clustering at psychological round numbers disappears after removing statutory thresholds, with all excess mass falling on statutory or accounting thresholds, attributed to rule-driven effects, with this leg reduced to a supporting role for the rule-driven account.

6.4.4 Evidentiary Status: Downgraded

The continuous fund-level micro-distribution of funding ratios needed for the main test (with De Nederlandsche Bank's policy ratio as the primary sample) is unavailable given the constraints on micro-data accessibility; moreover, public data is mostly bracketed rather than continuous (especially in the UK and US), which directly limits the precision of the discontinuity estimate and the construction of the counterfactual density. Accordingly, subject to this constraint, this paper does not report excess mass, does not report the counterfactual density.

The chapter has pre-fixed this granularity limitation as a granularity precondition: the main identification rests on the Netherlands' continuous policy ratio, with the UK and US brackets serving only as directional corroboration; if the only available continuous sample (the Netherlands) lacks sufficient power to discern excess mass at psychological round numbers, this is explicitly flagged as "limited by bracketed granularity, conclusion is suggestive," and clustering identification is not claimed. Given the data constraints preventing the execution of the bunching estimation, this leg stops at "design in place plus granularity precondition."

6.5 Third Leg: The Transfer of Reference-Point Ownership—The Disposition Effect Accelerating When It Should Least Do So

6.5.1 Claim and Mechanism

The third leg is the sharpest judgment of this chapter and the only one with an already-verified behavioral fingerprint. The claim is: as interest rates exited their low-level regime (from 2022), the bound reference point quietly transferred ownership from the trustee to the sponsor. The trustee's reference point is liability-side anxiety (volatility of the funding ratio, contribution pressure); the sponsor's reference point is an accounting and cash anchor (eliminating balance-sheet volatility, locking in the surplus). When rising rates turn the funding status from a deficit into a surplus, ownership of the binding reference point transfers, and what follows is a disposition effect—de-risking and buyouts accelerate, rather than decelerate, precisely in the window when rates are high, expected risk premia are thickest, and the mathematics of continued operation is most favorable.

"The mathematics of continued operation is most favorable" is the crux: in an environment of thick risk premia, continuing to hold the plan (run-on) and letting the surplus accumulate has a positive expected upside, and a pure expected-utility optimizer should tend to preserve this upside. But the observed behavior is the opposite—it is precisely in this window that sponsors accelerate buyouts, lock the surplus in with insurers, and forgo the positive-expected-value upside of continued operation. This is explained as a disposition effect under the "transfer of ownership" of the reference point (treating an already-banked surplus as an endowment and averting its volatility), rather than as a change in risk preference itself.

6.5.2 The Only Already-Verified Behavioral Fingerprint: The 2022-2023 Record-Level Buyout Wave

Unlike the first two legs, this leg's temporal claim has a publicly verifiable behavioral fingerprint, and it has already been verified at the record level, year by year:

• US pension risk transfer: 568 transactions in 2022, $51.8 billion in premiums, a historical record; single-premium group annuity purchases exceeded $45 billion in 2023, expected to remain above $50 billion annually from 2024 to 2026 (Aon 2023).
• UK buyouts or buy-ins: approximately £49-50 billion in 2023, a historical record (over 254 transactions, averaging about £190 million each; 11 transactions exceeding £1 billion; RSA Insurance Group's £6.5 billion and the British Steel Pension Scheme's £7.5 billion setting new single-transaction records) (LCP 2024).

This fingerprint substantiates this leg's core temporal claim: de-risking and buyouts indeed accelerated precisely in the 2022-2023 window when rates were high, risk premia were thick, and the mathematics of continued operation was most favorable. This is precisely the buyout wave triggered by the improvement in funding status following the LDI crisis.

Public discussion also circulates larger figures such as "a cumulative buyout wave of about $180 billion in the US since 2022," but its accounting basis (possibly a multi-year cumulative figure) cannot be directly reconciled with year-by-year records, and this paper does not adopt it; this paper adopts only verifiable year-by-year figures—the US's single-year figure of $51.8 billion in 2022 (Aon 2023), and the UK's approximately £49-50 billion in 2023 (LCP 2024).

6.5.3 The Strongest Counterargument: Rationally Accelerating De-Risking When Rates Are High

This leg confronts the strongest opposing hypothesis in the entire chapter, which offers two purely rational reasons for de-risking to accelerate after 2022, neither requiring "reference-point ownership transfer":

• Falling hedging costs: when rates are high and the curve shape improves, the cost of locking in liabilities with bonds and swaps falls, and rational funds naturally accelerate hedging at such times. The speed of de-risking rising smoothly with falling hedging costs is itself an optimization prediction.
• Mechanical narrowing of the duration gap: rising rates cause the present value of liabilities to fall (if duration is long) by more than the fall in assets, improving the funding ratio and narrowing risk exposure relative to liabilities; a rational ALM approach lowering growth-type weight as the gap narrows is entirely a first-order condition.

Therefore "de-risking accelerating when rates are high" is observationally equivalent to rational ALM. What is more thorny is that this rational narrative is itself verifiable: the 2023 buyout wave was explicitly narrated by the industry as "driven by improved funding following the LDI crisis" (LCP 2024)—that is, an explanation via rational ALM or falling hedging costs. The existence of this opposing hypothesis is publicly confirmed; this leg cannot refute it, and can only design a test capable of cutting it apart.

6.5.4 Resolution: The Interaction of Governance Structure and the Timing of Surplus Turning Positive

The way to cut it apart is the interaction of governance structure with the timing at which the surplus turns positive. With de-risking or buyout speed as the dependent variable, first control for hedging cost (cross-currency basis, curve shape, swap and bond costs) and the change in the duration gap (these two together absorb the entire rational-ALM channel), then test whether the residual is significantly explained by the interaction of "whether the sponsor dominates over the trustee (governance coding)" with "whether the fund has just turned from deficit to surplus."

The two hypotheses give opposite predictions: the rational hypothesis predicts that after controlling for hedging cost and the change in the duration gap, the interaction term should be zero—cheap hedging treats sponsor-dominated and trustee-dominated plans alike, and the magnitude of the reduction in growth-type weight from a narrowing duration gap can be fully explained by the change in the gap; the ownership-transfer hypothesis predicts that after controlling for these, excess de-risking is still significantly explained by "who dominates" and "whether the surplus has just turned positive"—under the same hedging cost and the same change in the gap, plans that have just turned from deficit to surplus and are sponsor-dominated de-risk faster (the disposition effect plus surplus endowment is at work).

The sharpest test point is the "excess" portion in which the sponsor, knowing full well that the mathematics of continued operation is superior (the forgone continued-operation upside is ex post estimable and systematically positive), still proceeds to buy out—this excess portion cannot be explained by rational optimization. If, after controlling for hedging cost and the change in the gap, the governance-dominance interaction term is not significant, the rational hypothesis wins and this leg retreats to "a restatement of rational acceleration."

