Daily Summary for 2026-11-02

daily

Transcript

Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.

Ruth: It's the second of November, twenty twenty-six, and this is the day's research.

Warren: 14 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 Estimating Network Spillovers under Dense Measurement Error.

The day's research: Ruth: This morning focused on improving estimates of network effects when data has errors.

Warren: That is important because understanding spillovers shows system movement.

Iris: The paper Estimating Network Spillovers under Dense Measurement Error introduces a regularization framework for latent networks.

Ruth: It proposes two estimators, one two-stage and one Generalized Method of Moments estimator.

Warren: Both methods show significant improvements over naive estimation techniques.

Iris: They reduce the root mean squared error of spillover estimates by about fifty to eighty percent.

Ruth: Another piece addresses how misclassified links in social networks invalidate standard statistical methods.

Warren: The paper Estimating Social Network Models with Link Misclassification proposes an adjusted two-stage least squares estimator.

Iris: This corrects for binary link misclassifications by estimating the misclassification rates themselves.

Ruth: That resolves issues where covariates become endogenous, relating to how we handle structural uncertainty in Choosing the Dictionary and Penalty for IV-LASSO.

Warren: Which derives a criterion for choosing dictionary and penalty pairs when estimating instrumental variables with LASSO.

Iris: We also have work on more complex modeling challenges like Distribution-Matching Posterior Inference for Incomplete Structural Models.

Ruth: That extends minimal econometric interpretation to generalized posterior inference by combining a structural prior with exact Dirichlet-multinomial scores.

Warren: This approach remains useful even when a model is stochastically singular.

Iris: Finally, On the Use of Design-Based Simulations analyzes how outcomes are fixed and variation is generated by resampling shocks.

Ruth: This study shows that these simulations can be misleading about spatial correlation in clustered experiments.

Warren: The work on calibrated forecasting and persuasion is relevant for communicating probabilistic information to decision-makers.

Iris: It addresses how experts can communicate probabilistic information by framing the problem as a static persuasion challenge.

Ruth: This establishes a benchmark for the value of information, comparing outcomes for both informed and uninformed experts.

Warren: This benchmark relates to how administrators manage funding relationships through an inspection process.

Iris: An inspection can reveal if an agency has diverted funds or if continued auditing sustains implementation without fully revealing the type of agency involved in Reputation and Delegated Public Spending.

Ruth: This suggests that monitoring mechanisms can be more valuable than just project delivery metrics alone.

Warren: Furthermore, the concept of exchangeability in quantity choices under dual submodularity provides a framework for understanding how aggregate decisions converge toward an equilibrium.

Iris: This is relevant when considering complex resource allocation problems in Monotone comparative statics under dual submodularity, with an application to aggregate deferred acceptance.

Ruth: This structural insight complements the behavioral aspects explored in how boundedly rational receivers design protocols to learn from biased senders in Asymmetric Rationality: Protocol Design and Behavioral Anomalies.

Warren: The paper Estimating Network Spillovers under Dense Measurement Error analyzes spillover effects in spatial models when network data is noisy and proposes estimators to reduce estimation error by denoising the network structure.

Iris: Estimating Social Network Models with Link Misclassification proposes an adjusted two-stage least squares estimator that corrects bias in social network models when reported links are misclassified using estimates of misclassification rates.

Ruth: Choosing the Dictionary and Penalty for IV-LASSO derives a criterion for choosing dictionary and penalty pairs to improve the performance of instrumental variables first stage estimation with LASSO.

Warren: Distribution-Matching Posterior Inference for Incomplete Structural Models extends posterior inference to generalized models by combining structural priors with exact scores to compare empirical and model distributions.

Iris: On the Use of Design-Based Simulations shows that design-based simulations can be misleading about spatial or serial correlation in clustered experiments when treatment effects are non-zero.

Ruth: Calibrated Forecasting and Persuasion characterizes the optimal strategy for an expert sending probabilistic forecasts to a decision-maker who verifies them using a calibration test.

Warren: Reputation and Delegated Public Spending studies how auditing agencies can sustain implementation by balancing the administrator's desire for funding against the agency's strategic diversion of advances.

Iris: Monotone comparative statics under dual submodularity, with an application to aggregate deferred acceptance develops monotone comparative statics for constrained quantity choices generated by convex costs satisfying dual submodularity.

Ruth: Asymmetric Rationality: Protocol Design and Behavioral Anomalies investigates how boundedly rational receivers design protocols that can lead to information disengagement and polarization from biased senders.

Warren: Tracks to Modernity: Railroads, Growth, and Civic Institutions in Denmark examines how railroad expansion shaped Denmark’s economic transformation and the diffusion of civic institutions.

Iris: On dynamic price formation in the course of capital reallocation driven by differential rates of profit: strict conservation of value supports Karl Marx's theory develops a three-sector model to show how capital reallocation under differential profit rates can lead to stable price trajectories.

