volatility-of-volatility risk in asset pricing 20171117 · model, vov is a state variable that...
TRANSCRIPT
Volatility-of-Volatility Risk in Asset Pricing
Te-Feng Chen, Tarun Chordia, San-Lin Chung, and Ji-Chai Lin*
November 2017
Abstract
This paper develops a general equilibrium model in an endowment economy with time-varying
uncertainty and provides empirical support for an outcome of the model that the market volatility-
of-volatility (VOV) predicts market returns. A second outcome is an asset pricing model that
incorporates the market, the market volatility and VOV as a pricing factors. The differential long-
short quintile portfolio return amounts to 72 basis points per month when sorting on the VOV beta.
The risk premium on VOV is robust across different testing strategies and different test assets.
Market risk and volatility risk are not priced in constant beta models but consistent with theory,
the conditional factor loadings on both the market risk and volatility risk are priced. The pricing
impact of VOV strengthens during market crashes while that of market volatility obtains during
normal times.
JEL Classification: G14; G12;
Contacts
Chen Chordia Chung Lin
Voice: 852-3400-3856 1-404-727-1620 886-2-3366-1084 852-3400-8454
Fax: 852-2330-9845 1-404-727-5238 886-2-2366-0764 852-2330-9845
E-mail: [email protected] [email protected] [email protected] [email protected]
Address: HK Polytechnic
University
Kowloon, Hong Kong
Goizueta Business School
Emory University
Atlanta, GA 30322
National Taiwan
University
Taipei, Taiwan
HK Polytechnic
University
Kowloon, Hong Kong
*We thank seminar participants at Hong Kong Polytechnic University.
1
Volatility-of-Volatility Risk in Asset Pricing
Abstract
This paper develops a general equilibrium model in an endowment economy with time-varying
uncertainty and provides empirical support for an outcome of the model that the market volatility-
of-volatility (VOV) predicts market returns. A second outcome is an asset pricing model that
incorporates the market, the market volatility and VOV as a pricing factors. The differential long-
short quintile portfolio return amounts to 72 basis points per month when sorting on the VOV beta.
The risk premium on VOV is robust across different testing strategies and different test assets.
Market risk and volatility risk are not priced in constant beta models but consistent with theory,
the conditional factor loadings on both the market risk and volatility risk are priced. The pricing
impact of VOV strengthens during market crashes while that of market volatility obtains during
normal times.
JEL classification: G12, G13, E4
1
1. Introduction
Turmoil in financial markets while episodic has become more frequent, leading to higher
uncertainty in markets.1 In order to capture the dynamics of market uncertainty, Bollerslev,
Tauchen, and Zhou (2009) build a general equilibrium model for a representative investor with
Epstein and Zin (1989) preferences and introduce the volatility-of-volatility (VOV) as an additional
source of uncertainty. Their model shows that VOV drives the time-varying variance risk premium
and its ability to predict stock market returns.
This paper develops a macroeconomic model that synthesizes the seminal long-run risk model
of Bansal and Yaron (2004) and the variance-of-variance model of Bollerslev, Tauchen, and Zhou
(2009). We solve the macro-finance model explicitly and derive the equilibrium aggregate asset
prices. The model establishes that an increase in VOV reflects higher uncertainty about market
volatility, and raises the market risk premium, implying an immediate price decline and higher
future returns. Further, VOV drives the contemporaneous covariance between market returns and
market volatility (denoted CVRV). CVRV is time-varying and becomes more negative as VOV
increases. Further, VOV and CVRV contain similar information for predicting market returns.
We use aggregate asset prices to characterize the macroeconomic risks, and transform the
underlying macro-based model to a market-based three-factor asset pricing model. The market-
based model has distinct advantages: (i) Financial data provide useful information because asset
prices tell us how market participants value risks; and (ii) Markets disseminate financial data in a
timely fashion. In the asset pricing model, the expected return of security i is determined by three
sources of risks: (i) the covariance with market returns, ov , , , ; (ii) the covariance
1As Mr. Olivier Blanchard, IMF’s chief economist, points out in The Economist in January 2009, “Crises feed uncertainty. And uncertainty affects behaviour, which feeds the crisis.”
2
with the market variance, ov , , , , where , ar , ; and, (iii) the
covariance with the variance of market variance, ov , , , , where ,
ar , . The first risk is the market risk of the classical capital asset pricing model (CAPM;
Sharpe, 1964; Lintner, 1965). The second risk corresponds to the volatility risk of Ang, Hodrick,
Xing, and Zhang (2006). The last risk, which is the main focus of our paper, is the volatility-of-
volatility, VOV, risk.
While the aggregate volatility risk is proxied by the volatility index, VIX, obtained from the
Chicago Board of Options Exchange (CBOE), we measure VOV using high frequency S&P 500
index option data. We first convert the tick-by-tick options data to equally spaced five-minute
observations and then use the model-free methodology (see Carr and Madan 1998, Britten-Jones
and Neuberger 2000, Bakshi, Kapadia, and Madan, 2003, and Jiang and Tian 2005) to estimate the
market variance implied by index option prices for each five-minute interval. For each day, we
estimate VOV by calculating the realized bipower variance from a series of five-minute model-free
implied market variance within the day. The bipower variation, introduced by Barndorff-Nielsen
and Shephard (2004), delivers a consistent estimator solely for the continuous component of VOV
while excluding the jump component.2 In other words, our empirical results are not affected by
the potential jump risk embedded in volatility as shown by Pan (2002), Eraker (2008), and
Drechsler and Yaron (2011).
