This study evaluates whether conditional biases in analysts’ earnings expectations are reflected in stock returns and post-earnings price dynamics in the Korean equity market. Building on Binsbergen, Han, and Lopez Lira (2023), a real time, statisti...
This study evaluates whether conditional biases in analysts’ earnings expectations are reflected in stock returns and post-earnings price dynamics in the Korean equity market. Building on Binsbergen, Han, and Lopez Lira (2023), a real time, statistically optimal and asymptotically unbiased benchmark for firms’ earnings expectations is constructed using a random forest model trained in rolling windows that combine public firm characteristics, macroeconomic variables, and analysts’ forecasts; Conditional bias is defined as the difference between analysts’ forecasts and the machine learning benchmark, scaled by previous-month closing price. This study finds that conditional bias serves as a powerful negative predictor of future returns, with an effect magnitude significantly larger than that documented in the U.S. market.
Using all firms listed on KOSPI and KOSDAQ from 2000 to 2024, three questions are addressed: (1) whether analysts’ conditional expectations in Korea exhibit horizon dependent optimism, (2) whether conditional bias forecasts the cross sectional returns after standard risk controls, and (3) whether post-earnings-announcement drift varies systematically with conditional bias. The results show that analyst expectations are, on average, upward biased, with the bias increasing in the forecast horizon. Moreover, conditional bias is a powerful negative predictor of future returns. Portfolios that are short high bias and long low-bias firms earn significant monthly spread of 1.62%. Bias drift–defined as a systematic, monotonic relation between conditional bias and post-announcement cumulative abnormal returns–is not supported in the Korean market, although conditional bias remains informative for the market response to extreme positive earnings surprise. These findings extend the external validity of the conditional bias framework beyond the U.S. setting and highlight the usefulness of the ML benchmark for studying the term structure of expectations, return predictability, and earnings-related price dynamics in Korea.