How do long-term relationships between banks and firms shape loan pricing and capital allocation? Using administrative data from Mexico’s credit registry, I provide stark evidence for an insurance view of relationship lending. When firms repeatedly borrow from the same bank, the pass-through of changes in their default risk to loan rates is nearly zero, and past risk assessments persistently influence credit terms. In contrast, switching to a new bank results in full risk pass-through, consistent with competitive market predictions. I rationalize this evidence in a structural model where banks compete for borrowers by offering optimal long-term contracts. Switching costs sustain commitment to banking relationships, enabling insurance. The estimated model replicates the observed pricing patterns and generates new predictions, which I validate in the data, regarding when firms receive cheap funding and when they are tempted to switch. At the macro level, switching costs enhance capital allocation by strengthening relationships, recovering over 10 percent of welfare losses from financial frictions. However, when embedded in a New Keynesian framework, relationships dampen monetary and fiscal policy pass-through, as banks optimally absorb part of these policy shocks.
What role have factors affecting female labor supply, such as social norms and discrimination, played in the decline of routine jobs in the U.S. since the 1970’s? While typically attributed to changes in labor demand, the decline in routine employment has been larger for women than men, reflecting a shift of female employment out of routine clerical jobs and into non-routine professions. This paper presents a quantitative analysis of the impact of falling labor market distortions faced by women in explaining the trend. One observable manifestation of these falling distortions is the Quiet Revolution, which refers to a shift in women’s life cycle labor force attachment from intermittent to continuous after 1970; it spurred the rise of female non-routine employment because these are long-term careers that reward experience. I develop and calibrate an equilibrium model of the labor market featuring the Quiet Revolution, discrimination, and improvement in automation. Counterfactual analyses reveal that the Quiet Revolution and reduced discrimination explain 21% and 59%, respectively, of the growth of non-routine relative to routine white-collar employment among women between 1970 and 2000. Together, they explain 36% of the aggregate increase, while automation explains 56%. Finally, the Quiet Revolution raised output per worker by 3% via increased female experience.
This paper establishes a new empirical finance puzzle, the retail alignment puzzle: aggregate retail trader purchases and sales are nearly perfectly correlated across time and in the cross section of equities despite retail traders representing a small fraction of exchange volumes and being commonly represented as displaying lopsided flow patterns. Consistent with this puzzle, retail purchases and sales in the cross section are linearly predicted by the same two attention-associated factors, recent return salience and recent volume, with regressions on purchases and sales possessing almost identical coefficients. Using both directly measured attention through Google Trends search volumes and common indirect measures of attention such as volumes and extreme returns, I show that surges in retail attention consistently generate both large trading volumes and proportionally limited net trading. I then use an equilibrium disagreement model to show analytically and through simulations that while positive shocks to retail attention, sentiment, and disagreement all increase price, only shifts in attention are capable of reproducing empirical volume and return patterns. This paper's results suggest that attention is one of the core drivers of retail volume in common stocks.
This paper studies the impact of industrial policies on technology competition and consumer welfare amid rising global trade disruption risks. Distilling key empirical features from novel data on the semiconductor foundry industry, I develop and estimate a dynamic oligopoly model that integrates step-by-step innovation, trade disruption risk, and industrial policies. While distortions from market power and technological externalities justify subsidies, their optimal levels depend on the magnitude of trade disruption risk: when the risk is low, the optimal subsidy rate remains low, as the welfare benefits are distributed globally, but the costs are borne exclusively by the subsidizing government. My quantitative model shows that a 35% trade disruption risk makes the 25% investment subsidy under the U.S. CHIPS Act optimal, resulting in a 6% welfare improvement for the U.S. The paper also analyzes the CHIPS Act's restrictions on investments in rival countries, intended to secure technological leadership against their firms. Its efficacy depends on the strength of technology spillover restrictions and the scale of the rival home market secured for rival firms.
I study optimal macroprudential policy when its effects on investment and productivity are taken into account. To do so, I introduce a tractable way of modeling misallocation that generates a link between investment and productivity and can be easily taken to the data. Because macroprudential policies affect investment, they lead to productivity losses. I show that, when the policymaker is constrained in their available instruments, this generates a policy trade-off between financial stability and productivity growth. I derive a sufficient statistics formula for the second-best policy, including its productivity costs. I leverage the tractability of my model to get a range of estimates for the latter using rich firm-level microdata for several European countries. The trade-off is quantitatively relevant: For baseline crisis probabilities, productivity losses switch optimal policy from a capital control to a foreign borrowing subsidy.