Empirically, VOV strongly affects the market return-volatility covariance (CVRV), and,
moreover, VOV and, to a lesser extent, CVRV predict market returns. VOV predicts market returns
even after controlling for the variance risk premium and jump risk, as documented by Bollerslev,
2 Measures of realized jump based on the difference between realized variation and bipower variation have been proposed by Barndorff-Nielsen and Shephard (2004), Huang and Tauchen (2005), and Andersen, Bollerslev, and Diebold (2007).
3
Tauchen, and Zhou (2009) and Drechsler and Yaron (2011), respectively. Thus, consistent with our
model, VOV is a state variable that drives the time-varying market risk premium. The next
question is whether VOV is a priced factor in the cross-section of asset returns.
We test the three-factor asset pricing model using stock returns on NYSE, AMEX, and
NASDAQ listed stocks over the period from 1996 to 2015. The main finding is that VOV is indeed
priced in the cross-section of stock returns with a return differential of -72 basis points per month
across quintile portfolios sorted on the VOV beta. The portfolio returns decrease with the VOV beta
because the market return decreases with VOV and so a stock whose return increases with VOV
i.e., a positive VOV beta, will provide a hedge against VOV risk and thus earn lower returns than a
stock with a negative VOV beta. The risk premium on VOV is robust in Fama-MacBeth (1973)
regressions when the test assets are 25 portfolios formed using independent quintile sorts on the
VOV beta and the VIX beta as well as when using individual stocks as test assets. The VOV premium
is robust in the presence of the Fama and French (1993) and the Carhart (1997) factors (SMB,
HML and UMD) as well as the market skewness and the market kurtosis factor of Chang,
Christoffersen, and Jacobs (2013) and the jump factor of Cremers, Halling and Weinbaum (2015).
Moreover, the VOV premium remains robust in the presence of the firm level characteristics
including firm size, the book-to-market ratio, past returns, the Amihud (2002) illiquidity, the
implied-realized volatility spread (IVOL-TVOL) and the call-put implied volatility spread (CIVOL-
PIVOL) of Bali and Hovakimian (2009) and Yan (2011).
The unconditional VIX beta and the MKT beta are not priced in our sample. However,
consistent with the theoretical model, the conditional component of the factor loadings of both,
VIX and the MKT, are priced. While Ang et al. (2006) and Adrian and Rosenberg (2008) show that
market volatility is a priced factor, Chang, Christoffersen, and Jacobs (2013) find mixed results.
4
Interestingly, we find that all the three risks (MKT, VIX, VOV) in the asset pricing model are
significantly priced during normal times when we remove the observations with large market
declines from our sample. However, during crash periods (with extreme negative market returns),
the price of VOV risk becomes even more significant while the premiums on MKT and VIX have
the wrong sign. The positive premium on VIX during market turmoil combined with the negative
premium during normal periods, results in an overall VIX premium that is indistinguishable from
zero. The results imply that market turmoil strengthens investors’ demand for compensation for
VOV risk.
Our study is related to several strands of the literature. Bollerslev, Tauchen, and Zhou (2009)
use the variance risk premium as an indirect measure of VOV to predict stock market returns.
However, Drechsler and Yaron (2011) show that jump shocks can capture the size and predictive
power of the variance risk premium. Further, Drechsler (2013) shows that model uncertainty also
has a large impact on the variance risk premium, helping explain its power to predict stock returns.
Unlike these studies, we focus directly on VOV. We empirically construct a VOV measure that
excludes jumps to show that it is a determinant of the variance risk premium and that it can predict
stock market returns. Moreover, we provide new evidence that VOV risk is important for pricing
the cross-section of stock returns.
Our paper is also related to the pricing model with higher moments of the market return as
risk factors, as proposed by Chang, Christoffersen, and Jacobs (2013). They find that market
skewness is a priced risk factor in the cross section of stock returns. Both our paper and their work
extend Ang et al. (2006) to extract implied moments from index option prices. Our results are
robust to the inclusion of the market skewness factor as well as the jump factor of Cremers, Halling,
and Weinbaum (2015). In fact, we find that VOV risk subsumes the pricing power of the higher
5
moment and jump betas documented in the literature. Our findings imply that, in the spirit of the
Merton (1973) ICAPM, the time-varying VOV seems better in capturing shifts in the investment
opportunity set than the time-varying market skewness and jumps. Finally, Baltussen, Van Bekkum,
and Van Der Grient (2013) show evidence that an ambiguity measure based on firm-level historical
volatility of option-implied volatility (vol-of-vol) is associated with future stock returns, which is
inconsistent with the rational pricing of uncertainty. Our study complements theirs as we show that
our VOV measure is in supportive of the rational pricing of VOV risk.
The remainder of the paper is organized as follows. The next section describes the economic
dynamics and develops our market-based three-factor model for the empirical implementation.
Section 3 constructs the measure of market volatility-of-volatility. Section 4 describes the data,
presents the summary statistics, and provides evidence for the VOV’s predictability on future stock
market returns. In section 5, we show empirical evidence on the pricing of VOV risk in the cross-
sectional stock returns and investigate the VOV risk-return tradeoff during market turmoil. Finally,
section 6 contains our concluding remarks.
2. An asset pricing model with VOV risk
This section develops an asset pricing model with volatility-of-volatility (VOV) risk starting
with a macroeconomic model that incorporates the seminal long-run risk model of Bansal and
Yaron (2004) and the variance-of-variance model of Bollerslev, Tauchen, and Zhou (2009). We
solve the macro-finance model explicitly and derive the equilibrium aggregate asset prices. Then,
we use the properties of aggregate asset prices to characterize the macroeconomic risks and
develop a market-based three-factor model for the cross-sectional asset prices.
