{"id":"6a22484b-f787-4eb4-bada-61b0a28102db","arxiv_id":"2607.27548","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Aggregate consumption inertia emerges endogenously in standard heterogeneous-agent models when Bayesian households with unobserved-component expectations misattribute equilibrium amplification to bigger shocks, through a 'belief multiplier' that is large when marginal propensities to consume are hig","lead":"This paper argues that standard heterogeneous-agent macro models can reproduce the slow response of consumer spending to shocks when survey-based expectations are used, and it offers a theory of why those expectations behave that way. The theory says households misread the economy's own amplification of shocks as larger shocks, which gradually unmoors their beliefs and delays adjustment.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Inertia is generated by the assumed truncation of the perceived law of motion, not by HA structure; the paper also does not verify its belief-multiplier condition at its own estimated EIS/MPC values.","rationale":"The reader's weakest-assumption identification is correct and is the most defensible reading of the paper: the inertia mechanism hinges on a perceived law of motion that is truncated relative to the actual law. The rational-learning benchmark in Section 5.4 confirms that removing this truncation eliminates inertia in the baseline, so the result is not an implication of heterogeneous-agent structure alone. My stress-test adds a sharper, quantitative version of the same concern: the paper's own estimated EIS/MPC values may place chi below the Proposition 1 threshold for plausible policy coefficients, meaning the theoretical mechanism may not even be operative in the region where the empirical inertia match is obtained. This is not an ad hominem or a demand for a different model; it is a request for the authors to make the link between their estimated model and their theory explicit. The paper is transparent about assumptions, and the semi-structural Section 4 result stands independently, so a conditional verdict remains appropriate. The proposed numerical check directly settles whether the theory applies to the estimated calibration.","tokens_in":38740,"tokens_out":16976,"duration_ms":193237,"concrete_test":"Reproduce the simple-learning equilibrium of Section 5.3 using the estimated parameters from Table 2 (MPC=0.04-0.05, EIS=0.08-0.09, so beta*omega = 1 - MPC), a standard real Taylor coefficient phi = 0.5, and the component persistences, variances, and Kalman gains used or implied in the paper. Report the belief multiplier chi and the Proposition 1 threshold X_e for each component shock, and plot the output impulse response to a transitory eta shock. If chi <= X_e or no hump appears, the central theory must be restricted to higher-MPC calibrations and does not rationalize the baseline estimated model's inertia.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central theoretical result depends on two ingredients: a positive belief multiplier chi = MPC - (1-MPC) sigma phi and a perceived law of motion that is truncated relative to the actual equilibrium law. The second ingredient is doing the load-bearing work. Section 5.4 shows that under rational learning, where the perceived law nests the actual law (Eq. 25), the baseline calibration produces no inertia (Fig. 13). The paper appeals to complexity or memory constraints but does not derive the specific truncation from those microfoundations; it is an assumed cognitive friction. This makes the inertia result a property of the assumption plus HA amplification, not an emergent property of HA structure alone. The quantitative connection to the empirical estimates is also missing. Table 2 reports MPC around 0.04-0.05 and EIS around 0.08-0.09. With chi = MPC - (1-MPC) sigma phi, chi is essentially zero or negative for any real Taylor coefficient phi >= about 0.5, and Proposition 1 requires chi > X_e > 0. The paper neither reports the chi used in Figures 9-11 nor checks that its own estimated region satisfies this inequality. If the simple-learning economy does not generate inertia at those parameters, the theoretical mechanism is not explaining the empirical inertia documented in Section 4; that match comes from directly feeding survey expectations into the model.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper makes two linked contributions. First, it develops a semi-structural method for evaluating consumption-savings models that uses survey expectations of income and interest rates in place of a model of expectation formation. Applying this to Känzig (2021) oil supply news shocks, the author estimates low EIS values (≈0.08–0.09) and quarterly MPCs of about 4–5% in perpetual-youth and standard incomplete-markets models, and argues these heterogeneous-agent (HA) models reproduce the empirically inertial consumption response, unlike a representative-agent benchmark. Second, it proposes a theory of inertia: agents learn output as an unobserved components process but perceive a truncated law of motion that ignores general-equilibrium