STRICTLY FOR STATISTICIANS
Dave Chamness wrote his own stepwise multiple regression program for Compass. To produce the S&P
500 forecast, he uses all eight inputs. The most highly correlated individual parameters for stocks are
dividend yield and the rate of change of interest rates. Interestingly, Chamness was not able to establish
any positive correlations for any trend-following parameters on a monthly basis. Not that there weren't
any significant technical parameters; there were. However, they were more in the nature of an
overbought/oversold type of parameter. Specifically, the difference between the S&P 500 and a moving
average was found to be significant—that is, the correlation coefficient for that predictor was positive.
When the market is below its average, the program becomes more bullish. When the market is above its
average, the program becomes more bearish.
The actual correlation coefficients for the within-sample forecasts were as follows: