{"id":"987a529f-7a6d-48d9-818b-34b54489de76","arxiv_id":"2508.20571","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A leading macroeconomist argues that rational expectations about equilibrium prices in heterogeneous-agent models are implausibly complex and should be replaced by simpler, empirically disciplined price forecasts.","lead":"This essay argues that rational expectations about prices force households in heterogeneous-agent models to forecast entire wealth distributions, a task too complex to be plausible. It proposes three tests that any replacement theory of price expectations must pass, and sketches learning-based alternatives that might pass them.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The paper's core step—'hard for modelers, therefore unrealistic for agents'—is never defended against the 'as if' reply, and footnote 7 concedes no replacement exists; this leaves the central claim conditional at best.","rationale":"The reader's weakest_assumption already identifies both load-bearing gaps: the inference from computational intractability to psychological implausibility, and the admitted absence of a concrete P satisfying Criteria 1–3. My stress-test agrees with that diagnosis and sharpens it. The mathematical content of the paper—Eq. (10), the non-Markovianity of prices in Section 3.5, and the taxonomy of equilibrium concepts—is sound and useful. The contested step is interpretive: the paper moves from 'modelers cannot easily solve the Master equation' to 'real agents cannot behave as if they did.' That inference is exactly what the Lucasian 'as if' tradition denies, and the essay does not refute it; it merely asserts that the distribution-forecasting implication 'seems self-evident' that real individuals do not do it. Likewise, the positive agenda is honest about its speculative status: footnote 7 and the conclusion frame the paper as posing a challenge rather than supplying a solution. For those reasons, I would not change the reader's CONDITIONAL verdict: the critique is worth taking seriously, but the central normative claim—'should be replaced'—is not fully established. A concrete computational experiment comparing RE solutions with price-only learning would be one way to reduce the uncertainty; if a simple price-forecasting heuristic achieves near-RE outcomes in a canonical heterogeneous-agent economy, the force of the paper's central objection would be substantially weakened, whereas if it fails, the objection would be strengthened.","tokens_in":44599,"tokens_out":6032,"duration_ms":66653,"concrete_test":"Take a calibrated Krusell-Smith economy with aggregate shocks and a nonlinear 'crisis' region (as in Section 3.4) and replace the Master-equation solution with a temporary equilibrium in which households forecast only the risk-free rate, using a simple recursively estimated AR(1) perceived law of motion for the price, with no distributional moments in the information set. Compare: (i) wall-clock time versus a Master-equation solver; (ii) welfare differences and time-series properties of prices relative to the RE solution; (iii) response of beliefs and prices to a policy change. If the price-only learner closely replicates RE dynamics with small welfare losses, the 'as if' defense survives and the complexity-based attack loses force.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that rational expectations about equilibrium prices is 'unrealistic and should be replaced' requires two premises: (A) that the distribution-forecasting in Eq. (10) is behaviorally implausible in a way that defeats the Lucasian 'as if' defense, and (B) that some alternative belief process P(p'|·) satisfies Criteria 1–3. Neither premise is established. For (A), Section 2.2 asks: 'if even our most advanced computational tools struggle with the Monster equation, how can we justify the assumption that real-world households and firms solve the associated decision problems?' This is a non-sequitur under the standard 'as if' interpretation: rational expectations is an equilibrium consistency condition, not a claim about agents' internal algorithms. The paper never engages the possibility that decentralized mechanisms—learning, selection, or simple heuristics—produce RE-consistent behavior without any agent literally solving a Master equation. Computational difficulty for a centralized modeler is not direct evidence about the behavioral or evolutionary feasibility of approximately rational outcomes. For (B), footnote 7 concedes 'I only know the problem, not the solution!' and Section 4's candidates are explicitly speculative: temporary equilibrium with measured expectations may fail Criterion 3b for policy counterfactuals, least-squares learning can be slow or non-convergent, and the author conjectures that RL beliefs may be 'close to' rational, which would undercut Criterion 2. Thus even granting the mathematical derivation in Section 2.2 and the non-Markov argument in Section 3.5, the normative conclusion 'should be replaced' is not supported by a demonstrated feasible alternative. The paper is best read as a research agenda, not an established verdict.