Anmerkungen:
Nach Informationen von SSRN wurde die ursprüngliche Fassung des Dokuments May 10, 2023 erstellt
Beschreibung:
Accounting parameters such as earnings response coefficients (ERC) are generally heterogeneous across firms. When panel data is available, the parameters are typically estimated using OLS with either pooled data which ignores parameter heterogeneity, or using firm-specific observations which tends to give noisy estimates. An alternative is to use Bayesian hierarchical models which preserve parameter heterogeneity but have the advantage of being less noisy than firm-specific OLS. Their advantage stems from the use of data about all other firms to form an informative prior about each firm’s parameter. In this paper, using a sample of 301 firms we compare the results from three Bayesian hierarchical models to OLS-based firm-specific ERCs. Our results show that the Bayesian models produce ERCs that reduce the number of negative ERCs from 48 to 6 and lower mean squared error in a hold-out sample by more than 90%. By enabling the estimation of less noisy firm-specific parameters, these models allow researchers to gain deeper insights into firm-level differences and their determinants. We show the formulation and estimation of these models step by step, with a discussion of the Stata commands involved. The Stata as well as R code is available online. We conclude by discussing the potential for such models to be applied to other contexts in accounting research