Oil financialization and volatility forecast: Evidence from multidimensional predictors

• Published date: 01 September 2019
• Author: Yan‐ran Ma,Jiaofeng Pan,Qiang Ji
• Date: 01 September 2019
• DOI: http://doi.org/10.1002/for.2577

RESEARCH ARTICLE

Oil financialization and volatility forecast: Evidence from

multidimensional predictors

Yan‐ran Ma

1,2

| Qiang Ji

1,2

| Jiaofeng Pan

1,2

1

Institutes of Science and Development,

Chinese Academy of Sciences, Beijing,

China

2

School of Public Policy and Management,

University of Chinese Academy of

Sciences, Beijing, China

Correspondence

Qiang Ji, Institutes of Science and

Development, Chinese Academy of

Sciences, Beijing 100190, China.

Email: jqwxnjq@163.com;

jqwxnjq@casipm.ac.cn

Funding information

National Natural Science Foundation of

China, Grant/Award Numbers: 91546109

and 71774152; Youth Innovation Promo-

tion Association of the Chinese Academy

of Sciences, Grant/Award Number:

Y7X0231505

Abstract

Using the generalized dynamic factor model, this study constructs three predic-

tors of crude oil price volatility: a fundamental (physical) predictor, a financial

predictor, and a macroeconomic uncertainty predictor. Moreover, an event‐

triggered predictor is constructed using data extracted from Google Trends.

We construct GARCH‐MIDAS (generalized autoregressive conditional

heteroskedasticity–mixed‐data sampling) models combining realized volatility

with the predictors to predict oil price volatility at different forecasting hori-

zons. We then identify the predictive power of the realized volatility and the

predictors by the model confidence set (MCS) test. The findings show that,

among the four indexes, the financial predictor has the most predictive power

for crude oil volatility, which provides strong evidence that financialization has

been the key determinant of crude oil price behavior since the 2008 global

financial crisis. In addition, the fundamental predictor, followed by the finan-

cial predictor, effectively forecasts crude oil price volatility in the long‐run fore-

casting horizons. Our findings indicate that the different predictors can provide

distinct predictive information at the different horizons given the specific

market situation. These findings have useful implications for market traders

in terms of managing crude oil price risk.

KEYWORDS

fundamental information, GARCH‐MIDAS model, generalized dynamic factor model, oil

financialization, volatility forecasting

1|INTRODUCTION

The recent, substantial fluctuations in crude oil prices can

attract the attention of market traders who manage crude

oil price risk and the attention of investors who invest in

crude oil. This is because crude oil price volatility can

influence each industrial department in the industrial

chain and thus significantly affect the global macroeco-

nomic system. Several studies suggest that the uncer-

tainty of the crude oil market can dampen investment

(Elder & Serletis, 2009), shock global financial markets

(Kilian & Park, 2009), and even cause a recession

(Gkanoutas‐Leventis & Nesvetailova, 2015). Therefore,

understanding crude oil price behavior is important for

pricing financial assets, implementing hedging strategies

and assessing regulatory proposals to restrict interna-

tional capital flows (Charles & Darné, 2017).

Given the volatility of crude oil prices, financial mar-

ket participants, policy market participants, and pro-

ducers and consumers of crude oil should understand

market uncertainty and avoid market risks by developing

a better understanding of crude oil price volatility and

making accurate predictions of crude oil price volatility.

Since 2004, the increased participation of hedge funds in

Received: 9 July 2018 Revised: 21 December 2018 Accepted: 3 February 2019

DOI: 10.1002/for.2577

564 © 2019 John Wiley & Sons, Ltd. Journal of Forecasting. 2019;38:564–581.wileyonlinelibrary.com/journal/for

the crude oil market has resulted in the financialization

of the market (Büyükşahin & Robe, 2014; Fattouh, Kilian,

& Mahadeva, 2013). Meanwhile, financial institutions

consider the crude oil market as a profitable alternative

investment to reduce their portfolios risk (Silvennoinen

& Thorp, 2013). Thus, determining how to predict crude

oil price volatility is an important research direction in

the field of energy risk management, and it is important

to explore how to accurately forecast crude oil price

volatility.

