單變量時間序列模型常能產生較為準確的長期能源需求預測(許志義等，1997)， 但因未納入其他因子與能源消費的互動關係，故無法考量可預見的經濟衝擊對能源需求的影 響，這些衝擊包括了能源危機、產業結構變動、環保法規和節約能源政策的施行等，致使長 期能源需求預測的可信度偏低。為兼顧預測的準確度及瞭解長期能源需求受經濟衝擊的影響 ， 本文採用我國煤、油、氣、電四大類能源消費量、GDP、工業生產指數，和能源價格指數 等資料，先以指數平滑法預測最終能源的長期需求，再以因果關係檢定能源和相關變數的領 先、落後關係，最後，建構一多變量向量自我迴歸 (vector autoregressive, VAR) 模型， 來進行能源消費的情境分析 (scenario analysis)，即所謂的政策模擬。並以模擬所得的數 據來調整單變量模型的預測結果，同時顯示有效的政策變數，供決策者參考。
More accurate forecasts of long-term energy demand are usually provided by univariate time series methods (Hsu et al., 1997). In this case, impact of foreseeable economic events on long-term energy demand is not able to be analyzed since the relationships between energy demand and other variables are unknown. Important economic incidents include energy shortage, industrial structure change, enforcement of environmental regulations and energy-conservation polices, etc. These incidents would make long-term forecasting of energy demand very unreliable. To solve the problem, this article first employs exponential smoothing methods to generate basic forecasts of Taiwan's final demand for coal, oil, natural gas and electricity. Then, the lead-lag relationships between energy demand and three policy simulation variables--GDP, industrial production index, and energy price index, are examined using robust causality tests. A vector autoregressive (VAR) model is built thereafter to perform scenario analyses. Finally, basic energy demand forecasts are adjusted according to the simulation results. Useful policy variables as well as the magnitude of economic impact on long-term energy demand are recealed in this study. The information should be valuable for policy makers.
