What Can Analysts Learn from Artificial Intelligence about Fundamental Analysis?

Abstract

We apply a machine learning algorithm to estimate Nissim and Penman’s (2001) structural framework that decomposes profitability into increasingly disaggregated profitability drivers. Our approach explicitly accommodates the non-linearities that precluded Nissim and Penman from estimating their framework. We find that out-of-sample profitability forecasts from our approach are generally more accurate than those of benchmark models. We use the profitability forecasts to estimate intrinsic values using the financial statement analysis design choices in Nissim and Penman’s framework and find that hypothetical investing strategies based on these value estimates generate risk-adjusted returns. Design choices that improve performance include increasingly granular disaggregation, a focus on core items, and long-horizon forecasts of operating performance. Perhaps surprisingly, we find only mixed evidence of benefits from incorporating historical financial statement information from beyond the current period.

Keywords: Financial Statement Analysis, Machine Learning, Earnings Forecasting

JEL Classification: C53, G10, M41