Risk-Free Rate Puzzle¶

The asset-pricing anomaly that a CRRA model calibrated to the observed equity premium predicts a real risk-free rate far above the ~1% seen — because the single parameter γ is overloaded as both risk aversion and the inverse elasticity of intertemporal substitution, so fitting one target misfits the other.

Core Idea¶

The risk-free rate puzzle (Philippe Weil, 1989) is a quantitative anomaly in consumption-based asset pricing: standard representative-agent models with CRRA (constant relative risk aversion) utility, calibrated to match the historically observed US equity premium of roughly 6 percentage points, simultaneously predict a real risk-free interest rate far higher than the historically observed level of roughly 1%. The puzzle is the companion anomaly to the equity premium puzzle of Mehra and Prescott (1985): the Mehra-Prescott finding is that the equity premium is implausibly large relative to what CRRA preferences imply for observed consumption-growth volatility; Weil's finding is that any attempt to resolve the equity premium puzzle by raising the coefficient of relative risk aversion also raises the model's predicted risk-free rate to implausible levels — often 10–20% or higher — while the historical risk-free rate stayed near zero.

The structural source of the tension is a mechanical coupling built into the CRRA preference specification. In a CRRA utility function with parameter γ, the same parameter governs both risk aversion (preference curvature across states of the world — the preference for consumption smoothing across contingencies) and the inverse of the elasticity of intertemporal substitution (preference curvature across time — the preference for consuming now versus later). When γ is raised to fit the large equity premium, it also implies that agents are highly averse to postponing consumption, which means the equilibrium risk-free rate at which agents are just willing to hold bonds must be very high — they require a large real return to be induced to save at all. One parameter is simultaneously overloaded with two empirically distinct roles, and calibrating it to match one target systematically misfits the other.

The puzzle's resolution menu consists of approaches that break this coupling or change the preference framework. Epstein-Zin-Weil recursive preferences (Epstein and Zin 1989; Weil 1989, 1990) introduce separate parameters for risk aversion and the elasticity of intertemporal substitution, allowing a high γ without forcing a high risk-free rate. Habit formation (Campbell and Cochrane 1999) adds a time-varying surplus consumption ratio that makes effective risk aversion vary over the cycle. Rare-disaster models (Rietz 1988; Barro 2006) add low-probability catastrophic consumption declines that agents rationally fear, raising the equity premium without raising the risk-free rate. Long-run risk models (Bansal and Yaron 2004) introduce persistent predictable components of consumption growth. Heterogeneous-agent models (Mankiw 1986) allow idiosyncratic uninsurable income risk that drives a wedge between the market interest rate and a representative-agent proxy. Each resolution succeeds insofar as it can simultaneously match the level of the risk-free rate and the size of the equity premium — the double target that the puzzle makes explicit.

Structural Signature¶

Sig role-phrases:

the representative-agent CRRA model — a consumption-based asset-pricing model with constant-relative-risk-aversion utility, the framework the anomaly is measured against
the overloaded parameter γ — the single CRRA parameter forced to serve as both the risk-aversion coefficient (curvature across states) and the inverse elasticity of intertemporal substitution (curvature across time)
the double empirical target — the historical ~6-point equity premium and the ~1% real risk-free rate level, which a successful model must match jointly
the equity-premium fix — raising γ to fit the large premium, the natural single-knob move
the risk-free-rate side effect — that higher γ makes the agent demand a large return to save at all, pushing the implied risk-free rate to implausible (10–20%) levels
the contrapositive diagnostic — any single-parameter resolution of the equity premium reintroduces the puzzle, so a high implied risk-free rate is the fingerprint of the coupling, not a calibration accident
the resolution menu — the companion family that breaks or supplements the coupling: recursive (Epstein-Zin-Weil) preferences splitting γ, habit formation, rare disasters, long-run risk, heterogeneous agents
the split-versus-supplement taxonomy — the structural question asked of every resolution: does it split the parameter outright or supplement the framework so the premium rises without dragging the risk-free rate up?

