原油價格組合預測模型之建構

Abstract

本文試以多種組合預測方法，評估多個文獻中曾提及的預測變數對原油月均價、月底價的預測效果，使用資料期間為1986:M2至2015:M10。根據Working (1960)闡述月均價格變動的前後期將會有相關性，本研究導入一個新變數用於預測原油月均價，其結果顯著優於no-change預測，使MSPE降低幅度達39%。此外，以過去文獻提及的預測變數來預測原油月底價時，預測效果集中於特定期間，無法於整體評估期間保持預測能力。同時，文獻上計算MSPE時有兩種算法，在預測效果於期間內並非均勻分布的情況下，將可能使兩種算法所得的MSPE Ratio有所差異，且此MSPE Ratio差異的大小亦與評估期間的選取有關。

This article uses monthly data from 1986:M2 to 2016:M10 to examine the predictive power for both monthly-averaged oil prices and end-of-month oil prices by multiple forecast combination methods. According to Working (1960) , our work derive a new predictor of monthly-averaged oil prices which can significantly reduce the MSPE by 39% compare with no-change forecast. When forecasting the end-of-month oil prices, the predictive power only exists in a certain period, can't hold the power for the whole evaluation period. Our work explain the relationship between MSPE calculated with two different ways. Because the predictive power only exists in a certain period, there would be some different between MSPE Ratio of two kinds of MSPE calculation ways.This difference would be correlated with the choice of evaluation period.