{"id":"096856d2-3cef-4441-8214-47117a6594f9","arxiv_id":"1908.10680","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Macroeconomic theories are replaced by tacit theoretical agreements rather than empirical falsification, because observationally equivalent models cannot be distinguished by data.","lead":"This economics essay argues that DSGE models replaced earlier macroeconomic theories through a tacit agreement among researchers, not because data settled the debate. It uses two examples, economic growth convergence and inflation persistence, to show that competing models often make identical predictions, so data cannot choose between them.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The universal observational-equivalence claim in §4.2 is broader than Pesaran's theorem allows; DSGE cross-equation restrictions are testable, so the claim that data cannot settle the controversy is unsupported.","rationale":"The paper's formal examples are correct and clearly presented, and they do establish that observational equivalence and underidentification occur in specific, influential specifications. The reader's weakest-assumption diagnosis is accurate: the paper generalizes Pesaran's theorem to all DSGE models without honoring the theorem's caveat about a priori restrictions on exogenous variables and disturbances. This is load-bearing because the abstract's headline explanation of the DSGE transition, 'not due to a settlement by nature', relies on the strong claim that data cannot discriminate between rational and adaptive expectations. If DSGE cross-equation restrictions are testable and can be rejected, then nature can in principle settle at least some controversies, and the historical claim must stand on citation-based evidence that the authors themselves admit is fragile. My proposed likelihood-ratio test would settle whether the universal equivalence claim fails for a representative estimated DSGE model; the existing literature already contains examples of rejected cross-equation restrictions, which supports the concern. That said, the paper does not collapse: the underidentification examples, the critique of identification opacity, and the historical narrative retain value as a conditional argument, and the reader's CONDITIONAL verdict already captures the need for qualification. I therefore recommend no change to the reader's verdict, while sharpening the specific caveat that the universal observational-equivalence statement in §4.2 should be replaced by a claim restricted to the model class where its conditions actually hold.","tokens_in":21797,"tokens_out":7885,"duration_ms":85752,"concrete_test":"Estimate a standard three-equation New Keynesian DSGE model (for example, Clarida-Galí-Gertler 1999 or a simplified Smets-Wouters 2007) and an unrestricted VAR(2) on the same observables by maximum likelihood, then compute a likelihood-ratio or posterior-odds test of the DSGE cross-equation restrictions. If the restrictions are rejected at conventional significance levels, the claim that every DSGE model has an observationally equivalent adaptive-expectations VAR that data cannot distinguish is false for this important class. A second check is to verify whether Pesaran's (1981) 'no a priori restrictions' condition is satisfied by DSGE models with AR(1) exogenous shocks; if not, the universal statement in §4.2 does not follow from the cited theorem.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 4.2 asserts 'For every linear rational expectations DSGE model, there is an observationally equivalent adaptive expectations VAR model' and concludes 'Data and econometrics cannot and will never be able to settle the controversy.' This overstates Pesaran (1981), whose theorem, as the paper itself quotes, requires 'the absence of a priori restrictions on the processes generating the exogenous variables and the disturbances.' Estimated DSGE models impose exactly such restrictions: exogenous shocks are typically AR(1) with iid innovations, policy rules are specified, and deep parameters enter the reduced-form coefficients nonlinearly. The result is that the DSGE reduced form is a restricted VAR, and an unrestricted adaptive-expectations VAR nests it; the cross-equation restrictions are in principle testable by likelihood-ratio or Bayesian methods. The paper's scalar inflation-persistence examples in §4.3 show non-identification because the structural parameter enters only the innovation variance, but that is a feature of those simplified models, not a general theorem. The paper even concedes in §4.1 that some specifications, such as RBC models, are misspecified relative to hump-shaped impulse responses, which is an example of data discriminating between models. Thus the universal claim that data cannot and never will settle rational-versus-adaptive expectations is not established for the class of DSGE models actually estimated. The central historical assertion that DSGE replaced its predecessors through a tacit agreement rather than a settlement by nature therefore rests on an overstated premise; the authors' own appendix warns that their citation evidence is not a reliable measure of academic impact, so the historical inference lacks independent documentary support.