Forecasting realized volatility of oil futures market: A new insight

• Author: Li Liu,Feng Ma,Yu Wei,Dengshi Huang
• Published date: 01 July 2018
• Date: 01 July 2018
• DOI: http://doi.org/10.1002/for.2511

RESEARCH ARTICLE

Forecasting realized volatility of oil futures market: A new

insight

Feng Ma

1

| Yu Wei

2

| Li Liu

3

| Dengshi Huang

1

1

School of Economics and Management,

Southwest Jiaotong University, Chengdu,

China

2

School of Finance, Yunnan University of

Finance and Economics, Kunming, China

3

School of Finance, Nanjing Audit

University, Nanjing, China

Correspondence

Yu Wei, School of Finance, Yunnan

University of Finance and Economics,

Kunming, China.

Email: weiyusy@126.com

Funding information

Natural Science Foundation of China,

Grant/Award Numbers: 71771124,

71671145, 71371157 and 71701170

Abstract

In this study we propose several new variables, such as continuous realized

semi‐variance and signed jump variations including jump tests, and construct

a new heterogeneous autoregressive model for realized volatility models to

investigate the impacts that those new variables have on forecasting oil price

volatility. In‐sample results indicate that past negative returns have greater

effects on future volatility than that of positive returns, and our new signed

jump variations have a significantly negative influence on the future volatility.

Out‐of‐sample empirical results with several robust checks demonstrate that

our proposed models can not only obtain better performance in forecasting

volatility but also garner larger economic values than can the existing models

discussed in this paper.

KEYWORDS

oil futures market,realized semi‐variances, signed jumpvariations, statistic and economic

evaluations, volatility forecasting

1|INTRODUCTION

Crude oil plays an essential role in the world economy.

Oil price uncertainty has important macroeconomic

effects (Hamilton, 1983, 2003; Kilian, 2009) and effects

on financial markets (Aloui & Jammazi, 2009; Kilian &

Park, 2009). Oil price volatility is a key input for risk

management, derivative pricing, portfolio selection, and

many other financial activities. Therefore, modeling and

forecasting the volatility of crude oil prices are critical

for researchers, market participants, and policymakers.

There are many works on forecasting oil price

volatility in the frameworks of generalized autoregressive

conditional heteroskedasticity (GARCH) and its various

extensions (see, e.g., Agnolucci, 2009; Charles & Darné,

2014; Efimova & Serletis, 2014; Nomikos & Pouliasis,

2011; Wei, Wang, & Huang, 2010). However, these models

are always applied to daily or lower‐frequency data,

which can result in a substantial loss of intraday trading

information (Carnero, Peña, & Ruiz, 2004; Corsi, 2009).

Recently, with the availability of high‐frequency data,

research on financial market volatility has taken new ave-

nues. The seminal works of Andersen and Bollerslev

(1998) and Andersen, Bollerslev, Diebold, and Ebens

(2001) propose the realized volatility or realized variance

(RV), which are defined as the sum of all available intra-

day high‐frequency squared returns. This volatility mea-

sure can enable researchers to better gauge the current

level of volatility and understand its dynamics. Corsi

(2009) propose a simple heterogeneous autoregressive

model of realized volatility (HAR‐RV) based on the hetero-

geneous market hypothesis, which can capture “stylized

facts”in financial market volatility, such as long memory

and multiscaling behavior, and have other advantages in

forecasting. Thus this model has garnered wide focus in

academia (see, e.g., Andersen, Bollerslev, & Diebold,

2007; Bekaert & Hoerova, 2014; Bollerslev, Patton, &

Quaedvlieg, 2015; Busch, Christensen, & Nielsen, 2011;

Corsi, Pirino, & Reno, 2010; Duong & Swanson, 2015;

Wang, Ma, Wei, & Wu, 2016; Wang, Pan, & Wu, 2017).

Received: 2 March 2016 Revised: 11 September 2017 Accepted: 2 December 2017

DOI: 10.1002/for.2511

Journal of Forecasting. 2018;37:419–436. Copyright © 2018 John Wiley & Sons, Ltd.wileyonlinelibrary.com/journal/for 419

Nevertheless, the HAR‐RV model is of great interest to

us here because there is an extremely limited strand of the

literature focusing on the oil price‐realized volatility fore-

casting using this model (Degiannakis & Filis, 2017;

Haugom, Langeland, Molnár, & Westgaard, 2014;

