Constructing Generalized Exponential Predictors Via Penalty Methods

In the opening part of this paper, we first offered an overview of the sup-ply-demand condition, the role of gold in world economy and the references of gold spot price and future price around the world, then elaborated the importan ce of making predictions of gold prices. We summarized the works of gold pri ce forecasting at the end of the first part.In the second part, we gave a brief introduction of generalized exponenti-al predictors, the model that we'd used in this paper.We described the parameter selection methods used in our model. The me-thods include:(1)ridge regression;(2)adding LASSO and SCAD penalties based on L1, L2 and LM loss which combines both L1 and L2, to select linear comb-inations of EWMA predictors of different parameters. This is where we made innovations. In this part, we gave detailed descriptions of how we implement our methods and discussed the advantages of using these methods.We did empirical analysis in the fourth part. Practical data showed that o-ur models had improved the single parameter EWMA model effectively and th-ey performed better than the models suggested by literatures. In addition, we made some research to see if we can improve our prediction by changing so-me given factors of our model, and we found that these changes had exactly improved the prediction.In the last part, we gave the conclusions and suggested some issues that need further discussions.