Ruth: Evaluating A/B Testing Methodologies via Sample Splitting: Theory and Practice develops a theoretical framework for sample splitting in A/B testing to understand the bias-variance trade-offs of different design choices.

Warren: This paper analyzes spillover effects in spatial models when network data is noisy and proposes estimators to reduce estimation error by denoising the network structure.

Iris: Estimating Social Network Models with Link Misclassification proposes an adjusted two-stage least squares estimator that corrects bias in social network models when reported links are misclassified using estimates of misclassification rates.

Ruth: Choosing the Dictionary and Penalty for IV-LASSO derives a criterion for choosing dictionary and penalty pairs to improve the performance of instrumental variables first stage estimation with LASSO.

Warren: Distribution-Matching Posterior Inference for Incomplete Structural Models extends posterior inference to generalized models by combining structural priors with exact scores to compare empirical and model distributions.

Iris: On the Use of Design-Based Simulations shows that design-based simulations can be misleading about spatial or serial correlation in clustered experiments when treatment effects are non-zero.

Ruth: Calibrated Forecasting and Persuasion characterizes the optimal strategy for an expert sending probabilistic forecasts to a decision-maker who verifies them using a calibration test.

Warren: Reputation and Delegated Public Spending studies how auditing agencies can sustain implementation by balancing the administrator's desire for funding against the agency's strategic diversion of advances.

Iris: Monotone comparative statics under dual submodularity, with an application to aggregate deferred acceptance develops monotone comparative statics for constrained quantity choices generated by convex costs satisfying dual submodularity.

Ruth: Asymmetric Rationality: Protocol Design and Behavioral Anomalies investigates how boundedly rational receivers design protocols that can lead to information disengagement and polarization from biased senders.

Lucky paper: 2607.19625: Ruth: We're looking at Estimating Network Spillovers under Dense Measurement Error today.

Warren: This paper tackles the problem of inaccurate spillover effect estimates when the adjacency matrix is corrupted by measurement errors from reporting or aggregation.

Iris: The authors introduce a regularization framework for latent network structure that handles contamination while capturing sparsity or low-rank properties.

Ruth: They propose two estimators to leverage this framework: a two-stage procedure and a Generalized Method of Moments estimator.

Warren: Both methods show significant improvements over naive estimation techniques, reducing the root mean squared error of spillover estimates by about fifty to eighty percent.

Iris: The core theoretical contribution is characterizing how the bias of standard spillover estimators depends on both the latent network structure and the nature of measurement error.

Ruth: They establish that ignoring these correlations can introduce non-vanishing bias even in dense regimes typical of economic networks.

Warren: They also derive convergence rates for both the spillover-parameter estimator and their estimated adjacency matrix.

Iris: The framework provides an interpretable decomposition of network connections into a low-rank component and a sparse component.

Ruth: This decomposition is used to provide a novel expression of network eigencentrality, quantifying the distinct contributions of each structural component.

Warren: They apply this decomposition to related concepts like the Leontief inverse, yielding a two-term approximation that captures localized effects versus dense network structures.

Iris: The empirical findings show strictly improved consistency rates for their spillover effect estimator compared to naive methods that ignore measurement error.

Ruth: Simulations strongly support this, showing that in noisy networks, their approach reduces the root mean squared error of spillover estimates by approximately fifty to eighty percent.

Warren: The application to international spillovers in economic growth and tax competition across U.S. states illustrates that denoising can restore Leontief stability and yield more plausible results.

Iris: Specifically, the low-rank plus sparse estimator, denoted as W c two was found to produce more moderate deviations from the raw benchmark.

Ruth: We see this structural decomposition applied to concepts like the Leontief inverse, where it yields a two-term approximation capturing localized shocks and dense network structures.

Warren: This structure allows for a more nuanced understanding of systemic risk by distinguishing between the centrality of highly connected hubs and the broader systemic structure.

Iris: The paper Estimating Network Spillovers under Dense Measurement Error provides a rigorous methodological advancement by integrating network regularization with spillover estimation.

Lucky paper: 2509.07343: Ruth: We're looking at Estimating Social Network Models with Link Misclassification today, which is paper number two of five this day.

Warren: This one tackles the problem where reported links in social networks are subject to random misclassification errors and conventional two-stage least squares estimators become inconsistent.

Iris: Conventional two-stage least squares fails because this misclassification introduces new sources of correlation between covariates and errors, making all covariates endogenous, specifically through the term lambda(G minus H)y.

Ruth: The paper proposes an adjusted two-stage least squares estimator that resolves this endogeneity by constructing a novel estimator of misclassification rates and using those estimates to both adjust endogenous peer outcomes and construct new instruments.