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This essay argues that rational expectations about equilibrium prices in heterogeneous-agent macro models are unrealistic and should be replaced. The argument is that, under rational expectations, households and firms must forecast equilibrium prices by forecasting the entire cross-sectional distribution of idiosyncratic states, which makes the distribution a state variable in individual Bellman equations. The paper formalizes this in the two-period example of Eq. (10), shows in Section 3.5 that equilibrium prices are generically non-Markov even in a finite-state setting, and reviews why this leads to an extreme curse of dimensionality in the resulting 'Master equation.' The constructive part proposes that agents instead forecast prices directly using subjective beliefs P(p'|·), subject to three criteria: computational tractability, consistency with empirical evidence, and endogeneity of beliefs to model reality (Lucas-critique immunity). Sections 4.1-4.5 discuss temporary equilibrium and internal rationality, survey-expectation approaches, least-squares learning, reinforcement learning, and heuristics as candidate replacements. The paper explicitly states in footnote 7 that it proposes criteria rather than a concrete alternative, and describes several of its own candidates as speculative.","tokens_in":44850,"tokens_out":4196,"duration_ms":42994,"significance":"If the central thesis were established, the paper would justify a reorientation of the heterogeneous-agent research program away from solving Master equations and toward models with disciplined subjective price beliefs. The paper has real strengths: Eq. (10) correctly places the future distribution G' inside the expectation, Section 3.5 gives a clean finite-state proof that prices are not Markov, and the taxonomy of equilibrium concepts in Appendix B (REE, SCE, RPE, CEE, IREE, TE) is useful and well-referenced. The paper is also commendably honest about the limits of its constructive proposal. However, the significance is conditional: the paper does not establish that any alternative belief process satisfies the three criteria simultaneously, and it does not engage the standard 'as if' defense of rational expectations. As a programmatic challenge the essay is valuable; as a demonstration that rational expectations 'should be replaced' it remains incomplete.","major_comments":[{"comment":"The paper's central inference from computational intractability for modelers to behavioral implausibility for agents is not defended against the standard Lucasian 'as if' reply. The text asks: 'If even our most advanced computational tools struggle with the Monster equation, how can we justify the assumption that real-world households and firms solve the associated decision problems?' But rational expectations is an equilibrium consistency condition, not a claim about the algorithms agents use internally. The manuscript does not discuss whether decentralized mechanisms—selection, evolutionary arguments, or simple learning heuristics—could produce approximately RE-consistent behavior without any agent literally solving a Master equation. Indeed, Section 4 later appeals to such mechanisms, but the connection to this objection is never made. As written, the computational difficulty supports a pragmatic research-priority claim, not the unconditional thesis of the abstract that rational expectations 'is unrealistic and should be replaced.' The authors should either weaken the thesis to a conditional challenge or directly address the as-if defense in the context of the full-distribution state.","section":"Section 2.2, 'The unrealism of rational expectations'"},{"comment":"The constructive half of the thesis requires the existence of a subjective belief process P(p'|·) satisfying Criteria 1, 2, and 3 simultaneously. Footnote 7 concedes 'I only know the problem, not the solution!', and Section 4 explicitly acknowledges that temporary equilibrium with measured expectations may fail Criterion 3b for policy counterfactuals (Section 4.2), that least-squares learning can be slow or non-convergent (Section 4.3), and that RL beliefs are conjectured to be 'close to' rational, which would make Criterion 2 hard to satisfy (Section 4.4). These admissions are honest, but they mean the paper's central claim rests on an unproved existence conjecture. The paper should be reframed as an explicitly conditional challenge, or it should provide at least one worked example of a belief process satisfying all three criteria in a nontrivial heterogeneous-agent model.","section":"Footnote 7 and Sections 3.3-4.5"},{"comment":"The claim that tracking prices rather than the distribution converts the agents' problem into a POMDP does not by itself deliver Criterion 1. In a POMDP, the optimal policy is a function of the posterior belief about the hidden Markov state—which is again an N-dimensional or infinite-dimensional object. Moreover, the price observation is generated by the same high-dimensional latent state. The paper does not explain how the POMDP formulation avoids the curse of dimensionality unless agents are assumed to use drastically simplified, low-dimensional belief-updating rules. This gap is load-bearing because 'computational tractability' is the first criterion and the main promised payoff of Section 3.1. The paper should either specify the class of approximate belief-updating rules that make the POMDP tractable or state explicitly that tractability is only conjectural under the POMDP interpretation.","section":"Section 3.5, last paragraph"}],"minor_comments":[{"comment":"Typo: 'evidendence' should be 'evidence' in the paragraph beginning 'The second criterion is...'