6.5.5 Model Design and Pre-Registration

• Main specification: a regression of de-risking or buyout speed on hedging cost, the change in the duration gap, the interaction of governance-dominant party with the timing of the surplus turning positive, and controls, with country and time fixed effects.
• Robustness checklist: first, within the same fund, a comparison of de-risking speed during the rate-rise period between sponsor-dominated and trustee-dominated cases; second, "excess de-risking" defined as actual de-risking minus the rational-ALM prediction, regressed on the governance-dominant party; third, sensitivity to the definition of governance coding (majority seats versus veto rights versus contribution share); fourth, a placebo—plans with pure trustee governance and no sponsor cash anchor (such as certain public or industry-wide plans) should not display excess acceleration after the surplus turns positive.
• Pre-registered expectations: the judgment supported (this leg holds) is that, after controlling for hedging cost and the change in the gap, the interaction of governance-dominant party with the surplus turning positive is significant, and the forgone continued-operation upside, estimated ex post, is systematically positive yet still traded away (the excess portion is greater than zero and is explained by sponsor dominance); the judgment falsified (this leg retreats) is that, after controlling for these, the interaction term is not significant, de-risking speed can be smoothly explained by hedging cost plus the change in the gap, unrelated to the governance-dominant party, with this leg retreating to a restatement of "rational acceleration when rates are high."

6.5.6 Evidentiary Status: Regression Downgraded, Behavioral Fingerprint Verified

The governance-dominant-party coding needed for the main regression (hand-constructed from charters, board composition, and consolidation conventions) is unavailable given the constraints on micro-data accessibility. This is the weakest link in the entire chapter: governance dominance is highly heterogeneous across P7, requires hand coding, carries large measurement error, and attenuation of the coefficient toward zero is a known threat. The chapter has pre-fixed this point as a measurement precondition: if coding noise causes the standard error of the interaction term to be too large to permit identification, this leg remains a "sharply stated, evidentially thin" theoretical proposition, explicitly labeled as unidentified, without overclaiming due to insufficient coding. Accordingly, subject to this constraint, this paper does not report the interaction-term coefficient.

But unlike the first two legs, this leg's key behavioral fingerprint (the acceleration of the 2022-2023 record-level buyout wave in the thick-premium window) has already been verified, consistent with the pre-registered temporal expectation. What needs to be precisely distinguished is: this verified fingerprint proves the phenomenon that "de-risking indeed accelerated precisely when it should least have accelerated," but it cannot, on its own, cut "reference-point ownership transfer" apart from "rational acceleration"—that cut requires the governance coding data that is unavailable given the data constraints. Therefore, the status of this leg is: the temporal phenomenon is verified, the core causal identification is downgraded.

6.6 Chapter Summary: The Evidentiary Status and Shared Boundary of the Three Legs

The chapter's core conclusion—"the triggering mechanism of liability-side de-risking is behavioral reference points, systematically deviating from first-order optimization"—is borne by three legs, each with its own clean discriminating identification, each corresponding to a fingerprint that a friction, rule, or rational model cannot generate: the asymmetric, path-dependent dead zone of the unidirectional ratchet; the clustering at psychological round numbers outside statutory thresholds for the round-number leg; and the excess de-risking dominated by governance after controlling for hedging cost for the reference-point-ownership-transfer leg. The three cuts (pair differencing, removal of statutory thresholds, and the governance-times-surplus interaction) are each independent; overturning any one does not implicate the others.

Under the data conditions this paper relies on, the micro-level identification of all three legs is downgraded (fund-level micro-data is unavailable), and per the pre-registered power, granularity, and measurement preconditions, they stop at the design stage, without overclaiming already-identified causality. The sole exception is the behavioral fingerprint of the reference-point-ownership-transfer leg: the acceleration of the 2022-2023 record-level buyout wave in the thick-risk-premium window has been verified, consistent with the expected timing—this is the hardest piece of empirical evidence in this chapter, but what it proves is the phenomenon, not the causal cut (the US's single-year figure of $51.8 billion in 2022, the UK's approximately £49-50 billion in 2023). To reiterate the shared boundary: this chapter claims only that the liability-side trigger is behavioral; the covariance repair behind the asset side's increased allocation to alternatives is explicitly labeled as optimization, serves as a control boundary, and is not included under the falsification umbrella—this boundary ensures that the "behavioral deviation" claimed by the three legs cannot be absorbed by the single sentence "the covariance just changed."

7. Chapter 5 The Boundary of Settlement Physics: What Pension Funds Can Hold and How Much Leverage They Can Add Is Determined by the Physical Constraints of Settlement and Collateral

7.1 From the Preference Layer to the Physical Layer: Where This Chapter Relocates the Constraint

The constraints addressed in the first four chapters all sit at the "preference layer" or the "signal layer"; they answer the question of "what pension funds want to hold, and how they react": institutional accounting refracts the signal (Chapter 2), the coefficient absorbs a common cause (Chapter 1), post-hedging carry constrains overseas allocation (Chapter 3), and a reference point distorts the liability-side reaction function (Chapter 4). This chapter turns to a layer that none of the previous chapters has touched: the physical layer — whether pension funds can settle, in what form they settle, and how settlement cost moves with price.

The substance of this pivot is to move the hard constraint on allocation forward. The traditional narrative places the LDI crisis at a later stage, and sets the allocation constraint at the preference or risk-budget layer ("how much duration is owed"). This chapter's judgment is: the true hard constraint lies neither in "whether one wants to" nor in "how much duration is owed," but in this physical settlement layer. The low-rate era (2009–2021) did not "entice" pension funds into reaching for duration by lowering the discount rate; rather, it redefined the feasible set of what pension funds can hold and how much leverage they can safely add through two settlement-layer mechanisms that each have an exogenous source of identification.

This chapter's causal chain connects strictly only two mechanisms — the cash variation-margin instrument-mismatch tax and collateral reflexivity — both of which operate at the terminal end of settlement, and both of which have genuine exogenous sources of identification. A third, related mechanism (the UK Debt Management Office's duration rationing) is retained as a supply-side background observation, but is explicitly labeled as unidentified, not admitted into the causal chain, not treated as a load-bearing pillar, and not carried into the core conclusions. This is not a matter of length but a matter of identification discipline: the two load-bearing legs identify a causality that terminates at settlement; the rationing mechanism cannot identify the connection from the issuance end to the settlement end, and this chapter refuses to pass off an unidentified segment as an identified causal chain.

7.2 The Judgment and Falsifiability of the Two Load-Bearing Legs

This chapter relocates the constraint to the physical settlement layer, borne specifically by two mechanisms:

First, the instrument-mismatch tax of cash variation margin (load-bearing, identification design already established). Central clearing mandates that variation margin be posted in cash, whereas pension funds naturally do not hold cash — their assets are long-duration bonds and real assets, and cash is the very last thing they wish to hold. Low interest rates amplify the opportunity cost of "holding cash to meet variation-margin calls" (roughly equal to the forgone duration return). As a result, what looks like an interest-rate hedge that is "long duration" is, at the settlement layer, reconstructed into an implicit short cash position that is "forced to minimize cash buffers and rely on collateral transformation (borrowing cash via repo)." The key structure here is: the intermediary rule (cash variation margin) is an institutional constant, low rates are the amplifier, and it is the interaction term between the two that is the neglected first-order variable — looking at the rule alone, or at rates alone, reveals nothing; the constraint only surfaces once they are multiplied together.

Second, collateral reflexivity (load-bearing, identification design already established). When the same class of asset (long-duration government bonds) is held both as a strategic holding and as hedging collateral, the procyclicality of central counterparty and bilateral margin models causes "the market value of collateral" and "the haircut or initial margin it is required to post" to move in a negatively reinforcing loop — as collateral falls in value, the very financing capacity it can lever up is simultaneously withdrawn, forming an endogenous amplification loop at the allocation layer that cannot be eliminated by diversification. During the low-rate period this loop was suppressed rather than eliminated by low volatility (models generated abnormally low initial margins, feeding implicit leverage), and was exposed all at once when volatility mean-reverted during the exit phase.