Warren: That is the core idea, replacing the original structural form with an adjusted feasible structural form where W(H, p0, p1) is a properly designed adjustment of the network measure H.

Iris: The paper relies on two key insights for instrument construction: first, by adjusting the noisy measure of peer outcomes Hy using the misclassification rates p0 and p1, one can restore the exogeneity of X in an adjusted feasible structural form.

Ruth: But even with that restored exogeneity, conventional instruments like HX or H2X remain invalid because the adjusted errors v still depend on H, so they need alternative functions of H and X like H prime X.

Warren: The paper then gives simple methods to identify and estimate those unknown misclassification rates p0 and p1 using variation in covariates X that affects actual link formation.

Iris: For scenarios involving two independent measures H(one) and H(two), the paper derives a system of six equations by conditioning on the function phiij (X) which relates to the distribution of Gij, and that system can be uniquely solved for all six unknown parameters including p(t)one and p(t)zero.

Ruth: The empirical application used data from Banerjee et al. from two thousand thirteen studying peer effects in household decisions to participate in a microfinance program in Indian villages.

Warren: And the results show that ignoring link misclassification leads to an upward bias of up to five percent in the estimates of peer effects.

Iris: The adjusted two-stage least squares estimator consistently estimates structural parameters with high accuracy, and the estimated peer effect is found to be around five point one five percent when linked to one more participating household.

Ruth: So, what does this mean for someone just listening? It means that if you are looking at social network data, and those links might be wrong because of misclassification errors, you need this adjusted method to get a reliable estimate of the peer effect.

Warren: It shows that ignoring that error introduces a noticeable upward bias of about five percent in your peer effect calculation.

Iris: The paper Estimating Social Network Models with Link Misclassification proposes adjusted two-stage least squares estimators that consistently estimate structural parameters, including peer effects, in social networks when the reported links are subject to random misclassification errors.

Ruth: It resolves the additional endogeneity issues caused by link misclassification by adjusting the endogenous peer outcomes and applying new instruments constructed from noisy network measures.

Warren: So it's about using those estimated rates p0 and p1 to fix both the peer outcomes and build better instruments for your estimation.

Iris: The empirical estimates suggest that peer effects are substantial in this application.

Lucky paper: 2609.07033: Ruth: Let's talk about Choosing the Dictionary and Penalty for IV-LASSO.

Warren: That paper tackles the difficulty of choosing instruments when you have a large set available for an instrumental variables model using LASSO.

Iris: First-order asymptotic theory doesn't give any guidance on which dictionary or penalty level to pick because any consistent implementation yields the same limiting distribution.

Ruth: But in finite samples, those choices can actually impact the structural parameter estimate quite a bit.

Warren: The paper uses first- and second-order Stein identities to derive an approximate mean squared error, or AMSE, for the instrumental variables LASSO estimator.

Iris: That AMSE reveals a bias versus variance trade-off: fitting more complex first stages approximates the conditional mean better, but it also increases correlation with structural errors.

Ruth: The paper says this complexity cost rises with the squared covariance between the first stage and structural errors, which is a quantity that neither plug-in nor cross-validation penalty rules account for.

Warren: Despite being derived in a Gaussian model, minimizing the feasible AMSE criterion delivers up to a one-third lower mean squared error compared to cross-validation and plug-in rules in Gaussian and non-Gaussian simulations calibrated to Gilchrist and Sands two thousand sixteen.

Iris: The paper develops a principled method by prespecifying a finite list of first-stage LASSO dictionary–penalty candidates and ranking them by the approximate mean squared error of their resulting IV-LASSO estimators.

Ruth: They define a criterion that ranks these candidates by analyzing second-order terms in the expansion of the resulting second-stage estimator's mean squared error.

Warren: Minimizing this criterion selects an estimator whose AMSE is as small as that of the best candidate in the list, up to vanishing relative error.

Iris: They construct an observable criterion that estimates it up to a candidate-independent centering term, which preserves its ranking asymptotically.

Ruth: The only extra inputs not obtained from the candidate fits are the structural error moments sigma squared epsilon and sigma ev.

Warren: They estimate those moments using residuals from a pilot candidate.

Iris: The simulation study shows that the proposed criterion adjusts first-stage complexity based on the level of endogeneity, which matches the increasing complexity cost seen in two thousand six hundred nine point zero seven zero three three one five.

Ruth: The paper also shows similarity between results in Gaussian and non-Gaussian regimes, with Table two showing the feasible criterion has the smallest average and maximum relative risk at both sample sizes.

Warren: And Table four shows the same behavior when using Laplace errors.

Iris: Simulation details show that the proposed criterion adjusts first-stage complexity based on endogeneity, consistent with the increasing complexity cost seen in two thousand six hundred nine point zero seven zero three three one five across multiple tables.

