In the early 1900s, telephone operation was among the most common jobs for American women, and telephone operators were ubiquitous. Between 1920 and 1940, AT&T undertook one of the largest automation investments in modern history, replacing operators with mechanical switching technology in over half of the U.S. telephone network. Using variation across U.S. cities in the timing of adoption, we study how this wave of automation affected the labor market for young women. Although automation eliminated most of these jobs, it did not reduce future cohorts’ overall employment: the decline in operators was counteracted by employment growth in middle-skill clerical jobs and lower-skill service jobs, including new categories of work. Using a new genealogy-based census-linking method, we show that incumbent telephone operators were most affected, and a decade later more likely to be in lower-paying occupations or no longer working.

We test for gender bias in promotions at financial institutions using two central predictions of Becker’s (1957, 1993) model: firms with bias will (1) raise the promotion bar for marginally promoted female workers, and (2) incur costs from forgoing efficient employment practices. We find support for both of these predictions using a new nationwide panel of mortgage loan officers and their managers encompassing approximately 72,000 workers from over 1,000 shadow banks from 2014 to 2019. Overall, our findings provide evidence that gender bias is an important factor in gender gaps at financial institutions.

Analyzing 48 foreign exchange (FX) rates and 1.2 million FX-related news articles over a 35-year period, using digital textual analysis, we find that a currency reversal investment strategy that buys (sells) currencies with low (high) media sentiment offers strong positive and statistically significant returns and Sharpe ratios. The results are robust and the strategy adds value over other currency premia determinants. Analysts’ forecasts systematically mispredict the reversal strategy. This is the first article to show that price reversals based on media sentiment are a well-defined feature of the FX market.

New technologies tend to be adopted slowly and – even after being adopted – take time to be reflected in higher aggregate productivity. One prominent explanation for these patterns is the need to reorganize production, which often goes hand-in-hand with major technological breakthroughs. We study a unique setting that allows us to examine the empirical relevance of this explanation: the adoption of mechanized cotton spinning during the First Industrial Revolution in France. The new technology required reorganizing production by moving workers from their homes to the newly-formed factories. Using a novel hand-collected plant-level dataset from French archival sources, we show that productivity growth in mechanized cotton spinning was driven by the disappearance of plants in the lower tail – in contrast to other sectors that did not need to reorganize when new technologies were introduced. We provide evidence that this was driven by organizational challenges such as developing optimal plant layout. A process of ‘trial and error’ led to initially low and widely dispersed productivity, and – in the subsequent decades – to high productivity growth as knowledge diffused through the economy and new entrants adopted improved methods of organizing production.

In this study, we examine auditors' reliance on artificial intelligence (AI) systems that are designed to provide evidence around complex estimates. In an experiment with highly experienced auditors, we find that auditors are more hesitant to rely on evidence from AI‐based systems compared to human specialists, consistent with algorithm aversion. Importantly, we also find that a small amount of control (i.e., providing input to specialists) can mitigate this aversion, though this effect depends on auditors' personal locus of control (LOC). Providing input increases reliance on evidence from AI systems for auditors who believe they have little control over their outcomes (i.e., an external LOC). In contrast, auditors with an internal LOC are particularly hesitant to rely on AI‐based evidence, and providing input has little impact on their reliance. Interviews with experienced auditors corroborate our findings and suggest auditors feel a greater sense of control working with human specialists relative to AI‐based systems. Overall, our results suggest perceived control plays an important role in auditors' aversion to AI and that auditors' individual traits can affect this aversion.

It is often desired to rank different populations according to the value of some feature of each population. For example, it may be desired to rank neighbourhoods according to some measure of intergenerational mobility or countries according to some measure of academic achievement. These rankings are invariably computed using estimates rather than the true values of these features. As a result, there may be considerable uncertainty concerning the rank of each population. In this paper, we consider the problem of accounting for such uncertainty by constructing confidence sets for the rank of each population. We consider both the problem of constructing marginal confidence sets for the rank of a particular population as well as simultaneous confidence sets for the ranks of all populations. We show how to construct such confidence sets under weak assumptions. An important feature of all of our constructions is that they remain computationally feasible even when the number of populations is very large. We apply our theoretical results to re-examine the rankings of both neighbourhoods in the U.S. in terms of intergenerational mobility and developed countries in terms of academic achievement. The conclusions about which countries do best and worst at reading, math, and science are fairly robust to accounting for uncertainty. The confidence sets for the ranking of the fifty most populous commuting zones by measures of mobility are also found to be small. These confidence sets, however, become much less informative if one includes all commuting zones, if one considers neighbourhoods at a more granular level (counties, census tracts), or if one uses movers across areas to address concerns about selection.

We study how the human capital embedded in teams is affected by, and reallocated through, corporate bankruptcies. After a bankruptcy, U.S. inventors produce fewer and less impactful patents. Moreover, teams become less stable. Consequently, compared to inventors that rely less on teamwork, the performance of team inventors deteriorates more. These findings point to the loss of team-specific human capital as a cost of resource reallocation through bankruptcy. Acquisitions by industrial firms and joint mobility of inventors with past collaborations limit these losses, suggesting that the labor market and the market for corporate control help preserve team-specific human capital in bankruptcies.

We modify the 1983 Diamond and Dybvig model so that banks offer liquidity services to depositors, raise equity funding, make risky loans, and invest in safe, liquid assets. Banks monitor borrowers to ensure that they repay loans and they are susceptible to depositor runs. We model the run decision by solving a novel global game. Relative to a social planner, banks opt for a more deposit-intensive capital structure, their assets may be more or less lending intensive, and the level of lending may be higher or lower. Correcting these three distortions requires a package of three regulations.

Gun violence is the most pressing public safety problem in U.S. cities. We report results from a randomized controlled trial (N = 2,456) of a community-researcher partnership called the Rapid Employment and Development Initiative (READI) Chicago. The program offered an 18-month job alongside cognitive behavioral therapy and other social support. Both algorithmic and human referral methods identified men with strikingly high scope for gun violence reduction: for every 100 people in the control group, there were 11 shooting and homicide victimizations during the 20-month outcome period. Fifty-five percent of the treatment group started programming, comparable to take-up rates in programs for people facing far lower mortality risk. After 20 months, there is no statistically significant change in an index combining three measures of serious violence, the study’s primary outcome. Yet there are signs that this program model has promise. One of the three measures, shooting and homicide arrests, declined 65% (p = .13 after multiple-testing adjustment). Because shootings are so costly, READI generated estimated social savings between $182,000 and $916,000 per participant (p = .03), implying a benefit-cost ratio between 4:1 and 18:1. Moreover, participants referred by outreach workers—a prespecified subgroup—saw enormous declines in arrests and victimizations for shootings and homicides (79% and 43%, respectively) which remain statistically significant even after multiple-testing adjustments. These declines are concentrated among outreach referrals with higher predicted risk, suggesting that human and algorithmic targeting may work better together.

Transaction costs have a first-order effect on the performance of currency portfolios. Proportional costs based on quoted bid–ask spread are relatively small, but when a fund is large, costs due to the trading volume price impact are sizable and quickly erode returns, leaving many popular strategies unprofitable. A mean–variance-transaction-cost optimized approach (MVTC) that accounts for costs in the optimization efficiently tackles the problem with only relatively minor negative implications on before-cost profitability. MVTC is robust even when the price impact of trading is severe. Finally, we introduce an accurate extrapolation approach to expand the sample of the realized Amihud measure of Ranaldo and Santucci de Magistris (2022) from 12 to 26 currencies and from 2012 back in time to 1986.