Zhang, Yunyi (2019) Forecasting real crude oil prices, their uncertainties and a Bayesian structural method for the world crude oil market. PhD thesis, University of Warwick.

Preview

PDF WRAP_Theses_Zhang_2019.pdf - Submitted Version - Requires a PDF viewer. Download (7Mb) | Preview

Request Changes to record.

Abstract

This thesis comprises three essays focusing on real crude oil price forecasting and structural analysis. The first essay (Chapter 2) begins by broadly reproducing Baumeister & Kilian's (2015) main economic findings, where an equal-weight combination of six econometric models outperforms a recursive mean squared predictor error weight based combination in oil-price point forecasting. The six models are an unrestricted global oil market vector autoregression, a commodity-price model, an oil-futures-spread model, a gasoline-spread model, a time-varying parameter product-spread model, and a random-walk model. I use their preferred measures of the real oil price and similar real-time variables. Remaining mindful of the importance of Brent crude oil as a global price benchmark and the divergence in oil price measures since 2010, I extend consideration to the North Sea-based measure and update the evaluation sample to 2016:12, finding that the combined forecasts for Brent crude oil are as accurate as the forecasts for other oil price measures. The extended sample employing the oil price measures adopted by Baumeister & Kilian (2015) yields similar results to those reported in their paper.

The second essay (Chapter 3) uses a Bayesian vector autoregression (BVAR) utilising time-varying parameters and stochastic volatility modelling time variation in forecasting real crude oil prices. An unrestricted global oil market vector autoregression and the equal-weight combination in Baumeister & Kilian (2015) are benchmarks. I extend the evaluation for model comparison purposes from standard statistical terms of point and density forecasts to an economic evaluation based xii on which specification would be more profitable in the crude oil futures market, and the forecast likelihood of extreme high and low real crude oil prices. For the same evaluation period as in Chapter 2, 1992:01{2016:12, the empirical results offer strong support for models using stochastic volatility in real crude oil price density forecasts relative to conventional VAR. Restricting time-varying parameters and allowing stochastic volatility can increase the probability of positive excess returns through utilising daily crude oil futures data, and can improve the calibration of the extreme high and low real oil price events forecasting. In conclusion, adding stochastic volatility and using the stochastic model specification search shrinkage prior of Eisenstat et al. (2016) are both important in ensuring reliable forecasts.

Finally, in the third essay (Chapter 4) I develop a parallel Metropolis{ Hastings (MH) algorithm to identify and compute Bayesian structural vector autoregressions (SVARs), which I refer to C-BSVARs. The motivation for this is the inefficiency of the traditional computation method for SVARs under certain types of identification. C-BSVARs extend Baumeister & Hamilton's (2015) method from only using sign restrictions to a broader set of identification assumptions and improve the computational efficiency relative to the traditional method for SVARs. Two specifications from the world crude oil market modelling are used to illustrate this. The first employs Kilian & Murphy's (2014) set of identification restrictions, while the second imposes an additional restriction on top of theirs | the uncertainty of a lower-bound on `the short-run oil demand elasticity for use'. C-BSVARs dramatically improve the acceptance rate for models deemed as admissible relative to the method used in Kilian & Murphy (2014), and it can narrow the critical intervals of Kilian & Murphy's (2014) impulse response functions. The additional restriction in the second specification enables precise estimates of oil demand elasticities, which is of importance for deciding the existence of crude oil price endogeneity and the relative weights of oil demand and supply shocks for driving the fluctuation of crude oil prices. To my knowledge, only the C-BSVARs approach in the existing SVARs literature is able to impose the restriction of uncertainties for elasticities, thereby providing a novel way of identifying key structural parameters that are non-linear.

Item Type:	Thesis (PhD)
Subjects:	H Social Sciences > HB Economic Theory H Social Sciences > HD Industries. Land use. Labor H Social Sciences > HG Finance
Library of Congress Subject Headings (LCSH):	Petroleum -- Prices, Petroleum -- Prices -- Econometric models, Economic forecasting, Prices -- Econometric models, Autoregression (Statistics)
Official Date:	September 2019
Dates:	Date Event September 2019 UNSPECIFIED
Institution:	University of Warwick
Theses Department:	Warwick Business School
Thesis Type:	PhD
Publication Status:	Unpublished
Supervisor(s)/Advisor:	Garratt, Anthony ; Vahey, Shaun P.
Format of File:	pdf
Extent:	xiii, 190 leaves : illustrations
Language:	eng

Request changes or add full text files to a record

Repository staff actions (login required)

View Item

Downloads