WORKING PAPERS

Expectations Matter: When (not) to Use Machine Learning Earnings Forecasts (Link )

with Katherine Wood, Zhongjin Lu, and John Campbell

Revise and Resubmit: Management Science

We comprehensively examine the usefulness of machine learning technology to predict a firm’s earnings and offer three main findings. First, while prior literature suggests machine learning can offer better earnings forecasts than analysts, we show that this result is highly sensitive to machine learning model specification choices (i.e., 80 percent of evaluated machine forecasts fail to beat analysts). Second, we examine why the most accurate machine learning forecast consistently beats analysts, finding that they correct for predictable analyst biases that are both linear and non-linear and largely relate to prior forecast errors, forecasted earnings levels, and the firm’s stock price. Finally, we find that investors’ earnings expectations, as revealed through stock prices, largely–but do not fully–correct for these predictable analyst biases, with delayed price realization up to nine months. In additional analysis, we find that optimal ML specification choices remain stable over time, and that while the machine’s outperformance narrows in recent periods, it remains substantial among small-cap stocks. Overall, our study moves beyond the question of whether machine forecasts are superior to human forecasts, and instead focuses on which machine forecast specifications matter, as well as when and why machine forecasts are most superior. In so doing, we provide code and estimates for the most accurate machine forecast specification and demonstrate that investors’ expectations appear to largely (but not fully) align with them.

Valuation Uncertainty and the Bounded Rationality of Investors’ Earning Expectation

with Katherine Wood, Zhongjin Lu, and Wang Renxuan

Information choice theories make ambiguous predictions about how investors acquire information under uncertainty: higher uncertainty may motivate more information acquisition but also increase processing costs. Using the return predictability of analysts’ conditional biases as a measure of investors’ information choice (more return predictability means less information acquisition and less rational market expectation), we find a positive relation between uncertainty and information acquisition in the time series but a negative relation in the cross-section. These results are robust across various measures of uncertainty and other earnings-related return predictors. We hypothesize that investors’ information processing costs are more driven by persistent firm-level uncertainty than time-series variations and show the hypothesis is empirically supported by direct measures of information scarcity and complexity. Our findings highlight the role of information costs in information choice theory and pose new challenges to alternative theories.

Quantifying The Maturity Risk Premium: Insight From Machine Learning 

Forward rates implied by the yield curve often deviate from reality because yields price in a premium for maturity risk. I use machine learning to create an ex-ante measure of this premium. Then, by subtracting this premium from forward rates, I create a powerful interest rate forecast which has strong statistical and economic predictive power. Finally, I use the machine learning model to dig into the determinants of the maturity risk premium, and find that markets charge a higher premium when opportunity costs are high, and when faced with higher uncertainty. This heterogeneity provides strong evidence against the weak expectations hypothesis’s assumption that the maturity risk premium is time invariant.