Content area
Full Text
This paper examines prediction errors in the general dividend discount model using a back-test method. The prediction errors are based on realized dividends, terminal stock prices, and estimates of time-varying discount rates. Models of varying lengths are examined in our tests. We include firms with a continuous record of dividend payments over the 20 years in our sample. We find that prediction errors vary from a high of approximately 55% to a low of 2% depending on the length of period used to calculate the estimated model. Reducing the number of dividends and the horizon date in the model reduces prediction error. Prediction error is also reduced by using time-varying equity discount rates when market volatility increases.
(ProQuest: ... denotes formulae omitted.)
In this paper, we conduct a detailed evaluation of the dividend discount model (DDM) using realized stock price and dividend information for a set of firms consistently paying dividends over the 1989 to 2008 period. This period is interesting since it corresponds with a significant run-up in stock prices followed by several bear markets occurring after 2001. We use a back-test method under the assumption of perfect foresight to examine this popular theoretical valuation model. Our method uses realized dividends and prices discounted with time-varying discount rates to calculate the intrinsic value. The discount period in models stretches from one to twenty years. The calculated values are compared to actual stock prices in order to calculate prediction errors for each firm in the sample. We recognize the generalized dividend discount model is based on expected dividends and expected future stock prices. However, comparing intrinsic value estimates to actual stock prices over time provides valuable insight into how the model may or may not perform under less ideal situations when dividends and a future stock price are unknown. The benefit of back-testing the model is that one source of estimation error is eliminated since we are not encumbered by the need to forecast future dividends, dividend growth or horizon values.
In an attempt to minimize reported prediction error, we estimate time-varying discount rates in all tests. Prior studies of the DDM including Hickman and Petry (1990), Penman and Sougiannis (1998), Foerster and Sapp (2005), and others rely on constant discount rates. In...