Parameter heterogeneity in the market-accounting relation: The need for dynamic firm-level modelling in capital market research

In this paper firm parameter heterogeneity in cross section regression analysis in capital market research (CMR) is investigated. Using panel data for 30 large US firms over the period 1955 to 2004, a well-specified common form of dynamic model for each firm is identified. Average parameter estimates from these models are compared to average parameter estimates from 50 annual cross section models having the same functional form. The dynamic parameters are mostly stable over time but variation in individual firm parameters is apparent. Analysis shows that even wellspecified annual cross section models using large samples of data cannot guarantee valid and reliable estimates of the parameters of interest. Firm-level dynamic analysis is necessary to avoid this problem. We show how a fixed effects panel analysis of the sample data can be used to approximate the average data generating process of the firms in the sample. Although the impact of accounting variables is slight, compared to the autoregressive component in market value, it is systematic. There is weak evidence of cointegration between market and accounting data in most firms in the sample. Consequently, it is possible to construct the cross section analogue of the dynamic error correction model. Book value of net assets is used to illustrate the role of accounting variables. Other variables could be used, but single variable, multiplicative noise models perform best when judged by joint explanatory power and forecast ability criteria.