Stock Prices, Earnings, and Expected Dividends

Long historical averages of real earnings help forecast present values of future real dividends. With aggregate U.S. stock market data (1871-1986), a vector-autoregressive forecast of the present value of future dividends is, for each year, roughly a weighted average of moving-average arnings and current real price, with between two thirds and three fourths of the weight on the earnings measure. We develop the implications of this for the present-value model of stock prices and for recent results that long-horizon stock returns are highly forecastable. IN THIS PAPER WE present estimates indicating that data on accounting earnings, when averaged over many years, help to predict the present value of future dividends. This result holds even when stock prices themselves are taken into account. The data are the real Standard and Poor Composite Index and associated dividend and earnings series 1871-1987. Our estimates indicate to what extent dividend-price ratios and returns on this index behave in accordance with simple present-value models, and allow us to shed new light on earlier claims that stock prices are too volatile to accord with such models (LeRoy and Porter [14], Shiller [20], Mankiw, Romer, and Shapiro [15], Campbell and Shiller [1, 2], and West [23]). It seems appropriate to consider earnings data for forecasting dividends, since earnings are constructed by accountants with the objective of helping people to evaluate the fundamental worth of a company. However the precise economic meaning of earnings data is not clearly defined; accounting definitions are complicated and change through time in ways that are not readily documented. Because of this, many studies of financial time series have avoided the use of earnings data and have thus omitted relevant information about fundamental value from the analysis.1 Our approach is to introduce earnings, measured either annually or as an average over a number of years, as an information variable in a vector-autoregressive (VAR) framework. Any errors in measurement in earnings are accounted for automatically by the estimation procedure, which allows earnings to enter the model only insofar as they are useful in forecasting. The VAR framework, * Princeton University and Yale University, respectively. An earlier version of this paper was presented at a joint session of the American Economic Association and the American Finance Association in Chicago on December 28, 1987. This research was supported by the National Science Foundation. 'There is a large accounting literature on the response of securities prices to earnings announcements; see Kormendi and Lipe [131 for a list of references. However, with a few exceptions, notably Kormendi and Lipe, this literature does not ask whether the response is consistent with a particular fundamental valuation model for the security price.