{"id":"49cd9898-d6ba-4910-8636-9063d0ffc19d","arxiv_id":"1908.07561","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A survey showing that the empirical content of major economic choice theories under risk, uncertainty, and intertemporal choice is a strengthened form of downward-sloping demand.","lead":"This paper surveys recent results in revealed preference theory for risk, uncertainty, and intertemporal choice. It shows how the testable implications of expected utility, subjective expected utility, maxmin expected utility, and exponential discounting all reduce to a common negative relation between prices and quantities.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Abstract's 'all theories imply inverse relation' overstates; Theorem 6 as stated is false, so the survey's common-thread claim needs the risk-aversion qualifier stated up front.","rationale":"The reader is right that the faithful restatement of cited theorems is the key assumption. My check found a concrete failure of that assumption in Theorem 6, and the abstract states the common thread without the risk-aversion qualifier that the paper itself provides in Section 4.4. The main theorems 3, 4, 5, and 8 are published results and, as presented, support the inverse-relation narrative for the restricted classes; so the paper's central idea is not wrong. But an ACCEPT verdict should be conditional on fixing Theorem 6 and explicitly qualifying the abstract. This is not an indictment of the survey's central contribution; it is a precise correction.","tokens_in":15936,"tokens_out":34591,"duration_ms":833733,"concrete_test":"Check the original definition in Polisson et al. (2017): if their 'on the data' set X includes all consumption levels that appear in budget constraints (here level 2), then the survey's simplified X={0} union X* is the error and Theorem 6 should be restated with that larger X. If the original also uses only observed chosen levels, then Theorem 6 is false and should be corrected or deleted. Either way, the abstract's unqualified inverse-relation claim should carry the Section 4.4 qualifier 'for risk-averse/convex versions.'","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim is only true for the risk-averse/convex specializations. Section 4.4 explicitly says the earlier tests are joint tests of the theory and risk aversion, so an unqualified abstract overstates the common thread. More concretely, the survey misstates its own Theorem 6. Let S={1,2}, mu=(1/2,1/2), and one observation p=(1,2), x=(0,1). With the survey's X={0} union X*={0,1}, the feasible grid points in the budget x1+2x2<=2 are (0,0), (1,0), and (0,1); setting u(0)=0 and u(1)=1 makes both (1,0) and (0,1) maximize, so the data are UEU-rational on the data. But full UEU-rationality fails: (2,0) is affordable and gives (u(2)+u(0))/2 > (u(1)+u(0))/2 because u is strictly increasing. Thus Theorem 6 is false as written. Since the survey's synthetic 'common thread' argument depends on faithful statements of the cited theorems, this is a genuine soft spot even though Theorems 3-5 and 8 appear to support the inverse-relation reading for the risk-averse versions of EU, SEU, MEU, and exponential discounting.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This survey reviews recent revealed preference characterizations for choice under risk, uncertainty, and intertemporal choice. It presents theorems showing that rationalizability by expected utility, subjective expected utility, maxmin expected utility, and exponential discounting is equivalent to strengthened versions of GARP taking the form of downward-sloping demand conditions on balanced or doubly balanced sequences of price-quantity pairs. It also reviews a computational method for expected utility without risk aversion, a necessary condition for probabilistic sophistication, and extensions to multiple physical goods. The paper's organizing claim is that all these theories imply an inverse relation between prices and quantities, with qualifications.","tokens_in":16266,"tokens_out":7021,"duration_ms":66211,"significance":"The survey is well organized and potentially useful as a synthetic reference. Its main value is showing that several recent characterization theorems share a common price-quantity monotonicity structure, and it is commendably transparent about limitations: the maxmin result is restricted to two states, the main tests are joint tests with risk aversion, and the characterization of probabilistic sophistication is open. The theorem statements are generally attributed clearly, and the paper does not claim proofs. However, the abstract overstates the common-thread claim, and Section 4.4 contains a theorem that is false as stated. These issues reduce the reliability of the survey until corrected, though the underlying risk-averse characterization theorems (Theorems 3, 4, 5, and 8) appear to support the inverse-relation reading.","major_comments":[{"comment":"The statement that a dataset is UEU-rational if and only if it is UEU-rational on the data is false as written. Let S={1,2}, mu=(1/2,1/2), and a single observation p=(1,2), x=(0,1). Then X={0,1}, and the feasible grid points in the budget x1+2x2<=2 are (0,0), (1,0), and (0,1). The strictly increasing function u(0)=0, u(1)=1 makes (0,1) a maximizer on the grid (tied with (1,0)), so