Forecasting The Spot Price In China Copper Market:Empirical Study Based On Time Series Models

This thesis is aimed at exploring methods to forecast the spot price in China copper market. We fit a Vector Autoregression (VAR) model and a Vector Error Correction (VEC) model to daily copper prices as well as two explanatory variables from September26,2007to March15,2013, and thereafter make some evaluation on the forecasting performance of each model. The empirical analysis can be divided into three parts.For the first part, we determine that two explanatory variables, the price and inventory of domestic copper futures, can help forecast copper price. We firstly find out nine possible explanatory variables and then get down to variable selection with methods including stationarity tests, cointegration tests, stepwise regression and Granger causality tests.For the next part, we fit a3-variable VAR model and a3-variable VEC model to the whole sample covering1080trading days. After several procedures including model specification, parameter estimation and model diagnose, we construct a VAR(6) model and a VEC(6) model where a sound goodness-of-fit is indicated and the white noise assumption of the residuals is confirmed. Furthermore, by methods of the Impulse Response Function and the variance decomposition, we obtain some information regarding the relationships among the three variables.For the last part of empirical analysis, we evaluate the forecasting performance to copper price of VAR(6) models and VEC(6) models by means ofrolling forecast, and show the results with daily, weekly and monthly data. At each forecast step, we compare a number of price forecasts with the corresponding actual prices. Both model display quite high accuracy rate at fewer forecast steps especially in overnight forecasting, and less accuracy at longer forecast steps or smaller sampling frequency. Comparatively speaking, VEC model is better at forecasting.