Innovations in statistical technology in functions including credit‐screening have raised concerns about distributional impacts across categories such as race. Theoretically, distributional effects of better statistical technology can come from greater flexibility to uncover structural relationships or from triangulation of otherwise excluded characteristics. Using data on U.S. mortgages, we predict default using traditional and machine learning models. We find that Black and Hispanic borrowers are disproportionately less likely to gain from the introduction of machine learning. In a simple equilibrium credit market model, machine learning increases disparity in rates between and within groups, with these changes attributable primarily to greater flexibility

Much of the extant literature predicts market returns with “simple” models that use only a few parameters. Contrary to conventional wisdom, we theoretically prove that simple models severely understate return predictability compared to “complex” models in which the number of parameters exceeds the number of observations. We empirically document the virtue of complexity in U.S. equity market return prediction. Our findings establish the rationale for modeling expected returns through machine learning

Using an experiment in which 3.1 million bank customers were encouraged to save, we explore the mechanisms behind coholding liquid savings and credit card debt. Theoretically, we show that the joint responses of spending, saving, and borrowing to the nudge differ across economic models of coholding. Using machine learning techniques, we find that the most responsive individuals reduce spending and increase savings by 4.9% (206 USD PPP per month) while their credit card debt remains unchanged. These individuals' marginal responses to the nudge are consistent with our model of coholding for the purpose of self‐ or partner‐control

We train a machine learning method on a class of informed trades to develop a new measure of informed trading, informed trading intensity (ITI). ITI increases before earnings, mergers and acquisitions, and news announcements, and has implications for return reversal and asset pricing. ITI is effective because it captures nonlinearities and interactions between informed trading, volume, and volatility. This data‐driven approach can shed light on the economics of informed trading, including impatient informed trading, commonality in informed trading, and models of informed trading. Overall, learning from informed trading data can generate an effective informed trading measure

Municipal bonds exhibit considerable retail pricing variation, even for same‐size trades of the same bond on the same day, and even from the same dealer. Markups vary widely across dealers. Trading strongly clusters on eighth price increments, and clustered trades exhibit higher markups. Yields are often lowered to just above salient numbers. Machine learning estimates exploiting the richness of the data show that dealers that use strategic pricing have systematically higher markups. Recent Municipal Securities Rulemaking Board rules have had only a limited impact on markups. While a subset of dealers focus on best execution, many dealers appear focused on opportunistic pricing

We develop a method that identifies the attention paid by earnings call participants to firms' climate change exposures. The method adapts a machine learning keyword discovery algorithm and captures exposures related to opportunity, physical, and regulatory shocks associated with climate change. The measures are available for more than 10,000 firms from 34 countries between 2002 and 2020. We show that the measures are useful in predicting important real outcomes related to the net‐zero transition, in particular, job creation in disruptive green technologies and green patenting, and that they contain information that is priced in options and equity markets

We construct a neural network algorithm that generates price predictions for art at auction, relying on both visual and nonvisual object characteristics. We find that higher automated valuations relative to auction house presale estimates are associated with substantially higher price‐to‐estimate ratios and lower buy‐in rates, pointing to estimates' informational inefficiency. The relative contribution of machine learning is higher for artists with less dispersed and lower average prices. Furthermore, we show that auctioneers' prediction errors are persistent both at the artist and at the auction house level, and hence directly predictable themselves using information on past errors

We assess the long‐term effects of managerial stress on aging and mortality. Using a difference‐in‐differences design, we apply neural network–based machine‐learning techniques to CEOs' facial images and show that exposure to industry distress shocks during the Great Recession produces visible signs of aging. We estimate a one‐year increase in “apparent” age. Moreover, using data on CEOs since the mid‐1970s, we estimate a 1.1‐year decrease in life expectancy after an industry distress shock, but a two‐year increase when antitakeover laws insulate CEOs from market discipline. The estimated health costs are significant, both in absolute terms and relative to other health risks

We build cross‐sections of asset returns for a given set of characteristics, that is, managed portfolios serving as test assets, as well as building blocks for tradable risk factors. We use decision trees to endogenously group similar stocks together by selecting optimal portfolio splits to span the stochastic discount factor, projected on individual stocks. Our portfolios are interpretable and well diversified, reflecting many characteristics and their interactions. Compared to combinations of dozens (even hundreds) of single/double sorts, as well as machine‐learning prediction‐based portfolios, our cross‐sections are low‐dimensional yet have up to three times higher out‐of‐sample Sharpe ratios and alphas

A war‐related factor model derived from textual analysis of media news reports explains the cross section of expected stock returns. Using a semisupervised topic model to extract discourse topics from 7,000,000 New York Times stories spanning 160 years, the war factor predicts the cross section of returns across test assets derived from both traditional and machine learning construction techniques, and spanning 138 anomalies. Our findings are consistent with assets that are good hedges for war risk receiving lower risk premia, or with assets that are more positively sensitive to war prospects being more overvalued. The return premium on the war factor is incremental to standard effects