the data are UEU-rational on the data. But full UEU-rationality requires a concave, strictly increasing u on R+, and any such u has u(2)>u(1), making (2,0), which is affordable, yield expected utility (u(2)+u(0))/2 > (u(1)+u(0))/2, contradicting optimality of (0,1). The theorem therefore needs additional hypotheses (for example, imposing concavity on the grid utility or a different definition of rationalization on the data); as stated, it is incorrect.","section":"Section 4.4, Theorem 6"},{"comment":"The abstract's unqualified claim that 'The theories all imply an inverse relation between prices and quantities' is too strong. Section 4.4 itself states that the tests presented earlier 'are really joint tests of the hypotheses that an agent is consistent with some particular theory of choice, and that the agent is risk averse.' The inverse-relation theorems (Theorems 2, 3, 4, 5, and 8) all assume concavity or risk aversion in the relevant sense, and the counterexample in my previous comment shows that the relation fails for expected utility without risk aversion. The abstract, and the survey's framing, should qualify the common-thread claim as applying to the risk-averse/convex specializations, or should present a dedicated, correct statement of what is known without risk aversion.","section":"Abstract and Section 4.4"}],"minor_comments":[{"comment":"The Strong Axiom of Revealed Additively Separable Utility refers to the 'risk-neutral downward-sloping property,' but Definition 4 defines only the 'downward-sloping demand property'; please harmonize the terminology and define the risk-neutral version explicitly.","section":"Section 3"},{"comment":"The display defining the Strong Axiom of Revealed Maxmin Expected Utility contains a garbled condition ('|I0,1| + |I1,1| - |I'_{1,1}| = |I'_{0,1}| + |I'_{2,1}| - |I2,1| <= 0'); please restate the intended inequality and verify the notation.","section":"Section 4.3"},{"comment":"In Statement (3), the term lambda^{k'} p^{k'}_{s'} s / pi_{s'} (x^k_s - x^{k'}_{s'}) appears to be a typographical error; the intended expression is presumably lambda^{k'} (p^{k'}_{s'}/pi_{s'}) dot (x^k_s - x^{k'}_{s'}).","section":"Section 6.1, Theorem 9"},{"comment":"There are several typographical errors: 'subjected expected utility' in the abstract should be 'subjective expected utility,' 'Pasandea' in the affiliation should be 'Pasadena,' 'sue to' in the footnote on Ellsberg should be 'due to,' and Section 4.5 uses both 'v(Fx)' and 'V(Fx')' for the same function.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The survey is within the scope of the Annual Review of Economics and, once Theorem 6 is corrected and the abstract is qualified, would be a solid contribution. The false theorem is localized but should not appear in print as stated; the rest of the survey appears to be a faithful and useful synthesis. The self-citations are appropriate given the author's central role in the literature surveyed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe survey is a genuinely useful synthesis. Echenique organizes a large recent literature on revealed-preference tests for expected utility, subjective expected utility, maxmin expected utility, and exponential discounting around a single idea: all these theories, in their risk-averse/convex forms, imply a negative relation between prices and quantities once you balance sequences of pairs appropriately. That framing is new and likely to shape how the area is taught. The main characterization theorems (3, 4, 5, 8) appear faithful to the cited papers, and the survey is admirably transparent about scope: two-state maxmin, joint tests with risk aversion, and no proofs.\n\nThe soft spots are real but localized. The stress-test counterexample against Theorem 6 holds up. With S={1,2}, uniform mu, a single observation p=(1,2), x=(0,1), the finite-grid check passes using u(0)=0, u(1)=1, but the dataset is not UEU-rational because the strictly increasing u would make the affordable bundle (2,0) better. So the claimed equivalence between UEU-rationality and \"UEU-rational on the data\" is false as stated, at least without additional conditions. Since Theorem 6 is presented as a computationally feasible test for the no-risk-aversion case, this is a substantive correction, not a typo. It does not, however, infect the rest of the survey: the risk-averse theorems stand on their own.\n\nAlso, the abstract's \"theories all imply an inverse relation\" is too bare. The inverse relation is qualified in the body—it holds for concave u, i.e., risk-averse preferences—and Section 4.4 explicitly says the earlier tests are joint tests with risk aversion. A reader skimming the abstract could easily carry away a stronger claim than the results support.\n\nWho is this for? Anyone working on nonparametric tests of choice under risk/uncertainty or intertemporal choice, and teachers looking for a unified way to present this literature. It deserves a serious referee: the survey is likely to be widely read and cited, and a referee can catch the Theorem 6 problem and any similar statement-level issues before publication. I'd accept with a request for revision, not reject.