{"id":"f4ae278d-ee95-4ebf-beaa-1c816e39c11f","arxiv_id":"2412.08657","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":6,"one_line_summary":"The paper argues that U.S. money velocity is partially ergodic and that a fitted log-ergodic model forecasts it with lower RMSE and MAE than a constant-velocity baseline.","lead":"This paper proposes modeling the velocity of money with log-ergodic stochastic processes, claiming that such models show stable long-term behavior and predictable mean reversion despite short-term fluctuations. It calibrates the model on U.S. GDP and money supply data from 1959 to 2008 and reports lower out-of-sample RMSE and MAE than a constant-velocity Quantity Theory benchmark.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 4.1 is unsupported: the EMO is not defined well enough to prove linearity or mean-ergodicity, and Eq. 4.5 makes the transformed process collapse to zero as T grows, so partial ergodicity is either unproved or vacuous.","rationale":"I agree with the reader that the unproved mean-ergodicity assumption is the load-bearing weakness. My stress-test adds a concrete check that the theorem may be vacuously true: the explicit expression for Z_vδ converges to zero for β>3/2, so the EMO does not reveal hidden ergodicity in the velocity series; it annihilates the process. This is an internal problem, not a disagreement with a consensus view, and it hits the central claim directly. I also note the empirical comparison in Table 2 does not support 'superior predictive power' as stated: the log-ergodic model has lower R-squared and higher SSE in-sample than QTM, and the out-of-sample RMSE/MAE improvements come with no uncertainty quantification. However, the theorem is the primary issue. The paper does provide a reproducible skeleton (MATLAB code and data sources), and the Monte Carlo framing is a legitimate way to illustrate the construction, but the code has placeholder parameters and missing data files, so it does not rescue the theorem. Because the central claim fails at Theorem 4.1, the reject verdict stands; no verdict change is needed.","tokens_in":16178,"tokens_out":6513,"duration_ms":58967,"concrete_test":"Re-derive Theorem 4.1 using the actual definitions of Dδ and Rδ from arXiv:2211.15637: first check whether ξβ is linear, i.e., whether ξ[YX−YM]=ξ[YX]−ξ[YM] follows from Definition 3.1; then compute the covariance integral in Definition 3.2 for Z_vδ with β=1.6 and T→∞. If the integral vanishes only because the terms in Eq. 4.5 are scaled by 1/T^β, the claimed partial ergodicity is a scaling artifact; if linearity fails, Theorem 4.1 is invalid. Either outcome settles whether the central claim holds.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proof of Theorem 4.1 is the only bridge from the geometric Brownian assumptions to the paper's central claim that money velocity is partially ergodic and long-term predictable. That proof does not go through from the manuscript alone. Definition 3.1 defines the EMO with symbols Dδ and Rδ that are never defined, and the key step 'Since the processes ~YX(δ) and ~YM(δ) are mean-ergodic' is imported from [Firouzi and Mamaghani(2023)] rather than proved. The proof also assumes EMO linearity, ξ[YX−YM]=ξ[YX]−ξ[YM], without derivation. Even if all those imports were granted, mean-ergodicity is not generally closed under subtraction: Definition 3.2 controls individual autocovariances, not the cross-covariance of the difference, so Z_vδ need not be mean-ergodic. More concretely, the displayed calculation gives Z_vδ = [(μX−μM)+((σM^2−σX^2)/2)] δ W_T/T^β + (σX−σM) Wδ/T^β. For β>3/2 and δ≤T, both terms converge to 0 almost surely as T→∞ because W_T=O(√T) and Wδ=O(√δ). Thus the process that is supposed to exhibit hidden mean-ergodicity is asymptotically degenerate; Definition 3.2 is satisfied only because the operator crushes the signal to zero. That makes the theorem vacuous rather than a statement about stable, predictable velocity. With Theorem 4.1 removed, the empirical 'superior predictive power' interpretation has no theoretical foundation.