feedback. The belief multiplier χ = MPC − (1−MPC)·EIS·φ controls the feedback loop; Proposition 1 states that inertia arises and peaks later when χ exceeds a threshold. The paper then derives policy implications, arguing that inertial Taylor rules and delayed fiscal financing become less effective under this learning friction.","tokens_in":39106,"tokens_out":6160,"duration_ms":70215,"significance":"If the results hold, the paper offers a tractable mechanism for endogenous aggregate consumption inertia without habit persistence, a new rationalization of extrapolation bias in survey expectations, and a novel caution about gradual policy regimes. The methodological contribution—using expectations data directly in sequence-space consumption functions—is valuable and carefully stated, and the paper is unusually transparent about its auxiliary assumptions. The theoretical analysis is formal, with propositions and proofs in the appendices, and the empirical section is explicit about identification and estimation. The main weaknesses are that the central theoretical mechanism is conditional on an assumed misspecification of the perceived law of motion, and that the quantitative connection between the estimated parameters and the theory's inertia region is not established. Both are addressable, but they currently limit the strength of the paper's central claims.","major_comments":[{"comment":"The inertia result in Proposition 1 is not an emergent property of heterogeneous-agent structure alone; it is driven by the truncated perceived law of motion in Eq. (23). Rational learning, Eq. (25), removes the misspecification wedge and produces no inertia in the baseline calibration (Fig. 13). The paper appeals to complexity or memory constraints but does not derive the specific truncation from those microfoundations. The abstract's claim that the theory works 'without ... any specific model of expectation underreaction' therefore overstates the theoretical finding. I recommend either deriving the truncation as an optimal constraint or explicitly reframing the contribution as showing that HA amplification magnifies a plausibly misspecified PLM, and discussing how the assumed PLM relates to primitive cognitive frictions.","section":"§5.3–5.4, Eqs. (23)–(25), Fig. 13"},{"comment":"The quantitative link between the empirical estimates and the theoretical mechanism is missing. With χ = 1 − βω − βωσφ = MPC − (1−MPC)σφ, the Table 2 estimates (EIS ≈ 0.08–0.09, MPC ≈ 0.04–0.05) imply χ is near zero or negative for any realistic Taylor coefficient φ (e.g., φ = 1.5 gives χ ≈ −0.075). Proposition 1 requires χ > X_e > 0. The paper never reports the χ used in Figures 9–11, nor computes X_e for its calibration, and does not show that the estimated parameter region actually produces inertia. Without this check, the theory is not shown to explain the empirical inertia documented in Section 4; that match is obtained by directly feeding survey expectations into the consumption function. A quantitative assessment of χ and X_e at the paper's own estimated parameters is needed.","section":"§5.3, Table 2, Fig. 11"},{"comment":"The empirical claim that HA models can match observed consumption inertia rests heavily on assumptions that are strong and only partially tested. Assumption 2c rules out correlation between expectation heterogeneity and idiosyncratic states; Assumption 5 imposes a shape restriction on the extrapolation of missing horizons; Assumption 6 requires forecaster expectations to be unbiased proxies for household expectations conditional on the instrument. The oil-shock choice is justified, but there is no direct evidence on Assumption 2c or on the household–forecaster mapping for income and interest-rate expectations. Please add robustness checks—for example using household-survey income expectations where available, and a wider range of auxiliary extrapolation models than Table 3—and discuss the likely direction of bias if Assumption 6 fails.","section":"§3.2, §4.2, Assumptions 2c, 5, 6"},{"comment":"Because the EIS (and ω) are estimated by minimizing the distance between the model-implied and empirical consumption impulse responses, the close visual match in Fig. 6 is partly mechanical. To substantiate the 'consistency with inertia' claim, the paper should report formal fit statistics (R², RMSE, or a test of overidentifying restrictions) and, ideally, a non-nested comparison with the representative-agent benchmark. The robustness table (Table 3) is useful but does not quantify fit, so the reader cannot judge whether the HA models' apparent success is meaningful beyond their flexibility.","section":"§4.5, Eq. (14), Fig. 6"}],"minor_comments":[{"comment":"The displayed equation beginning 'Y_t ∝ (1−βω−βωσφ)...' is referenced later as 'Equation (22)' but is not numbered in the text. Please number all displayed equations that are referenced.","section":"§5.2"},{"comment":"The timing convention for expectations is confusing: Eq. (21) updates E_t[λ_{t+1}] using information dated t, while Eq. (24) appears to use E_{t−1} in the same role. State the convention explicitly before