Campbell R. Harvey,
Fuqua School of Business, Duke University, Durham, NC
National Bureau of Economic Research, Cambridge, MA
However, assignment #3 is a prediction or forecasting exercise. It is a statistical exercise rather than something linked directly to asset pricing theory. Any variable can be included in the regression - as long as it is *lagged*. In my research, I have found that some variables that are global (or based in the U.S.) capture some of the predictability in country returns. I have also found that some local information variables (lagged) are also important. Hence, I expect your regression to have both common variables (probably U.S. variables, like lagged U.S. term structure) and local variables (like lagged dividend yields).
"In Ferson and Harvey (1994), you find that the most important variable accounting for country returns is the world market return - however, when we put the lagged world market return in the prediction regression it is insignificant"
Just because something "explains" returns in a risk model, does *not* imply that it *predicts* returns when it is lagged in a forecasting model. For example, in the U.S. we know that a substantial part of what happens to any stock's return is related to what happened in the market - at the same time. That is, what happened in January to IBM is linked to what happened in January in the market as a whole. It is a completely different issue as to whether the market return in December *forecasts* what will happen for IBM's stock return in January. Indeed, it is extremely unlikely that December S&P has any explanatory power for January IBM. It is extremely unlikely that the MSCI world return in December can explain the return in the German market in January by the same argument. I know this is a bit confusing because it is new. Bear with me. There is a payoff.