Daily Summary for 2026-11-05
daily
Transcript
Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.
Ruth: It's the fifth of November, twenty twenty-six, and this is the day's research.
Warren: 30 new papers came out today.
Ruth: I'm Ruth.
Warren: And I'm Warren, here with Iris.
Ruth: We'll take the day in one pass, then pull out the papers we're staying with.
Warren: Start us off with Partially identified heteroskedastic SVARs.
The day's research: Ruth: Generative AI at Work studies how using generative AI assistants helps customer support agents increase their productivity and improve worker skills.
Warren: Access to this assistance generally boosts productivity by about fifteen percent on average, though it varies by experience level.
Iris: Evidence shows AI helps workers learn new skills and improves English fluency, particularly for international agents.
Ruth: The biggest gains happen when dealing with problems that are relatively rare, where human agents often lack prior training.
Warren: This assistance also makes customer interactions better; customers become more polite and less likely to demand a manager.
Iris: This effect builds upon findings from Generative AI at Work regarding support agents across different skill levels.
Ruth: Fixed-smoothing Uniform Inference for Quantile Regression provides a more robust framework for analyzing heterogeneity in predictive effects.
Warren: That paper handles heteroskedasticity well when performing quantile regression inference in time series data.
Iris: The behavioral modeling suggests new regulatory targets are needed beyond simple volatility limits in automated markets.
Ruth: Modeling rational arbitrageurs, manipulators, and retail herd-followers found that rational agents exhibit more stable risk-adjusted performance.
Warren: This is crucial because it supports targeting behaviors rather than applying uniform limits on volatility or turnover.
Iris: Agentic reinforcement learning systems use distinct reward functions for each archetype in this behavioral modeling.
Ruth: The rational arbitrageur maximizes risk-adjusted return with penalties on realized volatility and turnover.
Warren: Conversely, the manipulator receives a bonus proportional to turnover and operates under a short discount horizon favoring rapid cycling.
Iris: Retail herd-followers are rewarded for aligning exposure with recent momentum in this framework.
Ruth: Pilot results on hourly cryptocurrency data showed rational agents have the highest baseline volatility reflecting continual exposure adjustments.
Warren: Manipulators exhibit sharp spikes during stress periods, contrasting with higher volatility not automatically implying more stress.
Iris: This work connects to causal inference by proposing logit-based IV estimators as alternatives to two-stage least squares in binary treatment settings.
Ruth: These estimators offer a causal interpretation because their probability limits can be represented as convex combinations of treatment effects with non-negative weights.
Warren: This contrasts with the traditional 2SLS estimator which may assign negative weights.
Iris: GenAI-assisted writing suggests this technology lowers publication barriers for researchers, potentially increasing output by about seventy-two percent.
Ruth: This productivity gain is largest among early career researchers and those from linguistically distant countries.
Warren: Today's papers include Logit-based alternatives to two-stage least squares proposing new estimators for binary treatment variables.
Iris: The research on GenAI-assisted writing shows using generative AI to write papers can increase publication output among social science researchers.
Ruth: Agentic AI in Automated Markets models different types of trading agents, such as arbitrageurs and manipulators, to understand their impact on market stress and volatility.
Warren: This study models these agents to understand their impact on market stress and volatility.
Iris: What Drives Energy Use? Prices, Policy, and Inefficiency uses an economic model to explain high U.S. energy use despite consumption declines.
Ruth: That paper uses an economic model to explain why U.S. energy use remains high despite national declines in consumption.
Warren: The pursuit of happiness analyzes how individuals maximize happiness through social interactions and revise attitudes based on others' happiness.
Iris: This paper analyzes how individuals maximize happiness through social interactions and how they revise their attitudes based on others' happiness.
Ruth: We look at AI and the research assistant evidence from economics working papers comparing those that disclose AI use to those that do not.
Warren: This study compares economic papers disclosing AI use to those without, finding operational AI is more common in junior labor projects.
Iris: Information Intermediaries in Monopolistic Screening investigates how information intermediaries shape consumer choices and affect market efficiency.