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a log-ergodic stochastic model for the velocity of money, defined as V(t)=X(t)/M(t) with X(t) nominal GDP and M(t) money supply. The model applies an 'ergodic maker operator' (EMO) to the logarithms of geometric Brownian motions for GDP and money supply, and claims that the resulting process is partially ergodic, mean-reverting, and topologically mixing. The authors calibrate drift and volatility parameters to U.S. data (1959-2024), run Monte Carlo simulations, and compare the log-ergodic model with the Quantity Theory of Money, claiming superior predictive power. The manuscript includes theoretical definitions, a theorem, a proposition, an empirical section, and a MATLAB appendix.","tokens_in":16558,"tokens_out":4183,"duration_ms":37160,"significance":"If the central claim were rigorously established, a log-ergodic framework for monetary velocity could offer a useful addition to stochastic monetary economics and policy modeling. The paper has some strengths: it uses publicly available data, provides simulation code, and explicitly acknowledges limitations such as vulnerability to structural breaks and the need for more complex jump-diffusion models. However, the theoretical foundation is not self-contained and the key theorem is not proved from the given definitions. The empirical comparison does not support the headline claim of superior predictive power. These are load-bearing issues, not presentational ones, so the manuscript cannot be accepted in its current form.","major_comments":[{"comment":"The ergodic maker operator is not well-defined: the quantities Dδ and Rδ appearing in Eq. (3.1) are never defined anywhere in the manuscript, and the operator is therefore not a function on the stated processes. The proof of Theorem 4.1 also assumes without derivation that ξβ is linear, ξβ[YX−YM]=ξβ[YX]−ξβ[YM]. Until these objects and properties are specified, the central derivation cannot be checked.","section":"Definition 3.1, Eq. (3.1)"},{"comment":"The key step \"Since the processes Y~X(δ) and Y~M(δ) are mean-ergodic\" is imported from [Firouzi and Mamaghani(2023)] and is not proved here. Even if that import is allowed, mean-ergodicity is not generally closed under subtraction: Definition 3.2 controls the autocovariance of individual processes, not the cross-covariance of the difference, so the conclusion that Zvδ is mean-ergodic does not follow.","section":"Theorem 4.1 and Section 4.1.1"},{"comment":"For β>3/2 and δ≤T, both terms in Eq. (4.5) converge to zero almost surely as T→∞ (W_T=O(√T) and W_δ=O(√δ)). The process Zvδ is therefore asymptotically degenerate; it satisfies Definition 3.2 only because the operator annihilates the signal. The theorem's claim of stable, predictable long-term velocity is vacuous under this construction.","section":"Eq. (4.5)"},{"comment":"The topological mixing proof is invalid: δ is an interval length, not a time-evolution parameter, and Zvδ is a family of random variables indexed by δ rather than a dynamical system on an event space Ω. No topology or measure on Ω is specified, and the Poincaré recurrence argument cannot be applied to this process. The claimed mixing property is therefore unsupported.","section":"Theorem 4.3"},{"comment":"The empirical claim of \"superior predictive power\" is not supported by the reported statistics: the log-ergodic model has lower R² (0.1576 vs 0.4001) and larger SSE (0.2531 vs 0.045) than QTM on the training sample, and the validation comparison reports only point estimates without uncertainty or significance tests. Additionally, the Monte Carlo prediction is generated from parameters fitted to the same data used to define the model, so it cannot independently validate the theoretical claim.","section":"Section 5.3, Table 2"}],"minor_comments":[{"comment":"The MATLAB code is not reproducible as written: variables such as \"No. of DATA\", \"Type Beta\", and \"Type here\" are placeholders, and the estimated parameters are computed from raw GDP levels while the diffusion uses log-differences, creating a mismatch between calibration and simulation.","section":"Appendix"},{"comment":"The formulas contain corrupted symbols (e.g., \"\\inte∅ral\" in Eq. (3.2)) and should be typeset correctly before any further review.","section":"Definition 3.2, Eq. (3.2)"},{"comment":"The sentence \"To ensure mean-ergodicity of the original processes in 4.3\" is misleading: the EMO is applied to the log processes, not to the original GDP and money supply processes.","section":"Section 4.1.1"},{"comment":"The paper claims both \"stable and predictable\" long-term behavior and, via Theorem 4.3, topological mixing with sensitivity to initial conditions; these implications conflict and should be reconciled.","section":"Sections 4.1.1 and 5.2"},{"comment":"The figures do not report units or axis definitions for the \"returns\" transformation, and the comparison plots do not include confidence bands, so the visual agreement between simulated and actual velocity cannot be assessed quantitatively.","section":"Figures 1-3"}],"recommendation":"reject","confidential_remarks":"The manuscript relies heavily on the authors' own prior arXiv preprint for both the definition of the EMO and the key mean-ergodicity property, so the current paper is not self-contained. The central theorem's asymptotic degeneracy and the invalid topological mixing proof are not fixable by minor edits; they undermine the main claim. The empirical comparison also does not establish superiority. A substantial theoretical rebuild would be needed before this could be considered for publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The new thing here is modest but real: applying the authors' own log-ergodic framework to money velocity, with a Monte Carlo exercise on U.S. data. That application is a reasonable idea, and the paper is honest enough to state its limitations in the discussion. But the central theoretical bridge, Theorem 4.1, does not hold up from the manuscript alone, and the empirical section does not rescue it.\n\nThe definition of the ergodic maker operator (Definition 3.1) uses D_δ and R_δ that are never defined. The proof of Theorem 4.1 then imports mean-ergodicity of the transformed processes from the authors' prior paper, asserts linearity of the EMO without derivation, and does not address whether mean-ergodicity is preserved under subtraction. Those are real gaps. The stress-test note adds a sharper point: Eq. (4.5) shows Z_v^δ decays to zero as T grows for β > 3/2 and δ ≤ T. If that is correct, the process is not exhibiting hidden ergodicity; it is asymptotically degenerate. That makes the theorem vacuous rather than a statement about stable, predictable velocity. I checked the displayed equations and the derivation of (4.5) appears to follow from their SDEs, so this concern lands.\n\nThe empirical comparison is also weaker than the abstract claims. Table 2 shows the log-ergodic model has lower in-sample R-squared than the QTM baseline (0.1576 vs 0.4001), though better out-of-sample RMSE. The chosen β is not reported for the main results, and the appendix code contains placeholder parameters and missing data files. So the 'superior predictive power' claim is not established by the evidence shown.\n\nWhat the paper does well: it identifies a real macroeconomic question, acknowledges the limitations of constant-velocity models, and the literature review is reasonably broad. The application to velocity is a legitimate direction, and the authors are transparent about relying on their prior framework. That transparency counts for something.\n\nBottom line: the paper deserves a serious referee because the application is relevant and the flaws are substantive but not obviously fatal to the broader research program. A referee could push the authors to either provide a self-contained proof or soften the claim to a conjecture plus empirical illustration. As it stands, the central theorem is not proven, and the empirical section needs better reporting. I would not cite it for the theorem, but I might cite it as an example of a plausible application if the framework gets fixed in revision.","headline":"A useful application idea undermined by an unsupported central theorem: the math does not close, and the empirical claim overreaches.","tokens_in":17147,"tokens_out":627,"would_cite":false,"duration_ms":97667,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["37A30","37H05","60G10","91B70","91G15"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims the stochastic velocity of money—nominal GDP divided by money supply—is partially ergodic after a log transform, making its long-run path stable and predictable despite short-run noise.","keywords":["log-ergodic process","velocity of money","partial ergodicity","mean-ergodicity","ergodic maker operator","geometric Brownian motion","monetary policy","stochastic processes"],"falsifier":"Compute the sample time average of $\\log V(t)$ over a long U.S. sample and compare its convergence to the ensemble average predicted by the calibrated $Z_\\delta^v$ process; if the time-average variance does not decay at the rate $T^{-2\\beta}$ implied by equation (4.5), or if the residual autocorrelation over 2008-2024 remains significant, the partial-ergodicity claim would fail. A direct check of Definition 3.2, evaluating whether $\\lim_{T\\to\\infty}\\frac1T\\int_0^T(1-\\tau/T)\\operatorname{Cov}_{yy}(\\tau)\\,d\\tau=0$ for the estimated parameters, would also settle the claim.","tokens_in":15894,"feed_emoji":"📈","tokens_out":6858,"duration_ms":54147,"temperature":0.7,"pith_summary":"The paper tries to establish that the velocity of money, defined as the ratio of nominal GDP to money supply, is not a static constant but a stochastic process whose logarithm becomes mean-ergodic after a particular transformation. If true, the long-run level of velocity is stable and predictable even while short-run movements look erratic. The authors build a stochastic model in which GDP and money supply follow geometric Brownian motions, apply their ergodic maker operator to the logarithms, and show that the transformed difference is mean-ergodic, mean-reverting, and topologically mixing. They then calibrate the model to U.S. data from 1959 to 2008 and report that it forecasts velocity from 2008 to 2024 with lower error than the quantity theory of money. The practical payoff claimed is a forecasting tool central banks can use to anticipate inflation and design monetary policy.","feed_headline":"Log-ergodic model makes money velocity's long-term path predictable","feed_subtitle":"On U.S. data from 2008-2024, the model out-forecasts the quantity theory of money, offering central banks a new tool.","key_machinery":"The load-bearing object is the ergodic maker operator (EMO), $\\xi_{\\delta,W_\\delta}^{\\beta}$, an operator that acts on a log-transformed stochastic process and is designed to expose mean-ergodic behavior hidden in the original process; the paper imports its definition and properties from earlier work. The argument works because the EMO is linear: applying it to $\\log(V(t))=Y_X(t)-Y_M(t)$ gives $Z_\\delta^v=\\tilde Y_X(\\delta)-\\tilde Y_M(\\delta)$, and if both transformed components are mean-ergodic, their difference is mean-ergodic. The explicit formula for the transformed velocity process is $Z_\\delta^v = [(\\mu_X-\\mu_M)+\\tfrac12(\\sigma_M^2-\\sigma_X^2)]\\,\\delta W_T/T^\\beta + (\\sigma_X-\\sigma_M)\\,W_\\delta/T^\\beta$, where $\\beta$ is the inhibition degree parameter. This formula is what links the estimated drift and volatility parameters of GDP and money supply to the claimed stability of velocity.","core_discovery":"The paper's central claim is Theorem 4.1: the stochastic velocity of money process $V(t)=X(t)/M(t)$ is partially ergodic. Partial ergodicity here means that after taking logarithms, $\\log V(t)=Y_X(t)-Y_M(t)$, and applying the ergodic maker operator $\\xi_{\\delta,W_\\delta}^{\\beta}$, the resulting process $Z_\\delta^v=\\tilde Y_X(\\delta)-\\tilde Y_M(\\delta)$ is mean-ergodic: its time average converges to its expected value, so long-run behavior is stable and predictable. The same transformed process is claimed to be mean-reverting and topologically mixing. Empirically, the paper reports that Monte Carlo simulations calibrated to U.S. GDP and money supply data from 1959 to 2008 track the actual velocity of money over 2008-2024 with lower root-mean-square error and mean absolute error than the quantity-theory benchmark, supporting the conclusion that the log-ergodic model has stronger predictive power for long-term monetary velocity.","pith_inferences":["The derivation suggests the result is largely structural: any two log-normal processes sharing the same time-scaling $T^{-\\beta}$ will produce a partially ergodic difference, so the same framework would apply to other ratios of positive economic variables, such as price-dividend ratios or output-per-capita measures.","A natural testable extension is to replace geometric Brownian motion with jump-diffusion processes, which the authors cite as already proved partially ergodic; one could check whether the jump terms degrade or improve the out-of-sample forecasts during financial crises.","The paper's own limitation statement acknowledges vulnerability to structural breaks; an applied extension would be to re-estimate the model on subsamples around known regime changes and see whether the mean-reversion parameter itself shifts.","Cross-country replication would clarify whether partial ergodicity of velocity is a property of the U.S. monetary regime or a general property of the ratio process."],"forward_implications":["Velocity of money has a stable long-run mean that can be estimated from historical data, so forecasts over horizons like five years are possible even when quarter-to-quarter