The relationship between stock market and macroeconomic variables.

摘要

Financial theory suggests that stock returns are predictable from macroeconomic variables. However, in much of the existing empirical literature this predictability is examined using linear models. In recent years some researchers have adopted the smooth transition autoregressive (STAR) model but their findings only show marginal improvements, if any, in forecasting stock returns.;This thesis contributes to the literature by presenting further evidence of nonlinearity in the stock returns-macroeconomic variables relationship. A set of macro variables that Granger-cause stock returns in a Vector Autoregressive (VAR) model is first identified. Using these macro variables (anticipated real output growth and past monetary base growth), two nonparametric techniques - Generalized Additive Model (GAM) and Multivariate Adaptive Regression Splines (MARS) - are investigated with the objective to improve the goodness of fit.;It is found that both GAM and MARS explain the in-sample stock returns better than the linear model and that MARS also provides more accurate out-of-sample forecasts than the linear model. Another finding is that the expected GDP growth calculated from ARIMA rolling forecasts has no significant influence on stock returns, while unexpected GDP growth has a significantly positive effect on current stock returns. Finally, the VAR estimation on monthly absolute volatility of stock market and macroeconomic conditions suggests a significant and positive relationship between stock market volatility and monetary growth volatility.