Ruth: That paper investigates how information intermediaries shape consumer choices and ultimately affect market efficiency and welfare.
Lucky paper: 2403.06879: Ruth: Alright everyone, let's talk about Partially Identified Heteroskedastic SVARs. This paper is dealing with a big problem in structural vector autoregressive models when the variance of the shocks changes over time and you can't tell them apart statistically.
Warren: It addresses that specific failure where point identification breaks down, and they propose a new strategy to keep identification going under empirically relevant conditions.
Iris: The core contribution seems to be their novel strategy that uses both heteroskedasticity assumptions along with zero and sign restrictions on model parameters or functions.
Ruth: They show this combined approach enables point identification even when shifts in shock variances are statistically indistinguishable from one another, which is where traditional methods usually fail.
Warren: That's interesting because it moves past just looking at the variances separately and looks at them together through these specific restrictions, like f ij equals m i minus j.
Iris: They also extend their topological analysis of identified sets from standard SVAR literature to this heteroskedastic setting, yielding results for both point identification and set identification with convex identified sets.
Ruth: And then they build a robust Bayesian estimation approach on top of that, adapting it from set-identified SVARs to handle these violations caused by heterogeneous variance shifts.
Warren: So they present three primary estimators for estimation: generalized least squares, maximum likelihood, and a derived Bayesian estimator in Appendix E.
Iris: The Bayesian approach uses likelihood functions transformed for Gibbs sampling and employs inverse Wishart priors when the covariance matrices are unknown to characterize the posterior distributions.
Ruth: That sounds like a solid framework for inference when you can't rely on simple assumptions about those covariance matrices being fixed.
Warren: It’s important because it gives a concrete algorithm for estimation and inference specifically on the identified set in this complex HSVAR setup.
Iris: From my side, I wonder how this practical algorithm performs when the underlying data generating process really deviates from their assumptions about those variance shifts.
Ruth: That's a fair question; we have to consider if these analytic derivations hold up when you look at real-world time series data.
Warren: The paper focuses heavily on the analytical conditions, establishing the necessary and sufficient conditions for point identification based on eigenvalue multiplicities and those non-redundant zero restrictions.
Iris: That theorem, Theorem one is key because it rigorously establishes exactly what those conditions are for point identification in this setting.
Ruth: It really shows how you can use those specific non-redundant zero restrictions to solve the problem of shifting variances being indistinguishable.
Warren: Building on that idea, they extend the topological analysis to handle convex identified sets too, which is a step beyond what's been done in standard SVAR literature.
Iris: So they’re not just solving point identification; they are also providing results for set identification when dealing with those convex identified sets.
Ruth: And the Bayesian estimation part seems designed to be robust specifically because it handles the violations caused by those heterogeneous variance shifts.
Warren: It's a comprehensive approach, moving from novel identification strategies to topological extensions and finally a robust Bayesian estimation algorithm.
Iris: I think the real power here is that they provide tools for analysis when you can’t trust the simpler assumptions about how those shock variances behave across time.
Lucky paper: 2105.10965: Ruth: Alright, we're moving into our second paper of five today, which is titled Inference on Quantile Treatment Effects with Small Treatment Groups from Marina Dias and Demian Pouzo at Amazon and UC Berkeley.
Warren: So the core idea here is developing a method for inference on conditional quantile treatment effects that stays asymptotically valid even when the treatment groups are small as the overall sample size gets bigger.
Iris: That sounds complicated because it combines estimating a conditional quantile function from a large reference group with an exact conditional Bernoulli representation of quantile indicators to control asymptotic size.
Ruth: Exactly, and the key assumption is a quantile analogue of selection on observables: conditional on past potential outcomes and observed characteristics, treatment selection preserves the quantile of interest of the unobserved rank.
Warren: That restriction needs to accommodate settings like randomized experiments or observational models satisfying conditional independence given past outcomes and observed characteristics.
Iris: The testing procedure involves three steps: first estimating the structural conditional quantile function from the large reference group, second evaluating that estimate in sample moments of those treatment-profile groups under the null hypothesis, and third comparing an aggregate of these moments with a critical value from their conditional Bernoulli distribution.