2.1. Economic dynamics and the equilibrium market risk premium
6
The setting of our model is a discrete-time endowment economy. The dynamics of the
consumption growth rate, , and the dividend growth rate, , , are governed by the
following processes:
, ,
, ,
, ,
, ,
, , ,
(1)
where , , , , , , , , , ~ 0,1 , represents the long-run consumption
growth, is the time-varying economic uncertainty, and is the economic volatility-of-
volatility, which is the conditional variance of the economic uncertainty. These features of the
long-run risk and the time-varying economic uncertainty have been proposed by Bansal and Yaron
(2004), while the additional feature of economic volatility-of-volatility has been introduced by
Bollerslev, Tauchen, and Zhou (2009).
The representative agent is assumed to have recursive preferences of Epstein and Zin (1989).
Thus, the logarithm of the Intertemporal Marginal Rate of Substitution (IMRS) is
log 1 , ,
where , is the return on consumption claim, is the time discount factor, is the risk-
aversion parameter, is the intertemporal elasticity of substitution (IES) parameter, and ≡
1 1 1/ . With 1, and 1, 0. Based on Campbell and Shiller’s (1988)
approximation, , , where is the logarithm of price–
consumption ratio, which in equilibrium is an affine function of the state variables, i.e.
.3 Substituting the equilibrium consumption return, , , into the
3 The equilibrium solutions are: / 0, / 0, and 0.
7
IMRS, the innovation in the pricing kernel is
, , , , , (2)
where 0, 1 0, 1 0, 1 0 .
The underlying parameters drive the prices for short-run risk ( ), long-run risk ( ), volatility
risk ( ), and volatility of volatility risk ( ). An analogous expression holds for the stock market
return, , , , , , , , where , is the log price–dividend
ratio, which in equilibrium is an affine function of the state variables, , , ,
, , .4 Since, 0, we have , 0, , 0, and , 0.
The innovation in market return can be expressed as
, , , , , , , , , ,
where , , , 0, , , , 0, and , , , 0 . The
conditional variance of market return is readily calculated as
, ≡ ar , , , , ,
and the process for innovations in market variance is , , , ,
, , , where , , and , , . Thus, innovations in market variance
are related to both the economic volatility shock and the economic volatility-of-volatility shock. It
follows that the market volatility-of-volatility (i.e., the conditional variance of market variance) is
, ≡ ar , , , , and the process for its innovations is ,
, , , , where , , . Note that, in our model, innovations to the
market volatility-of-volatility are solely determined by economic variance of variance shocks with
a scaling factor, , . Thus, the volatility of market VOV (i.e., the conditional variance of variance
4 The solutions are: ,
/
,, ,
. ,
,, and ,
. ,
,, where
, , and , , .
8
of market variance), , ≡ ar , , , is constant in our model.
It is straightforward now to derive the equity premium on the market portfolio. In the
following proposition, the expected market return is determined by the covariance risks with
respect to three sources of risks in the pricing kernel.
Proposition 1: The risk premium for the market, , , , is governed by three
covariance risks as follows:
, , 0.5 ar , ov , ,
ov , , ov , , ov , ,
, , , .
(3)
The first two terms in (3)represent the long-run risk premium and the volatility risk premium,
which are the same as in Bansal and Yaron (2004), while the last one represents the volatility-of-
volatility risk premium, as in Bollerslev, Tauchen, and Zhou (2009).5
2.2. Market return-volatility covariance and future return predictability
The contemporaneous market return and market volatility tend to be negatively correlated,
which is commonly referred to as the leverage effect in the literature (e.g. Black, 1976; Christie,
1982; among others). This covariance could also be understood as the volatility risk for the
aggregate market under ICAPM (Ang et al., 2006). The model endogenously generates a time-
varying contemporaneous correlation between the market return and the market volatility as well
as a negative contemporaneous correlation between the market return and the volatility-of-
volatility. Straightforward calculations yield the following proposition.
Proposition 2: The market return is negatively correlated with both the market volatility and the
5 Since we do not assume the square root process for the volatility-of-volatility as Bollerslev, Tauchen, and Zhou (2009) do, the volatility risk in the resulting equity premium does not confound with the volatility-of-volatility risk.
9
market volatility-of-volatility, as shown below:
, ≡ ov , , , , , , , 0,ov , , , , , 0.
(4)
The contemporaneous covariance between the market return and the market volatility is time-
varying and driven by the volatility-of-volatility. By contrast, the contemporaneous covariance
between the market return and the volatility-of-volatility is constant.
In the absence of the time-varying economic volatility-of-volatility i.e., when is constant
and =0, the second term of , is zero, resulting in a less negative value. Thus, the
economic volatility-of-volatility amplifies the impact of volatility on returns. Moreover, driven by
, the volatility risk for the aggregate market is time-varying and provides information about
future market returns since the time-varying market risk premium from (3) is driven by as
well. Further, the market volatility-of-volatility ( , ), which varies directly with the economic
volatility-of-volatility ( ), possesses similar information about future market returns. The
following proposition summarizes market return predictability of , through the channel
of the underlying economic volatility-of-volatility ( ).
Proposition 3. With the presence of time-varying economic volatility-of-volatility ( ), the
projection coefficient for the predictive regression of market returns on , is negative, i.e.,
, , , , with 0.
