The Econometrics of Energy Demand : identification and Forecast.

Summary The prevention of climate change is one of the priorities of the global energy policy, which aims to massively reduce greenhouse gas emissions. Faced with these challenges, it is striking that our knowledge of energy demand modeling remains imperfect because it is based largely on old empirical work and methodologies that are now outdated. The scientific objective of this thesis is twofold: to quantitatively analyze the economic determinants of energy demand and to develop new forecasting models. This thesis is structured in four chapters. The first chapter shows that natural gas consumption in France can be predicted with a simple model using only the information available to market participants. This chapter proves the existence of a long-term relationship between natural gas demand and the prices of other energies and provides estimates of their marginal impacts on observed demand levels. The second chapter empirically investigates the role of temperature in forecasting gas prices in the United States. It develops a methodology for constructing a new monthly temperature-based index. This index captures changes in the residual demand for natural gas in real time. It is used as an additional exogenous variable in structural VAR models to improve forecasts . and we show that these predictive models derived from structural models are improved by relying on real time data (not subject to revision). The third chapter proposes to use, in the case of oil, a structural model capturing expectations using non-causal VARs and to correctly identify the reactions of key oil variables to a news shock. The fourth chapter re-examines the predictive power of the convenience yield of oil and gas prices by incorporating expectations into an empirical specification, using a non-causal VAR based on storage theory that provides very competitive price forecasts in a simple bivariate framework.