Without imposing an ex ante functional form, the study re-examines the established decreasing trend in accounting value relevance to determine whether it reflects a genuine decline or a methodological bias. It compares the data-driven relationship between earnings and market value with the theoretical predictions of the dynamic real option model. Further, we hypothesize various consequences of this model for the value-relevance trend.
We employ random forest models on 7,884 firm-year observations from 657 companies traded on the Bombay Stock Exchange over 2012 and 2023. This nonparametric method lets the data determine the functional form and eliminates overfitting through out-of-bag predictions.
Random forest outperforms linear models. Aggregate value relevance has not declined; it exhibits an upward trend, mainly due to the rising relevance of book equity. The previously documented decline stems from shifts in sample composition and restrictive linear specification. The empirical earnings–price relationship is consistent with the dynamic real option model for positive earnings, and the hypotheses are consistent with this valuation framework.
We apply machine learning to address the limitations of linear models and integrate dynamic real option theory to explain changes in value relevance over time in an emerging market.