The projection coefficient for predictive regression of market returns on , is positive, i.e.,
, , , ,with 0.
Specifically, the projection coefficients are:
, , , ,
, ,0,
, , , ,
,
,
,0.
(5)
10
The negative association between the future market return and the volatility risk for the
market and a positive association between the future market return and the market volatility-of-
volatility are consistent with the feedback effect through the time-varying risk premium
documented in the literature (see, e.g. Campbell and Hentschel 1992; Bekaert and Wu 2000; Wu
2001; Bollerslev, Sizova, and Tauchen, 2012; among others). When increases, , also
increases (and , decreases), which leads investors to demand a higher risk premium,
resulting in a negative contemporaneous market return due to a higher discount rate pushing the
price lower, and leading to a higher future return to compensate for the increased risk.
Furthermore, Bollerslev, Tauchen, and Zhou (2009) show that the variance risk premium
(VRP), which is the proxy for the economic volatility-of-volatility, , in their equilibrium model,
can predict the future market return. However, Drechsler and Yaron (2011), challenge the role of
in Bollerslev, Tauchen, and Zhou, since the jump risk premium could be an alternative state
variable that contributes to both the variance risk premium and the market return predictability.
Thus, instead of using VRP, an indirect measure used by Bollerslev, Tauchen, and Zhou to proxy
for , we will use the direct measure - market volatility-of-volatility ( , . Further, we will
isolate the jump component of , from the continuous component and use only the latter.
2.3. A market-based three-factor model for pricing individual stocks
Based on the sources of risk in (1), we assume that the innovation in stock return i is6
, , , , , , , , , .
Given the pricing kernel in (2), the expected stock return can be written as
, , 0.5 ar ,
, , , . (6)
The expected stock return is determined by three sources of economic risks: economic long-run
6 We have not included the risk of consumption growth as the empirical support for it is weak.
11
risk ( , ), economic volatility risk ( , , and economic volatility-of-volatility risk ( , ). The
following proposition provides the risk premium in terms of market risks.
Proposition 4: The risk premium for stock i, , , , is governed by
, , 0.5 ar ,
ov , , , ov , , ,
ov , , , ,
(7)
where ,, ,
,, , ,
,. (8)
The proof of the above proposition uses the properties of the aggregate asset prices to
characterize the macroeconomic risks in (6). In equilibrium, the market volatility-of-volatility risk,
which is the return covariance with respect to the variance of market variance, is solely determined
by the economic volatility-of-volatility risk ( , ), i.e., ov , , , , , . Also,
the return sensitivities with respect to the market variance and with respect to the market return
provide additional information for the economic volatility risk and the long-run risk; that is,
ov , , , , , , , , and ov , , , , ,
, , , , . Substituting out the economic risks in (6), results in the market-based
three-factor model for individual stock excess returns in (7).
Thus, the expected return on stock i is determined by three sources of risks related to aggregate
asset prices. The first term measures the market risk of classical capital asset pricing model (CAPM;
Sharpe, 1964; Lintner, 1965). The second term corresponds to the aggregate volatility risk in Ang
et al. (2006). The last term, which is the main focus of this paper, measures the aggregate volatility
of volatility risk. The resulting three risk premiums in our market-based model, , , and ,
are related to the three economic risk premiums through a linear transformation.
12
The market-based model developed in this paper has several advantages. Financial data
provides useful information because asset prices tell us how market participants value risks.
Moreover, financial data conveys information to investors in a timely fashion. Further, the market-
based risk factors are easier to estimate. For instance, , is easier to estimate than . Also, the
empirical design of our model is compatible with a large literature of multi-factor models for
explaining cross-sectional stock returns (see, for instance, Fama and French, 1993, 2015; Ang et
al., 2006; Maio and Santa-Clara, 2012; Hou, Xue and Zhang, 2015; among others).
3. Estimating the variance of market variance
To test the proposed market-based three-factor model, we need three risk factors – the market
return, the market variance, and the variance of market variance. For the first two factors, we use
the CRSP value-weighted market index and the CBOE VIX index, respectively. The VIX index
has been used as a proxy for the market variance by Ang et al., (2006), Bollerslev, Tauchen, and
Zhou (2009), and Chang, Christoffersen, and Jacobs (2013) among others. We estimate the
variance of market variance by calculating the realized bipower variation from five-minute model-
free implied market variances, using the high-frequency S&P 500 index option data as follows.
First, we extract the model-free implied variance, using the spanning methodology proposed
by Carr and Madan (2001), Bakshi and Madan (2000), Bakshi, Kapadia, and Madan (2003), and
Jiang and Tian (2005). Bakshi, Kapadia, and Madan (2003) show that the price of a -maturity
return variance contract, _
≡ ℚ e , Log , which is the discounted conditional
expectation of the square of market return under the risk-neutral measure, and can be spanned by
a collection of out-of-money call options and out-of-money put options, i.e.,
13
_ 2 1 log ⁄;
2 1 log ⁄; ,
(9)
where ; and ; are the prices of European calls and puts at time written on the
underlying stock with strike price and expiration date at τ. The conditional variance of
market return can be calculated by e ,_
, where satisfies the risk-
neutral valuation relationship, which is related to the first four risk-neutral moments of market
returns as described in equation (39) of Bakshi, Kapadia, and Madan (2003).