Although there are various ways of forecasting crude oil

price volatility, the methods can be classified into two cat-

egories. First, many contributions use volatility data to pre-

dict the oil price volatility. The majority of the literature

uses information on history crude oil price to predict the

conditional volatility of oil prices based on generalized

autoregressive conditional heteroskedasticity (GARCH)‐

class models (Efimova & Serletis, 2014; Wang, Ma, Wei,

& Wu, 2016; Wang & Wu, 2012). Sadorsky and McKenzie,

(2008) show that GARCH‐type models produce more accu-

rate forecasts than any other competing models. Nomikos

and Pouliasis, (2011) use Mix‐GARCH and Markov regime

switching (MRS)‐GARCH models to forecast the 1‐day‐

ahead crude oil price volatility. Kang and Yoon, (2013)

combine autoregressive fractionally integrated moving

average (ARFIMA) models with GARCH models to pro-

duce 1‐,5‐, and 20‐day‐ahead forecasts. Meanwhile, given

that the ultra‐high‐frequency data are more information

rich (Andersen, Bollerslev, Diebold, & Labys, 2003;

Barndorff‐Nielsen & Shephard, 2004), some researchers

focus on predicting the realized volatility of oil prices based

on intraday data (Herrera, Hu, & Pastor, 2018; Liu, Ma,

Yang, & Zhang, 2018; Ma, Wahab, Huang, & Xu, 2017;

Ma, Zhang, Huang, & Lai, 2018).

This category of forecasting crude oil volatility involves

analyzing historical data regarding crude oil prices and

ignores the various external factors that comprehensively

affect the forecast target.

However, although historical market information

could be fully reflected by prices in the efficient market,

as proposed by Fama, (1970), Wang and Liu, (2010) find

that the crude oil market is not efficient in its weak form.

Pan, Wang, Wu, and Yin, (2017) also confirm that oil

prices do not reflect past information on macroeconomic

uncertainty. Meanwhile, Pan et al. increase the predictive

accuracy for oil price volatility forecasting by introducing

a macroeconomic uncertainty variable into the predictive

model. Like Pan et al., introducing driving factors of oil

price volatility into predictive models can increase predic-

tive accuracy for oil price volatility, which is the second

stream of academic studies.

Crude oil prices are driven by a large set of dynamic

and multidimensional factors that not only include

fundamental (physical) markets factors, but also financial

market factors and trading factors (Hamilton, 2009; Ji,

2012). Therefore, considering the driving factors of crude

oil prices in forecasting crude oil price volatility requires

the construction of a reasonable predictor. Most

researchers introduce efficient factors into forecasting

models to improve predictive accuracy.

However, given the imputing data at different frequen-

cies (i.e., the daily oil price volatility and monthly macro-

economic determinants and the ultra‐high‐frequency

data) in predicting models, most research studies use

mixed‐data sampling (MIDAS) regression models

(Ghysels, Santa‐Clara, & Valkanov, 2004) to address the

data frequency problem. Degiannakis and Filis, (2018)

use the MIDAS model combining high‐frequency finan-

cial information with oil market fundamentals to gain

incremental predictive accuracy. To overcome the draw-

back of GARCH‐class models, Engle, Ghysels, and Sohn,

(2013) input realized volatility (RV) and the economic

fundamentals into the GARCH model, forming the

GARCH‐MIDAS model, which prove macroeconomic

fundamentals play a significant role in forecasting. In

recent years, increasing numbers of researchers have

used this model or modified it to analyze and forecast

asset price volatility (Deng, Girardin, & Joyeux, 2018;

Kang, McIver, & Yoon, 2017; Mo, Gupta, Li, & Singh,

2018; Pan et al., 2017; Wei, Liu, Lai, & Hu, 2017). Wei

et al., (2017) use the GARCH‐MIDAS model and dynamic

model averaging combination method to forecast oil price

volatility.

Given the reasonableness of predictors, researchers use

different driving factors to predict crude oil price volatil-

ity on multi‐perspectives. Several studies rely on informa-

tion from fundamental (physical) markets to analyze and

forecast crude oil price volatility (Baumeister & Kilian,

Highlights

1. Three predictors are constructed using the

generalized dynamic factor model to

forecast oil price volatility.

2. An event‐triggered predictor is constructed

using data extracted from Google Trends.

3. Five GARCH‐MIDAS models with the model

confidence set are employed to test forecast

performance.

4. The financial predictor has the most

predictive power for crude oil volatility.

5. Fundamental information becomes more

important in the long‐run forecasting

horizons.

MA ET AL.565