Evolução do value relevance: uma abordagem Bayesiana

Abstract

Value relevance are lines of research that consider accounting measures capable of capturing relevant/useful information to explain stock prices. If the accounting measures cannot be shown to be relevant, it is difficult to argue that this information is necessary. As a general objective, the concern focused on investigating the effects of the evolution of the relevance of accounting information content over three decades, through the use of the Bayesian theory in Brazilian public companies. It was also verified: (i) the specific effects of IFRS adoption (before, during and after); (ii) the individual contribution of the twelve accounting measures; (iii) new economy (intensive use of intangibles and high growth rate); (iv) sector and (v) decades. In the methodology, the use of the same database to build the regression models (PL/LL and the complete model) was highlighted, using the two statistical theories, the Classical Theory and the Bayes Theory (using priori and log-distribution). normal), which allowed the comparison of model adjustments, in addition to the graphs of the PL/LL and complete model, verifying the trend of value relevance, with annual R² values. The sample used focused on Brazilian companies that are part of the Ibovespa index, covering the period from 1990 to 2020. Based on classic inferences, scientific research on value relevance raised the question whether accounting information would still be relevant. Considering our results, we emphasize that the use of Classical Theory brought very low R² values, with several results with less than 1% explanation, which coincided with some previous results found in the literature. We can highlight and compare the results with the use of Classical Theory and Bayes Theory that there was an increase in the explanatory power of R² in Bayesian inferences, highlighting in the PL/LL model the increase in R² was: 37% (1990-2019), 20% (new economy); 22% (sectors); 44% (decades) and 56% (IFRS), in the complete model showed an increase in R² of: 60% (decades) and 37% (IFRS). The results with the Bayesian inference brought evidence that allows us to conclude that there is relevance in the accounting information, as the R² values were medium and high, as opposed to the low values found with the Classical Theory. The use of regression models using the Bayesian theoretical input proved to be much superior, with more adequate adjustments, with higher estimates of the coefficient of determination and a greater number of significant independent variables. Considering more accounting measures as explanatory variables, in addition to PL and LL, the complete model proved to be more relevant, with greater explanatory power, making sense to consider other variables. The standardized values allowed the understanding of the individual contribution of each variable to explain the stock price in each regression model. We conclude that value relevance in Brazil is declining considering the period of three decades, when we consider the complete model there was a decline, but more attenuated than the PL/LL model, analyzing partial and total IFRS adoption we also found a decline. Checking the trend results with the support of the graphs, it can be seen that the points of drastic drop in the relevance of the accounting information content coincided with the periods of economic crises that affected Brazil. There was a contribution when analyzing the evolution of accounting relevance over three decades in Brazil considering more accounting measures (in addition to the traditional PL and LL) and contributed when analyzing the behavior of value relevance in the face of IFRS adoption, the practical contribution focuses on the results which can be sources of information for market analysts, regulatory bodies and standardsetters. There was a contribution in the construction of the posteriori to be used as a priori in future works on value relevance using a Bayesian approach. As a contribution of this study, the comparison of the results with the two statistical theories was highlighted, showing the superiority of the Bayes Theory, by bringing informational gain with higher values of estimates of R2 and more significant coefficients, thus proving the potential contribution of this theory to the large area of Applied Social Sciences.