Next, we use the model-free realized bipower variance, introduced by Barndorff-Nielsen and
Shephard (2004), to estimate the variance of market variance. Define the intraday stock return as
, ≡ log ⁄ log ⁄ , 1, . . . , , where M is number of five minute intervals
per trading day. Barndorff-Nielsen and Shephard (2004) provide two measures of realized
variation; the first one is the realized variance, ∑ , ,, and the second one is
the bipower variance, ∑ , , . Andersen, Bollerslev, and
Diebold (2002) show that the realized variance converges to the integrated variance plus the jump
contributions, i.e. →
∑ , , where is the number of
return jumps in day t+1 and , is the squared jump size. Moreover, Barndorff-Nielsen and
Shephard (2004) show that →
. In other words, the bipower
variance provides a consistent estimator of the realized variance but solely for the diffusion part.
Our measure for the variance of market variance is estimated from five-minute model-free
implied variances. The intraday model-free implied variances, ⁄ , 1, . . . , , are
calculated using equation (9). Since the process of market variance is a (semi-)martingale, we
14
apply the bipower variance formula on the changes in annualized model-free implied variances
and obtain a measure for the variance of market variance:
Δ2 1 , , (10)
where Δ , ≡ ⁄ ⁄ . Similarly, we estimate diffusion part of
the covariance between the market return and the market variance, CVRV, using the bipower
covariation formula of Barndorff-Nielsen and Shephard (2008):
14
Δ Δ (11)
Based on these measures, our empirical results will not be affected by volatility jumps (or the
return jumps embedded in the volatility).
4. Data and descriptive statistics
4.1. Data
To calculate implied volatility, we use the tick-by-tick quote data for the S&P 500 index (SPX)
options from CBOE’s Market Data Report (MDR) tapes over the time period from January 1996
to December 2015. The underlying SPX prices are also available on the tapes. Daily data for equity
options and S&P 500 index options are obtained from OptionMetrics. We use the Zero Curve file,
which contains the zero-coupon interest rate curve and the Index Dividend file, which contains the
dividend yield, from OptionMetrics. Daily and monthly stock return data are from the CRSP while
intraday transactions data are from TAQ. Financial statement data are from the COMPUSTAT.
Fama and French (1993) factors including their momentum factor, UMD, are obtained from the
15
online data library of Professor Ken French.7 The VIX index is obtained from the CBOE website.8
While we use the ‘new’ VIX index to calculate the market variance risk premium as proposed by
Bollerslev, Tauchen, and Zhou (2009), we also use the ‘old’ VIX, which is based on the S&P 100
options and Black–Scholes implied volatilities, as our volatility factor, following Ang et al. (2006).
We use index option prices from the Option Price file to replicate the market skewness factor and
the market kurtosis factor of Chang, Kristoffersen, and Jacobs (2013).
We follow the literature (see Jiang and Tian 2005; Chang, Kristoffersen, and Jacobs 2013;
among others) to filter out index option prices that violate arbitrage bounds. We eliminate all
observations for which the ask price is lower than the bid price, the bid price is equal to zero, or
the average of the bid and ask price is less than 3/8. We also eliminate in-the-money options (e.g.
put options with K/S>1.03 and call options with K/S<1.03) because prior studies suggest that they
are less liquid. We use the daily SPX low and high prices, downloaded from Yahoo Finance,9 to
filter out the MDR data that are outside the [low, high] interval.
To compute the market volatility-of-volatility, we first partition the tick-by-tick S&P500 index
options data into five-minute intervals. For each maturity within each interval, we linearly
interpolate implied volatilities for a fine grid of one thousand moneyness levels (K/S) between
0.01% and 300%. For moneyness levels below or above the available moneyness level in the
market, we use the implied volatility of the lowest or highest available strike price. The model-
free implied variance is estimated from (9), using out-of-money call and out-of-money put prices.
Linearly interpolated maturities are used to obtain the estimate at a fixed 30-day horizon. Each day,
the market volatility-of-volatility (VOV) is calculated by using the bipower variance formula of
equation (10) with the 81 within-day five-minute annualized 30-day model-free implied variance
7 http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/ 8 http://www.cboe.com/micro/vix/historical.aspx 9 http://finance.yahoo.com/q/hp?s=^GSPC+Historical+Prices
16
estimates covering the normal CBOE trading hours from 8:30 a.m. to 3:15 p.m. Central Time. We
estimate the daily bipower covariance, CVRV, between the market return and the market variance
using the intraday five-minute logarithmic return multiplied by 22 and the five-minute implied
variance for S&P 500 index.
Following Bollerslev, Tauchen, and Zhou (2009), we define the market variance risk premium
( , ) as the difference between the ex-ante implied variance ( , ) and the ex-post realized
variance ( , ), i.e. , ≡ , , . We focus on the 30-day variance risk premium.
Market implied variance ( , ) is the squared ‘new’ VIX index divided by 12. The sum of the
SPX five-minute squared logarithmic returns across the past 22 trading days is used to calculate
the one-month market realized variance ( , ).
To implement our empirical model, we construct innovations in market moments. Following
Ang et al. (2006), the innovation in market volatility (ΔVIX) is its first order difference, i.e.
. Chang, Kristoffersen, and Jacobs (2013) indicate that while the first
difference is appropriate for VIX, an ARMA(1,1) model is needed to remove the time-series
dependence in the skewness and kurtosis factors. Following their approach, the innovations in
market volatility-of-volatility (ΔVOV) is computed as the ARMA(1,1) model residual of the market
volatility-of-volatility. Similarly, an ARMA(1,1) model is used to obtain the innovations in the
bipower covariance, ΔCVRV.