The incremental contribution of these two judgments relative to the traditional framework is clear: the traditional framework contains no settlement-layer mechanism whatsoever. This chapter demonstrates that settlement-end collateral (reflexivity) and the settlement medium (instrument mismatch) are the true hard constraints. The core distinction is: the identification of the two legs does not depend on the issuance narrative of the rationing mechanism — regardless of the reason pension funds hold synthetic duration, settlement fragility holds independently. This is precisely why the two legs can stand on their own while the rationing mechanism can serve only as background.

Both judgments are falsifiable — a state of the world that would render each false can be written out:

• If the instrument-mismatch tax is false: were the variation-margin medium frictionlessly convertible (cash and government bonds instantaneously interchangeable, zero haircut, available on demand at all times), the mismatch tax would disappear and allocation would revert to a pure preference solution — there should be no systematic relationship between the share of cash buffers and "the lower the rate, the steeper the curve."
• If collateral reflexivity is false: were margin models sufficiently counter-cyclical (with floors, long lookback windows, and stress calibration such that initial margin is already elevated in calm periods and does not jump during crises), the reflexivity loop would be severed and safe assets would indeed be safe collateral — within the same stress window, the decline in government bond prices should show no positive co-movement with the rise in initial margin or haircuts, and forced sales should not intensify with the degree of overlap between "assets" and "collateral."

7.3 First Leg: The Cash Variation-Margin Instrument-Mismatch Tax

7.3.1 Mechanism: The Three-Step Transmission of the Intermediary Tax

The identifiable causal chain (at the design layer) of the instrument-mismatch tax is a three-step transmission:

First, from settlement-medium rule to collateral-demand structure: central clearing mandates cash variation margin, forcing pension funds to trade off between "holding government bonds to earn duration return" and "holding cash to meet variation-margin calls."

Second, from low rates to a worsening trade-off: in a negative-carry environment, the cost of holding cash is high (the forgone duration return is large), pushing pension funds toward the extreme form of "minimizing cash buffers while relying on collateral transformation (borrowing cash via repo)."

Third, this form quietly converts interest-rate risk into settlement-liquidity risk — the effectiveness of the hedge now depends on whether the pipeline for "converting government bonds into cash variation margin on demand" remains open during periods of stress.

During the 2022 exit phase, this pipeline became clogged, and the implicit tax was made manifest all at once.

7.3.2 Verified Institutional Roots and the Manifestation Event

The mechanism of the instrument-mismatch tax is publicly verified both at the level of institutional fact and at the level of the manifestation event:

• The institutional root is verified: under the European Market Infrastructure Regulation, all counterparties trading in the EU must post variation margin in cash (rather than in bonds); pension funds do not hold large amounts of cash, making it especially difficult for them when huge cash variation-margin calls arrive during a stress period. This is precisely the institutional constant behind the instrument-mismatch tax (Macfarlanes 2019). Low rates are the amplifier, the cash variation-margin rule is the constant, and it is their interaction — the mechanism's institutional-fact layer is solidly established.
• The manifestation event is verified: after the "mini-budget" of September 23, 2022, government bond yields spiked sharply, triggering margin and collateral calls of over £70 billion on LDI funds and pension funds; pension funds sold government bonds to meet the margin calls, driving prices into a downward spiral; the Bank of England intervened with emergency targeted purchases of government bonds to restore order and to buy LDI funds time to recapitalize (Federal Reserve Bank of Chicago 2023; Bank of England 2022). This margin surge of over £70 billion is precisely the macro-level manifestation event of "the implicit tax being made manifest all at once."

One further point of common confusion in public discussion needs clarifying: the figure of "£65 billion" is not the scale of the margin calls, but the total purchase ceiling of the Bank of England's intervention on that occasion (with a daily purchase cap of £5 billion at the start of the operation); it is a different measure from the margin-call scale of over £70 billion and the two should not be conflated (Bank of England 2022).

7.3.3 The Strongest Counterargument: The Mismatch Tax Merely Renames a Duration Preference

The opposing view holds: the low cash buffers and high leverage of pension funds can be fully explained by the preference- or liability-layer story of "widening duration gaps under low rates leading to leverage to hedge duration"; even if a change in collateral-medium composition is observed around the expiry of the European Market Infrastructure Regulation exemption, this could merely be a coincident phenomenon in which the same-period path of interest rates drove changes in duration-hedging demand, rather than a causal effect of "the tightness of the cash variation-margin constraint." In other words, this leg is observationally equivalent to the "pure duration preference" solution, and the intermediary rule is nothing more than a surface veneer over the duration story.

7.3.4 Treatment: A Quasi-Natural Experiment Using the European Market Infrastructure Regulation Pension Clearing Exemption

The treatment approach is to use the expiry or extension of the European Market Infrastructure Regulation pension clearing exemption as a quasi-natural experiment, decoupling the intermediary constraint from the duration preference on the basis of exogenous variation, and pre-registering the falsification conditions.

The source of identification is the exogenous switch, from the exemption "being in place" to "approaching lapse," in the tightness of the cash variation-margin constraint — the extension or expiry of the exemption is determined by the EU's legislative calendar, orthogonal to any single plan's duration gap, and it therefore moves "whether government bonds can be used to satisfy variation margin" rather than "whether one wants to hedge duration." In the difference-in-differences setup, only the portion that remains significant after controlling for the duration gap is counted as the net effect of the instrument-mismatch tax; if the coefficient goes to zero after controlling for duration, the opposing view of "merely renaming a duration preference" wins and this leg concedes (a pre-registered concession threshold). The placebo group consists of plans that do not hedge duration, or that are not bound by the clearing obligation; their collateral-medium composition and leverage should show no response around the exemption's turning point; if the placebo group jumps in the same way, identification is contaminated by a common shock and this leg does not hold.

The exogenous source for this quasi-natural experiment has been precisely verified: on June 9, 2022, the European Commission adopted a delegated regulation extending the exemption by one year (originally set to expire on June 18, 2023); the EU's European Market Infrastructure Regulation exemption ultimately expired on June 18, 2023 (with no further extension); the UK Treasury extended the UK and European Economic Area pension exemption by two years to June 18, 2025 (effective June 12, 2023) (Norton Rose Fulbright 2023). Thus the exogenous source of the treatment timing for this leg's difference-in-differences design (exemption in place versus approaching lapse) is precisely verified — it is set by the EU's or the UK's legislative calendar, orthogonal to any single plan's duration gap, and the repeated two-year extensions plus the 2025 expiry provide multiple points of identification.

7.3.5 Modeling Design and Pre-Registration

• Difference-in-differences specification: regress collateral-medium composition or leverage on approaching exemption lapse, duration gap, plan fixed effects, and time fixed effects; the net mismatch tax equals the coefficient on approaching exemption lapse after controlling for the duration gap. This coefficient should be insignificant for the placebo group.
• Pre-registered expectations (one-sided sign predictions, each with an attached concession threshold): First (cross-section), jurisdictions or plans with a higher share of cash variation margin (subject to a stronger clearing obligation) should have higher LDI leverage multiples and greater sell-off intensity during stress periods, with the sign fixed as positive; the concession threshold is that if the sign is zero or reversed, the judgment on the instrument-mismatch tax is falsified. Second (time series), hedges established at points when the policy rate is lower and the term spread is wider should have a lower share of cash buffers, with the sign fixed as negative; the concession threshold is that if the buffer share shows no systematic relationship with "low rates times a steep curve," the "low-rate amplifier" mechanism is falsified, and this leg reverts to a pure institutional constant with no low-rate interaction increment. Third (the core of the difference-in-differences), after controlling for the duration gap, "approaching exemption lapse" should still have a significant positive effect on the decline in the share of government bonds usable to meet margin or the rise in the share of cash variation margin, with no such effect in the placebo group; the concession threshold (critical) is that if the coefficient goes to zero after controlling for the duration gap, this leg is observationally equivalent to the "pure duration preference" solution and the judgment concedes to being merely a mechanism hypothesis; if the placebo group jumps in the same way, identification is contaminated by a common shock and this leg does not hold.