What It Is Not¶

Not a real-world mystery about why interest rates are low. The puzzle is a model-failure finding: the CRRA representative-agent model, calibrated to the equity premium, predicts a risk-free rate of 10–20% that the historical ~1% contradicts. The anomaly is the model's misfit against the data, not an unexplained fact about the economy itself.
Not an independent puzzle from the equity premium puzzle. It is the companion anomaly: the natural fix for the equity premium (raise the risk-aversion coefficient γ) is exactly what drives the implied risk-free rate to implausible levels. The two are bound by the same parameter, and that linkage is the whole point — resolving one naively breaks the other.
Not curable by recalibrating γ. The defect is the CRRA coupling — one parameter forced to serve as both risk aversion (curvature across states) and the inverse elasticity of intertemporal substitution (curvature across time) — not a wrong value of γ. Any single-parameter resolution of the equity premium reintroduces the risk-free rate puzzle; a high implied rate is the fingerprint of the coupling, not a calibration accident.
Not a claim that the data are mismeasured. The historical ~1% real risk-free rate is taken as a target the model must hit, jointly with the ~6-point premium. The puzzle's contribution is precisely to impose that double target — matching the spread alone is not success if the level is missed — not to dispute the measurement.
Not the general parameter-overloading or identifiability pattern. The substrate-spanning lesson — one knob cannot satisfy two independent targets that depend on it through different channels, and unintended coupling of distinct quantities in one formal device — is carried by a parameter_overloading / device_overcoupling pattern with identifiability as its nearest home. The macrofinance scaffold (CRRA, the representative agent, the T-bill rate, the resolution menu) that makes this the risk-free rate puzzle does not travel; off macrofinance, name that pattern instead.

Scope of Application¶

The risk-free rate puzzle lives within asset pricing and macrofinance; it operates wherever a consumption-based representative-agent model is calibrated against historical returns, and it is discipline-bound — there is no biological or physical "risk-free rate puzzle," and the generalizable lesson (one parameter cannot serve two channels) belongs to a parameter_overloading / identifiability pattern, not here. It pairs throughout with the equity premium puzzle.

Asset-pricing theory — the home turf and load-bearing motivation: the double target (a ~6-point premium and a ~1% risk-free level jointly) is the empirical goal of Epstein-Zin-Weil recursive preferences, Campbell-Cochrane habit, Rietz-Barro rare disasters, and Bansal-Yaron long-run risk.
Calibration of representative-agent macro models — any quantitative macro model with consumption-based asset pricing must check whether its calibration yields a plausible risk-free rate, with the puzzle defining what "plausible" means.
Behavioral finance — the puzzle is a canonical anomaly offered as evidence the representative-agent expected-utility framework needs supplementation (myopic loss aversion, ambiguity aversion, narrow framing).
Long-horizon discounting and climate policy — the puzzle informs the social-discount-rate debate (Stern versus Nordhaus on climate-policy discounting) and Lucas's welfare-cost-of-business-cycles calculation, where the right risk-free rate is contested.
Sovereign-debt and pension economics — the structurally adjacent social-discount-rate question for discounting future generations' welfare draws directly on this puzzle's literature.

Clarity¶

Naming the risk-free rate puzzle imposes a discipline that the equity premium puzzle alone leaves invisible: a successful asset-pricing model must match not just the spread between equity and bond returns but the level of the risk-free rate as well. Before the puzzle was articulated, the natural reading of Mehra and Prescott was that the equity premium could be rescued simply by cranking up risk aversion; Weil's result makes legible that this "fix" silently breaks a second target, pushing the implied risk-free rate to 10–20% against a historical 1%. The clarifying force is to convert a one-dimensional anomaly into an explicit double target — premium and level jointly — which is precisely what makes the asset-pricing puzzle space hard and what justifies the entire post-1989 enrichment of preference structures. A modeler who has internalized the puzzle no longer asks "can I generate a 6% premium?" but the sharper "can I generate a 6% premium while holding the risk-free rate near 1%?"

Beneath the double target, the puzzle localizes the structural culprit with unusual precision: the single CRRA parameter γ is overloaded, made to serve simultaneously as the coefficient of risk aversion (curvature across states) and as the inverse elasticity of intertemporal substitution (curvature across time). Holding those two roles apart — which the puzzle forces the field to do — is what turns a vague sense that "the model fits badly" into a clean diagnostic: any single-parameter resolution of the equity premium reintroduces the risk-free rate puzzle, because the coupling itself, not the value of γ, is the defect. That diagnostic immediately organizes the otherwise-sprawling resolution menu by what each approach does to the coupling — recursive preferences split the parameter outright, while habit formation, rare disasters, long-run risk, and heterogeneous agents supplement the framework so the premium can rise without dragging the risk-free rate with it. The puzzle thus lets a practitioner read each proposed model as an answer to one structural question: how does it stop one knob from having to turn two dials at once?

