Recent studies have shown that composite forecasting produces superior forecasts when compared to individual forecasts. This paper extends the existing literature by employing linear constraints and robust regression techniques in composite model building. Security analysts forecasts may be improved when combined with time series forecasts for a diversified sample of 261 firms with a 1980-1982 post-sample estimation period. The mean square error of analyst forecasts may be reduced by combining analyst and univariate time series model forecasts in constrained and unconstrained ordinary least squares regression models. These reductions are very interesting when one finds that the univariate time series model forecasts d o not substantially deviate from those produced by ARlMA (0,1,1) processes. Moreover, security analysts' forecast errors may be significantly reduced when constrained and unconstrained robust regression analyses are employed.
KEY WORDS Biased regression Time series Portfolio analysis Security analysis
The majority of the literature supports the conclusion that earnings forecasts prepared by security analysts are more accurate than time series model forecasts (Fried and Givoly, 1982; Armstrong, 1983); however, not all of the economic studies have supported the forecasting efficiency of analysts (Cragg and Malkiel, 1968; Elton and Gruber, 1972; Guerard, 1987). The purpose of this study is to develop models combining analyst and time series forecasts to more effectively forecast corporate earnings. Fried and Givoly (1982) used a linear correction technique and discredited such composite modeling; however, Guerard (1 987) addressed multicollinearity and composite model building using ridge regression and found support for including both analyst and time series forecasts. The majority of researchers have analyzed the annual earnings generating process and found that a random walk with a first-order moving average operator best describes the series (Albrecht, Lookabill and McKeown, 1977; Watts and Leftwich, 1977; Ball and Watts, 1979). Composite models are estimated using ordinary least squares and robust-weighting schemes. Furthermore, the issue of constrained and unconstrained schemes in estimating the regression models is addressed. Granger and Ramanathan (1984) proposed a method of combining forecasts with no restrictions on the weights. Moreover, a constant term was estimated. The unrestricted weighting scheme of combining forecasts with a constant term produces an unbiased forecast and will produce the lowest estimated mean square error. Clemen (1 986), estimating combined models