7.3.6 Evidentiary Status: Downgraded

The plan-level panel data on cash variation-margin medium composition (share of cash variation margin, or share of government bonds usable to meet margin) and LDI leverage multiples required for the main test are unavailable owing to constraints on micro-data availability (the Pensions Regulator's LDI survey does not disclose plan-level distributions). Owing to this constraint, this paper does not report this coefficient and does not report the placebo test.

A precise distinction must be drawn between what has and has not been verified for this leg: the exogenous source of the treatment variable (the exemption's timing) is precisely verified (see above), the design is in place and carries its own falsification conditions (if the coefficient goes to zero after controlling for duration, this is observationally equivalent to the pure duration-preference solution and the leg concedes; if the placebo group jumps synchronously, identification is contaminated). The one-sided signs and concession thresholds have been fixed ex ante through pre-registration. However, "the net effect of approaching exemption lapse after controlling for the duration gap" requires a plan-level cash variation-margin medium panel that is unavailable owing to data constraints; the downgrade path (using net repo positions or cash-holding ratios as a proxy) likewise requires non-public plan-level data and has not been executed.

7.4 Second Leg: Collateral Reflexivity

7.4.1 Mechanism: The Four-Ring Reflexive Loop

The identifiable causal chain (at the design layer) of collateral reflexivity is a four-ring reflexive loop: first, government bonds are simultaneously an asset and a form of collateral; second, a rise in volatility causes margin models (value-at-risk type, with short lookback windows) to mechanically push up initial margin or haircuts; third, meeting the additional margin call requires selling government bonds or borrowing more via repo, which drives government bond prices down further and pushes volatility up again; fourth, the loop returns to the second ring.

During the low-rate period, prolonged low volatility caused models to generate abnormally low initial margins, and pension funds built up a structural dependence on "low margin" (supporting the largest possible hedge with the least possible collateral) — this is precisely the implicit leverage fed by the model's low volatility. During the exit phase, as volatility mean-reverted, the procyclicality of the models caused margins to jump from abnormally low to normal, exposing the implicit leverage instantaneously.

The instrument-mismatch tax and collateral reflexivity sit at different levels of the collateral pipeline: the former concerns "what is used to pay" (the medium layer: the instrument mismatch tax of cash versus government bonds), while the latter concerns "how much must be paid, and how that changes with the price of collateral" (the dynamic layer: model procyclicality and the reflexivity of the same asset). The two are complementary but each has its own independent exogenous source of identification, and neither is a necessary precondition for the other.

7.4.2 Verified Mechanism Direction and Policy Acknowledgment

• The macro-level manifestation of the reflexive loop is verified: the Bank of England explicitly documented the "vicious spiral of collateral calls and forced government bond selling" — this is precisely the macro-level manifestation of the second through third rings of the reflexive loop (additional margin calls, selling government bonds, further declines, further calls) (Bank of England 2022).
• Policy acknowledgment of margin procyclicality is verified: margin procyclicality has been listed as an international policy concern, with the Basel Committee and others pushing to improve the transparency of central counterparty initial margin models and requiring central counterparties to further assess the procyclicality of their initial margin models (Bank of England 2022). Both the mechanism direction of collateral reflexivity — "margin-model procyclicality causes financing capacity to be withdrawn precisely as collateral falls in value" — and its policy acknowledgment are verified.

7.4.3 The Strongest Counterargument: This Is Merely a Broad Market Sell-Off, Not a Separable Reflexivity

The opposing view holds: the 2022 sell-off was the result of a broad market decline combined with a one-directional sharp rise in interest rates; "margin-model procyclicality reflexivity" is neither necessary nor separable — singling it out is merely pinning a mechanism label on a macro interest-rate shock. After the "mini-budget," everyone holding government bonds was losing money and every leveraged position was facing margin calls; forced selling can be explained without invoking a "same-asset reflexive loop" at all; moreover, "collateral falling in value" and "margin rising" both share the same origin — the interest-rate shock — and their positive co-movement is a spurious correlation produced by a common cause (interest rates), from which "reflexive amplification" cannot be separated out from "a broad market sell-off."

7.4.4 Treatment: Cross-Sectional Heterogeneity in Margin-Model Parameters

The treatment approach is to use cross-sectional heterogeneity in central counterparty or bilateral margin-model parameters as the source of identification, differencing "reflexive amplification" out from "a broad market sell-off." The model parameters of different central counterparties or bilateral agreements (lookback window length, presence or absence of a floor, presence or absence of an anti-procyclicality tool) are cross-sectionally heterogeneous under the same shock (the 2022 "mini-budget"): comparing the margin jump and forced selling of similar plans under "more procyclical models" versus "smoother models" arrangements, the broad market decline is a common term shared by both groups (and is thus differenced out), and what remains in the residual — "the more procyclical group jumps more sharply and sells off more" — is the net contribution of reflexivity. The placebo is a collateral arrangement using a fixed haircut (one that does not vary with volatility) — if the fixed-haircut group is likewise forced to sell, the driving force is the broad decline rather than model procyclicality, and this leg concedes. On the low-volatility-feeding side, cross-verification comes from the positive correlation between plan leverage and the "degree to which the model's initial margin runs low" (implied initial margin from 2015 to 2021 relative to a stress-calibrated benchmark), which is a second observational facet of the same mechanism as the reflexive loop.

7.4.5 Modeling Design and Pre-Registration

• Event-study specification: regress forced selling or margin jumps on the degree of model procyclicality, the degree of overlap between "assets" and "collateral," and their interaction, plus shock fixed effects; the net reflexive amplification equals the coefficient on the degree of procyclicality plus the interaction term (predicted sign greater than zero). The coefficient on procyclicality should be insignificant for the fixed-haircut placebo group.
• Pre-registered expectations (one-sided sign predictions, each with an attached concession threshold): First (strength of reflexivity), within the same stress window, the decline in government bond prices should positively co-move with the rise in the applicable haircut or initial margin, and this co-movement should be stronger at plans with a higher degree of "asset equals collateral" overlap (interaction term greater than zero), with the sign fixed as positive; the concession threshold is that if the co-movement is insignificant or does not strengthen with the degree of overlap, the reflexive loop is falsified. Second (low-volatility feeding), from 2015 to 2021 the model-implied initial margin should be significantly below the stress-calibrated benchmark, and plan leverage should be positively correlated with "the degree to which the model's initial margin runs low"; the concession threshold is that if the implied initial margin is not systematically below the benchmark, or if leverage shows no positive correlation with the degree of underestimation, "low volatility feeding implicit leverage" is falsified. Third (procyclicality causality), smoother central counterparties or jurisdictions that introduce a margin floor or a longer lookback window should show a smaller margin jump and less member selling during stress periods (coefficient on the degree of procyclicality greater than zero), and the fixed-haircut placebo group should not show selling that jumps with volatility; the concession threshold is that if the smoother group and the more procyclical group show no difference in jump size or selling, or if the fixed-haircut group is likewise forced to sell, the driving force is the broad market decline rather than model procyclicality, and this leg concedes.