.","section":"Section 3.3, Criterion 2"},{"comment":"Typo: 'macroeonomies' should be 'economies' in the sentence 'Heterogeneous-agent economies (and indeed real-world macroeonomies)...'.","section":"Section 4.4"},{"comment":"If the manuscript is published in its current form, permission to reproduce Figure 1 from Sutton and Barto (2018) must be secured; the figure appears to be a direct reproduction from a copyrighted textbook.","section":"Section 4.4, Figure 1"},{"comment":"The conclusion states that developing alternatives 'holds two main promises' and that realism and simplicity 'could be a rare case in which the two go hand in hand.' This is fine as a conjecture, but the conditional phrasing could be carried into the abstract, which currently states the thesis unconditionally.","section":"Section 5"}],"recommendation":"major_revision","confidential_remarks":"This is a well-written programmatic essay that will likely generate useful discussion. My main concern is the mismatch between the abstract's categorical claim ('should be replaced') and the body's explicit admission that no replacement is known. If the author reframes the paper as a challenge with explicit conditional claims and addresses the POMDP dimensionality point in Section 3.5, I would be comfortable supporting publication. The paper's scope seems consistent with a general-interest economics venue, though it is more of an agenda-setting piece than a conventional research article."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, don't read this for new mathematics; the formal pieces—Eq. (10) and the non-Markov result in Section 3.5—are standard and the paper says so, citing Cardaliaguet et al., Krusell and Smith, and Den Haan. Second, do read it if you care about where HANK is going. Moll has done the field a service by putting the 'distribution in the expectation' problem in plain language and refusing to let the Master equation be buried in methods appendices.\n\nThe historical section is genuinely good: it reminds us that rational expectations was sold as a way to make price expectations operational, and it shows how the same promise fails once the state is an infinite-dimensional distribution. The three criteria—tractability, empirical discipline, Lucas-critique immunity—are a sensible checklist, and the survey of temporary equilibrium, survey expectations, least-squares learning, and reinforcement learning is a useful map of the available exits. No data are fitted and no theorem is new, but none are promised.\n\nThe soft spot is the one the reader flagged. The paper's central leap is from 'we cannot solve this equation' to 'therefore real agents do not solve it.' That leap is a behavioral claim, not a mathematical one, and the standard 'as if' defense—rational expectations as an equilibrium consistency condition, not a description of mental arithmetic—is never engaged. Computational difficulty for a centralized modeler is not direct evidence about decentralized mechanisms like learning, selection, or heuristics that might produce RE-consistent behavior. The author knows this, which is why footnote 7 says 'I only know the problem, not the solution,' but that concession sits uneasily with the abstract's 'should be replaced.' The constructive part is a research agenda, not a demonstrated alternative; Criterion 3b is where any candidate will live or die, and the paper's own discussion concedes that survey-based beliefs risk failing it and RL beliefs may be 'close to' rational.\n\nBottom line: this is a provocative essay from someone who knows the literature, with clean mathematics where it bothers to show any, an honest limitation statement, and a bibliography that is contextual rather than self-promotional. It deserves to be engaged by a serious referee—not desk-rejected—but the referee should ask Moll to either soften the normative claim or deliver a candidate belief process that actually satisfies all three criteria. I'd assign it to a macro field journal and ask for a revision that separates the critique (strong) from the agenda (still open).","headline":"A clearly-written and honest research agenda that restates a standard point about the Master equation, but its central 'hard for us, therefore unrealistic for agents' inference remains undefended against the as-if