Proposition 1 to avoid ambiguity.","section":"§5.3"},{"comment":"The acronym FIRE is used before it is defined. Define it at first use.","section":"§2"}],"recommendation":"major_revision","confidential_remarks":"This is an ambitious paper that will likely have impact, but the advertised claim that inertia emerges 'without ... any specific model of expectation underreaction' is too strong given that the theoretical mechanism hinges on an assumed truncated perceived law of motion. The more serious fixable issue is the absence of any numerical check of Proposition 1 at the paper's own estimated EIS and MPC values. I would not reject on these grounds, but the authors should either reframe the theoretical claim or supply the missing quantitative evidence. I have no concerns about attribution or scholarly conduct."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, the semi-structural exercise—plugging Bluechip survey expectations into sequence-space HA consumption functions and estimating EIS/MPC by matching the oil-shock consumption response—is a real step forward. It cleanly shows that HA models with measured expectations can reproduce the sluggish consumption response where the RA model cannot. Second, the theoretical mechanism (belief multiplier χ = MPC − (1−MPC)·σ·φ) is neat, but the inertia result is not an emergent property of HA structure; it comes from an assumed truncation of the perceived law of motion, and the paper never checks whether the mechanism operates at its own estimated parameter values.\n\nThe empirical documentation of extrapolation bias, and the failure of cognitive discounting/sticky expectations to match delayed overreaction, is well done. The policy counterfactuals (gradual monetary rules, delayed fiscal financing) are interesting and worth taking seriously.\n\nWhere it's soft: rational learning (Section 5.4) removes inertia in the baseline (Figure 13), so the friction is load-bearing. That is not by itself fatal—complexity or memory constraints could justify it—but the paper neither derives the specific truncation nor verifies that χ at the estimated EIS (≈0.08) and MPC (≈0.04–0.05) exceeds the Proposition 1 threshold. For any plausible Taylor coefficient φ ≥ 0.5, χ is essentially zero or negative, meaning the theoretical channel would be inactive in the empirically relevant region. The paper should report the χ used in Figures 9–11 and confront it with the estimates. Also, the semi-structural fit is achieved by estimating parameters against the same consumption impulse response used for validation; that's consistency, not prediction. No replication package is provided.\n\nOverall: a serious, transparent paper with a novel combination of methods and a clear, if conditional, theory. The main missing piece is the quantitative link between the estimated region and the theoretical condition. I would send this to a competent referee and ask the author to address the χ gap and share code.","headline":"A serious, methodologically novel paper whose empirical semi-structural exercise is the real contribution; the theory is elegant but rests on an assumed misspecification and is not shown to operate at the paper's own estimated parameter values.","tokens_in":39564,"tokens_out":3190,"would_cite":true,"duration_ms":36772,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91B55","91B51","91B64"],"pacs":[],"model":"deepseek-v4-flash","headline":"Standard heterogeneous-agent models can produce consumption inertia without habit formation, by way of a single belief multiplier that governs how much equilibrium amplification distorts expectations.","keywords":["macroeconomic inertia","heterogeneous agents","consumption","expectations","Bayesian learning","unobserved components","belief multiplier","monetary policy"],"falsifier":"Estimate agents' beliefs about the equilibrium feedback channel—for example, by eliciting households' expected income responses to a shock that is amplified by their own spending. If beliefs already incorporate general-equilibrium amplification (so the perceived and actual laws of motion coincide), the theory predicts no inertia; observing strong inertia among such fully informed agents would refute it. Alternatively, a cross-section of economies with different MPCs should show systematically different inertial peaks in consumption responses to common shocks.","tokens_in":38592,"feed_emoji":"🔄","tokens_out":3672,"duration_ms":38534,"temperature":0.7,"pith_summary":"The paper argues that sluggish aggregate consumption responses—long a puzzle for benchmark models—can arise endogenously in standard heterogeneous-agent economies, without habit formation or imposed underreaction. It first shows, using survey expectations of income and interest rates, that heterogeneous-agent consumption functions reproduce observed inertia, while the representative-agent model does not. The paper then develops a theory: when households learn about unobserved shock components with a perceived law of motion that ignores general-equilibrium feedback, equilibrium amplification is misread as larger shocks, expectations progressively