Can the forecast of the cotton price be improved using a model based upon economic variables?

Abstract: The purpose of this thesis is to find a model, which is based on economic variables that can forecast the cotton price better than commonly used benchmark models. A vector error correction model is used because of the existence of non-stationary variables and one cointegration relation in the data. Two types of forecasting methods are used for out-of sample predictions. The dynamic forecasting in this thesis is used to predict the cotton price six days ahead and the static forecast only predicts one day ahead. Three different types of estimation windows are used to see which gives the best forecasting results. The residuals are then used to calculate the root mean squared error, RMSE, enabling the comparison with random walks and autoregressive processes. The static forecasts did result in significant better forecasts than the benchmark models while the dynamic forecasts did not produce significantly better nor worse results than the benchmark models. Including economic variables when predicting the cotton price only significantly improves static forecasts of one-day ahead predictions. A sign prediction test was conducted in order to test the static method for one-day speculative purposes, and significant results were found.