7.4.6 Evidentiary Status: Downgraded, With an Unaddressed Residual Threat Noted

The specific central counterparty margin-model parameters (lookback window length, presence or absence of a floor, anti-procyclicality tools), pension sub-account margin data, and plan leverage required for the main test are unavailable owing to constraints on micro-data availability (central counterparty model parameters and pension sub-account margin are largely non-public). Owing to this constraint, this paper does not report the coefficient on the degree of procyclicality, does not report the interaction term, and does not report the degree of underestimation of implied initial margin.

Here, one residual concern that has not previously been explicitly pre-registered as a threat must be honestly noted: namely, "whether more procyclical central counterparties or venues systematically cleared higher-risk exposures." If venue selection is not orthogonal to plan risk (a more procyclical model's venue happens to have cleared riskier exposures), the coefficient on the degree of procyclicality would overstate reflexive amplification. This paper's judgment is: margin-model parameters are primarily a choice of the risk framework made by the central counterparty or the regulator (the European Market Infrastructure Regulation's anti-procyclicality technical standards, lookback windows), applied uniformly to all members of the same product rather than tuned to any individual pension fund; and because government bond clearing venues are few (the choice is essentially limited to "which central counterparty to use," largely determined by the product), venue-level model attributes are approximately orthogonal to individual plan risk — the combination of "comparable-plan controls plus overlap-degree interaction plus fixed-haircut placebo" constitutes a defensible identification response, with no specific channel of breakdown, though it does not amount to a proof or a concession threshold. However, owing to data constraints, there is no joint micro-data on central counterparty parameters times plan risk with which to verify orthogonality, so this is honestly flagged as such: falsifying or confirming this threat would require this joint micro-data, which is unavailable owing to data constraints; the design-layer response holds, but has not been executed at the empirical layer; if the coefficient on the degree of procyclicality is contaminated, the direction of the bias is toward overstating reflexive amplification.

7.5 Treatment of the Rationing Mechanism: Supply-Side Background, Explicitly Labeled Unidentified, Excluded From the Chain

This chapter contains one related mechanism that must be handled with extreme discipline: the UK Debt Management Office's duration rationing. It constitutes an appealing supply-side narrative, but this chapter refuses to elevate it to an identified causal status.

The descriptive background of supply contraction runs as follows: as rates exited the low regime, DB plans matured or their surpluses improved, demand for hedging at the ultra-long end became more rigid, ultra-long-end yields were pushed down, the Debt Management Office observed that the ultra-long end was "expensive" and that domestic DB was shrinking, it shortened the weighted maturity of issuance, net investable duration supply contracted, and (a background observation, not an identified cause) pension funds were forced to substitute synthetic duration for physical bonds, feeding into the settlement chain described above. As a motivational backdrop explaining "why net investable duration supply contracted," this narrative holds. The descriptive fingerprints supporting it include: a negative correlation between ultra-long-end yields and the Debt Management Office's share of ultra-long issuance, with the Debt Management Office's side responding with a lag; a downward adjustment in the UK Debt Management Office's share of long-dated gilt issuance, and a record inversion between the 2073 maturity bond and the 30-year bond; a sharp marginal decline in DB net purchases of physical bonds, alongside a monotonic rise in the share of incremental hedging achieved through synthetic duration.

But this chapter explicitly declines to elevate this to a causal status: the causal link from "issuance contraction" to "synthetic duration substitution" is not identified — the Debt Management Office's objective function is unobservable, and there is no exogenous shock that would move duration demand independently while leaving the Debt Management Office's objective unchanged; whether the downward adjustment in the ultra-long share reflects active rationing, a lagged response to demand, a third factor, or a reverse supply-scarcity premium cannot be disentangled with the tools currently available. This rationing mechanism therefore remains forever a background observation; it is not elevated to identification, not admitted into the causal chain, not treated as a load-bearing pillar, and not carried into the core conclusions. The evidence it relies on (the Debt Management Office's issuance plans and consultations, fiscal term-structure reports, the curve inversion) supports a correlational fingerprint rather than a causal one; accordingly, this chapter sets no falsification threshold for the rationing mechanism, applying only an existence test for the supply-side background — even were it to hold in full, it would not be elevated to identified causation.

7.6 Chapter Summary: Causation at the Terminal End of Settlement and Honest Identification Boundaries

This chapter's core conclusion — that "the physical constraints of settlement (the instrument-mismatch tax plus reflexivity), rather than liability duration, determine what pension funds can hold and how much leverage they can add" — relocates the hard constraint on allocation from the preference layer to the physical settlement layer, borne by two legs each with its own independent exogenous source of identification: the cash variation-margin instrument-mismatch tax (the medium layer) and collateral reflexivity (the dynamic layer). Neither leg's identification depends on the issuance narrative of the rationing mechanism, which is precisely why each can stand on its own.

Under the data conditions available to this paper, the following are all verified: the manifestation event (the 2022 margin surge of over £70 billion plus the Bank of England's intervention); the institutional root of the instrument-mismatch tax (the European Market Infrastructure Regulation's cash variation margin); the exogenous source of the treatment timing for the instrument-mismatch tax (the exemption's extension in June 2022, its EU expiry in June 2023, and its UK expiry in June 2025); and the policy acknowledgment of the procyclicality behind collateral reflexivity. However, the plan-level or central-counterparty-parameter-level identification regressions for both legs are downgraded (micro-data unavailable), remaining at the design layer per the pre-registered one-sided signs and concession thresholds, with no coefficients reported. The residual threat (the venue-selection threat to reflexivity) is honestly noted as "the design-layer response holds, but has not been executed at the empirical layer," with the direction of any contamination to the procyclicality coefficient being toward overstatement. The rationing mechanism is labeled unidentified throughout, treated as supply-side background, and excluded from the core conclusions. The scale of the 2022 LDI margin calls was over £70 billion (the £65 billion figure often cited in public discussion is the total purchase ceiling of the Bank of England's intervention, a different measure).

8. Conclusion: The Dichotomy Itself Is Misspecified — Four Overlooked Drivers

8.1 Returning to the Core Question

The core question posed at the outset of this paper is: through what mechanism does a prolonged period of low interest rates systematically alter pension funds' strategic asset allocation function? The judgment focus throughout has been an apparently either/or dichotomy — is reaching for yield dominant, or is asset-liability management (ALM) pressure dominant? This matters because the two mechanisms carry opposite policy implications: if it is reaching for yield, regulators should manage risk appetite; if it is ALM pressure, they should manage discount-rate rules. It is therefore, at bottom, not a question of regression coefficients, but a judgment about the very nature of pension-fund behavior.

