Modelling and Forecasting of Commodity Trading Price

Abstract

This paper proposes and improves a model for China's cotton reserves trading market, generates a more accurate price level table in contrast to the widely used China Cotton Association(CCA)'s table, and predicts the future trading price with this model. Data used is all 29895 trade records of cotton reserves from May, 2016 to Sept, 2016. In this paper, we firstly give a briefly introduction to the data as well as basic knowledge of cotton, especially the CCA's price level table as the target to improve accuracy. We then present a multiple linear regression with dummy variable model, which can reduce the error on predicting the trading price of current month, but still remains some problems such as the result shows a batch of cotton with better quality may get a lower price. So we finally advance the model with multidimensional isotonic regression and more factors taken into consideration. Our last model could be used to predict the trade price of both current month and next month with approximately 4% mean absolute percentage error(MAPE).