Níveis de governança corporativa e desempenho: uma validação por meio de redes neurais artificiais

Abstract

Studies have long been conducted that analyze the impact that corporate governance has on corporate performance. However, there is still no convergence in the results about this relationship, which may be generating investor skepticism about the influence that governance would have on performance. It has been common in these surveys to construct a representative governance index to associate it with performance, which can lead to distorted results if the indicator created does not accurately reflect the market's perception of corporate governance. In this context, the brazilian stock market makes it possible to analyze the relationship between these factors without the need for subjective index creation, since the brazilian official stock exchange, BM&FBovespa, performs corporate segmentation at corporate governance levels, studies between corporate governance and performance based on objective data. However, a critique of the relationship between corporate governance and performance through BM&FBovespa governance levels concerns the fact that adherence to them is voluntary, so the results obtained could be distorted by disregarding companies that do in fact Corporate governance but did not request adherence to levels with the brazilian stock exchange. Therefore, in addition to analyzing the relationship between corporate governance and performance through levels of governance, this research carried out a new classification of corporate governance, so that companies not listed in the governance segment were included in any of the groups: traditional segment, Level 1 (N1), Level 2 (N2) or New Market (NM). To perform the classification, due to the non-linearity of the data, the artificial neural networks technique was used. The results showed that, during the period from 2001 to 2015, the entire stock market listing period, based on a sample of 572 companies and 8,580 observations, adherence to corporate governance levels had no positive relation with any performance variables analyzed: ROA, ROE, Ebitda / AT, Ebitda / PL and Q of Tobin. When comparing the results based on stock market classification and the neural network for the year 2015 results, there was a change in the statistical significance of the N1 and NGOV variables (which represented the three levels together). In relation to N2 and NM, there was no change in statistical significance. For future work, it is suggested to identify variables obtainable in databases and financial statements that may represent corporate governance, thus reducing the need to use hard-to-obtain data and subject to the company's sincerity in sending information.