Manages Complexity¶

The post-1989 macrofinance literature presents, on its face, a bewildering proliferation: recursive preferences, habit formation, rare disasters, long-run risk, heterogeneous agents with uninsurable income shocks, ambiguity aversion, narrow framing — each a distinct apparatus with its own machinery, its own calibration, its own derivation. Confronting this menu without an organizing principle, an asset-pricing modeler would have to evaluate each proposal on its own terms, re-deriving from scratch what it does and whether it works. The risk-free rate puzzle compresses that sprawl by fixing the defect every model must address to a single mechanical fact: in CRRA utility one parameter, γ, is overloaded, serving simultaneously as risk aversion (curvature across states) and as the inverse elasticity of intertemporal substitution (curvature across time), so any single-knob fit of the equity premium drags the implied risk-free rate the wrong way. The modeler no longer studies each model as a free-standing construct; the modeler asks one question of all of them — how does this proposal stop one knob from having to turn two dials?

That reduction also fixes the success criterion to a single low-dimensional target. The whole field's empirical ambition collapses to two numbers held jointly: a roughly 6-point equity premium and a roughly 1% risk-free rate level, the second of which the equity premium puzzle alone never made visible. A candidate model is then read off against that double target rather than re-litigated case by case: recursive preferences pass by splitting γ outright into separate risk-aversion and intertemporal-substitution parameters; habit formation, rare disasters, long-run risk, and heterogeneous agents pass by supplementing the framework so the premium can rise without the risk-free rate following. The high-dimensional "which of these elaborate preference structures is the right resolution?" problem becomes a low-dimensional diagnostic: identify how each model treats the γ coupling, then check whether it hits premium and level together — two parameters tracked, one structural question asked, the qualitative verdict on each model read off without re-deriving its internals.

Abstract Reasoning¶

The risk-free rate puzzle licenses a focused set of moves in consumption-based asset pricing, all turning on the parameter-overloading diagnosis and the double-target it imposes.

Diagnostic (trace a model's misfit to the overloaded γ, not to a wrong value of γ). The defining move is to attribute the failure of a CRRA representative-agent model not to a poorly chosen risk-aversion coefficient but to a structural coupling: the single parameter γ governs both risk aversion (curvature across states) and the inverse elasticity of intertemporal substitution (curvature across time), so it cannot satisfy two independent targets that depend on it through different channels. The signature inference is contrapositive and unusually sharp — any single-parameter resolution of the equity premium will reintroduce the risk-free rate puzzle, because raising γ to fit the premium necessarily raises the implied risk-free rate (a more impatient agent demands a high return to save at all). The analyst therefore reads a high implied risk-free rate as the fingerprint of the coupling, not as a calibration accident, and diagnoses the defect as lying in the preference specification's overloading of one knob rather than in its setting.

Interventionist (decouple the knob, or supplement the framework, and predict the effect on both targets). Because the defect is localized to the γ coupling, the interventionist content is to ask of any proposed model the single question "how does this stop one knob from turning two dials?" — and to predict its success by what it does to the coupling. Splitting γ outright into separate risk-aversion and intertemporal-substitution parameters (recursive preferences) is predicted to allow a high effective risk aversion without forcing a high risk-free rate. Supplementing the framework rather than splitting the parameter — habit formation making effective risk aversion vary over the cycle, rare disasters adding feared catastrophic consumption declines, long-run risk adding persistent consumption-growth components, heterogeneous agents adding uninsurable idiosyncratic risk — is each predicted to let the equity premium rise without dragging the risk-free rate up with it. Each resolution is read as an intervention on the coupling, and its predicted effect is scored on whether it relieves the misfit at the risk-free-rate target while preserving the premium.

Boundary-drawing (a successful model must hit premium and level jointly). The puzzle's central contribution to reasoning is a boundary the equity premium puzzle alone leaves invisible: a candidate asset-pricing model may not be judged by the return spread it generates alone — it must simultaneously match the level of the risk-free rate. The move "I produced a 6-point equity premium, so the model works" is ruled out of bounds unless the risk-free rate is also held near its historical level; matching one target while missing the other is treated as failure, not partial success. This converts a one-dimensional ambition ("generate the premium") into an explicit double target ("generate the premium while holding the risk-free rate near 1%"), and the boundary is what makes the post-1989 enrichment of preference structures necessary rather than optional — a model that clears only one target falls outside the admissible set.