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper argues that the replacement of 1980s macroeconomic theories (monetarism, real business cycles, disequilibrium macroeconomics, and ad hoc rational expectations models) by New-Keynesian DSGE theory was not an empirical victory but the result of a \"tacit agreement\" among leading macroeconomists around 1997. It claims that observational equivalence between rational expectations DSGE models and adaptive expectations VAR models, together with chronic identification failures, prevents data from settling macroeconomic controversies. These claims are illustrated with two examples: cross-country convergence regressions (Section 3.4) and inflation persistence models (Section 4.3). The paper also proposes criteria (microfoundations, parsimony, identification, interpretability) for selecting among observationally equivalent models and predicts that the New-Keynesian DSGE approach will undergo creative destruction.","tokens_in":21975,"tokens_out":5809,"duration_ms":57014,"significance":"If its claims held in their full generality, the paper would be an important contribution to the methodology and historiography of macroeconomics, connecting long-standing identification results (Pesaran 1981; Canova and Sala 2009; Koop et al. 2013) with the recent critique of DSGE modeling by Romer (2016) and Blanchard (2016). The paper's strengths include a clear synthesis of a large literature, explicit worked examples of non-identification and observational equivalence in convergence and inflation persistence, and a welcome emphasis on the gap between normative and positive economics. The claims are not circular: they rely on external published theorems rather than fitting parameters to reach conclusions. However, the paper's central assertion that \"data and econometrics cannot and will never be able to settle the controversy\" goes beyond what the cited theorems support and is internally inconsistent with the paper's own acknowledgment that RBC models are misspecified against hump-shaped impulse responses. The historical \"tacit agreement\" narrative is asserted without direct evidence. These issues materially weaken the force of the argument as currently written.","major_comments":[{"comment":"The paper asserts \"For every linear rational expectations DSGE model, there is an observationally equivalent adaptive expectations VAR model\" and concludes \"Data and econometrics cannot and will never be able to settle the controversy.\" This universal claim is not supported by Pesaran (1981), whose theorem, as quoted by the paper, requires \"the absence of a priori restrictions on the processes generating the exogenous variables and the disturbances.\" Estimated DSGE models such as Smets-Wouters (2007) impose exactly such restrictions: exogenous shocks are AR(1) with iid innovations, policy rules are specified, and deep parameters enter the reduced form nonlinearly, generating cross-equation restrictions that are in principle testable against an unrestricted VAR. The paper's own Section 4.1 concedes that RBC models are misspecified because they fail to match hump-shaped impulse responses, which is an example of data discriminating between models; this is inconsistent with the \"never\" claim. The valid examples in Section 4.3 show non-identification for specific scalar models because the structural parameters appear only in the innovation variance, but that does not generalize to the entire class of estimated DSGE models. Since the paper uses the \"cannot and never will\" claim as a central pillar for its thesis that the DSGE takeover was not due to empirical success, this overstatement is load-bearing and needs to be substantially revised.","section":"§4.2"},{"comment":"The historical claim that New-Keynesian DSGE theory \"is a bargain on two elementary theoretical assumptions and three other core elements between leaders of opinions in theoretical macroeconomics around 1997\" is presented as an established fact, but the only evidence offered is interpretive secondary literature (De Vroey 2016; Duarte 2015). No archival documents, interviews, or meeting records are cited to demonstrate a conscious \"tacit agreement\" or \"bargain.\" The reconstruction may be plausible, but the paper's abstract elevates it to the central alternative to \"settlement by nature.