Study On Forecasting Of Gold Price Based On Varying-coefficient Regression Model

For a long period, gold has attracted much attention as a measure of value, circulation means, storage means, payment, world monetary functions and so on, how to predict the gold price becomes an important academic and practical research topic. For the research of gold price forecast, the multiple linear regression model has been greatly studied at present, as it assumes that the impacts of variables on the price of gold remain invariability throughout time, which obviously does not accord with the actual situation. To avoid this problem, varying-coefficient regression model is applied to predict the gold price in this paper, the model has dynamic response to the various variables influence the price of gold, so it has greatly improved the prediction accuracy.Many factors affect the price of gold and they are confused, the articles mainly explored correlation analysis in U.S. dollar index, commodity prices, European stock markets, the Asia-Pacific stock markets and world economic situation and other aspects of the factors affecting the gold price, and identify the main factors. Finally U.S. dollar index, oil prices, silver prices, DOW index, OECD leading index and the CRB index are chosen as main factors affecting the price of gold, and regard them as variable coefficient model parameters.In addition, the weighted least squares is adopted as an estimation of the parameters, corrects the traditional least squares method defect which assumes the sample data weights equal points to the prediction, making sample weights larger closer with prediction points. In the choice of weighting function, the paper uses cross validation, chooses the gold price data from January 1990 to December 2009 as the sample, selects different smoothing parameters, and calculates the corresponding residual sum of squares values and the smoothing parameter value corresponding minimum residual sum of squares, to achieve the optimization for the entire sample space. Then predicts gold prices from January 2000 to December 2009 used the variable coefficient regression model and multiple linear regression model to simulate, the analysis found that the residual sum of squares of varying-coefficient regression model is less than multiple linear regression model, and the latter's error rate is often higher than the former. So this article predicts the 12 months gold prices from January 2010 December 2010 applies varying-coefficient regression model. The results have much theoretical and practical value.