Using forecasting criteria to identify value relevance in the relationship between accounting numbers and market value

We explain and demonstrate a disciplined and systematic approach to repeatable modelling using forecast criteria, in addition to the usual statistical estimation criteria, to identify value relevance in regressions of the market-accounting relation. The method was used in Cooke et al. (2009). It is illustrated here in the case of a single firm over a 59-year period. Market and accounting data for the U.S. firm Abbott Laboratories Inc. from 1955 are modelled using a testing-down, error correction approach. Hold-out samples of 10 to 15 years are used to assess forecasting performance relative to a random walk. Emphasis is placed upon the use of simple, directly observable and theory-independent model variables that can be replicated with other sample data. In this case, logarithmic transformations of all variables have to be computed in order to achieve correct statistical specification, implying a multiplicative relationship in the raw data. The strongest cointegrating accounting variable with forecasting ability for Abbott's market value is earnings. The model parameters exhibit long-run stability and the accounting regressor marginally improves forecasts of market value compared to a random walk, demonstrating ‘value relevance’.