Ruth: The oracle statistic has an exact conditional distribution, which establishes the asymptotic validity of this feasible test.
Warren: They also prove asymptotic size control and derive an explicit lower bound on power, which allows treatment-group sizes to remain bounded.
Iris: The feasibility of the test relies on Assumption four which requires sequences K of functions such that P Tb n,K(n) minus T 0,K(n) is greater than epsilon n as n goes to infinity.
Ruth: That condition controls the estimation error introduced by replacing the unknown reference-profile quantile function with mu b n, K zero.
Warren: The Monte Carlo simulations illustrate how these factors affect the ability to detect treatment effects, showing that holding expected treatment-group size fixed, the feasible test approaches its oracle behavior as the reference group grows.
Iris: They also confirm that its null rejection probability is largely insensitive to the shape and tails of the disturbance distribution.
Ruth: The empirical application used temperature shocks and GDP growth to illustrate inference on effects associated with distinct treatment histories.
Warren: Conventional significance assessments can overstate the strength of evidence for economically important effects when the number of observations informative about those effects is small, which is a big caveat there.
Iris: The procedure does not reject the absence of an effect, indicating that statistical evidence for these estimated growth losses is sensitive to how inference accounts for that small treatment group.
Ruth: So, even with this method, if the reference group itself is small, the procedure becomes conservative because it relies on that approximation.
Warren: That's a fair point; the power is constrained by the information and discreteness inherent in extremely sparse groups.
Iris: The paper also shows that holding expected treatment-group size fixed, its power approaches oracle behavior as the reference group grows, but this still requires a sufficiently accurate reference-profile estimator.
Ruth: We can look at how they construct confidence regions for finite-dimensional parameters indexing the structural quantile treatment effects.
Warren: The coverage result shows that the infimum of n as it goes to infinity of the probability that theta zero is in the confidence region n alpha is greater than or equal to one minus alpha.
Iris: That corollary can be used to compute confidence intervals for finite-dimensional parameters indexing those structural quantile treatment effects.
Ruth: The oracle conditional distribution being binomial means this procedure doesn't require homoskedasticity or a parametric specification of the residual distribution.
Warren: It also confirms that the procedure becomes conservative when the reference group is itself small, which is something we need to keep in mind for deployment.
Lucky paper: 2609.05883: Ruth: So we're looking at Fixed-smoothing Uniform Inference for Quantile Regression, and the core idea here is developing inference methods that are robust against both heteroskedasticity and autocorrelation in time series quantile regression.
Warren: The paper introduces a uniform approach across all quantile levels tau, which explicitly accounts for dependence both over time and across different quantiles.
Iris: That uniformity allows the construction of uniform confidence bands, Wald tests, and Sup-t tests for joint hypotheses as well as formal tests to assess shape restrictions on the quantile effects.
Ruth: The authors developed two main techniques: the Uniform-in-tau method and the Stack-Wald method.
Warren: The Uniform-in-tau method estimates the covariance structure and simulates these distributions to address issues related to non pivotal limiting distributions under weak dependence.
Iris: And the Stack-Wald method provides pivotal fixed smoothing inference for tests involving a finite collection of quantile levels, establishing asymptotic validity for both approaches.
Ruth: Simulation results show that these fixed-smoothing methods substantially improve size control compared to existing Heteroskedasticity and Autocorrelation Consistent procedures while maintaining good statistical power.
Warren: Empirical application to predictive quantile regressions for stock returns reveals significant heterogeneity in predictive effects across both quantiles and various forecast horizons.
Iris: The paper finds that the stock variance exhibits its strongest predictive effect at short horizons, particularly within the tails of the return distribution, and this effect diminishes as the forecast horizon increases.
Ruth: Regarding those inference tests, they compare Wald versus Sup-t tests and find that the Wald test tends to exhibit higher null rejection rates as the number of restrictions m increases.
Warren: Conversely, the Sup-t test is relatively insensitive to the choice of discretization grid points n used for quantile ranges and is comparatively more robust to an increasing number of restrictions under the null hypothesis.