4.2. Descriptive statistics
Figure 1 plots the daily S&P 500 logarithmic return (rSPX) and changes in market volatility
(ΔVIX) over the time period from January 1996 to December 2015. There are clear spikes on the
graph—the Asian financial crisis in 1997, the LTCM crisis in 1998, September 11, 2001, the
WorldCom and Enron bankruptcies in 2001 and 2002, the subprime loan crisis in 2007, the recent
17
financial crisis in 2008, the flash crash in 2010, the battle over the fiscal cliff in 2011 and 2013,
and the recent sovereign debt crisis in Europe.
Table 1 reports descriptive statistics for the daily factors used in this paper, including the four
factors (MKT, SMB, HML, and UMD) of Fama and French (1993) and Carhart (1997), the market
variance risk premium (VRP), the volatility index (VIX; Ang et al., 2006), our measure of variance
of market variance (VOV), the market skewness factor (SKEW) and market kurtosis factor (KURT)
of Chang, Kristoffersen, and Jacobs (2013), the zero beta straddle factor of Coval and Shumway
(2001), and the options return volatility factor (VOL) and the options return jump factor (JUMP)
of Cremers, Halling, and Weinbaum (2015).
In our sample, the mean of 30-day market variance risk premium (VRP) is 0.155%, which is
slightly smaller than 0.183% in Bollerslev, Tauchen, and Zhou (2009) sample. The mean VOV is
0.047%, and its standard deviation is 0.497%. The mean SKEW is -1.864 and the mean KURT is
10.971, suggesting that the risk-neutral distribution of the market return is asymmetric and has fat
tails. The mean delta-neutral straddle return (STR) is -0.346%; the mean delta-neutral, gamma-
neutral, vega positive straddle return (VOL) is -0.039%; the mean delta-neutral, vega-neutral,
gamma positive straddle return (JUMP) is -0.209%. These sample estimates are consistent with
the findings of Cremers, Halling, and Weinbaum (2015) that the STR documented by Coval and
Shumway (2001) is in large part due to the jump component (JUMP).
Panel B reports the Spearman correlations between the daily factors, where the non-return
based state variables are measured by their innovations. As expected, MKT is strongly negatively
correlated with ΔVIX (-0.786), but much less with ΔVOV (-0.054). However, the correlation of
MKT and ΔVOV increases to -0.160 on days when the S&P 500 index declines by more than 1.91%,
(the 5th percentile value of daily rSPX), while the correlation of MKT and ΔVIX increases to -0.581.
VRP is positively correlated with ΔVOV (0.140) and JUMP (0.090), consistent with Bollerslev,
18
Tauchen, and Zhou (2009) and Drechsler and Yaron (2011), respectively. ΔKURT and ΔSKEW are
highly correlated at -0.888, which is comparable to the -0.83 reported by Chang, Kristoffersen,
and Jacobs (2013). Further, STR has a much higher correlation with JUMP (0.929) than that with
VOL (0.263), which is similar to the findings reported by Cremers, Halling, and Weinbaum (2015).
Interestingly, ΔVOV has low correlations with ΔVIX (0.065), ΔSKEW (-0.006), ΔKURT (-0.016),
STR (0.065), VOL (0.055), and JUMP (0.055), which suggests that the pricing power of VOV, if
any, is unlikely to be due to these market moments proposed by previous studies.
While ΔVOV has a low correlation with ΔVIX, VOV determines the market volatility risk.
According Proposition 2, VOV dictates the extent to which the contemporaneous market return and
market volatility covary. Indeed, the results of the following two simple regressions show that the
daily bipower covariance, CVRV, between the market return and the market variance (using the
intraday 5-min logarithmic return multiplied by 22 and the 5-min implied variance for S&P 500
index) is significantly and negatively related to VOV:
0.05310.8
0.45245.656
, 0.293
0.0020.6
0.39351.42
, 0.344
The regression results suggest that the correlation between market returns and the market
variance becomes more negative as VOV increases. Figure 2 depicts VOV and -CVRV (i.e., the
negative of CVRV) over time. The two time-series strongly co-move together. Furthermore, the
results of the following two simple regressions show that the market variance risk premium (VRP)
is significantly and positively related to VOV and significantly and negatively related to CVRV.
0.17317.3
0.145 4.33
, 0.069
0.16316.5
0.277 6.48
, 0.146
19
The results are consistent with our model in which an increase in VOV makes the market more
uncertain, and leads investors to demand a higher risk premium, resulting in a lower stock price
and a negative contemporaneous market return. If VOV is indeed a determinant of the equity risk
premium, the lower stock price now paves the way for a higher future return, which means that
future stock returns are predictable by VOV. We examine this prediction in the next subsection.
4.3. VOV’s predictability of future market returns
Table 2 reports the results of the one-period return predictability regression of daily S&P 500
logarithmic returns (rSPX) in excess of the risk-free rate (rf) on the lagged bipower covariance
(CVRV), market volatility (VIX), variance of market variance (VOV), variance risk premium (VRP),
market skewness (SKEW), market kurtosis (KURT), and jump risk (JUMP). We use robust Newey-
West (1987) t-statistics with the optimal lags to account for autocorrelations.
Panel A presents the regressions at a daily frequency. Consistent with Proposition 3, in
columns [1] through [3], we find that VOV positively predicts one-period ahead daily market return
in all of the specifications that include the aforementioned control variables (except for CVRV).
Consistent with Bollerslev, Tauchen, and Zhou (2009) and Drechsler and Yaron (2011), VRP
significantly predicts future stock market returns. Nevertheless, VOV contains information about
future stock market returns beyond that in VRP.