This paper's answer is: the dichotomy itself is misspecified. The misspecification operates at two levels. First, at the level of measurement — the observed coefficient supporting either side is itself untrustworthy: Chapter 1 demonstrates that treating the naive "interest rate → allocation" coefficient as causal ignores population aging as a common cause that simultaneously drives the equilibrium real rate and duration demand; the coefficient absorbs an effect that properly belongs to demographics, and its direction is contaminated. Using an untrustworthy coefficient to adjudicate the dichotomy is building a tower on sand, whichever way it is decided. It should be clarified that the untrustworthiness of the observed coefficient has two distinct sources: the debunking of an identified demographic common cause (this paper's load-bearing claim, and the sole basis on which this paper adjudicates magnitude here), and a theoretical-level, unidentified demand-side reflexive collusion (in which the causal arrow reverses to "demand → environment," retained throughout only as a theoretical proposition, never passed off as identified) — the conclusion's adjudication of magnitude relies solely on the former; the latter merely reinforces the warning that "this coefficient cannot be read as exogenously causal" and does not enter into any claim about magnitude. Second, at the level of mechanism — treating the dichotomy as a cleanly separable opposition is itself an error: the mechanisms that actually drive allocation are entirely overlooked by this dichotomy and are scattered across four layers, and the dichotomy's other pole, ALM, is not superseded by these four layers but rather is dispersed precisely within them (the accounting motives of liability-driven investment (LDI), post-hedge carry, and the narrowing of duration gaps can all be understood as components of ALM broadly construed). Hence "who dominates" is not a question that can be cleanly adjudicated by an observed coefficient.

8.2 The Four Overlooked Driver Layers

The paper's four empirical chapters each reveal a layer of genuine driver obscured by the "reaching for yield versus ALM" dichotomy (for a comparison of the core claims, real variables, and evidentiary status of the four mechanism layers, see Table 1):

Layer One: Accounting conventions. What enters the allocation function is never "the interest rate" but rather the shadow rates refracted through four misaligned institutional accounting regimes; allocation drift is driven by the opening and closing of the accounting wedge. The identifiable core evidence treats IAS19R (effective 2013, which eliminated the corridor method and the discretionary expected-return assumption, and adopted the net-interest method) as a quasi-exogenous rate shock — functionally equivalent to an institutional downward shift in expected returns that severed two accounting subsidies to equity holding; the intensity of de-equitization varies with sponsor balance-sheet fragility rather than plan duration. The institutional facts are fully verified (IFRS Foundation 2011); published literature (Anantharaman and Chuk 2018's Canadian sample, "IAS19R shifted equity to bonds, primarily due to the elimination of the expected-return assumption") supports the core claim directionally; micro-level dose identification is downgraded.

Cross-cutting layer: Identification unreliability. This layer does not assert a positive claim; it only debunks: it warns that any of the aforementioned "interest rate → allocation" coefficients is contaminated by a demographic common cause. Two demographic upstream channels — population aging's contribution to the equilibrium real rate (approximately 1.25 percentage points since 1980, "much, if not all") and plan maturity's effect in raising duration or long-bond demand via — are both verified (Gagnon, Johannsen, and López-Salido 2016; OECD 2024; IMF 2025). The debunking judgment that "the interest-rate coefficient absorbs an omitted demographic effect" thus has public support at the mechanism level. Honest boundary: the magnitude of the lower bound of the bias cannot be bounded (because demographics' influence on the equilibrium real rate includes a genuine discount-rate channel, controlling for demographics would simultaneously remove part of the genuine interest-rate effect); the core conclusion claims only that "the observed coefficient is unreliable," not any magnitude.

Layer Two: Post-hedge carry. The real constraint on overseas allocation is not "home-currency interest rates monotonically driving overseas allocation" (the Bank for International Settlements' 1-percentage-point-to-3.4-percentage-point naive correlation plus the reaching-for-yield narrative is precisely the correction target), but rather post-hedge carry and regulatory corner solutions. When short-end inversion causes hedging costs to eat into the spread and post-hedge carry turns negative, overseas or hedged foreign debt gets de-allocated while home-currency long-end holdings remain low — this is directionally consistent with already-verified facts such as the Government Pension Investment Fund (GPIF) of Japan's 2014 allocation revision (each of four asset classes at 25%, domestic bonds cut down from roughly 60%), GPIF's 2022 sell-off of foreign bonds, and the Dutch civil-service pension fund's halving of equity dollar hedging from 50% to 25% (GPIF 2014; Council on Foreign Relations 2023; IPE 2023); the regulatory corner solution (European Market Infrastructure Regulation cash variation margin, Solvency II capital charges for currency mismatch) is institutionally verified. Identification at the fund level or index level for both legs is downgraded; the correction target — the Bank for International Settlements' 1-percentage-point-to-3.4-percentage-point figure — is verified verbatim (BIS 2024).

Layer Three: Behavioral reference points. The triggering mechanism for liability-side de-risking is not first-order optimization of a yield gap, but a behavioral reference point — a one-way ratchet of loss aversion, the perceptual salience clustering of psychological round numbers, and the transfer of the reference point from trustees to sponsors. Each of the three legs has a clean discriminating identification cut (matched difference-in-differences, exclusion of statutory thresholds, governance-by-surplus interaction); micro-level identification is fully downgraded across all three, except that the behavioral fingerprint of the reference-point-transfer leg — the record-level buyout boom of 2022–2023 accelerating within a thick risk-premium window (the United States' single-year total of USD 51.8 billion in 2022, the United Kingdom's roughly GBP 49–50 billion in 2023) — is verified (Aon 2023; LCP 2024). The boundary is: only the liability-side trigger is claimed to be behavioral; the covariance repair from adding alternatives on the asset side is explicitly marked as optimization, serving as a contrasting boundary case.

Layer Four: Settlement physics. What pensions can hold, and how much leverage they can add, is ultimately determined by the physical constraints of settlement and collateral — the cash variation margin instrument-mismatch tax and collateral reflexivity, both of which have genuine exogenous sources and both of which act at the settlement end. Low interest rates do not "induce" reaching for duration through the discount rate; rather, they redefine the feasible set through these two settlement-layer mechanisms. The manifesting events (the 2022 margin surge of over GBP 70 billion plus the Bank of England's intervention), the institutional roots in the European Market Infrastructure Regulation, the timing of exemption treatment, and the acknowledgment of procyclical policy are all verified (Bank of England 2022; Federal Reserve Bank of Chicago 2023; Macfarlanes 2019; Norton Rose Fulbright 2023); identification for both legs is downgraded; the rationing mechanism is explicitly marked as unidentified and serves only as supply-side background, not entering the causal chain.

It should be noted that the cross-cutting layer is a cross-cutting identification warning rather than a parallel fifth layer: it invalidates the premise that "the observed coefficient can adjudicate the dichotomy," thereby clearing space for the four genuine mechanism layers — accounting conventions, post-hedge carry, behavioral reference points, and settlement physics.

8.3 Policy Implications: What to Regulate Depends on Which Layer Dominates

If the "reaching for yield versus ALM" dichotomy is misspecified, then the policy either/or of "manage risk appetite or manage discount-rate rules" collapses along with it. This paper's four-layer mechanism offers a more granular policy mapping — what to regulate depends on which layer dominates in a given jurisdiction at a given time:

• If accounting conventions dominate, the policy lever is the accounting standard itself (rules for recognizing expected returns, the corridor method, treatment of other comprehensive income) — de-equitization is a product of standard-setting change, and neither managing risk appetite nor managing discount rates hits the target.
• If post-hedge carry dominates, what matters is the depth and cost of the hedging market, along with the regulatory capital rules that pin overseas hedging at a corner solution (European Market Infrastructure Regulation, Solvency II) — simply lowering home-currency rates will not linearly drive overseas allocation.
• If behavioral reference points dominate, the policy space lies in governance and information architecture (who dominates investment decisions, how reference points are set and presented), rather than in direct constraints on risk appetite.
• If settlement physics dominates, the policy lever lies in settlement-medium rules (whether to retain or abolish the cash variation margin exemption) and in the countercyclical design of margin models (floors, long lookback windows) — this is the direct lesson of the 2022 LDI crisis.