unanchor, and output responds sluggishly. The strength of this feedback loop is governed by a single 'belief multiplier' that is large when marginal propensities to consume are high. If true, gradual monetary and fiscal policy regimes transmit more slowly than standard rational-expectations analysis suggests.","feed_headline":"One belief multiplier generates consumption inertia in standard models","feed_subtitle":"High marginal propensities to consume make expectations unanchor, slowing policy transmission.","key_machinery":"The belief multiplier χ = 1 − βω − βωσϕ, which summarizes the net sensitivity of current consumption to expected future income relative to expected future interest rates. In the perpetual-youth formulation it can be rewritten as MPC − (1 − MPC) · EIS · ϕ. When χ is positive and large, the equilibrium feedback of beliefs into output exceeds the direct decay of the shock; the same χ appears in the Kalman update as a misspecification wedge, converting standard belief updating into a self-reinforcing loop that delays the peak response.","core_discovery":"The central claim is that inertia in aggregate consumption is not a failure of heterogeneous-agent structure but a consequence of it. Proposition 1 formalizes the result: if the belief multiplier χ = 1 − βω − βωσϕ exceeds a threshold, output exhibits inertia and the inertial peak lengthens as χ increases. The multiplier is high when the current marginal propensity to consume is high and the intertemporal substitution elasticity is low—precisely the features that distinguish heterogeneous-agent from representative-agent economies. Inertia emerges because agents' perceived law of motion is truncated relative to the true equilibrium law of motion, so the same equilibrium feedback that amplifies","pith_inferences":["The truncation assumption could be micro-founded by bounded memory or state-space complexity limits; if those frictions vary across agents, the model predicts heterogeneity in inertia by household type.","A direct test: in a laboratory or survey setting, inform agents explicitly that their own spending feeds back into aggregate income; the extrapolation bias documented here should shrink or disappear when they are trained on this feedback.","The same unobserved-components learning structure could be extended to inflation and investment, potentially generating inertial responses in those aggregates without separate frictions."],"forward_implications":["Habits, sticky expectations, or other ad hoc frictions are unnecessary to reconcile heterogeneous-agent models with observed consumption inertia.","Gradual (inertial) Taylor rules and delayed deficit financing are less effective at anchoring expectations, so transmission lags grow.","The framework predicts that inertia strengthens with household MPC and weakens with more responsive monetary policy, up to a destabilization threshold.","The semi-structural method offers a way to evaluate any consumption-savings model against impulse responses without committing to a model of expectations.","Low estimated elasticities of intertemporal substitution (near 0.1) are consistent with the mechanism rather than a puzzle."],"fun_headline_variants":["Belief multiplier generates consumption inertia in HA models","Inertia emerges from heterogeneous-agent expectations, not habits","One belief multiplier explains sluggish consumption responses","Survey expectations make standard HA models show inertia"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The inertia result depends on agents' perceived law of motion being truncated relative to the actual equilibrium law of motion; if households fully internalized how their beliefs feed back into output, the baseline calibration produces no inertia.","fun_headline_variants_meta":{"raw":{"variants":["Belief multiplier generates consumption inertia in HA models","Inertia emerges from heterogeneous-agent expectations, not habits","One belief multiplier explains sluggish consumption responses","Survey expectations make standard HA models show inertia"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000189,"raw_usage":{"total_tokens":1155,"prompt_tokens":711,"completion_tokens":444,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":455,"completion_tokens_details":{"reasoning_tokens":386}},"tokens_in":455,"tokens_out":444,"duration_ms":4718,"temperature":1.0,"reasoning_tokens":386,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T05:52:55.677397+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Estimate agents' beliefs about the equilibrium feedback channel—for example, by eliciting households' expected income responses to a shock that is amplified by their own spending. If beliefs already incorporate general-equilibrium amplification (so the perceived and actual laws of motion coincide), the theory predicts no inertia; observing strong inertia among such fully informed agents would refute it. Alternatively, a cross-section of economies with different MPCs should show systematically different inertial peaks in consumption responses to common shocks.","supporting_citations":[],"review_version":1}