Moreover, column [4] reports that CVRV negatively predicts one-period ahead daily market
returns. That is, future market returns will be higher when the market return and the market
variance are more negatively correlated (i.e., when VOV is higher). Thus, CVRV also impacts future
market returns and this impact is driven by VOV. In column [5], we find that VOV and CVRV
become less significant when both are in the regression model, suggesting that they contain similar
information about future market returns. Since VOV is still marginally significant and CVRV is
20
insignificant in the regression, it suggests that VOV is the main driving force for the market return
predictability of the covariance of the market return and the market variance.
However, the market variance itself, as measured by VIX, is insignificant in all the
specifications. Thus, while VIX is inversely correlated with the contemporaneous market return, it
contains little information about future market returns.
In Panel B of Table 2, we use the monthly S&P 500 logarithmic returns (rSPX) in excess of
the risk-free rate (rf) as the dependent variable, and the independent variables are sampled at the
end of the previous month. We similarly find that VOV positively predicts one-period ahead
monthly market returns in all of the specifications, except when CVRV is included as one of the
independent variables. Overall, the test results support our proposition that VOV is an important
determinant of the market risk premium, and a state variable that affects the aggregate asset prices.
5. VOV risk in the cross-sectional stock returns
This section examines how VOV risk affects cross-sectional stock returns. Following Ang et
al. (2006), we use daily returns to estimate betas each month t for each firm i listed on the NYSE,
AMEX, and NASDAQ with more than 17 daily observations in a given month.
, , , , , , . (12)
We then use the pre-ranking betas obtained from the above regression in the previous month to
form the test assets. Specifically, we independently sort the pre-ranking betas, , and , ,
each into quintile portfolios to obtain 25 (5x5) test portfolios. After portfolio formation, we
calculate the value-weighted daily and monthly stock returns for each portfolio. Using these 25
portfolio test assets, we test our proposition that, holding other things constant, the expected return
on portfolio p is determined by its three systematic risks implied by our proposed model,
21
, , , , (13)
where , , , , and , are the post-ranking betas obtained from the following time
series regression estimated over the entire sample period,
, , , , , , . (14)
In addition, since our theoretical model (see Proposition 4) results in factor loadings that are related
to time-varying volatility and time-varying volatility-of-volatility, we follow Avramov and
Chordia (2006) to estimate a conditional beta model as follows,
, , , , , , , ,
, , , , ,
, , , , , ,
(15)
Thus, the expected return on portfolio p is determined by
, , , , , , ,
, , , , , . (16)
The above specification facilitates the identification of two time-varying systematic risks,
, , , , , , , and , , , , , ,
which not only nest the constant betas of , and , examined by Ang et al. (2006) but
also incorporate the time-varying betas of , , , , , , and , , implied by our
theoretical model.
5.1.Portfolios results
We first use the , and , sorted portfolios to test how portfolio returns vary with
their post-ranking VIX betas and VOV betas and report the results in Table 3. Specifically, we sort
stocks into quintile portfolios based on , , from the lowest (quintile 1) to the highest (quintile
5), and also independently sort stocks into quintile portfolios based on , . For each of the
quintile portfolios sorted on , , we average across the five , portfolios that intersect
22
with the , portfolio, resulting in , quintile portfolios controlling for , . A similar
approach yields , quintile portfolios controlling for , .
We expect the portfolio returns to decrease with VOV beta for the following reason.
Proposition 2 shows that the market return decreases with VOV and so a stock whose return
increases with VOV i.e., a positive , will provide a hedge against the VOV risk, thus earning
a lower return than a stock with a negative , . Panel A of Table 3 reports that controlling for
, , the quintile 5 stocks (whose returns co-move more positively with VOV beta) have lower
average returns than the quintile 1 stocks with a negative VOV beta, by 72 basis points per month
(t-statistic = -2.25). Controlling for Fama-French (1993) and Carhart (1997) four factor model, the
“5-1” long-short portfolio yields a significant alpha of -0.86 percent per month with a t-statistic of
-2.79. The test results suggest that VOV risk is priced independent of volatility risk and the widely
used factors –market, size, B/M, and momentum factors.
The post-ranking beta, , , lines up with the pre-ranking beta, , and increases from
quintile 1 to the quintile 5 portfolio. The , -sorted “5-1” long-short portfolio has a significant
positive post-ranking , of 0.18 (t=2.26), an insignificant , of 0.01 (t=0.23) and an
insignificant , of 0.01 (t=0.23). Furthermore, it is clear that the post ranking betas are time-
varying. Consistent with Proposition 4, we find that , varies with the market variance and
the VOV and , varies with VOV.10 In terms of the firm characteristics, the quintile 5 firms
are the larger more liquid firms and have lower past month returns as compared to the quintile 1
firms.
Controlling for , , we also find that the return difference between stocks with high
volatility risk and stocks with low volatility risk is significantly negative, at -0.42 percent per
10 Proposition 4 actually shows that , varies with the economic volatility and the economic volatility of volatility but these are in turn related to market volatility and VOV.
23
month with a t-statistic of -1.77. The , -sorted “5-1” long-short portfolio has significant
positive post-ranking , of 0.10 (t=3.97) as well as significant , of 0.24 (t=4.49) and
significant , of 0.03 (t=1.89). The negative impact of , on the cross-section of stock
returns is also consistent with our model and with the results in Ang et al. (2006). Proposition 2
shows that the market return and variance are negatively correlated. Thus, a stock whose return is
positively correlated with the market variance will provide a hedge and earn negative returns.
