国际石油价格预测模型研究

【作者】 邓美玲；

【导师】 李小明；

【作者基本信息】 西南石油大学 ， 应用数学， 2009， 硕士

【摘要】 石油是现代工业最基本的原材料和整个经济发展的生命线。石油价格变动与经济起落息息相关,历史上的三次石油危机,导致世界经济全面衰退和多种形式的社会危机。近几年来的国际油价持续大幅度上涨和高位震荡,也已对各国经济增长带来多方面影响。2008年7月油价的暴涨暴跌令政府、企业、专家、投资者瞠目结舌,无所适从。石油价格冲击降低石油进口国的经济增长率,甚至降低绝对产出水平造成经济衰退;抬高物价水平,并可能引起通货膨胀。可见,研究油价发展的规律性,并用这个规律动态地预测未来的国际油价意义非常。本文首先研究了国际油价的背景、常用的预测方法,分析归纳了国际油价的五大主要影响因素。在此基础上对油价序列进行了奇异性分析,将时间段分为两种类型:含奇异点时段;不含奇异点时段。分别在不同类型的时段对延迟因变量回归模型、(AR-R)双回归模型、(AR)神经网络模型、滞后一期神经网络模型进行了预测效果评价,经过比较得出:无论哪种类型的时段下,模型简单、效果也比较理想的是滞后一期神经网络模型。其中,(AR-R)双回归模型是通过对传统的多元回归模型进行改进得到的新模型,通过具体算例证实改进的(AR-R)双回归模型效果优于传统的多元回归模型。另外,本文利用SAS和MATLAB编写了模型的拟合、预测程序,完成了对实例的分析和研究工作,并通过实证研究及结果对比分析论证了以下模型的可行性,为寻求效果更理想的预测方法做出了有益的尝试。1.延迟因变量回归模型。它是非平稳序列随机分析的一种方法,所需数据量少,操作简单易行,易于对模型进行直观解释,通过两种时段对几种一元非平稳序列随机分析方法的比较得出它是其中拟合效果最好的方法。2.(AR-R)双回归模型。方法的改进之处在于预测因子的决定上。在预测因子中加入了因变量的自身历史数据,与当期因变量自相关系数最大的几个保留;在外在影响因素因子的选择上,保留与因变量运动同步的因子。3.(AR)神经网络模型。与传统神经网络不同之处在于预测因子是外在影响因素(与油价同期的)及油价自身的历史数据,比较适于不含奇异点时段的拟合预测。4.滞后一期神经网络模型。这种模型体现了其它因素对经济环境变化的敏感早于油价,它对油价的涨跌预测的极为准确,并通过具体算例证实这种模型比较适合油价的拟合预测。最后,得出结论:作拟合预测时最好分时段进行,在含奇异点时段时最好不要纳入内生变量作为预测因子,效果比较好;而在不含奇异点时段时最好纳入内生变量作为预测因子,效果比较好;无论哪种类型的时段,就预测的前兆性及数据的采集难度而言,滞后一期神经网络效果最好,其误差也可以接受。就预测的精度而言(AR)神经网络模型在不含奇异点时段效果最好,传统神经网络模型在含奇异点时段效果最好;就模型的解释能力而言,(AR-R)双回归模型在不含奇异点时段效果最好;上述模型都可以判断出油价的涨跌,模型简单、效果也比较理想的是滞后一期神经网络。这些模型各有优劣,具体使用哪种模型作预测,要根据实际情况,具体情况具体对待。当然也可以同时使用上述几种模型进行预测,对结果进行综合分析。更多还原

【Abstract】 Oil is the most basically raw material of modern industry and the lifeline of entire economical development. The fluctuation of the oil price and the change of the economics are closely linked, the three historical oil shocks causes comprehensive decline of world economics and all kinds of social crises. The international oil price in the last few years continues to rise and it shocks in the largest scale, as well as it has brought various influences to economic growth of various countries. The oil price rises suddenly and falls suddenly in July, 2008, it causes that the government, the enterprise, the expert and the investor are all at a loss. The impact of the oil price reduces the economic growth rate of the oil importing countries, and it even reduces the absolute output level that it causes the economic recession; it drives up prices level and it possibly causes the inflation. Obviously, we research the development’s regularity of the oil price, and forecast dynamically the future international oil price with this rule. Its significance will be unusual.First the background of the international oil price, the common forecasting technique are have studied, and five major influencing factor about the international oil price are have induced in this article. And the strangeness analysis to the oil price sequence is carried on, the time section is divided into two types: time interval of including singular point; time interval of not including singular point. It is evaluated about the forecast effect of delay dependent variable regressive model, (AR-R) double-regressive model, (AR) neural network model, lag 1 step neural network model in the different time interval, which the model is simple and its effect is also ideal is lag 1 step neural network model with the two time intervals. And, (AR-R) double-regressive model is a new improved model on the traditional multiple regressive model, it is confirmed that the effect of (AR-R)-double regressive model surpasses the effect of traditional multiple regressive model by concrete example.In addition, the model’s fitting and forecasting procedure is compiled with SAS and MATLAB in this article, the analysis and the research work about the example has been completed, and the feasibility of the following model has been proved by the empirical study and the contrastive analysis on their results, the beneficial attempt to seek more ideal forecasting technique is made.1. Delay dependent variable regressive model. It is a random analysis method of nonstationary series, the data quantity is few, its operation is easy and feasible, the explanation of the model is easy and its effect is the best in the several monad nonstationary series in the two time intervals by comparison.2. (AR-R) double-regressive model. The improvement of the method lies in the decision of the forecasting factor. It has been joined in the historical data of the dependent variable on the forecasting factor, the several delay dependent variable series which coefficient of autocorrelation is the biggest retain; the factor which his movement is at the same time with dependent variable in the choice of external influencing factor retains.3. (AR) neural network model. The difference between it and the traditional neural network is that it embodys the inertia of the sequence’s selfmovement, it is more suitable to forecast on the not including singular point time interval.4. Lag 1 step neural network model. This model manifested that other factors are earlier sensitively to the change of the economic environment than the oil price, the rise and drop of oil price is forecasted extremely accurate, and this model is quite suitable to the forecast of oil price is confirmed by the concrete example.Finally, the conclusion is drawn as follows:The fitting and forecasting is made by different time interval is the best, the effect is the best when dependent variable factor should better not to be as independent variable factor on including singular point time interval and but the effect is the best when dependent variable factor should better to be as independent variable factor on not including singular point time interval. The effect of lag 1 step neural network is the best and its error may also is accepted as to the omen of the forecasting and the difficulty of the data gathering on the two different time interval.The effect of (AR) neural network model is the best on not including singular point time interval, the effect of tradition neural network model is the best on including singular point time interval, speaking of the forecasting precision; The effect of (AR-R) double-regressive model is the best on not including singular point time interval, speaking of explanatory ability of the model; The rise and drop of oil price may be judged by the above model, which the model is simple and the effect is also ideal is lag 1 step neural network.Each model has different merit and shortcoming respectively, which kind of model is used to make forecasting that must be according to the situation or generalized analysis is made according to their results.更多还原