Informativeness of Performance Measures: Coefficients or R-Squareds?

Measuring the informativeness of earnings is of fundamental importance to accounting research. Both coefficients and R-squareds have been proposed as candidates for measuring the informativeness of earnings. However, recent research has focused substantially more on using coefficients, rather than R-squareds, to draw inferences. This paper first documents in a small theoretical model that under some circumstances, R-squareds map more closely to informativeness than coefficients. Second, this paper documents that in archival data, coefficients and R-squareds can draw opposite inferences regarding the informativeness of earnings and other performance measures up to 50% of the time. Third, this paper proposes an approach to provide statistical inference using R-squareds. Taken together, this paper suggests that rather than solely relying on coefficients, as is common in prior literature, R-squareds can also be used to measure the informativeness of earnings and other performance measures.