This paper examines whether teachers’ unions affect student achievement in Wisconsin. First, I establish several facts about which teachers are voluntary union members. In particular, I find that union members appear negatively selected by teacher value added. Second, using the staggered decertification of district unions over time, I find increases in both student test scores and attendance rates. These effects are not driven by compositional changes within the teaching workforce; rather, I find evidence suggesting that teachers’ productivity improved in decertified districts. Together, the results imply that union efforts to insulate workers may adversely affect the quality of public services.
This paper investigates how differences in subjective beliefs about stock returns contribute to wealth inequality through portfolio choice. I argue that this channel significantly contributes to wealth inequality in the US. Using the Michigan Survey of Consumers, I find that (1) subjective beliefs about future stock returns are widely dispersed, (2) optimistic households are more likely to participate in the stock market and invest more in stock, and (3) subjective beliefs are persistent over time. Motivated by these findings, I develop and calibrate a heterogenous-agent model that matches data on income inequality, beliefs distribution and portfolio choice. Compared to a model without dispersion in beliefs, the model with belief heterogeneity generates an additional 0.12 in the Gini coefficient of wealth inequality and 10 percent more wealth owned by the top 15% of the households. This suggests that the dispersion in subjective beliefs about stock returns could be an important factor in explaining the size of wealth inequality in the US.
Program evaluations are motivated in part by a desire to improve policy effectiveness. Yet there is limited empirical evidence on the efficacy of evaluation itself. This paper examines the systematic relationship between program evaluations and changes in policy spending, in the context of Conditional Cash Transfers in Latin America and the Caribbean. Using a novel dataset of 128 program evaluations mapped to spending on the corresponding evaluated programs, I find a robust zero relationship between research results and spending. This holds for several definitions of evaluation outcomes: more statistically significant, larger magnitude, more surprising, or more positively framed results, do not correspond with larger increases in spending. As policymakers may learn from cumulative evidence rather than individual studies, I then use a Bayesian hierarchical approach to aggregate evaluations. I find a zero association between a country’s cumulative evidence base and its spending. Finally, I explore mechanisms for this result by considering heterogeneous responses to evaluations that are more credible, actionable, or generalizable. I find that credibility and generalizability are unrelated to spending, but evaluations conducted quickly (within four years of the effect year) and attributable to the political party in power, are significantly predictive of spending. Thus, timeliness may be an overlooked aspect of the evidence-to-policy pipeline.
I study how financial technology reshapes competition among banks. I exploit quasi-random variation in exposure to the introduction of Brazil's Pix, an instant payment system, and show that instant payments increase deposit competition. Small bank deposits rise relative to large banks because Pix allows small banks to offer payment convenience more similar to large banks. Since they become more competitive providing payment services, small banks reduce deposit rates relative to large banks. Finally, I estimate a deposit demand model and find that depositors' welfare increases with Pix. These findings suggest that universally available payment systems can foster banking competition.
The rise of Artificial Intelligence (AI) has the potential to reshape the knowledge economy by enabling problem solving at scale. This paper introduces a framework to analyze this transformation, incorporating AI into an economy where humans form hierarchical firms to use their time efficiently: Less knowledgeable individuals become "workers" solving routine problems, while more knowledgeable individuals become "solvers" assisting workers with exceptional problems. We model AI as a technology that transforms computing power into "AI agents," which can either operate autonomously (as co-workers or solvers/co-pilots) or non-autonomously (only as co-pilots). We show that basic autonomous AI displaces humans towards specialized problem solving, leading to smaller, less productive, and less decentralized firms. In contrast, advanced autonomous AI reallocates humans to routine work, resulting in larger, more productive, and more decentralized firms. While autonomous AI primarily benefits the most knowledgeable individuals, non-autonomous AI disproportionately benefits the least knowledgeable. However, autonomous AI achieves higher overall output. These findings reconcile seemingly contradictory empirical evidence and reveal key tradeoffs involved in regulating AI autonomy.