On the predictability of crude oil market: A hybrid multiscale wavelet approach

• Author: Stelios Bekiros,Gazi Salah Uddin,Jose Arreola Hernandez,Ahmed Taneem Muzaffar
• DOI: http://doi.org/10.1002/for.2635
• Date: 01 July 2020
• Published date: 01 July 2020

RESEARCH ARTICLE

On the predictability of crude oil market: A hybrid

multiscale wavelet approach

Stelios Bekiros

1

| Jose Arreola Hernandez

2

| Gazi Salah Uddin

3

|

Ahmed Taneem Muzaffar

4

1

Department of Economics, European

University Institute (EUI), Florence, Italy

2

Department of Accounting and Finance,

Rennes School of Business, Rennes,

France

3

Department of Management and

Engineering, Linköping University,

Linköping, Sweden

4

Social Protection Department,

International Labour Organization,

Geneva, Switzerland

Correspondence

Jose Arreola Hernandez, Department of

Accounting and Finance, Rennes School

of Business, Rennes 35000, France.

Email: jose.arreola-hernandez@rennes-sb.

com

Abstract

Past research indicates that forecasting is important in understanding price

dynamics across assets. We explore the potentiality of multiscale forecasting in

the crude oil market by employing a wavelet multiscale analysis on returns

and volatilities of Brent and West Texas Intermediate crude oil indices

between January 1, 2001, and May 1, 2015. The analysis is based on a shift-

invariant discrete wavelet transform, augmented by an entropy-based method-

ology for determining the optimal timescale decomposition under different

market regimes. The empirical results show that the five-step-ahead wavelet

forecast that is based on volatilities outperforms the random walk forecast, rel-

ative to the wavelet forecast that is based on returns. Optimal wavelet causality

forecasting for returns is suggested across all frequencies (i.e., daily–yearly),

whereas for volatilities it is suggested only up to quarterly frequencies. These

results may have important implications for market efficiency and predictabil-

ity of prices on the crude oil markets.

KEYWORDS

crude oil markets, forecasting, multiscale, wavelets

1|INTRODUCTION

The past decade witnessed fundamental changes in the

composition of commodity markets. In the wake of the

2008 global financial crisis (GFC) and the 2009–2010

recession period that followed, predicting the price of

crude oil, arguably the most important commodity in the

world today, is of considerable interest to economists,

policymakers, and investors, as the business cycle of crude

oil prices is heterogeneous and volatile among the com-

modity classes.

1

The sharp rise and decline in the price of

oil since the oil price shock of 2007–2008, which by any

measure could be categorized as one of largest shocks to

the international price of oil on record, and its conse-

quences have renewed the debate on understanding the

fundamental behavior of future oil prices (Hamilton, 2009).

Forecasting crude oil price is important to

policymakers because oil is a main source of energy, and

fluctuations in its prices affect government planning and

policymaking decisions with respect to taxation and sub-

sidies in traditional and renewable energy sectors, as well

as in the agricultural sector. The issue of stability in oil

Average oil prices, for instance from 2003 to 2008, rose each year,

reaching a historical high of $147 per barrel in nominal terms in mid-

July of 2008. However, in 2013 real oil prices plummeted from about

$115 per barrel in 2011 to around $50 per barrel in 2015. On the upside,

the US International Energy Outlook 2016 (IEO, 2016) reported that

world oil prices were expected to reach $252 per barrel by 2040—123%

higher from 2012 and almost 79% higher than projected in the so-called

reference case ($141 per barrel). On the downside, the IEA in 2016

projected that world oil prices would decline to around $58 per barrel in

2020 and then would rise to $76 per barrel in 2040 (IEO, 2016).

Received: 3 October 2018 Revised: 20 May 2019 Accepted: 28 October 2019

DOI: 10.1002/for.2635

Journal of Forecasting. 2020;39:599–614. wileyonlinelibrary.com/journal/for © 2019 John Wiley & Sons, Ltd. 599

prices is also linked to inflation and its effect on the cost

of consumer goods and production in the industrial

sector, which in turn may critically impact the political

order of oil-intensive economies. Moreover, the behavior

of agricultural commodity prices and its relationship with

the price of oil matters for global food policy and sensitiv-

ity of various economic agents, to whom oil is the

primary input in the production process. From the per-

spective of investors, the forecasting nature of future

returns and volatility on oil markets is also fundamen-

tally important for determining asset pricing, including

derivative pricing, hedging and risk management.