reply.","tokens_in":45482,"tokens_out":3360,"would_cite":true,"duration_ms":33112,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This essay argues that rational expectations about equilibrium prices should be dropped from heterogeneous-agent macroeconomics because they force agents to forecast entire cross-sectional distributions.","keywords":["rational expectations","heterogeneous agents","Master equation","curse of dimensionality","price expectations","temporary equilibrium","least-squares learning","reinforcement learning"],"falsifier":"Survey or laboratory evidence that ordinary households' forecasts of interest rates and wages are as accurate as the objective forecasts computed from the full cross-sectional distribution of wealth—rather than from a simple function of current prices—would undercut the paper's central implausibility claim. A more direct test: in an experimental economy with a known distribution process, if subjects' price forecasts match the rational-expectations benchmark better than a price-only autoregressive rule, the claim that agents do not forecast via distributions is falsified.","tokens_in":44332,"feed_emoji":"📈","tokens_out":5639,"duration_ms":50136,"temperature":0.7,"pith_summary":"This essay argues that rational expectations about equilibrium prices—interest rates, wages, and similar price-like variables—should be abandoned in heterogeneous-agent macroeconomics because the assumption is conceptually implausible and computationally prohibitive. Under rational expectations, a household forecasting tomorrow's interest rate must first forecast the entire cross-sectional distribution of income and wealth, since equilibrium prices depend on that distribution. The resulting Bellman equation, called the Master equation or 'Monster equation,' has an infinite-dimensional state and is extremely hard to solve even with frontier methods. The essay proposes replacing it with models in which agents forecast prices directly using subjective beliefs, and offers three criteria for disciplining those beliefs: computational tractability, consistency with empirical evidence, and endogeneity of beliefs to model reality. A sympathetic reader would care because this reorientation could let heterogeneous-agent models address aggregate non-linearities such as financial crises.","feed_headline":"Rational expectations forces agents to forecast whole economies","feed_subtitle":"A new essay says the 'Monster equation' is unrealistic and proposes criteria for replacing the assumption.","key_machinery":"The load-bearing object is the Master equation, a Bellman equation for the value function $V(x, G, z)$ whose state includes the infinite-dimensional cross-sectional distribution $G$ of idiosyncratic states; the paper nicknames it the 'Monster equation.' It is what shows that rational expectations about prices force agents to forecast distributions: because equilibrium prices are functions of $G$ and $G$ evolves via a Chapman-Kolmogorov equation, prices alone are not Markov, and the rational expectation in equation (10) must integrate over $G'$. The proposed alternative machinery is a subjective price-belief distribution $P(p'|\\cdot)$, which lets agents solve lower-dimensional Bellman equations in prices and idiosyncratic states, sidestepping the curse of dimensionality.","core_discovery":"The paper's central claim is that in any generic heterogeneous-agent model, rational expectations about equilibrium prices imply that decision makers solve a dynamic program in which the cross-sectional distribution $G$ is a state variable, because prices satisfy $p_t = \\mathcal{P}^*(G_t, z_t)$ and $G$ is the Markov state while prices are not. Equation (10) makes the mechanism explicit: the expectation in the household's Bellman equation is taken over future distributions $G'$, so forecasting prices requires forecasting the entire distribution. The author argues this is not just computationally hard for modelers but psychologically implausible for real households and firms, who do not directly care about the distribution at all. The constructive proposal is to replace the rational expectation with a subjective probability distribution over future prices, $P(p'|\\cdot)$, and to discipline that distribution by three criteria. If the paper is right, the field should shift from solving Master equations to computing temporary equilibria with disciplined subjective price beliefs.","pith_inferences":["My inference: the argument cuts against Krusell-Smith style moment forecasting as well, since forecasting moments of the distribution to forecast prices is conceptually similar to forecasting the distribution itself; the paper hints at this but stops short of a full rejection.","My inference: if tractability is taken seriously as a criterion, it rules out any behavioral model that is defined as a 'twist' on rational expectations, such as diagnostic expectations or cognitive discounting, because those still require agents to compute the rational-expectations benchmark.","My inference: a testable extension would be to elicit