The lead of oil price rises on US equity market beliefs and preferences

• Published date: 01 November 2021
• Author: Jonathan Dark
• Date: 01 November 2021
• DOI: http://doi.org/10.1002/fut.22247

J Futures Markets. 2021;41:1861–1887. wileyonlinelibrary.com/journal/fut © 2021 Wiley Periodicals LLC

|

1861

Received: 15 February 2021

|

Revised: 24 June 2021

|

Accepted: 25 June 2021

DOI: 10.1002/fut.22247

RESEARCH ARTICLE

The lead of oil price rises on US equity market beliefs

and preferences

Jonathan Dark

Department of Finance, Faculty of

Business and Economics, University of

Melbourne, Carlton, Victoria, Australia

Correspondence

Jonathan Dark, Department of Finance,

Faculty of Business and Economics,

University of Melbourne, Level 12, The

Spot, 198 Berkeley St, Carlton, VIC 3010,

Australia.

Email: jdark@unimelb.edu.au

Abstract

We find that oil price rises from a strengthening global economy have a state‐

dependent lead on US equity market beliefs and preferences. When equity

volatility is low, rising oil prices lead higher cashflow expectations, lower long‐

run (LR) volatility, and increased LR risk aversion. When volatility is high,

markets focus on the contractionary effects of higher input costs, with rising

oil prices leading decreased cashflow expectations, higher LR volatility, and

decreased LR risk aversion. Findings suggest important refinements for asset

pricing, portfolio choice, and models that link financial markets to the

macroeconomy.

KEYWORDS

crude oil, risk‐neutral density, risk preferences, threshold regression

JEL CLASSIFICATION

C22, C30, G10, G13

1|INTRODUCTION

Decades of research have documented the significant effects of oil price shocks on the macroeconomy (Hamilton, 1983;

Hamilton, 2011; Hamilton & Herrera, 2004; Kilian, 2009; Kilian & Murphy, 2014). The macroeconomic outlook has

important implications for equity valuation, and so the relation between crude oil and equity markets has also been

well studied. Kilian and Park (2009), for example, show that oil market fundamentals are an important determinant of

US equity returns, with 22% of the long‐run (LR) variation in the US equity market driven by oil price shocks. Oil price

shocks also have significantly different impacts across industries, which has important implications for risk man-

agement and asset allocation. See Maghyereh et al. (2016) for a review.

Despite this significant body of research, the direction of the relation between oil price shocks and equity returns is

unclear. Evidence suggests it is either negative (Jones & Kaul, 1996; Sadorsky, 1999), insignificant (Chen et al., 1986;

Huang et al., 1996), or state dependent, switching between a positive and negative relation over time (Aloui &

Jammazi, 2009; Conrad et al., 2014; Jammazi & Aloui, 2010). The literature has focused on the relation between oil and

ex post equity returns and volatility. However the relation between oil and forward‐looking equity market beliefs

(expected volatility, skewness, and kurtosis) and preferences (or investor risk aversion) is not well understood.

We make a number of contributions to the literature. To our knowledge, we are the first paper to examine the lead of oil

prices on forward‐looking equity market beliefs and preferences. This includes an examination into the lead of oil prices on

expected higher‐order moments (skewness and kurtosis) as well as time‐varying risk preferences. Drawing on component

models that link macroeconomic fundamentals to LR volatility and correlation (Conrad et al., 2014;Engle&Rangel,2008;

Engle et al., 2013), we are also the first paper to consider the lead of oil prices on LR beliefs and preferences.

Previous oil research has paid little attention to skewness and kurtosis, despite their significance in asset pricing,

portfolio choice, risk management, and Value at Risk (Chung et al., 2006; Conrad et al., 2013; Jondeau & Rockinger,

2012). It has also failed to examine the relation between oil prices and preferences, which are a fundamental building

block of choice theory. Preferences have important implications for many individual decisions as well as macro-

economic outcomes in the labor market, saving and investment, total factor productivity, and political outcomes. See

Schildberg‐Hörisch (2018) for further discussion.

