Creative Commons

This dissertation attempts to investigate the potential causal relationships between, Google search queries, data for crude oil relative keywords, and certain international crude oil price benchmarks and their cumulative volatility. Additionally we extend our testing to two ‘artificial’ Energy sector volatility indexes provided by the FRED website. Our data consist of: West Texas Intermediate (WTI) Cushing, Oklahoma and Brent, Europe (spot prices), crude oil by-products price benchmarks: Conventional gasoline prices, New York harbor, Kerosene-Type Jet Fuel Prices, U.S. Gulf Coast and No. 2 Heating Oil Prices, New York Harbor. Volatility Indexes: CBOE Crude Oil ETF Volatility Index and CBOE Energy Sector ETF Volatility Index. All data were available through the F.R.E.D. website. We used weekly data for the sample period spanning from January 1, 2004 till October 13, 2013. For the creating of the Google Trends Index (GTI), co-integration methods (Engle Granger and Johansen co-integration tests) were used, in order to arrive, with the help of a regression equation, at a fitted equation, weighting in all the selected keywords. Additionally in order to extract volatility from all five examined crude price benchmarks, a GARCH(1.1) model was implemented. Upon the implementation of further tests, we examined all our variables for unit root existence, through many of the most known tests at hand, such as, ADF, DF-GLS, PP and KPSS. Granger Linear causality tests were then employed, between Google search series and the volatility extracted from all the crude oil related series. Furthermore two frequency domain non-linear causality tests [Breitung and Candelon, 2006 and Lemmens et al., iv 2008] were implemented as well, in order to identify the nature of causality. Firstly, we examined the linear causal relationships between our series and found a unidirectional causality running from the Google Trends Index towards the crude oil benchmarks (apart from WTI). Linear causality was also detected from GTI toward the Crude oil Volatility index, but not for the Energy sector Volatility index. Moreover we implemented nonlinear causality tests for all frequencies. As for the conclusions, our findings revealed the existence of unidirectional causal relationships in both short- and long-run between Google Trends Index and most of our crude oil benchmarks. Full spectrum causality revealed, for all the crude oil benchmarks, except perhaps for WTI. As far as the CBOE volatility indexes, the findings were identical to those of linear causality tests.