\n\nBest,\n[Name]","headline":"A useful and readable survey of revealed-preference tests for risk, uncertainty, and intertemporal choice, organized around a persuasive 'downward-sloping demand' lens, but Theorem 6 is false as stated and the abstract overreaches without the risk-aversion qualifier.","tokens_in":16732,"tokens_out":3532,"would_cite":true,"duration_ms":34670,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91B06","91B08","91B16","91B42"],"pacs":[],"model":"deepseek-v4-flash","headline":"This survey shows that expected utility, subjective expected utility, maxmin expected utility, and exponential discounting are all exactly testable through price–quantity monotonicity conditions built on Afriat's GARP.","keywords":["revealed preference","expected utility","subjective expected utility","maxmin expected utility","exponential discounting","downward-sloping demand","risk and uncertainty","intertemporal choice"],"falsifier":"The common thread would be broken by a single finite dataset that a theory can rationalize but that violates the corresponding Strong Axiom—for example, an expected-utility-rational dataset failing the Strong Axiom of Revealed Objective Expected Utility. Because each theorem claims an equivalence, checking any published budget-choice dataset against the relevant linear inequalities would settle it; such a counterexample is a concrete, computable falsifier.","tokens_in":15783,"feed_emoji":"📉","tokens_out":9000,"duration_ms":79365,"temperature":0.7,"pith_summary":"This survey argues that the testable content of four workhorse economic theories—expected utility under risk, subjective expected utility under uncertainty, maxmin expected utility, and exponentially discounted utility—can be stated in one language: choice data from linear budgets must display an inverse relation between prices and quantities, with each theory adding its own qualification. The paper reviews characterization theorems showing that Afriat's classical test of rational choice (GARP) can be strengthened, by balancing price–quantity comparisons across observations in particular ways, to test each theory exactly. If these characterizations are right, researchers with a finite dataset of budget choices can determine whether observed behavior is consistent with each of these foundational models by checking systems of linear inequalities. The survey matters because these are the theories most used in macroeconomics, finance, and experimental economics, and the common thread promises one unified empirical framework for all of them.","feed_headline":"All four foundational choice theories share one price-quantity law","feed_subtitle":"Expected utility, ambiguity, and time discounting each impose downward-sloping demand in budget-choice data.","key_machinery":"The load-bearing object is a balanced sequence of pairs of observed quantities, $(x_{l_i}^{k_i}, x_{l'_i}^{k'_i})$, in which each observation $k$ appears on the left exactly as often as on the right; a doubly balanced sequence additionally requires each good (state or period) $l$ to appear on the left exactly as often as on the right. Economically, balancedness is what lets the analyst 'mix and match' price and quantity comparisons across different observations without accumulating spurious information. The first-order conditions of concave utility then convert the requirement that larger chosen quantities be associated with cheaper prices into a product-of-price-ratios inequality, the downward-sloping demand property; each theory's functional form determines exactly which pairs must be balanced.","core_discovery":"On the paper's own terms, the central discovery is that expected utility, subjective expected utility, maxmin expected utility, and exponential discounting all have the same revealed-preference skeleton: after the right adjustment, each theory says that when one chosen quantity exceeds another, the corresponding price ratios must multiply to at most one. Expected utility compares risk-neutral prices (price divided by objective probability); subjective expected utility allows subjective probabilities and therefore requires doubly balanced sequences of pairs; maxmin expected utility has a tight characterization only when there are two states, because then only two extreme beliefs matter; exponential discounting requires the time periods on the left of each comparison to be, on net, no earlier than those on the right. Each characterization is a strengthening of GARP, and each is equivalent to satisfying a 'Strong Axiom' built from balanced pairs. The paper presents these as necessary and sufficient conditions for a finite dataset to be rationalized by the respective theory.","pith_inferences":["The inverse price–quantity thread suggests that all four theories admit a unified implementation as linear inequality systems, so a single software routine could report which theories a dataset passes; the survey does not itself make this implementation claim.","The two-state limitation for maxmin may be a real boundary rather than a technical gap: with more than two states the set of beliefs has more than two extreme points, so the argument that one extreme belief explains each choice breaks down; a multi-state characterization may need new conditions or be computationally harder.","The same balanced-sequence machinery could plausibly test other intertemporal and non-expected-utility models, such as rank-dependent utility or quasi-hyperbolic discounting, by