Hence this paper's policy implication is not a single prescription but a diagnostic framework: first identify which layer dominates, then select the corresponding lever. Reducing every situation to the "manage risk appetite versus manage discount rates" dichotomy is precisely the policy-level version of the misspecification this paper aims to dismantle.

8.4 The Mechanism Chain This Paper, as the First in the Series, Provides for Subsequent Research

This paper is the first of five progressively building papers (the theoretical-and-empirical foundation piece), with no upstream dependencies. What it provides for downstream papers is precisely this "interest-rate-to-allocation transmission mechanism chain" plus a set of stylized facts:

• It decomposes the seemingly simple "interest rate to allocation" transmission into four overlooked driver layers plus a cross-cutting identification warning (the demographic common cause); any subsequent paper studying allocation behavior in a particular jurisdiction, asset class, or under a particular shock can use this framework to locate which layer dominates.
• It offers a set of already-verified stylized facts (P7 allocation of roughly 48% equities, 31% bonds, 19% other, 3% cash, defined contribution (DC) share of roughly 63%; the Bank for International Settlements' 1-percentage-point-to-3.4-percentage-point figure; the timing and content of IAS19R; demographics' contribution of roughly 1.25 percentage points to the equilibrium real rate; GPIF's 2014 and 2022 actions; the 2022 LDI margin surge and Bank of England intervention; the timing of the European Market Infrastructure Regulation exemption; the record of the buyout boom, etc.), forming an evidentiary base that subsequent papers can cite directly.
• It brings forward the settlement-physics mechanism of the LDI crisis into this foundational piece — settlement constraints are not a special case that appears only during a crisis; they are a standing constraint that is continuously suppressed by low volatility during the low-rate period and only becomes manifest during the decumulation (payout) phase.

8.5 Honest Limitations

The limitations of this paper must be stated honestly, and they have been fixed in advance in each chapter rather than rationalized after the fact:

First, micro-level identification awaits data. The micro-level identification data required by the design of each load-bearing leg are unavailable given this paper's data constraints: a cross-standard panel of pension footnote disclosures (for the standards-based difference-in-differences), fund-level hedged/unhedged breakdowns and monthly hedge ratios (for the post-hedge-carry leg), plan-level detail on the cash-variation-margin instrument (for the instrument-mismatch tax), central counterparty margin-model parameters and sub-account-level margin (for collateral reflexivity), coding of the governance-dominant party (for reference-point transfer), the continuous micro-level distribution of funding ratios (for the round-number threshold), and a panel of recovery events (for the one-way ratchet). For all of these, this paper explicitly downgrades to "identification design in place plus literature synthesis," presenting the design, the pre-registered direction, and the support or contradiction from published literature as they stand, and flagging them as downgraded owing to the limits of micro-data availability. Owing to this constraint, this paper reports no regression coefficient, significance level, or sample size that has not actually been obtained.

Second, honest grading of evidentiary status. Three statuses are never conflated: first, publicly verifiable aggregate facts (with sources cited); second, literature-level directional support (always annotated as "literature-level, not micro-identified," never passed off as identified causation); third, downgraded items (honestly labeled as such). Every piece of directional evidence this paper is able to verify is consistent with the corresponding pre-registered direction, or substantiates the correction target; not a single verified fact contradicts the direction of any load-bearing leg. But this only amounts to the direction not having been refuted by evidence — it does not amount to the load-bearing leg having been identified.

Third, two explicit reservations, and unidentified items that do not overstep their bounds. The magnitude of the lower bound of bias at the identification layer cannot be bounded (the direction is confirmed, the magnitude is not obtainable); the site-selection threat to collateral reflexivity at the settlement layer has a defensible response at the design level but has not been executed at the empirical level — both are honestly noted as supplementary, without inflating any figures. Propositions explicitly marked as unidentified — the rationing mechanism, cross-border reverse pricing, demand-side reflexivity, distributional politics — are treated throughout only as background or literature discussion, never reported as causal, and never enter the core conclusions; the pre-registered fallback thresholds were all fixed prior to data, and this paper neither triggers nor evades any of them (for a comparison of the strongest counterarguments, identification strategies, and pre-committed falsification thresholds for each load-bearing mechanism, see Table 2).

These limitations do not diminish this paper's core contribution. This paper's contribution does not lie in estimating some numerical value for an allocation elasticity, but in demonstrating that the widely accepted dichotomy of "reaching for yield versus ALM" is misspecified — it treats an opposition that cannot be cleanly separated as an either/or choice adjudicable by a single coefficient, when that observed coefficient has already been contaminated by a demographic common cause and demand-side reflexivity, and ALM itself is dispersed across the accounting, hedging, and duration layers rather than standing as an independent option parallel to and mutually exclusive with the four layers — and, on this basis, offers a framework that reconstructs allocation transmission as four overlooked mechanisms, together with a set of falsifiable identification designs. This is not a claim that ALM is irrelevant, but a claim that the very question of "who dominates, reaching for yield or ALM" rests on an untrustworthy coefficient, and that ALM's role has already been dispersed into the four overlooked layers. Filling in the numbers awaits the availability of micro-data; the reframing of the framework is, at this point, already complete.

Numbered Figures and Tables

Table 1 Comparison of Core Claims Across the Four Mechanism Layers

Table 2 Comparison of Competing Hypotheses and Identification Strategies for Each Load-Bearing Mechanism

Table 3 Data Sources and Availability

9. Data Availability and the Boundaries of the Research

This section consolidates the downgrade statements scattered throughout the preceding chapters into a single, formal statement of the research's limitations.

Methodologically, this paper takes the position that judgment precedes data and that identification design is the core; the direct consequence is that this paper strictly distinguishes the evidentiary status of three types of claims and honestly labels every data boundary, never passing off a test that cannot be executed owing to data constraints as an already-identified empirical result. The three evidentiary statuses are: first, aggregate facts traceable directly to public sources, which this paper cites in-text with parenthetical references; second, micro-level identification data that a load-bearing mechanism requires but that are unavailable owing to limits on micro-data availability, which this paper explicitly downgrades while honestly presenting the identification design and the pre-registered direction; and third, pre-registered expected directions fixed in writing prior to seeing the data, which this paper explicitly labels as pending estimation or as directional predictions. Owing to the limits of micro-level identification data availability, this paper reports no regression point estimate, significance level, or sample size that has not actually been obtained.

Specifically, the micro-level identification data required by each load-bearing mechanism but unavailable are summarized as follows (for a summary of each data source and its availability status, see Table 3): the cross-standard-regime pension footnote panel required for the standards-based difference-in-differences; the fund-level hedged/unhedged breakdown flow panel and monthly hedge ratios required for the post-hedge-carry leg; the index of regulatory penalty intensity on hedged foreign debt and the same-country pension-versus-insurer allocation-alignment panel required for the regulatory-corner-solution leg; the plan-level detail on cash-variation-margin instruments and the LDI leverage-multiple panel required for the instrument-mismatch-tax leg; the central counterparty margin-model parameters and pension sub-account margin required for the collateral-reflexivity leg; the coding of the governance-dominant party required for the reference-point-transfer leg; the fund-level continuous micro-distribution of funding ratios required for the round-number-threshold leg; and the recovery-event panel required for the one-way-ratchet leg. For each of these, this paper has already presented, in the corresponding chapter, the complete identification design, the pre-registered direction fixed prior to seeing the data, the pre-committed falsification threshold, and the support for the corresponding direction from published literature or public facts, and has clearly labeled each as downgraded owing to limits on micro-data availability.

