KTH Matematik |

This paper is an econometric study of five publically attainable macroeconomic and market variables from the US market. The aim is to study the behaviour and dependencies of GDP growth, inflation, yield curve, equity indices and foreign exchange rates in order to determine a forecasting model with horizon of one year. Such a model could be used to study possible future macroeconomic scenarios as well as to increase the understanding of how these variables react one to another. Very little is assumed of the series prior to modelling and relations predicted by econometric theory are expected to reveal themselves through the process of data analysis. The forecasting model is kept as simple as possible and the complexity of a model is only increased if it results in significant improvements of forecasting accuracy. Univariate ARMA models as well as multivariate VAR models that allow for international variable dependencies are tested, assuming either normal, t(3) or GARCH-gauss distribution of error terms. Models are ranked by comparing RMSE forecasting values and error terms behaviour. Results indicate that it is hard to find models that outperform the random walk although diagnostic tests indicate that many series share similar patterns of historical movements. It is therefore of interest to study further whether the models can be improved by identifying a common cointegrating vector. |

Sidansvarig: Filip Lindskog Uppdaterad: 25/02-2009 |