The effect of oil prices on stock prices: fresh evidence from asymmetric causality tests

ABSTRACT

This article investigates the causal impact of oil prices on stock prices in each G7 market as well as in the world market. An asymmetric causality test developed by Hatemi-J is used for this purpose. Since the underlying data appears to be non-normal with time-varying volatility, we use bootstrap simulations with leverage adjustments in order to produce more reliable critical values than the asymptotic ones. Based on symmetric causality tests, we find no causal effect of oil prices on the stock prices of the world market or any of the G7 countries. However, when we apply an asymmetric causality test, we find that increasing oil prices cause stock prices to rise in the world, the U.S. and Japan while decreasing oil prices cause stock prices to fall in Germany. This may imply that the world, the U.S. and Japanese stock markets consider increases in oil prices as an indicator of good news as this may mean that there is an increase in oil demand due to an expected growth in the economy while the German stock market treats decreasing oil prices as a signal of an expected contraction in the economy.

KEYWORDS:

• Oil prices
• stock markets
• G7
• asymmetric causality

JEL CLASSIFICATION:

• G15
• C30

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ The number of observations is 2027.

² For causality testing between negative cumulative shocks, the vector $X_t^{-}=(S_t^{-},O_t^{-})$ is used.

³ The VAR model was originally suggested by Sims (1980). For dealing with deterministic trend parts see Hatemi-J (2014) and Hatemi-J and El-Khatib (2016). For transforming the data with deterministic trend parts the software component produced by Hatemi-J and Mustafa (2016) can be utilized.

⁴ The HJC has been suggested by Hatemi-J (2003, 2008). The conducted Monte Carlo simulations by the mentioned author have shown clearly that the HJC is effective in finding the optimal lag order even when there is ARCH. This information criterion has also good forecasting capabilities based on the simulation results.

⁵ An extra lag that is unrestricted was included in the VAR model in order to account for the effect of one unit root as suggested by Toda and Yamamoto (1995).

⁶ The initial values are assumed to be available (see Lutkepohl, 2005).

⁷ Details on leverage adjustment are provided by Davison and Hinkley (1999) for univariate analysis and Hacker and Hatemi-J (2006, 2012) for multivariate analysis.

⁸ The unit root tests results are not presented to save space. The results are available on request however. The graphical illustrations of the variables are also in line with unit root process.

Additional information

Funding

This work was supported by The UAE National Research Fund via the UAE University: [Grant Number # 31B028].