household interest-rate and wage forecasts under different hypothetical aggregate states and compare them with the objective forecasts implied by a solved Master equation; the paper's claim predicts large, systematic gaps."],"forward_implications":["If the argument is correct, the heterogeneous-agent research program should redirect effort from solving Master equations toward models with disciplined subjective price beliefs.","Such models can be computed as temporary equilibria, which are only modestly harder than stationary equilibria, because there is no fixed point between beliefs and actual prices at the level of the individual decision problem.","Candidate belief models must satisfy three criteria: tractability, consistency with survey and experimental evidence on expectations, and approximate consistency with model-generated price dynamics to retain some immunity to the Lucas critique.","Departing from rational expectations may allow heterogeneous-agent models to study aggregate non-linearities—financial crises, boom-bust cycles—that are currently out of reach.","Least-squares learning and reinforcement learning are both stochastic approximation algorithms, so convergence results from that theory can be used to analyze learning about equilibrium prices."],"supporting_citations":[{"why":"Establishes the heterogeneous-agent model with aggregate risk and the original observation that the cross-sectional distribution enters agents' forecasting problem.","marker":"Krusell and Smith (1998)"},{"why":"Early companion work noting the same curse of dimensionality in heterogeneous-agent models with aggregate uncertainty.","marker":"Den Haan (1996)"},{"why":"Supplies the name and formal theory of the Master equation from mean field games.","marker":"Cardaliaguet et al. (2019)"},{"why":"Shows how rational expectations about equilibrium prices are built into recursive competitive equilibrium, the target of the critique.","marker":"Lucas and Prescott (1971)"},{"why":"Introduces rational expectations as a hypothesis about price expectations, the assumption the essay argues should be replaced.","marker":"Muth (1961)"},{"why":"Defines internal rationality, the framework the essay uses for temporary equilibrium with subjective beliefs.","marker":"Adam and Marcet (2011)"},{"why":"Articulates the 'wilderness of non-rational expectations' problem that motivates the three criteria.","marker":"Sargent (2008)"}],"fun_headline_variants":["Why rational expectations fail in heterogeneous-agent models","The 'Monster equation' that makes rational expectations implausible","Replacing rational expectations: a challenge for macroeconomics","Macro's rational-expectations flaw: agents can't solve the Monster equation","When forecasting prices means forecasting whole economies"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument depends on treating the computational difficulty of the Master equation for professional economists as evidence that real households and firms cannot behave as if they solved it, which is exactly the inference the 'as if' defense of rational expectations denies.","fun_headline_variants_meta":{"raw":{"variants":["Why rational expectations fail in heterogeneous-agent models","The 'Monster equation' that makes rational expectations implausible","Replacing rational expectations: a challenge for macroeconomics","Macro's rational-expectations flaw: agents can't solve the Monster equation","When forecasting prices means forecasting whole economies"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000221,"raw_usage":{"total_tokens":1443,"prompt_tokens":930,"completion_tokens":513,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":546,"completion_tokens_details":{"reasoning_tokens":433}},"tokens_in":546,"tokens_out":513,"duration_ms":4675,"temperature":1.0,"reasoning_tokens":433,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T16:43:33.245000+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Survey or laboratory evidence that ordinary households' forecasts of interest rates and wages are as accurate as the objective forecasts computed from the full cross-sectional distribution of wealth—rather than from a simple function of current prices—would undercut the paper's central implausibility claim. A more direct test: in an experimental economy with a known distribution process, if subjects' price forecasts match the rational-expectations benchmark better than a price-only autoregressive rule, the claim that agents do not forecast via distributions is falsified.","supporting_citations":[{"cited_title":"Heterogeneity, Aggregate Uncertainty, and the Short-Term Interest Rate,","cited_arxiv_id":null,"evidence_quote":"Early companion work noting the same curse of dimensionality in heterogeneous-agent models with aggregate uncertainty."},{"cited_title":"Rational Expectations and the Theory of Price Movements,","cited_arxiv_id":null,"evidence_quote":"Introduces rational expectations as a hypothesis about price expectations, the assumption the essay argues should be replaced."}],"review_version":2}