\" The authors should either soften the claim to something like \"can be reconstructed as a tacit agreement\" or provide direct evidence; otherwise the historical thesis is not falsifiable and the argument appears as narrative rather than documented history.","section":"§2.2"}],"minor_comments":[{"comment":"In the paragraph beginning \"We highlight a number of facts,\" \"DGSE\" should be \"DSGE\".","section":"§1"},{"comment":"\"the method of undetermined coefficient\" should be \"the method of undetermined coefficients.\"","section":"§4.2"},{"comment":"\"Inverting and is observational equivalent\" should be \"Inverting and are observationally equivalent.\"","section":"§4.3.1"},{"comment":"The title \"The Chiasm of the New-Keynesian Synthesis\" is unclear; please clarify whether \"chiasm\" is intended as \"schism\" or \"chiasmus.\"","section":"Table 1"},{"comment":"In the text, \"Reinhard and Rogoff 2010\" should be \"Reinhart and Rogoff 2009\" (see also the reference list entry).","section":"References"},{"comment":"The displayed formulas for the convergence parameter in the Solow and Ramsey-Cass-Koopmans models have garbled subscripts and superscripts; please ensure all structural parameters are legibly defined.","section":"Equations (1)-(2)"}],"recommendation":"major_revision","confidential_remarks":"The paper is an essay in the history of economic ideas rather than a technical contribution, so the bar for formal proof is not the appropriate standard. However, the universal observational-equivalence claim in §4.2 is a central argument and it is too strong relative to the cited theorem; the authors will need to either narrow the claim to the class of models for which it holds or explicitly acknowledge the testability of cross-equation restrictions in estimated DSGE models. The historical narrative in §2.2 is engaging but should be framed as interpretation unless direct evidence is provided. With these revisions, the paper could make a useful contribution to the ongoing methodological debate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things. First, this is a readable and unusually honest methodological critique; the formal results are borrowed, but the framing is the authors' own and worth arguing with. Second, the universal observational-equivalence claim in §4.2 is broader than Pesaran's theorem allows, and the paper's own examples do not support it.\n\nThe paper earns credit. It shows clearly why the Solow–Ramsey convergence test cannot identify utility curvature separately from production curvature. It lays out the Lubik–Schorfheide and Beyer–Farmer inflation-persistence examples in a way students can follow. It offers four sensible criteria for choosing among observationally equivalent models. There is no circularity: nothing is estimated or normalized to force the conclusion. The appendix is candid that the citation counts are not reliable measures of impact.\n\nThe soft spot is the leap from 'some DSGE models have non-identified parameters' to 'data cannot and will never settle the controversy.' Pesaran's theorem, as the paper itself quotes, requires the absence of a priori restrictions on exogenous processes and disturbances. Estimated DSGE models impose such restrictions—AR(1) shocks with iid innovations, specified policy rules, nonlinear cross-equation restrictions. That makes the DSGE reduced form a restricted VAR, and an unrestricted adaptive-expectations VAR nests it. The cross-equation restrictions are in principle testable, even if weak identification makes tests fragile. The paper's scalar examples show non-identification when a structural parameter enters only the innovation variance; that is a feature of those examples, not a general theorem. The stress-test note is right on this point.\n\nThe historical claim—that DSGE won by tacit agreement rather than by evidence—is plausible but under-supported. It relies on secondary histories and selected recollections, and the authors admit their citation data cannot carry the weight. Calling it a 'tacit agreement' is a useful rhetorical interpretation, not a documented finding. If the universal equivalence claim were true, the historical story would be less exposed; once that claim is scaled back, the historical inference has to stand on thinner evidence.\n\nThe modest thesis—macro theory has not been cumulative, identification problems are real and under-reported, parsimony should discipline observationally equivalent models—holds up. The immodest thesis does not. The paper is for historians of macro thought and methodologically inclined economists, and it makes for a good reading group. If it crosses my desk, I'd send it out; the referee prompt should focus on §4.2 and on separating the historical narrative from the formal results.","headline":"A provocative, clearly written methodological critique of DSGE macroeconomics whose formal examples are sound but whose universal 'data can never settle it' claim overreaches Pesaran's theorem; worth peer review, not a desk