There are many ways to extract preferences. Constant risk preferences have been estimated via expected utility and

quadratic functions (Quiggin & Chambers, 2004), pricing kernels (PKs) (Grith et al., 2013), or ambiguity and un-

awareness models (Karni & Vierø, 2013). Time‐varying risk preference estimation adds considerable complexity, with

methods including consumption‐based asset pricing models (Campbell & Cochrane, 1999), time‐varying generalized

autoregressive conditional heteroscedasticity (GARCH) in mean models (Dias, 2017), or extraction via option prices

(Aıt‐Sahalia & Lo, 2000; Jackwerth, 2000; Rosenberg & Engle, 2002) and realized volatility (Bollerslev et al., 2011).

We examine the lead of oil prices on equity market beliefs and preferences using option prices for a number of

reasons. Higher‐order moments based on a time series of the underlying asset are sensitive to outliers, noisy, and more

susceptible to estimation error. They are also usually poor estimates of expected future higher moments, as accurate

estimates require a long time span. Option prices overcome these issues as a large number of prices are available cross‐

sectionally for a given underlying asset across strikes. This permits the construction of expected density functions for the

underlying asset that are less prone to estimation error. Multiple option maturity dates also mean expected densities and

preferences can be extracted for various horizons. This is not the case for many alternative methods based on underlying

asset returns. Risk preference estimation via GARCH in mean models, for example, typically employs daily data, and so

little if anything can be inferred about forward‐looking preferences beyond the data frequency employed. Further,

consumption‐based asset pricing models suffer from imprecise measurement of aggregate consumption, and time‐varying

methods based on the underlying asset require selection of state variables where there is little consensus.

The risk‐neutral distribution (RND) extracted from option prices combines beliefs about the probability of future

returns via the physical probability density function (PDF), with preferences towards those returns via the PK or

stochastic discount factor. The PK explains most of the nonnormality in the RND, and so much of the variation in

higher‐order RN moments is due to changing preferences (Han, 2007).

To allow for time‐varying preferences, we employ the method of Rosenberg and Engle (2002). The physical PDF is

obtained via simulation from a fitted model within the GARCH class of processes. The PK which is specified as a

flexible generalized fourth‐order Chebyshev polynomial is then estimated using option prices. There is no need to

select state variables and the PK provides a better fit than monotonically decreasing PKs which are rarely supported

empirically (Cuesdeanu & Jackwerth, 2016).

Research examining the relation between oil and equity markets is confronted with many econometric issues.

While exogeneity of oil prices is commonly assumed, more recent research recognizes the same forces may be driving

oil, equity markets, and macroeconomic aggregates, with reverse causality also possible (Kilian, 2009; Kilian & Park,

2009). The source of oil price shocks can also be demand or supply side driven. An oil price rise from strengthening

global demand, for example, provides a very different signal to an oil price rise from a decrease in supply. Finally, the

impact of oil price shocks on equity markets is state dependent (Aloui & Jammazi, 2009; Conrad et al., 2014; Jammazi

& Aloui, 2010).

We are cognizant of these issues and employ multivariate systems of equations that allow for regime switching via

threshold regression. A vector of LR RND moments (volatility, skewness, and kurtosis) is regressed against monthly

lagged oil returns and macroeconomic controls. A second threshold regression disentangles the RND into its com-

ponents, regressing a vector of LR physical PDF moments (volatility, skewness, and kurtosis) and LR preferences

(proxied by the slope of the PK) against the same variables. Lagged (as opposed to contemporaneous) oil returns and a

wide variety of macroeconomic controls help alleviate endogeneity concerns. The reduced form means we can only

examine if oil prices contain any additional predictive information beyond that contained in macroeconomic

aggregates. The paper therefore remains silent with respect to causality.

By focusing on LR predictability we overcome the need for a structural model that distinguishes between oil price

changes from demand and supply side effects.

1

This is because production disruptions cause small and transitory oil price

rises, while demand increases result in price rises that are large and sustained (Kilian, 2009). In fact, the global business

1

This distinction is important in the short run (SR) though, where demand and supply side shocks impart different effects (Ready, 2018).

1862

|

DARK