1. Chi-squared Goodness-of-fit Tests for Regression Models
2. All coefficient signs remain the same, with E/P still significant, and the relative magnitude of the coefficients on E/P and Y in bivariate regressions remains the same. In other words, the longterm results are weaker, but hold up quite well without Japan. Of course, while it's always interesting to know the sensitivity of a regression to influential outlier points, the fact that Japan was selling for a very high relative P/E (low E/P) in 1987 and subsequently had terrible relative returns (despite the low starting bond yields favored by the Fed Model) is a pretty nice data point in the Traditional Model's favor! References Arnott;D Robert;Excluding Japan from my tests, the R-squared of the univariate E/P regression falls from 86.7% to 43.7% and the bivariate regression from 87.3% to 51.1%,2002