Knowledge that Transforms

Abstract Why are job polarization and income inequality higher in large U.S. cities? I offer a new explanation: when house prices grow faster in large cities, middle-income households increasingly cannot afford to own a house there. They move to smaller cities and the middle of the income distribution in large cities hollows out, making them more polarized and unequal. I document that (1) cities with higher price growth experienced larger polarization and increase in inequality since 1980 and (2) middle-income households migrate more often to cheaper locations for housing-related reasons than low- or high-income households. Using a spatial equilibrium model with tenure choice and skill heterogeneity, I find that excess growth of prices relative to incomes and rents in large cities accounts for nearly all of the gap in polarization and almost one-half of the gap in inequality growth between large and small cities from 1980 to 2019.

Abstract Regulatory concerns over a tension between expanding financial access and limiting default have led to significant restrictions on contract terms in a number of countries, despite limited evidence on their effectiveness. We use a large nation-wide RCT to examine new borrower responses to changes in interest rates and minimum payments for a credit card that accounted for 15% of all first-time formal loans in Mexico. Default rates were 19% over the 26 month experiment and a 30 pp decrease in interest rates decreased default by 2.5 pp with no effects on the newest borrowers. Doubling minimum payments increased default by 0.8 pp during the experiment but reduced it by 1 pp afterwards, possibly by reducing debt. Matching the experimental sample to their formal employment histories we find that the effect of job separation—more common among new borrowers—on default is seven times larger than the effect of the 30 pp interest rate change. We provide a simple framework for interpreting the experimental results, and rationalize the smaller contract term effects by their limited effects on cash flow rather than by differences in per-peso impacts.

Abstract Human decision-makers frequently override the recommendations generated by predictive algorithms, but it is unclear whether these discretionary overrides add valuable private information or reintroduce human biases and mistakes. We develop new quasi-experimental tools to measure the impact of human discretion over an algorithm on the accuracy of decisions, even when the outcome of interest is only selectively observed, in the context of bail decisions. We find that 90% of the judges in our setting underperform the algorithm when they make a discretionary override, with most making override decisions that are no better than random. Yet the remaining 10% of judges outperform the algorithm in terms of both accuracy and fairness when they make a discretionary override. We provide suggestive evidence on the behaviour underlying these differences in judge performance, showing that the high-performing judges are more likely to use relevant private information and are less likely to overreact to highly salient events compared to the low-performing judges.

Abstract A seller with commitment power sets prices over time. Risk-averse buyers arrive to the market and decide when to purchase. We show that it is optimal for the seller to choose a constant high price punctuated by occasional episodes of sequential discounts that occur at random times. This optimal price-path has the property that the price a buyer ends up paying is independent of his arrival and purchase times, and only depends on his valuation. Our theory accommodates empirical findings on the timing of discounts.

Abstract We quantify micro and macro effects of changes in the potential benefit duration (PBD) in unemployment insurance. In Poland, the PBD is 12 months for the newly unemployed if the previous year’s county unemployment rate is more than 150% of the national average, and 6 months otherwise. We exploit this cut-off using regression discontinuity estimates on registry data containing the universe of unemployed from 2005 to 2019. For those whose PBD is directly affected by the policy rule, benefit recipients younger than 50, a PBD increase from 6 to 12 months leads to 13% higher unemployment. A decomposition analysis reveals that 12 months after an increase in the PBD, only half of the increase in unemployment is due to the effect on search effort (the micro effect) while the other half is due to increased inflows into unemployment. The total effect on unemployment, which includes equilibrium effects, is entirely explained by the increase in unemployment of workers directly affected by the policy change. We find no evidence of spill-overs on two distinct groups of unemployed whose PBD is unchanged and no effect on measures of labour market tightness.

Abstract This article provides a unified framework for studying the identification of counterfactual parameters in a general class of discrete outcome models, allowing for endogenous regressors and multidimensional latent variables, all without parametric distributional assumptions. Our main theoretical result is that, when the covariates are discrete, the infinite-dimensional latent variable distribution can be replaced with a finite-dimensional version that is equivalent from an identification perspective. The finite-dimensional latent variable distribution is constructed in practice by enumerating regions of the latent variable space with a new and efficient cell enumeration algorithm for hyperplane arrangements. We then show that bounds on a certain class of counterfactual parameters can be computed by solving a sequence of linear programming problems, and show how the researcher can introduce additional assumptions as constraints in the linear programmes. Finally, we apply the method to a mobile phone choice example with heterogeneous choice sets, and to an airline entry game example.

Abstract When selling goods like wireless spectrum or electricity contracts, designers often opt for core-selecting mechanisms—i.e. those that induce outcomes in the core—in order to balance revenue and efficiency goals. But increasingly, auctions—such as the Federal Communications Commission’s Incentive Auction and those explored for natural resources—seek to reallocate goods, not just sell them. We show that when bidders can both buy and sell, substitutability among goods is no longer sufficient or necessary for core selection. In particular, in these environments, core selection can fail even with a single good and positive revenue, and can succeed even when some or all bidders view goods as complements. Instead, we show that the key feature that determines core selection is heterogeneity among the bidders. With too much heterogeneity, reallocation mostly realises pre-existing gains from trade among the bidders, and core selection fails. With limited heterogeneity, most gains from trade among the bidders are created by the quantity auctioned, and a core-selecting mechanism is possible.

Abstract When inferring causal effects from correlational data, a common practice by professional researchers but also lay people is to control for potential confounders. Inappropriate controls produce erroneous causal inferences. I model decision-makers (DMs) who use endogenous observational data to learn actions’ causal effect on payoff-relevant outcomes. Different DM types use different controls. Their resulting choices affect the very correlations they learn from, thus calling for an equilibrium analysis of the steady-state welfare cost of bad controls. I obtain tight upper bounds on this cost. Equilibrium forces drastically reduce it when types’ sets of controls contain one another.

Abstract We revisit the recent revealed preference analysis of sophisticated quasi-hyperbolic consumers by Blow et al. [(2021), “Non-parametric Analysis of Time-Inconsistent Preferences”, The Review of Economic Studies, 88, 2687–2734] (BBC). We show that BBC’s revealed preference test is too lax. There are non-rationalizable data that would pass their test. A basic problem with their test is that it requires finding a certain endogenous elasticity, without regard to the rationalizing utility. Their approach motivates a more stringent test, also based on first-order conditions, that would connect the endogenous elasticity and utility: We show that this test is also too lax. Aside from testing, we also discuss the possibility of recovering model parameters. We show that, even when discount factors are exactly identified, the approach followed in BBC allows for incorrect parameter values to lie in their identified set.

Abstract I study repeated games with anonymous random matching where players can add or remove signals from their records. The ability to manipulate records introduces monotonicity constraints on players’ continuation values, under which sufficiently long-lived players will almost never cooperate. When players’ expected lifespans are intermediate, their ability to sustain cooperation depends on (i) whether their actions are complements or substitutes and (ii) whether manipulation takes the form of adding or removing signals.