Lucky paper: 2508.09004: Ruth: We're moving on to paper four of five, and it's called Dividing a cake for the irrationally entitled by Florian Brandl and Andrew Mackenzie.

Warren: That paper tackles the communication cost associated with solving fair division problems when agents have private information about their preferences.

Iris: They formalize this using a "tidy kitchen" analogy where agents have private measures of their preferences over a perfectly divisible cake and available knives.

Ruth: The mediator's goal is to construct an allocation that respects a given entitlement profile while minimizing the worst-case query cost needed to learn about those private preferences.

Warren: A key concept they introduce is "clonage," which acts as an index for entitlement profiles, defined as infinity when any entitlement in the profile is irrational.

Iris: They establish relationships between clonage, precision, and fineness indices, showing that clonage C(e) is at least P(e) and P(e) is at least F(e).

Ruth: The central finding focuses on irrational entitlements; they prove that if an entitlement profile e contains even one irrational share, no e-proportional protocol can be bounded in terms of query complexity.

Warren: They show this by analyzing the adversary game associated with the problem, which exhibits a value of minus infinity but lacks any Nash equilibrium.

Iris: That lack of equilibrium means the mediator's task effectively degrades into a one-player game where the adversary can select an arbitrarily large number of queries, meaning no bounded protocol exists.

Ruth: They also provide quantitative lower bounds on communication costs based on these indices with Theorem one establishing a direct relationship between clonage and cost.

Warren: Specifically, if the clonage level C(e) is greater than c for any setting with at least two agents, then the required cost C(e) must satisfy C(e) > two(2c * (one/n - one)).

Iris: That result is further supported by lower bounds from precision and fineness indices, such as Proposition one showing that for a given precision level p, the cost must exceed two(2p).

Ruth: They synthesize these into concrete complexity results with Corollary CF providing bounds on the supremum of the worst-case query cost.

Warren: That corollary shows that supcost(c) equals O(2c) and also supcost(c) equals (3c), which is equivalent to two(3c).

Iris: Complementary results, Corollary two establish a lower bound on the infimum cost, indicating that infcost(c) equals two(2c).

Ruth: The combination of those upper and lower bounds confirms they are tight, meaning no further improvement is possible in this setting.

Warren: So what does this mean for someone who just listens to the show? It means if your preferences have any irrational component, you can't find a solution without an arbitrarily large number of questions.

Iris: Exactly. The paper Dividing a cake for the irrationally entitled rigorously demonstrates that irrational entitlements fundamentally change the nature of the fair division problem by making standard e-proportional protocols unbounded in terms of query complexity.

Lucky paper: 2603.11381: Ruth: Alright, we're on segment seven of seven, and we're looking at the paper On the Use of Design-Based Simulations by Bruno Ferman from Sao Paulo School of Economics – FGV.

Warren: This paper deals with design-based simulations which hold realized outcomes fixed and generate variation by resampling treatment assignment or shocks.

Iris: Standard simulations that fix outcomes and resample shocks may rely on a data-generating process that is not aligned with the true one, which confounds true treatment effects with error dependence.

Ruth: The core problem in simulation design is that holding realized outcomes fixed means the true data-generating process wouldn't be given by those fixed outcomes if treatment effects are present.

Warren: If potential outcomes are symmetric and single-peaked, a design-based simulation might consider a data-generating process where the distributions of potential outcomes may be dual peaked, which differs fundamentally from the true data generating process.

Iris: This implies that whether resulting rejection frequencies are informative about validity depends critically on how closely the data generating process used in simulations relates to the true one along dimensions relevant for inference.

Ruth: The paper proposes alternative simulation designs to circumvent this problem of confounding treatment effects with error dependence.

Warren: Specifically, they show that these alternative simulations can be more informative about distortions due to spatial correlation than tests based on the statistical significance of regressions using pre-treatment outcomes.

Iris: The analysis of empirical applications reinforces that standard inference methods should be used with caution in shift-share applications given the possibility of spatial correlation.

Ruth: Standard design-based simulations tend to overstate the magnitude of size distortions attributable to spatial correlation because they confound true treatment effects with error dependence.

Warren: The choice of simulation design can materially affect conclusions about the reliability of competing inference methods, which is a big deal for anyone trying to make sense of these results.

Iris: Design-based simulations are shown to be more informative about spatial correlation problems than checking whether there are significant effects in placebo regressions using pre-treatment outcomes.

Ruth: The simulations can flag spatial correlation problems whenever there is sufficient variation in the shocks so that some treatment assignments would generate large values of tau one minus tau zero even if tau one minus tau zero is small for the realized treatment assignment.

Warren: Overall, the usefulness of design-based simulations depends on how closely the simulated data generating process aligns with the true one along dimensions that matter for inference.

Iris: Careful construction and interpretation are therefore essential when using these methods.

More episodes

← Home