چکیده:
In general, energy prices, such as those of crude oil, are affected by deterministic events such as seasonal changes as well as nondeterministic events such as geopolitical events. It is the nondeterministic events which cause the prices to vary randomly and makes price prediction a difficult task. One could argue that these random changes act like noise which effects the deterministic variations in prices. In this paper, we employ the wavelet transform as a tool for
smoothing and minimizing the noise presented in crude oil prices, and then investigate the effect of wavelet smoothing on oil price forecasting while using the GMDH neural network as the forecasting model . Furthermore, the Generalized Auto-Regressive Conditional Hetroscedasticity model is used for capturing time varying variance of crude oil price. In order to evaluate the proposed hybrid model, we employ crude oil spot price of New York and Los Angles markets . Results reveal that the prediction performance improves by more than 40% when the effect of noise is minimized and variance is captured by Auto-Regressive Conditional Hetroscedasticity model.
خلاصه ماشینی:
"Forecasting Crude Oil Prices: A Hybrid Model Based on Wavelet Transforms and Neural Networks Nafiseh Behradmehr1, Mehdi Ahrari2 Received: 5/12/2012 Accepted: 27/1/2014 Abstract In general, energy prices, such as those of crude oil, are affected by deterministic events such as seasonal changes as well as non- deterministic events such as geopolitical events.
Crude Oil Price Forecasting; Group Method of Data Handling (GMDH) Neural Networks; Wavelet Transform; Generalized Auto-Regressive Conditional Hetroscedasticity 1.
Naseri and Ahmadi (2006) use a hybrid artificial intelligence model for monthly crude oil price forecasting by means of feed forward neural networks, genetic algorithm and k – means clustering and they reveal that this framework is so effective.
In this way, the coefficients of each quadratic function Gi are obtained to optimally fit the output in the whole set of input–output data pair, that is: (رجوع شود به تصویر صفحه) In the basic form of the GMDH algorithm, all the possibilities of two independent variables out of total n input variables are taken in order to construct the regression polynomial in the form of Equation (12) that best fits the dependent observations / in a least-squares sense.
0052 We have studied 5 cases, based on the inputs provided to the GMDH neural network: Case one: 5 and 50 days moving average, obtained from non-smoothed oil price data sets.
We observe that removing noise form the data set results in better forecasting performance, as evident from column 2 of Table 2, in which the RMSE indicator for both New York and Los Angeles oil prices falls with respect to the benchmark model."