Predicting Earnings With Sub-Entity Data - Some Further Evidence

In a prior article appearing in this journal, Kinney [1971] presents some preliminary evidence concerning the relative predictive ability of segment versus consolidated data in estimating future total-entity earnings of diversified companies. His study was motivated by the continuing debate about whether segment revenue and profitability figures are useful in assessing the future earnings prospects of multiproduct firms. There is some basis for questioning the reliability and usefulness of such data due to the inconsistencies across firms in defining segments, differences in intersegment transfer pricing policies, and many arbitrary cost allocations. Analyzing twenty-six firms that disclosed segment revenue and profit data in their annual reports, Kinney found that segment-based predictions of 1968 and 1969 consolidated earnings had significantly smaller average absolute prediction errors than did predictions based on historical consolidated earnings figures. In summarizing his findings, however, Kinney warned against generalizing the results to the total population of multisegment firms since his sample was limited to a relatively small group of companies that voluntarily disclosed segment data. He cautions that there may be some characteristic peculiar to the reporting firms which explains their willingness to disclose voluntarily their segments' earnings. The generality of Kinney's findings is also limited by the fact that only two forecasting procedures were employed in conjunction with the consolidated data. These models were chosen with no explicit reference to empirical research concerning the time series properties of accounting income (Beaver [1970], and Ball and Watts [1972]). Market-association studies (Ball and