Iris: There's also a trade-off between size distortion and power across smoothing parameters, where the Wald test shows strong power for small numbers of restrictions m but that power declines substantially as m increases.
Ruth: The choice of grid for quantile levels matters too; specifically, a tail-based grid yields noticeably lower predictive power than a center-based grid.
Warren: That confirms that inference becomes more challenging when focusing on tail quantiles in this Fixed-smoothing Uniform Inference for Quantile Regression paper.
Iris: This framework successfully enables simultaneous and uniform inference for quantile effects by explicitly modeling dependence across quantiles, offering a unified methodology for robust inference in time series QR.
Ruth: The practical utility is shown through the empirical findings, especially that net equity expansion demonstrates a uniformly positive trend with stability across quantiles where confidence bands suggest statistical significance over a broad range of distributions.
Warren: That's interesting because it shows the effect is stable regardless of which quantile you are looking at in terms of trend direction.
Iris: What this means for us is that we have a unified way to study heterogeneity in quantile effects by explicitly modeling dependence across quantiles, which was previously very difficult.
Ruth: We also see comparisons between different methods revealing trade-offs, like how the Wald test versus the Sup-t test performs under different restriction counts m.
Warren: The paper confirms that for large numbers of restrictions, the Sup-t test is a more useful alternative because it handles that increase better than the Wald test.
Iris: So, even when you have many hypotheses to test about quantile effects, this framework provides a more stable statistical inference method.
Lucky paper: 2505.22909: Ruth: So we're looking at Learning to Charge More, which explores how Q-learning agents might learn to charge supracompetitive prices in infinite repeated games.
Warren: It introduces a framework based on stochastic games with bounded memory to analyze the learning process for n firms competing over an infinite time horizon.
Iris: The paper extends classical fixed-point theory to establish the existence of one-memory subgame perfect equilibria in this setting, which allows for dynamic collusion patterns.
Ruth: They study a variant of Q-learning with bounded experimentation to see how firms adapt their pricing policies based only on observed profits.
Warren: They show that when the game admits both a one-stage Nash equilibrium price and a collusive-enabling price, and when the Q function satisfies certain inequalities at the end of experimentation, firms learn to consistently charge supracompetitive prices.
Iris: This learning process is formalized through Algorithm three which combines softmax based Q learning with bounded experimentation to model that adaptation.
Ruth: The research identifies three types of policy profiles that can support this learned supracompetitive behavior.
Warren: These include naive collusion where firms always choose the collusive enabling price, grim trigger policies involving credible threats and dynamic escalation patterns, and increasing strategies.
Iris: I'm curious about the mechanics behind how those policy profiles interact to sustain that higher pricing structure over time.
Ruth: That is what the paper gets into when it breaks down those three specific types of behavior in detail.
Warren: The theoretical framework itself is built on analyzing dynamic collusion supported by those specific policy profiles mentioned earlier in Learning to Charge More.
Iris: From a modeling perspective, does this imply that the learning process itself is inherently biased towards these collusive outcomes if the conditions are met?
Ruth: That depends entirely on those inequalities regarding the Q function satisfaction at the end of experimentation, which dictates what happens next.
Warren: It’s a very specific condition that governs whether they converge to supracompetitive prices or some other equilibrium.
Iris: Thinking about implementation, if a firm uses a naive collusion policy, how does that compare in terms of stability to the grim trigger policies?
Ruth: The paper suggests the grim trigger policies offer dynamic escalation patterns, which implies a more adaptive approach than just sticking to one price.
Warren: That dynamic escalation is what makes them different from simple static collusive pricing.
Iris: What about the role of bounded memory in this framework; how does that constraint affect the complexity of the learning process for these firms?
Ruth: The bounded memory means firms only use information from the current state and previous prices when choosing their next price, which limits what kind of complex strategy they can pursue.
Warren: So even with infinite time horizons, they are constrained by a one-memory policy structure in this model.
Iris: If we look at the practical implications for market structure, does this suggest that firms might unintentionally develop these collusive pricing behaviors under certain learning conditions?