Considering its huge importance on the global,

regional, and country economies, and to minimize the

negative impact of oil price fluctuations, a growing body

of literature in recent years has paid attention to model-

ing and forecasting the volatility of oil prices. For

instance, Ma, Ji, and Pan (2019) constructed four predic-

tors of crude oil price volatility: namely, a fundamental

(physical) predictor, a financial predictor, a macroeco-

nomic uncertainty predictor, and an event-triggered pre-

dictor, using the generalized dynamic factor model. The

study then employed GARCHMIDAS (generalized auto-

regressive conditional heteroskedasticity–mixed data

sampling) models to identify the best predictor of crude

oil price volatility, through the analysis of the predictive

performance of each predictor with the help of the model

confidence set (MCS) test. The sample period of the study

ran from September 1, 2007, to June 1, 2017. The findings

of the study revealed that, among the four indexes, the

financial predictor had the most predictive power for

crude oil volatility, providing strong evidence that

financialization had been the key determinant of crude

oil price behavior since the 2008 Global Financial Crisis.

In addition, the fundamental predictor, followed by the

financial predictor, effectively forecast crude oil price vol-

atility in the long-run forecasting horizons. The authors

noted that the different predictors could provide distinct

predictive information at the different horizons given the

specific market situation.

Zhang and Zhang (2018) claimed that a hybrid fore-

casting method significantly improves forecasting accu-

racy of crude oil price volatility, particularly in the longer

time horizon. Ryan and Whiting (2017) criticized the tra-

ditional model selection process that identified a single

“best model”from a set of candidate models. They mea-

sured the performance of multimodel inference (MMI)

forecasts compared to the predictions made from a single

model for crude oil prices. The authors forecast the West

Texas Intermediate (WTI) crude oil spot prices using total

OECD petroleum inventory levels, surplus production

capacity, the Chicago Board Options Exchange Volatility

Index, and an implementation of a subset autoregression

with exogenous variables (SARX). The study covered the

period between April 2002 and July 2012. The authors

argued that coefficient and standard error estimates

obtained from SARX, determined by conditioning on a

single “best model”, ignored model uncertainty and

resulted in underestimated standard errors and over-

estimated coefficients. The findings of the study suggest

that the MMI forecast outperforms a single-model fore-

cast for both in-sample and out-of-sample datasets over a

variety of statistical performance measures. They also

found that weighting models according to the Bayesian

information criterion generally yielded superior results

both in-sample and out-of-sample when compared to the

Akaike information criterion.

Dutta (2017) investigated whether the crude oil vola-

tility index (OVX) in the realized volatility (RV) models

improved the accuracy of predictions. The author used a

sample period spanning from May 10, 2007 to June

30, 2016. The findings of the study demonstrate that the

information content of crude OVX helps to provide more

accurate volatility predictions in comparison to the base-

line RV model, which contains only historical oil volatil-

ities. Moreover, the study's forecasting encompassing test

further suggests that the modified RV model (when OVX

is introduced in the baseline RV model) forecast encom-

passes the conventional RV forecast in majority of the

cases.

In a similar vein, an earlier study by Haugom,

Langeland, Molnár, and Westgaard (2014) examined the

information content of the OVX when forecasting real-

ized volatility in the WTI futures market. The study also

investigated whether other market variables such as vol-

ume, open interest, daily returns, bid–ask spread, and

the slope of the futures curve contained predictive power

beyond what was embedded in the implied volatility.

The sample period considered in this study spanned

between May 16, 2007, and May 15, 2012. The authors

noted that in out-of-sample forecasting econometric

models based on realized volatility could be improved by

including implied volatility and other variables. The

findings suggest that including implied volatility signifi-

cantly improves daily and weekly volatility forecasts.

However, the authors pointed out that including other

market variables significantly improved not just daily

and weekly volatility forecasts but also monthly volatility

forecasts.

Oil price forecasting is regarded as “an intractable

task due to the intrinsic complexity of the oil market

mechanism”(Jammazi & Aloui, 2012) and the multiple

and interdependent global forces—for example, OPEC

production output, global demand, shale and biofuel oil

production, weather conditions, crisis events such as the

2008 global financial crisis, wars—affecting its demand

600 BEKIROS ET AL.