movements are noisy.","Central banks can use the calibrated $Z_\\delta^v$ process as an indicator of the intensity of contractionary or expansionary policy, and can anticipate inflation trends from predicted velocity.","Because the transformed process is mean-reverting, policy interventions have temporary effects and velocity tends to return to its long-run average, which matters for interest-rate and quantitative-easing decisions.","The topological mixing property implies that the system is highly sensitive to initial conditions, so policymakers should expect complex interactions and prefer adaptive frameworks over rigid rules.","On U.S. data, the log-ergodic model beats the quantity theory of money on out-of-sample RMSE and MAE for 2008-2024, so the approach is a practical alternative for forecasting monetary velocity."],"supporting_citations":[{"why":"Supplies the definition of the ergodic maker operator, the log-ergodic process concept, and the asserted mean-ergodicity and mean-reversion results that Theorem 4.1 imports.","marker":"[Firouzi and Mamaghani(2023)]"},{"why":"Provides the ergodic theory foundations, Poincaré recurrence, and topological mixing background used in Theorem 4.3.","marker":"[Viana and Oliveira(2016)]"},{"why":"Motivates the relevance of ergodicity to economics by contrasting time averages with ensemble averages in financial systems.","marker":"[Peters(2019)]"},{"why":"Supplies the quantity theory of money model used as the comparison baseline for predictive accuracy.","marker":"[Jung(2024)]"},{"why":"Supports the claim that ergodic properties matter for dynamic economic models and gives conditions for ergodicity in such settings.","marker":"[Kamihigashi and Stachurski(2016)]"},{"why":"Provides the mean ergodic theorem that underlies the log-ergodic definition's convergence statement.","marker":"[Coudène(2016)]"}],"fun_headline_variants":["Log-ergodic model beats quantity theory in money velocity forecast","Log-ergodic model out-forecasts classic money velocity theory","Money velocity long-term path predictable via log-ergodic model","Central banks gain from log-ergodic money velocity forecasts"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof takes as given, from the authors' earlier paper, that applying the ergodic maker operator to the logarithms of GDP and money supply makes both transformed processes mean-ergodic; the operator's own symbols $D_\\delta$ and $R_\\delta$ are never defined here, so the mechanism that produces this mean-ergodicity is not visible in this paper.","fun_headline_variants_meta":{"raw":{"variants":["Log-ergodic model beats quantity theory in money velocity forecast","Log-ergodic model out-forecasts classic money velocity theory","Money velocity long-term path predictable via log-ergodic model","Central banks gain from log-ergodic money velocity forecasts"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000908,"raw_usage":{"total_tokens":3867,"prompt_tokens":870,"completion_tokens":2997,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":486,"completion_tokens_details":{"reasoning_tokens":2926}},"tokens_in":486,"tokens_out":2997,"duration_ms":19338,"temperature":1.0,"reasoning_tokens":2926,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T10:59:32.750229+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the sample time average of $\\log V(t)$ over a long U.S. sample and compare its convergence to the ensemble average predicted by the calibrated $Z_\\delta^v$ process; if the time-average variance does not decay at the rate $T^{-2\\beta}$ implied by equation (4.5), or if the residual autocorrelation over 2008-2024 remains significant, the partial-ergodicity claim would fail. A direct check of Definition 3.2, evaluating whether $\\lim_{T\\to\\infty}\\frac1T\\int_0^T(1-\\tau/T)\\operatorname{Cov}_{yy}(\\tau)\\,d\\tau=0$ for the estimated parameters, would also settle the claim.","supporting_citations":[{"cited_title":"Log-ergodicity: A New Concept for Modeling Financial Markets","cited_arxiv_id":"2211.15637","evidence_quote":"Supplies the definition of the ergodic maker operator, the log-ergodic process concept, and the asserted mean-ergodicity and mean-reversion results that Theorem 4.1 imports."},{"cited_title":"Seeking ergodicity in dynamic economies","cited_arxiv_id":null,"evidence_quote":"Supports the claim that ergodic properties matter for dynamic economic models and gives conditions for ergodicity in such settings."}],"review_version":1}