| Popis: |
Commodity prices are very volatile which presents substantial challenges and uncertainty for the global economy. In 2014 oil prices experienced substantial decline due to a combination of economic and geo-political factors. Organization of the Petroleum Exporting Countries (OPEC) currently supplies a little over 30% of global oil market, a significant decline from its 50% market share back in the 1970s. OPEC’s market stabilisation power continues to erode, which, coupled with a lack of consensus between Open members, will continue to contribute to volatility of oil prices going forward. The traditional methods of oil price forecasting include supply/demand modelling, use of econometric time-series analysis and other tools. However, due to non-linear and complex behavior of oil prices, these methods showed poor historical performance even during periods of stable oil prices, like the 1980s (Behmiri and Manso 2013). Whilst some machine learning techniques have been previously applied to the forecasting of financial markets, this was mostly in the equity and foreign exchange domains. Limited literature exists on the application of machine-learning techniques to the commodity markets in general, including on forecasting of prices for such an important commodity such as oil. This thesis introduces machine learning techniques such as SVM regression to the task of predicting commodity prices and compares performance of machine learning algorithm with that of traditional techniques such as econometric models like ARIMA and GARCH. Working with continuous futures time series we develop, implement and back-test a real trading strategy using trading signals generated by both traditional econometric models and SVM regression. Our research suggests that there is significant potential in the application of advanced machine learning techniques to the forecasting of commodity prices and commodity trading/risk management. |
