VIX futures pricing with conditional skewness
Published date | 01 September 2018 |
Date | 01 September 2018 |
DOI | http://doi.org/10.1002/fut.21925 |
Author | Xinglin Yang,Peng Wang |
Received: 5 July 2017
|
Revised: 22 March 2018
|
Accepted: 1 April 2018
DOI: 10.1002/fut.21925
RESEARCH ARTICLE
VIX futures pricing with conditional skewness
Xinglin Yang
|
Peng Wang
Institute of Chinese Financial Studies,
Southwestern University of Finance and
Economics, Chengdu, China
Correspondence
Xinglin Yang, Institute of Chinese
Financial Studies, Southwestern
University of Finance and Economics,
555 Liutai Avenue, 611130 Chengdu,
Sichuan, China.
Email: xinglinyang@126.com
Funding information
National Natural Science Foundation of
China, Grant/Award Number: 71473200
We develop a closed‐form VIX futures valuation formula based on the inverse
Gaussian GARCH process by Christoffersen et al. that combines conditional
skewness, conditional heteroskedasticity, and a leverage effect. The new model
outperforms the benchmark in fitting the S&P 500 returns and the VIX futures
prices. The fat‐tailed innovation underlying the model substantially reduced
pricing errors during the 2008 financial crisis. The in‐and out‐of‐sample pricing
performance indicates that the new model should be a default modeling choice
for pricing the medium‐and long‐term VIX futures.
KEYWORDS
closed‐form formula, conditional skewness, GARCH, pricing kernel, VIX futures
JEL CLASSIFICATION
G13
1
|
INTRODUCTION
The VIX, introduced by the Chicago Board Options Exchange (CBOE) in 1993, is an important risk indicator. It is often
referred to as the “fear index”by leading financial publications and business news shows. The purpose of the VIX is to
measure S&P 500 volatility over the next month as implied by stock index option prices. The VIX has become a
particularly popular measure of the U.S. stock market volatility. CBOE introduced exchange‐traded VIX futures and
options as tradable assets in March 2004 and February 2006, respectively. In the decade since their launch, their
combined trading activity has grown markedly, now totaling approximately 800,000 contracts per day.
The fast development of the financial market has increased the importance of speculation and hedging against
unexpected changes in volatility. The crucial roles of VIX derivatives in such hedging have been studied widely in the
academic literature; see, for example, Grünbichler and Longstaff (1996), Dotsis, Psychoyios, and Skiadopoulos (2007), and
Mencía and Sentana (2013). To obtain protection from volatility risk, strategies for managing the risk should be built on
reliable VIX futures and option valuation models that describe the empirical properties of asset returns adequately.
Compared with the approaches of options valuation, the literature still offers relatively few pricing formulas of
futures valuation. However, several meaningful valuation models for VIX futures have been proposed. Zhang and Zhu
(2006) price VIX futures assuming stochastic instantaneous variance in the diffusion process of the S&P 500 returns.
Considering the importance of jump diffusion, Lin (2007) constructs a valuation model of VIX futures with
simultaneous jumps in the price and variance processes. Zhu and Lian (2012), combining stochastic volatility and
random jumps, derive a closed‐form and exact formula to evaluate VIX futures. To avoid pricing distortions during the
financial crisis, Mencía and Sentana (2013) find that a VIX process combining central tendency and stochastic volatility
has a better pricing performance for VIX futures and options.
The GARCH process can simultaneously capture the correlation of volatility with returns and the path dependence
in volatility, and it has a more parsimonious description in many situations (Bollerslev, 1986; Heston & Nandi, 2000).
Compared with continuous models, GARCH‐type models readily compute volatility through historical asset prices and
J Futures Markets. 2018;38:1126–1151.wileyonlinelibrary.com/journal/fut1126
|
© 2018 Wiley Periodicals, Inc.
conveniently implement maximum likelihood estimation (MLE). Therefore, GARCH‐type models have been used
extensively for options valuation (Christoffersen, Feunou, Jacobs, & Meddahi, 2014; Christoffersen, Heston, & Jacobs,
2006; Dorion, 2016; Heston & Nandi, 2000). To the best of our knowledge, only Wang, Shen, Jiang, and Huang (2017)
have proposed a closed‐form pricing formula for VIX futures based on the Heston–Nandi (HN) GARCH process.