Comparative reasoning across the resolution menu. Once every proposed model is read as an answer to the same structural question (what it does to the γ coupling) and scored against the same double target (premium and level jointly), the analyst can compare the sprawling menu — recursive preferences, habit, disasters, long-run risk, heterogeneous agents — on common terms rather than re-deriving each from its internals. The puzzle thus licenses ranking and contrasting resolutions by mechanism (split the parameter versus supplement the framework) and by whether each hits both numbers, turning an unstructured proliferation of preference innovations into a comparison along two fixed axes.

Knowledge Transfer¶

Within asset pricing and macrofinance the risk-free rate puzzle transfers as mechanism — more precisely as a load-bearing diagnostic and double-target — across the subfields that share its apparatus. The same diagnosis (one CRRA parameter γ overloaded as both risk aversion and the inverse elasticity of intertemporal substitution, so any single-knob fit of the equity premium drags the implied risk-free rate the wrong way) and the same success criterion (match a ~6-point premium and a ~1% risk-free level jointly) carry intact into asset-pricing theory (it is the empirical target motivating Epstein-Zin-Weil recursive preferences, Campbell-Cochrane habit, Rietz-Barro rare disasters, Bansal-Yaron long-run risk, and heterogeneous-agent models), the calibration of any consumption-based representative-agent macro model, behavioral finance (offered as evidence the expected-utility framework needs supplementation), and the long-horizon discounting debates — the social discount rate, the Stern-versus-Nordhaus climate-policy dispute, the welfare cost of business cycles, sovereign-debt and pension economics — where what "plausible risk-free rate" means is defined by this puzzle. The diagnostics carry with the vocabulary — the representative agent, CRRA, the stochastic discount factor, the risk-aversion/EIS coupling, the double target, the split-versus-supplement taxonomy of resolutions — wherever there is a consumption-based asset-pricing model to calibrate against historical returns. It pairs throughout with its companion, the equity premium puzzle.

Beyond macrofinance the honest reading is the shared-abstract-mechanism case (B); there is no biological or physical "risk-free rate puzzle," and stripped of the macrofinance particulars (CRRA utility, the representative agent, the historical US T-bill rate, the stochastic discount factor) no recognizable pattern remains in the named anomaly. What does generalize is the structural lesson underneath it, at two levels. The looser lesson is that a single parameter cannot in general satisfy two independent empirical targets that depend on it through different channels — an under-determination fact already covered by identifiability and parsimony. The sharper, and more distinctive, lesson is unintended coupling of conceptually distinct quantities within a single formal device: the CRRA case (risk aversion mechanically equal to the inverse of the EIS) is one instance, alongside the conflation of mean and variance in mean-variance optimization under non-elliptical returns, the entanglement of structural shocks in SVAR identification, and the conflation of correlation and causation in an OLS regressor. That pattern — a candidate parameter_overloading / device_overcoupling prime, with identifiability as its nearest existing home — is the thing that genuinely recurs across substrates, and the cross-domain lesson should be carried by it, not by "the risk-free rate puzzle."

The home-bound cargo is the entire macrofinance scaffold: the representative agent, CRRA preferences and the specific γ that ties risk aversion to intertemporal substitution, the consumption-growth time series, the equity-premium and risk-free-rate empirical targets, and the resolution menu (recursive preferences, habit, disasters, long-run risk, heterogeneous agents). None of that survives extraction, because the puzzle is a specific empirical finding that a particular model predicts a particular quantity diverging from a particular measurement — strip any of those particulars and the puzzle dissolves. So invoking "a risk-free rate puzzle" outside consumption-based asset pricing is essentially never apt; the honest move is to name the parameter-overloading pattern (or identifiability) for the structural point. One discipline travels usefully wherever that pattern is recognized: the puzzle's real payoff is the contrapositive diagnostic — a persistent misfit on a second target is the fingerprint of a coupling in the model's formal device, not of a badly chosen parameter value, so the fix is to decouple the device (here, split γ) rather than to recalibrate. That habit — read a stubborn two-target misfit as evidence of overcoupling, and intervene on the device rather than the setting — generalizes to any underdetermined model with one knob doing two jobs. Mechanism within macrofinance (anomaly plus double-target discipline), parameter-overloading / identifiability recurrence beyond — the profile Structural Core vs. Domain Accent makes precise.

