Has accounting information lost relevance in the Brazilian capital market? Three decades of evidence across different measurement scales

Abstract

This study examines whether the findings of Barth, Li, and McClure (2023) replicate on the B3 over a 31-year period. The research investigates whether the observed trend reflects the true value relevance of accounting information or is merely an artifact of scaling measures on a per-share basis. To test this hypothesis, a sample of 7,884 non-financial firm-years was analyzed. Share price was modeled using a tree bagging algorithm incorporating 16 accounting items and ten industry indicators. Value relevance was measured by out-of-bag R² and the reduction in this metric when items and blocks were reassigned relative to an exchangeable null distribution. The results reveal that the per-share R² in Brazil declines by 1.31 percentage points per year, moving in the opposite direction of the pattern observed in the United States. This decline, however, does not prove a loss of accounting value relevance, as it stems from the mechanical effect of dividing by the number of shares. Reverse stock splits, concentrated among low-priced firms, compressed price dispersion, whereas the dispersion of log market value remained stable. When per-share deflation is removed, the average R² reaches 73% with no statistically detectable downward trend. On the other hand, the metric becomes heavily dependent on firm size and allows for a potential decline of up to 17.5 percentage points over 30 years. Alternative deflators yield ambiguous results. Deflating by total assets, for instance, shows a localized jump in 2010. The evolution of accounting value relevance in Brazil remains inconclusive. No loss of earnings weight was detected, the increase in the importance of other financial groups proved fragile, and only three tests, all focused on the balance sheet, were distinguishable from noise after correcting for multiple hypothesis testing. As its primary contributions, this work delivers the first replication of this research design in Brazil, demonstrates that per-share scaling distorts R² comparisons over time, and provides a comparison between commercial databases and open-source data.