It is worth emphasizing the importance of capturing nonnormality and asymmetry in economics and finance, which
has been widely noted (Patton, 2004, 2006; Patton & Timmermann, 2007). However, under the local risk‐neutral
valuation relationship (RNVR) framework proposed by Duan (1995), the existing GARCH‐type models with Gaussian
innovations may not sufficiently allow for all the mass in the tails and the asymmetry in the asset return distribution.
Motivated by the stylized fact of conditional skewness in asset returns, we propose a closed‐form formula for pricing
VIX futures using the discrete‐time IG GARCH model in which the innovation follows an inverse Gaussian distribution
with a nonzero third moment. The IG GARCH process reinforces the pricing accuracy for the VIX futures through
capturing the conditional skewness of the S&P 500 returns. Unlike the approach that obtains risk‐neutral dynamics by
directly assuming the form of the state‐price density, our framework uses only the no‐arbitrage assumption and some
technical conditions on the investment strategies, without attempting to characterize the preferences underlying the
RNVR or fully describe the economic environment. Following the theoretical framework by Christoffersen, Elkamhi,
Feunou, and Jacobs (2010), first we propose a pricing kernel to calculate the equivalent martingale measure (EMM)
coefficient. Then, we derive the risk‐neutral dynamics of the inverse Gaussian process using an EMM coefficient that
fills the gap between the physical and the risk‐neutral measures.
We test the performance of the IG GARCH model and that of the most related benchmark, the HN GARCH model,
using the daily S&P 500 returns and the VIX futures data. First, we estimate the models by fitting the S&P 500 returns
using the maximum likelihood method. The empirical results show that the IG GARCH model outperforms the HN
GARCH model in fitting the returns, in forecasting 1‐day and 1‐week volatility, and in absorbing the heteroskedasticity
of returns. Second, we optimize the likelihood function defined on the VIX futures errors and analyze the out‐of‐sample
performance. We find that, overall, the IG GARCH model that incorporates conditional skewness has a better in‐and
out‐of‐sample VIX futures valuation performance than the HN GARCH model. When we further dissect the pricing
performance of the VIX futures across maturity, bias, and VIX level, the IG GARCH model outperforms the HN
GARCH model for small bias, for high VIX level, and particularly for medium‐and long‐term VIX futures contracts.
Using the futures contracts of 2017 as an out‐of‐sample experiment, the IG GARCH model has futures pricing errors
that are 19% below those of the benchmark model for medium‐and long‐term VIX futures contracts. The empirical
findings in the joint subsamples that lie at the intersection of bias, maturity, and VIX levels indicate that the IG GARCH
model is significantly superior to the HN GARCH model in all categories. Third, the fat‐tailed innovation of the IG
GARCH model offers an improvement of 11.78% on average on the extreme days of the 2008 financial crisis. Fourth, we
jointly fit the S&P 500 returns and the VIX futures. The joint results once again reconfirm that the IG GARCH model
with the nonnormal innovation reduces the pricing errors of the benchmark model with the normal innovation.
The remainder of the paper is organized as follows: Section 2 introduces the inverse Gaussian GARCH process and
the nested HN GARCH process; Section 3 provides the estimation results on daily returns; Section 4 develops the
theoretical framework of risk neutralization and the closed‐form formula for VIX futures valuation; Section 5 analyzes
the in‐and out‐of‐sample pricing performance with the volatility index and the VIX futures data; Section 6 estimates the
models jointly on returns and futures; and Section 7 concludes the paper.
2
|
THE MODEL
2.1
|
The return process
The daily returns Rt+
1
and the conditional variance
h
t+
1
specified by the dynamic inverse Gaussian GARCH model are:
≡æ
è
ç
ç
ç
ö
ø
÷
÷
÷
÷
RS
Srζh ηylog = + +
,
tt
t
tt
+1 +1 +1 +1 (1)
hwbhcy
ah
y
=+ + +
,
tt
t
t
t
+1
2
(2)
YANG AND WANG
|
1127
To continue reading
Request your trial