Selecting an Accounting-based Valuation Model

We evaluate the relative valuation accuracy of six accounting-based valuation models. We begin by demonstrating that there is no single best model. The model with the lowest mean unsigned valuation error does not have the lowest median unsigned valuation error; and, the model with the lowest mean (median) unsigned valuation error is the most accurate for only 19.56 (14.22) percent of our sample observations. In light of these facts, we evaluate whether firm-level attributes can be used to identify the most accurate model ex ante. Using fitted probabilities from a set of ordered logistic regressions we are able to identify the most accurate model for 40.10 percent of the observations in our sample. Moreover, our model-selection algorithm is superior to several alternative algorithms. We also provide evidence regarding the factors underlying the firm-level attributes that we use to predict relative accuracy; and, we show that a lifecycle factor has strong predictive power.