| Stocks are frequently classified on the basis of industry, and investors generally expect that stocks within an industry sector, such as transportation, utilities, or electronics, will behave similarly over time. This thesis bases stock classification on stock performance. Using monthly returns generated by NYSE stocks over twenty years, cluster analysis is used to verify that well-defined groupings of stocks exist. These groups are shown to be more homogeneous in performance than standard industrial groupings. A mixture model for monthly stock returns is developed to classify stocks and characterize the behavior of the resulting stock groups. The stock groups and the parameters that are estimated for an autoregressive mixture model are shown to have predictive value. The predictions can be used to construct a series of portfolios that outperform a market portfolio proxy. The portfolios resulting from the autoregressive mixture model, however, do not significantly outperform momentum-based portfolios selected on the basis of past returns of individual stocks. The phenomenon of return continuation, or momentum, is shown to be present in our data, suggesting a revised model that includes moving average parameters. A version of a momentum investment strategy that utilizes forecasts derived from the improved mixture model outperforms the standard momentum investment strategies based on past returns for one-month holding periods. |
