INTRODUCTION TO COMPLEXITY SCIENCE (original) (raw)
Related papers
Assessing the Predictability of Oil Prices
SSRN, 2019
Oil prices are volatile. They fluctuate due to several demand and supply characteristics. Several macroeconomic factors, may be used to assess the direction of oil prices. However, the data on these variables are often annual, and cannot be used for short term forecasting. As a result, speculators and retail traders often rely upon econometric time series models to produce forecasts. Early models for univariate forecasting include, the Autoregressive Integrated Moving Average (ARIMA), and the Exponential Generalized Autoregressive Conditional Heterscedasticity (EGARCH). These models are often criticized for their linearity. Recent machine learning models have become popular in the forecasting discipline. In fact, the Artificial Neural Network (ANN), and the Wavelet Transform have been increasingly used for forecasting. This study uses the ARIMA, EGARCH, ANN, and Wavelet Transform (Daubechies level 2 order 3)-ARMA models to forecast oil prices. Data on oil prices over the Jan 02, 1986 to June 10, 2019 period is considered. To complement the analysis, fundamental analysis is also used to forecast the direction of oil prices. Surprisingly, the fundamentals, based on the US oil inventories seem to have a higher predictive accuracy than the aforementioned models.
2017
Our results show that over the two cycles that characterize the 2003-2016 period a significant change in the working of oil markets occurs. Our pricing investigation, based on a three-agent model (hedgers, fundamentalist speculators and chartists), find that from 2009 onwards traditional analysis of supply and demand forecasts, loses its explanatory power and hence its credibility. The sharp and unexpected fluctuations in oil prices, compounded by unpredictable political factors and technological break-troughs (e.g. tight sands/shale oil) strongly raises uncertainty and reduces the effectiveness of customary forecasting techniques.