Examples¶

Canonical¶

Weil (1989) worked the consumption-Euler pricing of a riskless bond under CRRA utility and lognormal consumption growth, where the real risk-free rate satisfies r_f ≈ −ln β + γ·μ_c − ½·γ²·σ_c². Using Mehra and Prescott's US data — mean annual consumption growth μ_c ≈ 1.8%, volatility σ_c ≈ 3.6% — reproducing the observed ~6-point equity premium already demands a coefficient of relative risk aversion γ of several dozen, far above the γ ≤ 10 usually judged plausible. Weil's finding was that once such a large γ is plugged into the risk-free-rate equation, matching the historical ~1% real rate can be salvaged only by assuming a subjective discount factor β greater than one — agents who value the future strictly more than the present, a negative rate of time preference. One knob, tuned to the premium, forces an economically absurd patience assumption on the rate.

Mapped back: The lognormal-consumption bond-pricing setup is the representative-agent CRRA model; the single γ appearing in both the premium expression and the risk-free-rate equation is the overloaded parameter, serving as risk aversion in one and inverse intertemporal-substitution curvature in the other. Fitting the ~6% premium is the equity-premium fix; the forced β>1 to keep the rate near 1% is the risk-free-rate side effect, and reproducing both numbers at once is the double empirical target.

Applied / In Practice¶

Bansal and Yaron's (2004) long-run risk model is a working deployment of the resolution menu inside quantitative asset pricing. They replace CRRA with Epstein-Zin-Weil recursive preferences — separating the risk-aversion coefficient (calibrated near 10) from the elasticity of intertemporal substitution (set above 1) — and add a small, persistent predictable component to expected consumption growth plus time-varying volatility. With risk aversion and intertemporal substitution now free to move independently, the model generates a sizeable equity premium from the persistent-growth and volatility risks while keeping the implied risk-free rate low and smooth, matching both empirical targets simultaneously. The framework has since become a standard building block in macro-finance and dynamic stochastic general equilibrium models used to price long-horizon risk.

Mapped back: The switch to Epstein-Zin-Weil preferences is the split branch of the split-versus-supplement taxonomy, prying apart the two roles bundled into the overloaded parameter γ; the added long-run-risk and volatility components are the supplement branch working alongside it. That the model is judged by whether it hits a large premium and a low, stable rate together is the double empirical target, and the whole design is the prescribed move of decoupling the device rather than recalibrating a single knob.

Structural Tensions¶

T1: The coupling as defect versus the coupling as parsimony (the same single knob is CRRA's economy and its flaw). The overloading of γ — one parameter serving as both risk aversion and the inverse elasticity of intertemporal substitution — is what the puzzle indicts. But that same single-parameter economy is precisely what made CRRA the canonical, tractable, disciplined workhorse: one number to estimate, a tight cross-equation restriction, no freedom to fit each moment separately. Splitting the parameter (Epstein-Zin-Weil) dissolves the puzzle at the cost of that discipline — a second free parameter buys the fit but weakens the restriction that made the framework falsifiable. The tension is that the coupling is not a sloppy modeling error to be excised at no cost; it is the parsimony that gave the model its bite, and relaxing it trades explanatory economy for empirical reach. A model that can hit any two moments proves less by hitting them. Diagnostic: Does the proposed decoupling earn the extra parameter with independent identification, or does it merely restore fit by surrendering the restriction that made the misfit informative?

T2: Splitting the device versus supplementing the framework (diagnosing the defect or curve-fitting to a target). The resolution menu forks. Recursive preferences split γ, naming the coupling as the defect and prying the two roles apart. Habit formation, rare disasters, long-run risk, and heterogeneous agents supplement the framework, adding structure so the premium can rise without dragging the rate up. Both hit the double target, but they make different epistemic claims: the split says "the formal device was overcoupled," while the supplement says "the environment contains a risk (disasters, persistent growth) the baseline omitted." The tension is that a supplement can be read either as discovering a real feature of the world or as inserting whatever mechanism is needed to match two numbers — and the double target, once rich enough machinery is admitted, cannot by itself distinguish the two. Passing the test is necessary; it is not evidence the added risk is real. Diagnostic: Is the resolution claiming the model's device was miscoupled, or asserting a feature of the world — and is that feature independently observable, or inferred only from the moments it was built to match?

