Assessing Linear Models of Value Relevance: Do They Capture What They Should?

This study aimed to investigate the quality and impact of value relevance models of financial information using quantile regression (QR) compared to the ordinary least squares (OLS) methods. Following the principles and foundations of Ohlson (1995), Feltham and Ohlson (1995) and Ohlson and Kim (2015), it was possible to use a comparison parameter between models for evaluating the relevance of accounting information. Therefore, we applied two tests (A and B), with two models each as in Ohlson and Kim (2015), one with the dependent variable as net income in the following period and, second, as company market value in the current period. Given this theme, quantile regression showed to be more efficient and have less possibilities for estimation errors than OLS, at least under the strict conditions of this work. Therefore, we recommend the estimation of quantile regression in models that use accounting and financial information, since heteroscedasticity and outliers are commonly found in these types of data, and because this estimation method is less sensitive and more robust to such conditions typically displayed by the data of this research field.