choosing the appropriate balancing rule; the survey mentions some of these but does not develop a fully general recipe.","An experiment with three or more states could directly probe whether the two-state maxmin test under-rejects or over-rejects relative to a future multi-state characterization, since current two-state tests cannot distinguish genuine ambiguity aversion from the state-count limitation."],"forward_implications":["Expected utility with risk aversion can be tested exactly on finite budget-choice data by checking the Strong Axiom of Revealed Objective Expected Utility, using risk-neutral price ratios.","Subjective expected utility demands a weaker test than objective expected utility but a stronger one than GARP: every doubly balanced sequence must satisfy downward-sloping demand.","Maxmin expected utility is exactly testable in two-state datasets; outside two states or under other ambiguity models, available characterizations are limited to risk neutrality.","Exponential discounting is testable by counting whether later-period quantities in a balanced sequence are consistently larger only when the corresponding price ratios are lower.","Any dataset that violates GARP fails all four theories, so the strong axioms are refinements of the same rational-choice core."],"supporting_citations":[{"why":"Supplies Afriat's theorem and the GARP equivalence that all later strong axioms refine.","marker":"Afriat (1967)"},{"why":"Provides the GARP terminology and the nonparametric demand-analysis framework used throughout.","marker":"Varian (1982)"},{"why":"Gives the Strong Axiom of Revealed Objective Expected Utility characterization for expected utility under risk.","marker":"Kubler et al. (2014)"},{"why":"Proves the doubly balanced sequence characterization of subjective expected utility and separates it from probabilistic sophistication.","marker":"Echenique & Saito (2015)"},{"why":"Proves the two-state maxmin expected utility characterization and related ambiguity-model results.","marker":"Chambers et al. (2016a)"},{"why":"Gives the Strong Axiom of Revealed Exponentially Discounted Utility and extensions to other discounting models.","marker":"Echenique et al. (2019b)"},{"why":"Provides the finite-data rationalization method that tests expected utility without risk aversion.","marker":"Polisson et al. (2017)"},{"why":"Supplies the cyclic monotonicity characterization for intertemporal choice with multiple physical goods and one observation.","marker":"Browning (1989)"}],"fun_headline_variants":["One price-quantity law unifies four choice theories","Expected utility, ambiguity, time: same revealed-preference skeleton","Revealed preference: risk, uncertainty, and time obey one law","Balanced pairs: the unified test for four theories","A common axiom behind all major choice theories"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The unified common-thread story rests on the correctness of the cited characterization theorems; in particular the maxmin result is proven only for two states and the expected-utility tests assume risk-averse convex preferences.","fun_headline_variants_meta":{"raw":{"variants":["One price-quantity law unifies four choice theories","Expected utility, ambiguity, time: same revealed-preference skeleton","Revealed preference: risk, uncertainty, and time obey one law","Balanced pairs: the unified test for four theories","A common axiom behind all major choice theories"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000826,"raw_usage":{"total_tokens":3550,"prompt_tokens":821,"completion_tokens":2729,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":437,"completion_tokens_details":{"reasoning_tokens":2648}},"tokens_in":437,"tokens_out":2729,"duration_ms":20913,"temperature":1.0,"reasoning_tokens":2648,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:03:48.185400+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"The common thread would be broken by a single finite dataset that a theory can rationalize but that violates the corresponding Strong Axiom—for example, an expected-utility-rational dataset failing the Strong Axiom of Revealed Objective Expected Utility. Because each theorem claims an equivalence, checking any published budget-choice dataset against the relevant linear inequalities would settle it; such a counterexample is a concrete, computable falsifier.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies Afriat's theorem and the GARP equivalence that all later strong axioms refine."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the GARP terminology and the nonparametric demand-analysis framework used throughout."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the Strong Axiom of Revealed Objective Expected Utility characterization for expected utility under risk."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Proves the doubly balanced sequence characterization of subjective expected utility and separates it from probabilistic sophistication."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the finite-data rationalization method that tests expected utility without risk aversion."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the cyclic monotonicity characterization for intertemporal choice with multiple physical goods and one observation."}],"review_version":1}