T3: The double target as discipline versus as an overfittable pair (stringency that richer models can game). The puzzle's signature contribution is to insist a model hit premium and level jointly — a discipline that exposes any single-knob fix as breaking the second target. That stringency is real and it is what justified the whole post-1989 enrichment. Yet the very move that rules out cheap fixes invites expensive ones: with enough free preference structure, matching two moments becomes routine, so clearing the double target stops discriminating among models even as it keeps disqualifying naive ones. The tension is that the double target is simultaneously a genuine filter against under-specified models and a low bar for over-specified ones — two moments cannot adjudicate among frameworks flexible enough to fit any two moments. The discipline is asymmetric: sharp against too little structure, blunt against too much. Diagnostic: Is hitting both numbers ruling out a model that could not otherwise fit, or is it a foregone conclusion for a framework with enough free parameters to match any premium-and-level pair?

T4: Fitting the target versus abandoning the framework (what a stubborn anomaly is evidence for). The puzzle takes the ~1% historical rate as a target the model must hit and reads persistent misfit as a call to enrich the preference structure. But the same anomaly licenses the opposite conclusion — that a representative-agent expected-utility model of a monetary economy is simply the wrong object, and behavioral finance reads it exactly so, as evidence for myopic loss aversion, ambiguity aversion, and narrow framing rather than for a cleverer utility function. The tension is that the puzzle does not itself decide whether it is a problem within the paradigm to be solved by better preferences or a symptom of the paradigm to be solved by leaving it. Every added epistle of preference machinery is a bet that enrichment, not abandonment, is the right response — a bet the anomaly cannot settle. Diagnostic: Is the misfit being treated as a specification to be fixed inside the representative-agent framework, or as a verdict against that framework calling for a different class of model?

T5: A named anomaly of its own versus the flip side of the equity premium puzzle (independent finding or bound twin). The risk-free rate puzzle earns its own name because it makes visible a target the equity premium puzzle never surfaces — the level of the rate, not just the spread. Yet it is not independent: it is the companion anomaly, bound to the equity premium puzzle by the very same γ, so the natural fix for one (raise γ) is exactly what breaks the other. The tension is that treating the two as a single "asset-pricing puzzle" loses the distinct level-discipline that is the risk-free rate puzzle's whole contribution, while treating them as fully separate anomalies misses that they are two readings of one overloaded parameter and cannot be resolved in isolation. It is a distinct diagnostic that has no independent existence. Diagnostic: Is the analysis using the level target as a separate constraint the equity premium puzzle omits, or double-counting one coupling as two unrelated failures?

T6: Autonomy versus reduction (a specific macrofinance anomaly or an instance of parameter overloading that travels). The risk-free rate puzzle is a named, precisely dated empirical finding — a particular model (CRRA representative agent) predicting a particular quantity (the real T-bill rate) that diverges from a particular measurement (~1%). Strip those particulars and the named puzzle dissolves; there is no biological or physical "risk-free rate puzzle." What genuinely recurs across substrates is the structure underneath: a single parameter overloaded to serve two conceptually distinct roles through different channels, so fitting one target misfits the other — the same shape as mean/variance conflation in mean-variance optimization or structural-shock entanglement in SVAR identification, best carried by a parameter_overloading / device_overcoupling pattern with identifiability as its home. The tension is between a self-contained anomaly worth studying in situ and the recognition that its portable lesson — read a stubborn second-target misfit as a coupling in the device, and decouple rather than recalibrate — belongs to that more general prime. Diagnostic: Resolve toward the parameter-overloading / identifiability pattern when asking what carries beyond macrofinance; toward the named puzzle when calibrating a consumption-based model against the historical premium and rate in situ.

Structural–Framed Character¶

The risk-free rate puzzle sits at the mixed position on the structural–framed spectrum, with a genuine tension between two of its criteria: an evaluatively neutral structural core (one parameter mechanically cannot satisfy two independent targets) housed inside something that is entirely an artifact of a modeling tradition. On evaluative weight it is structural: the puzzle praises and blames nothing — it is a positive report that a CRRA model calibrated to the equity premium predicts a risk-free rate the data contradict, and the resolution menu is a taxonomy of preference structures, not a normative verdict on anyone. On import-vs-recognize it is also structural within its range: across asset-pricing theory, macro-model calibration, behavioral finance, and the discounting debates the identical diagnosis and double-target are recognized and applied literally, one finding meaning the same thing throughout, and the entry is explicit that the portable lesson underneath (a stubborn second-target misfit fingerprints a coupling in the device) is a substrate-independent structural fact.

