Austin Chapter of the American Statistical Association
Anthony (Andy) Waclawski
In corporate America a forecast is often required whenever a decision is made. This is especially true for decisions regarding management of large international portfolios. Philosophically, the ultimate purpose of describing past return on investment is to acquire baseline knowledge that will enable one to predict future performance. Data of potential value in the formulation of policy frequently occur in the form of time series. Questions of the following kind often arise: "Given a known intervention (i.e. change in the investment portfolio), is there evidence that change in the series of the kind actually expected actually occurred, and, if so what can be said of the nature and magnitude of the change?" From one perspective, one can argue that it is highly questionable that past time periods are statistically valid random sample's of all immediately succeeding future time periods. Although we may be able to precisely describe the past pattern of variation in our criterion variable there is absolutely no theoretical assurance that it will occur in the future, if at all. Fortunately, empirical investigation of time series data suggests that future events are often a reflection of historical trends. The Central Theorem holds that all physical phenomena regress towards their arithmetic mean over time and is often cited as the theoretical construct that allows one to employ stochastic variables to successfully forecast the future. Researchers have found that fundamental economic factors can be used to forecast security returns. However, what factors to include and how to model the relationship remain open questions. Financial economists have carefully selected and tested a small set of variables suggested by economic theory. At the other extreme Morillo and Pohlman (2002) forecasted equity market returns by applying the dynamic factor model of Stock and Watson (1998) to large set of macroeconomic variables. In this article we apply the latest data mining techniques to forecasting equity market returns. The results are economically significant.