reject.","tokens_in":22623,"tokens_out":4084,"would_cite":false,"duration_ms":41973,"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":"Macro data cannot settle which theory is right: the leading models are observationally equivalent.","keywords":["macroeconomic theory","DSGE models","observational equivalence","parameter identification","economic growth","convergence","inflation persistence","creative destruction"],"falsifier":"Estimate a specific New-Keynesian DSGE model and its adaptive-expectations VAR counterpart on the same data with identifying restrictions derived from the microfoundations, then compare their out-of-sample forecasts and likelihoods. If one model is decisively rejected while the other survives across multiple datasets, or if a preference parameter such as the intertemporal elasticity of substitution is locally identified from macro data without prior restrictions, the general observational-equivalence claim would be broken.","tokens_in":21544,"feed_emoji":"🔄","tokens_out":8680,"duration_ms":86994,"temperature":0.7,"pith_summary":"This paper argues that the eclipse of 1980s macroeconomic theories by dynamic stochastic general equilibrium (DSGE) models was not an empirical victory: data could not settle the dispute, and the paper says no data ever will. The reason is observational equivalence, the fact that a rational-expectations DSGE model and a simpler adaptive-expectations statistical model can be written to produce the same predictions, plus persistent under-identification of structural parameters. Two examples carry the argument: the cross-country convergence regression cannot separate the curvature of utility from the curvature of production, and the New-Keynesian Phillips curve cannot separate intrinsic inflation persistence from serial correlation in shocks. The paper concludes that DSGE dominance is a social and institutional outcome—a negotiated combination of assumptions that were previously treated as incompatible—and that macroeconomics is therefore not yet a cumulative science.","feed_headline":"No data can settle macro theory wars, paper argues","feed_subtitle":"If true, DSGE's dominance is a negotiated bargain between researchers, not an empirical victory.","key_machinery":"The workhorse is observational equivalence: the theorem that, without a priori restrictions on the processes generating exogenous variables and disturbances, a rational-expectations model and an adaptive-expectations distributed-lag model will fit the same data and cannot be told apart. In the paper's examples this equivalence bites at the level of reduced-form parameters—the convergence coefficient is the same weighted combination of four technology parameters (Solow) or six technology-and-preference parameters (optimal growth), and the inflation autocorrelation cannot separate the effect of lagged inflation from the effect of an autocorrelated shock. The mechanism shows why microfoundations do not rescue the DSGE model: they add structural parameters without adding identifying information, making weak identification and non-identification opaque rather than absent.","core_discovery":"The paper's central claim is that the New-Keynesian DSGE model, which replaced monetarism, real business cycle theory, disequilibrium models and non-optimizing rational-expectations models, is a 'bargain' among leading macroeconomists rather than a 'settlement by nature.' It combines five previously separate components: exogenous autocorrelated shocks, staggered price setting, optimal-saving microfoundations, an interest-rate policy rule responding to inflation and output, and a forward-looking solution condition. Because of observational equivalence, any such rational-expectations model has a counterpart adaptive-expectations VAR that cannot be distinguished empirically, and the structural parameters that would give the model its economic meaning are frequently not identified. The paper demonstrates this for growth convergence and inflation persistence, and concludes that the controversy between rational and adaptive expectations, and between rival macro theories more generally, cannot be resolved by additional data.","pith_inferences":["Taken further, the equivalence argument implies that monetary policy conclusions drawn from rational-expectations models can be re-derived from backward-looking models by choosing the right shock processes; so policy advice is underdetermined by data to an even greater degree than the paper's examples show.","A testable extension would be a systematic audit of published DSGE estimates for local identification, using rank conditions and sensitivity of posteriors to priors; if under-identification is as widespread as the paper claims, such an audit would change reporting standards.","The argument suggests a prediction about the next macro consensus: it will emerge through coalition-building and journal networks rather than through a decisive empirical test, and the new models will face the same observational-equivalence problem.","One possible escape route the paper does not explore is that micro data, natural experiments, and panel data with many cross-sectional units may identify parameters that macro time series cannot; if so, the pessimistic conclusion applies specifically to aggregate time-series macroeconomics, not to all empirical economics."],"forward_implications":["Data cannot settle the rational-versus-adaptive expectations debate, because the two families of models can be specified to produce identical observations.","Convergence regressions cannot distinguish exogenous-saving growth models from optimal-saving growth models, since the same estimated convergence speed is consistent with many technology-preference pairs.","New-Keynesian estimates of inflation persistence are not evidence for intrinsic persistence: a hybrid Phillips curve with a serial-correlated shock is observationally equivalent to a simpler two-lag adaptive model.","The DSGE takeover is explained by a negotiated recombination of assumptions and by institutional research incentives, not by superior empirical performance.","If the paper's forecast is right, the New-Keynesian consensus will undergo a similar creative destruction, and the underlying controversies will remain unresolved."],"supporting_citations":[{"why":"Supplies the observational-equivalence theorem: rational expectations and general distributed-lag adaptive models cannot be distinguished without a priori restrictions on exogenous processes and disturbances.","marker":"Pesaran (1981)"},{"why":"Provides the hybrid DSGE solution structure used in Section 4.2 to show the rational-expectations model is observationally equivalent to an adaptive-expectations VAR.","marker":"Koop et al. (2013)"},{"why":"Documents weak and non-identification of structural parameters in estimated DSGE models, supporting the argument that identification opacity protects the theory.","marker":"Canova and Sala (2009)"},{"why":"Gives the convergence-coefficient formulas for both the constant-savings-rate and optimal-growth models, making the growth-theory under-identification example explicit.","marker":"Barro and Sala-I-Martin (2004)"},{"why":"Shows the determinate solution of the forward-looking New-Keynesian Phillips curve is observationally equivalent to the hybrid specification and that its parameters are not identified.","marker":"Lubik and Schorfheide (2006)"},{"why":"Extends the Bayesian analysis of DSGE models and confirms the observational equivalence and non-identification results for inflation persistence.","marker":"An and Schorfheide (2007)"},{"why":"Demonstrates that an indeterminacy (sunspot) solution is observationally equivalent to the determinate solution, so multiple-equilibria models cannot be distinguished by data.","marker":"Beyer and Farmer (2007)"}],"fun_headline_variants":["Macro theory wars: data can't settle the score","DSGE's reign is a pact, not an empirical win","Observational equivalence blocks macro theory verdicts","No data will end the macro theory feud","Macro's non-cumulative science: a negotiated bargain"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the observational-equivalence theorem applies to the DSGE models actually estimated—meaning that no informative restrictions are placed on how exogenous variables and disturbances evolve; if real estimated models do impose such restrictions, data could in principle discriminate between rational and adaptive expectations.","fun_headline_variants_meta":{"raw":{"variants":["Macro theory wars: data can't settle the score","DSGE's reign is a pact, not an empirical win","Observational equivalence blocks macro theory verdicts","No data will end the macro theory feud","Macro's non-cumulative science: a negotiated bargain"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000168,"raw_usage":{"total_tokens":1205,"prompt_tokens":833,"completion_tokens":372,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":449,"completion_tokens_details":{"reasoning_tokens":295}},"tokens_in":449,"tokens_out":372,"duration_ms":3876,"temperature":1.0,"reasoning_tokens":295,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:35:27.649199+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Estimate a specific New-Keynesian DSGE model and its adaptive-expectations VAR counterpart on the same data with identifying restrictions derived from the microfoundations, then compare their out-of-sample forecasts and likelihoods. If one model is decisively rejected while the other survives across multiple datasets, or if a preference parameter such as the intertemporal elasticity of substitution is locally identified from macro data without prior restrictions, the general observational-equivalence claim would be broken.","supporting_citations":[],"review_version":1}