What holds it back to mixed rather than mixed-structural are human-practice-bound and institutional origin, both of which lean framed, and reinforced by a vocab-travels that fails. Unlike a mechanism that runs in the world observer-free, the risk-free rate puzzle exists only as a misfit of a posited model against measured data: with no representative-agent CRRA framework there is no γ to overload, no prediction to diverge, and so no puzzle — it is constituted by the practice of consumption-based asset-pricing modeling and dissolves the instant that practice is removed. Its institutional origin is likewise framed: it is a named, precisely dated anomaly of a discipline (Weil 1989, companion to Mehra–Prescott 1985), taxonomic furniture of macrofinance rather than a fact of nature that theory merely names. And its operative vocabulary — CRRA, the representative agent, the stochastic discount factor, the elasticity of intertemporal substitution, the T-bill rate, the split-versus-supplement taxonomy — is irreducibly macrofinancial and renames or dissolves the moment it leaves that substrate.

The portable structural skeleton is parameter overloading / device overcoupling: a single formal knob forced to serve two conceptually distinct roles through different channels, so fitting one target systematically misfits the other. That skeleton is precisely what the puzzle instantiates from its umbrella pattern — the candidate parameter_overloading / device_overcoupling prime with identifiability as its home — and the entry names the co-instances (mean/variance conflation in mean-variance optimization, structural-shock entanglement in SVAR identification, correlation/causation conflation in OLS) that carry that parent as mechanism in their own right, none of them risk-free-rate-puzzle instances. The cross-substrate reach belongs to that parent; the puzzle's distinctive cargo — the CRRA specification, the representative agent, the ~6-point premium and ~1% rate targets, the resolution menu — is macrofinance furniture that neither does nor should travel. Its character: evaluatively neutral and structurally recognized within its field, but constituted by a modeling practice and stated in irreducibly macrofinancial vocabulary, structural only in the parameter-overloading skeleton it borrows from its umbrella — mixed, not a free-floating prime.

Structural Core vs. Domain Accent¶

This section decides why the risk-free rate puzzle is a domain-specific abstraction and not a prime: a portable parameter-overloading skeleton sits at its core, but the macrofinance scaffold that makes it the risk-free rate puzzle is discipline furniture that does not lift.

What is skeletal (could lift toward a cross-domain prime). Strip the macrofinance and one clean structure survives: a single formal knob is forced to serve two conceptually distinct roles through different channels, so tuning it to fit one target systematically misfits the other. One parameter, two independent empirical targets that depend on it through separate channels, and a coupling that makes both unhittable at once. That skeleton is genuinely substrate-portable and recurs as co-instance — mean/variance conflation in mean-variance optimization under non-elliptical returns, structural-shock entanglement in SVAR identification, correlation/causation conflation in an OLS regressor — which is why the puzzle instantiates a candidate parameter_overloading / device_overcoupling prime, with identifiability (and parsimony) as its nearest existing home. But that overcoupling structure is the core it shares, not what makes the risk-free rate puzzle distinctive.

What is domain-bound. Almost all of the concept's working content is macrofinance furniture, and none of it survives extraction: the representative-agent CRRA model; the specific parameter γ that mechanically ties risk aversion to the inverse elasticity of intertemporal substitution; the stochastic discount factor and consumption-growth time series; the double empirical target (the ~6-point equity premium and the ~1% real risk-free rate); and the resolution menu (Epstein-Zin-Weil recursive preferences, Campbell-Cochrane habit, Rietz-Barro rare disasters, Bansal-Yaron long-run risk, heterogeneous agents) with its split-versus-supplement taxonomy. These are the worked apparatus and empirical cases (Weil's β>1 result, the Bansal-Yaron long-run-risk model) of asset pricing. The decisive test: the puzzle is a specific empirical finding that a particular model predicts a particular quantity diverging from a particular measurement — strip any of those particulars (CRRA, the T-bill rate, the historical premium) and the named puzzle dissolves, leaving only the bare overcoupling structure with none of its macrofinance content. There is no biological or physical "risk-free rate puzzle."