Ruth: It suggests that if the game admits those two price points, the learning dynamics push them toward charging more than they would in a standard Nash equilibrium.
Warren: That is a direct consequence of how Algorithm three formalizes the softmax based Q learning within the bounded experimentation setting.
Iris: This connects back to our earlier discussion on behavioral modeling in automated markets; are these firms behaving like those we modeled, or are there other factors at play?
Ruth: The study focuses specifically on the learning dynamics driven by profit observation rather than external market shocks.
Warren: So it isolates the effect of internal pricing policy adaptation as the main driver here.
Iris: In terms of what this means for understanding economic complexity, does this suggest that simple learning rules can lead to complex emergent behaviors like collusion?
Ruth: Yes, the paper formalizes how those simple learning rules interact with the game structure to produce these more sophisticated outcomes.
Warren: It shows a mechanism for how firms can move away from competitive pricing structures under specific algorithmic conditions.
Iris: That’s interesting because it moves beyond just static collusion and looks at how dynamic, learned behavior emerges from constrained learning algorithms.
Ruth: Indeed, Learning to Charge More lays out the theoretical basis for that emergent behavior in repeated games.
Lucky paper: 2606.01687: Ruth: We're moving into paper five of five today, and this one is titled Information and voting: Evidence from Peru's two thousand twenty-six general elections.
Warren: It looks like this study uses a Bayesian-updating model for multi-candidate plurality voting frames to predict vote reallocation toward the three candidates deemed viable.
Ruth: They are exploiting a natural experiment where one hundred eighty-seven polling tables failed to install on April twelve and voting was extended until Monday the 13th for about fifty-five thousand electors.
Iris: The model uses a multinomial logit where each voter’s choice probability is interacted with the posterior viability of a candidate and their expressive match to him.
Ruth: The paper identifies the treatment as a change of Blackwell experiment over the state space Theta in Section four point two two six, which sets up how they look at this data.
Warren: The empirical section uses that JNE ruling to identify treated polling locations and builds a new dataset from INEI Redatam census data and two thousand twenty-one results.
Iris: The key finding is that the flash estimate raised perceived viability above the prior for three candidates: Nieto, López Aliaga, and Sánchez.
Ruth: The basic prediction is that the viable candidates gain votes while non-viable ones lose votes, and this reallocation was overwhelmingly concentrated on Nieto.
Warren: Specifically, Nieto had a seventeen point point gain compared to both named candidates in either estimate for that candidate.
Iris: That ordering shows that the reallocation tracks affinity rather than just national viability because the ordering is Nieto greater than López Aliaga greater than Sánchez.
Ruth: The paper does show robustness across every exercise of their sensitivity battery, including leave-one-treated-location-out and Rosenbaum bounds.
Warren: They also found that a candidate can advance nationally on a signal but move few local votes because the reallocation is governed by local affinity while viability travels with the signal.
Iris: The main limitation they flag is the small number of treated locations, which comes from thirteen clusters where they have independent causal information.
Ruth: They also noted that sensitivity to unobserved confounding remains a concern because they only match observed covariates.
Warren: The conclusion confirms that information has a significant robust effect and that the flash raised shares for Nieto, López Aliaga, and Sánchez while lowering non-viable candidates like H5.
More episodes
- Daily Summary for 2026-11-19
- 2608.21274-Recommendation Quality and the Concentration of Consumption: Experimental Evidence from Netflix
- 2408.02757-A nonparametric test for diurnal variation in spot correlation processes
- 2306.16393-High-Dimensional Canonical Correlation Analysis
- 2605.17299-Geometric Brownian motion with intermittent entries and exits
- 2607.04753-Fooling Yourself: how narratives shape beliefs
- 2408.14671-Inference on Treatment Effects with Repeatedly Mismeasured High-Dimensional Controls
- 2201.06898-Difference-in-Differences Estimators for Treatments Continuously Distributed at Every Period
- 2501.15307-Influence Functions: A Unified Differentiation-Based Framework
- The day in conversation for 2026-11-19