Knowledge that Transforms

This paper develops methods for detecting discrimination by individual employers using correspondence experiments that send fictitious resumes to real job openings. We establish identification of higher moments of the distribution of job‐level callback rates as a function of the number of resumes sent to each job and propose shape‐constrained estimators of these moments. Applying our methods to three experimental data sets, we find striking job‐level heterogeneity in the extent to which callback probabilities differ by race or sex. Estimates of higher moments reveal that while most jobs barely discriminate, a few discriminate heavily. These moment estimates are then used to bound the share of jobs that discriminate and the posterior probability that each individual job is engaged in discrimination. In a recent experiment manipulating racially distinctive names, we find that at least 85% of jobs that contact both of two white applications and neither of two black applications are engaged in discrimination. To assess the potential value of our methods for regulators, we consider the accuracy of decision rules for investigating suspicious callback behavior in various experimental designs under a simple two‐type model that rationalizes the experimental data. Though we estimate that only 17% of employers discriminate on the basis of race, we find that an experiment sending 10 applications to each job would enable detection of 7–10% of discriminatory jobs while yielding Type I error rates below 0.2%. A minimax decision rule acknowledging partial identification of the distribution of callback rates yields only slightly fewer investigations than a Bayes decision rule based on the two‐type model. These findings suggest illegal labor market discrimination can be reliably monitored with relatively small modifications to existing correspondence designs.

This paper draws parallels between the principal components analysis of factorless high‐dimensional nonstationary data and the classical spurious regression. We show that a few of the principal components of such data absorb nearly all the data variation. The corresponding scree plot suggests that the data contain a few factors, which is corroborated by the standard panel information criteria. Furthermore, the Dickey–Fuller tests of the unit root hypothesis applied to the estimated “idiosyncratic terms” often reject, creating an impression that a few factors are responsible for most of the nonstationarity in the data. We warn empirical researchers of these peculiar effects and suggest to always compare the analysis in levels with that in differences.

Waitlists are often used to ration scarce resources, but the trade-offs in designing these mechanisms depend on agents' preferences. We study equilibrium allocations under alternative designs for the deceased donor kidney waitlist. We model the decision to accept an organ or wait for a preferable one as an optimal stopping problem and estimate preferences using administrative data from the New York City area. Our estimates show that while some kidney types are desirable for all patients, there is substantial match-specific heterogeneity in values. We then develop methods to evaluate alternative mechanisms, comparing their effects on patient welfare to an equivalent change in donor supply. Past reforms to the kidney waitlist primarily resulted in redistribution, with similar welfare and organ discard rates to the benchmark first come first served mechanism. These mechanisms and other commonly studied theoretical benchmarks remain far from optimal. We design a mechanism that increases patient welfare by the equivalent of an 18.2 percent increase in donor supply.

A firm selects applicants to hire based on hard information, such as a test result, and soft information, such as a manager's evaluation of an interview. The contract that the firm offers to the manager can be thought of as a restriction on acceptance rates as a function of test results. I characterize optimal acceptance rate functions both when the firm knows the manager's mix of information and biases and when the firm is uncertain. These contracts may admit a simple implementation in which the manager can accept any set of applicants with a sufficiently high average test score.

Centralized markets reduce search for buyers and sellers. Their “thickness” increases the chance of order execution at nearly competitive prices. In spite of the incentives to consolidate, some markets, securities markets and on‐line advertising being the most notable, are fragmented into multiple trading venues. We argue that fragmentation is an inevitable feature of any centralized market except in special circumstances.

We compare sparse and dense representations of predictive models in macroeconomics, microeconomics, and finance. To deal with a large number of possible predictors, we specify a prior that allows for both variable selection and shrinkage. The posterior distribution does not typically concentrate on a single sparse model, but on a wide set of models that often include many predictors.

Policymakers frequently use price regulations as a response to inequality in the markets they control. In this paper, we examine the optimal structure of such policies from the perspective of mechanism design. We study a buyer‐seller market in which agents have private information about both their valuations for an indivisible object and their marginal utilities for money. The planner seeks a mechanism that maximizes agents' total utilities, subject to incentive and market‐clearing constraints. We uncover the constrained Pareto frontier by identifying the optimal trade‐off between allocative efficiency and redistribution. We find that competitive‐equilibrium allocation is not always optimal. Instead, when there is inequality across sides of the market, the optimal design uses a tax‐like mechanism, introducing a wedge between the buyer and seller prices, and redistributing the resulting surplus to the poorer side of the market via lump‐sum payments. When there is significant same‐side inequality that can be uncovered by market behavior, it may be optimal to impose price controls even though doing so induces rationing.

We study deep neural networks and their use in semiparametric inference. We establish novel nonasymptotic high probability bounds for deep feedforward neural nets. These deliver rates of convergence that are sufficiently fast (in some cases minimax optimal) to allow us to establish valid second‐step inference after first‐step estimation with deep learning, a result also new to the literature. Our nonasymptotic high probability bounds, and the subsequent semiparametric inference, treat the current standard architecture: fully connected feedforward neural networks (multilayer perceptrons), with the now‐common rectified linear unit activation function, unbounded weights, and a depth explicitly diverging with the sample size. We discuss other architectures as well, including fixed‐width, very deep networks. We establish the nonasymptotic bounds for these deep nets for a general class of nonparametric regression‐type loss functions, which includes as special cases least squares, logistic regression, and other generalized linear models. We then apply our theory to develop semiparametric inference, focusing on causal parameters for concreteness, and demonstrate the effectiveness of deep learning with an empirical application to direct mail marketing.

The location of individuals determines their job and schooling opportunities, amenities, and housing costs. We conceptualize the location choice of individuals as a decision to invest in a “location asset.” This asset has a current cost equal to the location's rent, and a future payoff through better job and schooling opportunities. As with any asset, savers in the location asset transfer resources into the future by going to expensive locations with high future returns. In contrast, borrowers transfer resources to the present by going to cheap locations that offer few other advantages. Holdings of the location asset depend on its comparison to other assets, with the distinction that the location asset is not subject to borrowing constraints. We propose a dynamic location model and derive an agent's mobility choices after experiencing income shocks. We document the investment dimension of location and confirm the core predictions of our theory using French individual panel data from tax returns.