Why this does not clear the prime bar. A prime's vocabulary travels and its transfer is recognition of the same mechanism, not analogy. The puzzle's transfer is bimodal. Within asset pricing and macrofinance the diagnosis and the double-target travel intact — the overloaded-γ diagnosis, the joint premium-and-level success criterion, and the split-versus-supplement taxonomy mean the same thing across asset-pricing theory, macro-model calibration, behavioral finance, and the long-horizon discounting debates, because all share the consumption-based representative-agent apparatus. Beyond macrofinance, invoking "a risk-free rate puzzle" is essentially never apt: what recurs is the general overcoupling pattern, not the named anomaly. And when the bare structural lesson — read a stubborn second-target misfit as the fingerprint of a coupling in the device, and decouple rather than recalibrate — is needed cross-domain, it is already carried, in more general form, by the parameter_overloading / device_overcoupling pattern and identifiability. The cross-domain reach belongs to that parent; the puzzle's CRRA specification, representative agent, and premium-and-rate targets are the domain accent that stays home in macrofinance.

Relationships to Other Abstractions¶

Current abstraction Risk-Free Rate Puzzle Domain-specific

Parents (1) — more general patterns this builds on

Risk-Free Rate Puzzle presupposes Equity premium puzzle Domain-specific

The Risk-Free Rate Puzzle arises when raising CRRA risk aversion to repair the Equity Premium Puzzle drives the same model's risk-free-rate prediction implausibly high.

Hierarchy path (1) — routes to 1 parentless root

Risk-Free Rate Puzzle → Equity premium puzzle → Calibration Anomaly

Not to Be Confused With¶

The equity premium puzzle (Mehra–Prescott). The companion anomaly, bound to this one by the same parameter: that the historical equity premium is implausibly large relative to what CRRA preferences imply for observed consumption volatility. The risk-free rate puzzle is not independent — it is what the natural fix for the equity premium (raise γ) does to the level of the risk-free rate, pushing it to 10–20%. Tell: is the anomaly about the return spread being too big for plausible risk aversion (equity premium puzzle), or about the risk-free level going implausibly high once γ is raised to fit that spread (risk-free rate puzzle)?
A real-world mystery about why interest rates are low. The puzzle is a model-failure finding — the CRRA model predicts a high rate the ~1% data contradict — not an unexplained empirical fact about the economy (secular stagnation, safe-asset demand, a liquidity trap). Those are substantive claims about the world; the puzzle is a misfit of a posited model against measured data. Tell: is the question why observed rates are low as an economic fact (a real-world low-rate debate), or why a calibrated CRRA model predicts a rate the low observed one contradicts (the puzzle)?
The resolution menu (Epstein-Zin-Weil, habit, rare disasters, long-run risk). These are the fixes, not the puzzle: preference structures that break or supplement the γ coupling so a model can hit both targets. Epstein-Zin-Weil splits γ into separate risk-aversion and intertemporal-substitution parameters; the others supplement the framework. Conflating a resolution with the anomaly loses the double-target discipline the puzzle imposes. Tell: is it the anomaly a model must resolve (the puzzle), or a preference structure proposed to resolve it (a resolution)?
Identifiability. The broader statistical/econometric concern of whether parameters can be uniquely pinned from the data at all. The risk-free rate puzzle is a specific instance-adjacent case — one parameter forced to serve two channels so it cannot satisfy two targets — and identifiability is its nearest catalog home, but identifiability is the general under-determination concept, not this named macrofinance anomaly. Tell: is the issue whether parameters are recoverable from data in general (identifiability), or the specific CRRA finding that γ cannot hit the premium and the rate at once (the puzzle)?
The parameter-overloading / device-overcoupling umbrella (parent). The substrate-neutral skeleton the puzzle instantiates — a single formal knob forced to serve two conceptually distinct roles through different channels, so fitting one target misfits the other — which recurs as co-instance in mean/variance conflation in mean-variance optimization, structural-shock entanglement in SVAR identification, and correlation/causation conflation in OLS. Tell: off macrofinance, the overcoupling pattern is the umbrella (treated in a later section); "a risk-free rate puzzle" invoked elsewhere is borrowing the name for what is really that parent, stripped of CRRA and the T-bill rate.

Neighborhood in Abstraction Space¶

Risk-Free Rate Puzzle sits in a moderately populated region (45^th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.

Family — Macroeconomic Equilibria & Consumer Demand (19 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-07-12