Haugom, Lien, Veka, & Westgaard, 2014; Haugom, Veka,

Lien, & Westgaard, 2014; Liu & Wan, 2012; Ma, Wahab,

Huang, & Xu, 2017; Prokopczuk, Symeonidis, & Wese

Simen,2016; Sévi, 2014). Recently, Patton and Sheppard

(2015) used the realized semi‐variances proposed by

Barndorff‐Nielsen, Kinnebrock, and Shephard (2010),

which decompose the realized variance into a component

that relates only to positive high‐frequency returns

(“good”volatility) and into a component that relates only

to negative high‐frequency returns (“bad”volatility). They

demonstrate that the previous literature usually uses the

square (absolute) intraday returns, which can result in

losing the information that may be contained in the sign

of these returns. They further find that those models that

include the realized semi‐variances can improve the fore-

casting performance. Owning to special information on

signed intraday returns, realized semi‐variances have

been given attention by many scholars, such as Chen

and Ghysels (2010), Baruník, Kočenda, and Vácha

(2016), Duong and Swanson (2015), and Audrino and

Hu (2016).

It is well known that the decomposition between the

continuous and the jump components may help to

obtain better forecast accuracy (Andersen et al., 2007;

Corsi et al., 2010; Sévi, 2014). Thus, inspired by the

above‐mentioned reference, we further decompose the

realized semi‐variances into continuous and discontinu-

ous jump components, which are similar to realized vol-

atility. In other words, the positive (negative) realized

semi‐variance can divide the positive (negative) contin-

ued sample path realized semi‐variance (hereafter

denoted CRS) and positive jump components. In accor-

dance with Patton and Sheppard (2015), we have new

signed jump variations, which include the jump tests.

We use those variables and combine them with HAR‐

RV and its various extensions to construct the new

models, labeled as HAR‐RV‐type. In our study, we

empirically investigate whether our proposed models

have better performance than do the existing models in

predicting the future volatility.

Our main contributions are as follows. The first contri-

bution of this paper is to introduce new continuous real-

ized semi‐variances and signed jump variations. We can

use those variables to gain new insights and first investi-

gate the impact of those variables on forecasting oil price

volatility. The second contribution is to propose several

novel HAR‐RV‐type models based on those proposed var-

iables and evaluate their forecasting performance using

the statistical and economic significance. Our papers are

closely related to the studies by Sévi (2014) and Patton

and Sheppard (2015); therefore, we compare our new

HAR‐RV‐type models with their models by the statistical

and economic significance. Regarding forecasting perfor-

mance, the majority view is that statistical significance is

not sufficient to prove the superiority of a specific model,

because market investors are more interested in the eco-

nomic value of volatility models. Therefore, we follow

the literature by considering a mean–variance utility

investor who allocates his or her assets between stock

and the risk‐free Treasury bill, where the optimal weight

of stock in the portfolio is ex ante determined by volatility

and mean forecasts of the stock return (see, e.g., Guidolin

& Na, 2006; Neely, Rapach, Tu, & Zhou, 2014; Rapach,

Strauss, & Zhou, 2010). To the best of our knowledge,

the economic value of realized volatility forecasts has

been considered in very few existing studies beyond the

notable work of Fleming, Kirby, and Ostdiek (2003).

Therefore, in our study, we seek to answer the following

question: Do our new HAR‐RV‐type models help inves-

tors obtain more economic benefits?

The major findings from this study are threefold. First,

the new negative CRS has a stronger impact on the future

volatility of the oil futures market than on that of the pos-

itive CRS. Second, the continuous signed jump variations

have significant negative effects on future realized

volatility, and the effect is larger than that of the existing

sign jump variation proposed by Patton and Sheppard

(2015). Depending on the Wald test, we find that our

new continuous realized semi‐variances and signed jump

variations show a strong asymmetric “leverage effect”on

future volatility. Third, based on those new variables, we

extend several new HAR‐RV‐type volatility models. Our

out‐of‐sample results demonstrate that the newly

proposed variables and extended HAR‐RV‐type models

not only attain better performance in forecasting the

volatility of the oil futures market but also garner larger

economic values than traditional HAR‐RV‐types. These

conclusions are robust and reliable for different bench-

mark and forecasting windows.

The remainder of this paper is organized as follows.

Section 2 provides the descriptions of the volatility

measures and models. Section 3 presents the data and

the preliminary analysis. The empirical forecasting results

are presented in Section 4. Section 5 concludes the paper.

2|VOLATILITY MEASURES,

JUMPS AND HAR‐TYPE MODELS

In this section we provide a brief description of several

popular volatility measures using intraday data and the

420 MA ET AL.