In addition, this paper honestly notes two supplementary reservations concerning robustness, neither of which oversteps its bounds or overstates any figure: first, the magnitude of the lower bound of bias for the debunking leg at the identification layer cannot be bounded — because the channel through which demographics affect the equilibrium real rate includes a genuine discount-rate component, controlling for demographics would simultaneously remove part of the genuine interest-rate effect, and so this paper asserts only the directional debunking warning that "the observed coefficient is unreliable," and asserts no point estimate of any magnitude of shrinkage; second, the collateral-reflexivity leg at the settlement layer carries a residual venue-selection threat that has not been explicitly pre-registered — if more procyclical clearing venues systematically cleared riskier exposures, the corresponding coefficient would overstate reflexive amplification. This paper judges that this threat has a defensible response at the design level (margin-model parameters are a risk-framework choice made by the central counterparty or the regulator, applied uniformly to all members holding the same product), but confirming or refuting it would require joint micro-data and has not been executed at the empirical level owing to data constraints.

Finally, this paper explicitly labels several propositions as unidentified — the supply-side rationing mechanism, cross-border reverse pricing, demand-side reflexivity, and distributional-politics refusal — treating them throughout only as background observations, theoretical propositions, or qualitative case illustrations, never reporting them as causal, and never admitting them into the core conclusions. This paper's pre-registered fallback thresholds were all fixed prior to data; under this paper's data conditions, none of them is either triggered or evaded, and the paper faithfully stops at the honest boundary of "identification design in place, plus directional support (from literature or public facts), plus micro-level identification downgrade." Every piece of directional evidence this paper is able to verify is consistent with the corresponding pre-registered direction, or substantiates the correction target; not a single verified fact contradicts the pre-registered direction of any load-bearing mechanism. But this only amounts to the direction not having been refuted by evidence — it does not amount to the load-bearing mechanism having been identified.

10. References (Works Cited)

Anantharaman, Divya, and Elizabeth Chuk. "The Economic Consequences of Accounting Standards: Evidence from Risk-Taking in Pension Plans." The Accounting Review, vol. 93, no. 4, 2018, pp. 23–51. https://publications.aaahq.org/accounting-review/article/93/4/23/4041/.

Aon. Seizing Opportunity in a Booming Pension Risk Transfer Market. Aon plc, 2023. https://www.aon.com/en/insights/articles/seizing-opportunity-in-a-booming-pension-risk-transfer-market.

Bank of England. Financial Stability Report — December 2022. Bank of England, 2022. https://www.bankofengland.co.uk/financial-stability-report/2022/december-2022.

Bank for International Settlements (BIS). Pension Contributions and Tax-Based Incentives: Evidence from the TCJA. BIS Working Papers No. 863, 2021. https://www.bis.org/publ/work863.htm.

Bank for International Settlements (BIS). Pension Funds and the Search for Yield. BIS Papers No. 172, 2024. https://www.bis.org/publ/bppdf/bispap172.pdf.

Bartik, Timothy J. Who Benefits from State and Local Economic Development Policies? W.E. Upjohn Institute for Employment Research, 1991. https://research.upjohn.org/up_press/77/.

Black, Fischer. "The Tax Consequences of Long-Run Pension Policy." Financial Analysts Journal, vol. 36, no. 4, 1980, pp. 21–28.

Council on Foreign Relations. The Disappearing Japanese Bid for Global Bonds. Council on Foreign Relations, 2023. https://www.cfr.org/articles/disappearing-japanese-bid-global-bonds.

Federal Reserve Bank of Chicago. "The 2022 Crisis in the UK LDI Market." Chicago Fed Letter, no. 480, 2023. https://www.chicagofed.org/publications/chicago-fed-letter/2023/480.

Gaertner, Fabio B., Daniel P. Lynch, and Mary E. Vernon. The Effects of the Tax Cuts & Jobs Act of 2017 on Defined Benefit Pension Contributions. Working Paper, 2018. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3185711.

Gagnon, Etienne, Benjamin K. Johannsen, and David López-Salido. Understanding the New Normal: The Role of Demographics. Finance and Economics Discussion Series 2016-080, Board of Governors of the Federal Reserve System, 2016. https://www.federalreserve.gov/econresdata/feds/2016/files/2016080pap.pdf.

Goldsmith-Pinkham, Paul, Isaac Sorkin, and Henry Swift. "Bartik Instruments: What, When, Why, and How." American Economic Review, vol. 110, no. 8, 2020, pp. 2586–2624. https://www.aeaweb.org/articles?id=10.1257/aer.20181047.

Government Pension Investment Fund (GPIF). Adoption of New Policy Asset Mix. GPIF, 2014. https://www.gpif.go.jp/en/performance/pdf/adoption_of_new_policy_asset_mix.pdf.

Holston, Kathryn, Thomas Laubach, and John C. Williams. Measuring the Natural Rate of Interest: International Trends and Determinants. Federal Reserve Bank of San Francisco Working Paper 2016-11, 2016. https://www.frbsf.org/wp-content/uploads/wp2016-11.pdf.

IFRS Foundation. IAS 19 Employee Benefits (2011): Project Summary and Feedback Statement. IFRS Foundation, 2011. https://www.ifrs.org/content/dam/ifrs/about-us/our-history/2011-feedback-19-employee-benefits.pdf.

International Monetary Fund (IMF). Pension Funds and Financial Stability. Global Financial Stability Notes No. 2025/001, IMF, 2025. https://www.imf.org/en/publications/global-financial-stability-notes/issues/2025/03/06/pension-funds-and-financial-stability-562805.

Investment & Pensions Europe (IPE). "Should Pension Funds Increase Their Dollar Hedge Now?" IPE, 2023. https://www.ipe.com/analysis/should-pension-funds-increase-their-dollar-hedge-now/10129755.article.

Konradt, Maximilian. Do Pension Funds Reach for Yield? Evidence from a New Database. Working Paper, 2023. https://mpra.ub.uni-muenchen.de/116209/1/MPRA_paper_116209.pdf.

Lane Clark & Peacock (LCP). New Entrants Push 2023 Buy-in/out Volumes to New Record of Nearly £50bn. LCP, 2024. https://www.lcp.com/en/media-centre/press-releases/new-entrants-push-2023-buy-inout-volumes-to-new-record-of-nearly-50bn.

Macfarlanes. Pension Scheme Arrangements: Clearing Exemption Extended. Macfarlanes LLP, 2019. https://www.macfarlanes.com/what-we-think/102eli5/pension-scheme-arrangements-clearing-exemption-extended-to-2022-102h2j3.

Norton Rose Fulbright. EMIR Clearing Exemption for Pension Scheme Arrangements. Norton Rose Fulbright, 2023. https://www.nortonrosefulbright.com/en/knowledge/publications/477b1168/.

Organisation for Economic Co-operation and Development (OECD). Pension Markets in Focus 2024. OECD Publishing, 2024. https://www.oecd.org/en/publications/pension-markets-in-focus-2024_b11473d3-en.html.

Oster, Emily. "Unobservable Selection and Coefficient Stability: Theory and Evidence." Journal of Business & Economic Statistics, vol. 37, no. 2, 2019, pp. 187–204.

Tepper, Irwin. "Taxation and Corporate Pension Policy." The Journal of Finance, vol. 36, no. 1, 1981, pp. 1–13.

Willis Towers Watson (WTW), Thinking Ahead Institute. Global Pension Assets Study 2026. WTW, 2026. https://www.thinkingaheadinstitute.org/research-papers/global-pension-assets-study-2026/.