Note that the , -sorted portfolios are accompanied by dispersed post-ranking factor
loadings with respect to not only the VIX factor but also the MKT and the VOV factors. Further,
unlike in the case of the , -sorted portfolios, the betas are no longer time-varying. In terms of
the firm characteristics, the quintile 5 firms are smaller than the quintile 1 firms.
5.2. Controlling for firm characteristics, high moment risks, and jump risk
We now check whether our results are robust to other competing systematic risks and also
firm characteristics. Table 4 presents results on , -sorted portfolios, controlling for market
capitalization (Size), book-to-market ratio (B/M), past 11-month returns (RET_2_12), past one-
month return (RET_1) and Amihud’s illiquidity (ILLIQ). We also control for other competing
systematic risks including the Chang, Christoffersen, and Jacobs (2013) market skewness risk
( , ) and market kurtosis risk ( , ), the Coval and Shumway (2001) volatility risk with
respect to the zero beta straddle factor ( , ), the Cremers, Halling, and Weinbaum (2015)
volatility risk ( , ) and jump risk ( , ), and the Kelly and Jiang (2014) return tail risk
( , ). As before, we independently sort stocks into quintile portfolios based on , and on
the other factors or characteristics. We then average the returns across the , portfolios.
Panel A of Table 4 presents the post-ranking portfolio alphas for , quintile portfolios
controlling for the firm characteristics and other potential sources of systematic risk. The Fama-
24
French (1993) and Carhart (1997) four factor alpha of the “5-1” long–short portfolio remains
significant at -0.40 percent (t=-1.94) after controlling for Size; at -0.68 percent (t=-2.48) after
controlling for B/M; at -0.54 percent (t= -2.08) after controlling for RET_2_12; at -0.61 percent
(t=-2.16) after controlling for RET_1; and at -0.42 percent (t=-1.98) after controlling for ILLIQ.
Thus, the lower returns on higher , stocks are not driven by the well-known firm
characteristics that impact the cross-section of stock returns.
Panel A further shows that the four-factor alpha of the “5-1” long–short portfolio sorted by
, also remains significant after controlling for the competing systematic risks. Specifically,
the alpha is -0.63 percent (t=-2.34) after controlling for , ; -0.58 percent (t=-2.14) after
controlling for , ; -0.67 percent (t=-2.35) after controlling for , ; -0.71 percent (t=-2.40)
after controlling for , ; -0.78 percent (t=-2.75) after controlling for , ; and -0.70 percent
(t=-2.37) after controlling for , . Again, the results imply that the lower returns on higher
, stocks are not likely due to the existing high-moment risks, jump risk, or tail risk.
Table 4 also presents results for the quintile portfolios sorted on each of the competing
systematic risks. Panel C controls for , , while Panel B does not. Panel B shows that, in our
sample period, only , and , are significantly priced, with their four factor “5-1”
alphas at -0.60 (t=-2.50) and at -0.55 (t=-1.65), respectively. Controlling for , , as shown in
Panel C, the four factor “5-1” alpha for , remains significantly negative at -0.45 (t = -2.13),
but the four factor “5-1” alpha for , is no longer significant. This indicates that a substantial
part of the jump risk premium is due to VOV risk.
In summary, the return differential associated with , cannot be explained by either firm
characteristics or other competing systematic risks. On the other hand, while , and ,
are the only two other competing systematic risks that carry significant return differentials in our
sample period, and moreover , subsumes the pricing effect of , .
25
5.3. Cross-Sectional Regressions
We now run Fama-MacBeth (1973) regressions to estimate the risk premium on VOV risk
while controlling for the other risk factors including the Fama-French and Carhart factors (SMB,
HML, and UMD), the Chang, Christoffersen, and Jacobs (2013) market skewness factor (ΔSKEW)
and market kurtosis factor (ΔKURT), and the Cremers, Halling, and Weinbaum (2015) jump factor
(JUMP). Our main set of test assets are the 25 portfolios formed on intersection of , and
, quintile portfolios. Since , and , are the only two other competing
systematic risks that carry significant return differentials in our sample period as shown in Table
4, we also form 125 (5x5x5) portfolios by sorting on , , β , , and , and yet another
set of 25 portfolios by sorting on , and , . For each portfolio, we use the post-
formation daily value-weighted returns to obtain the post-formation factor loadings. We then run
the cross-sectional regressions to estimate the premiums. Robust Newey and West (1987) standard
errors with six lags that account for autocorrelations are used.
Panel A of Table 5 presents the constant beta results while Panel B presents results for the
time-varying betas. In column [1] of Panel A, the estimated price for VOV beta, , is -3.35 (t
= -2.51), implying that investors are willing to accept a lower return to the tune of 3.35% on their
investments for a unit increase in VOV beta. To apply this price of VOV risk to the “5-1” hedge
portfolio in Table 3, a VOV beta difference of 0.18 between quintile portfolio 5 and quintile
portfolio 1 contributes 0.60 percent lower (i.e., -3.35×0.18=-0.60) per month to the hedge
portfolio’s expected return, which accounts for about 83% of the hedge portfolio’s return of -
0.72% per month. Column [2] shows that adding other risks to the regressions does not materially
change the risk premium on VOV. In fact, none of the other risk factors are significant. In column
[3] where we use the 125 portfolios independently sorted on , , , , and , , we
26
find that the estimate of is significantly negative at -1.60 with a t-statistic of -6.51, but the
estimates of and are insignificant with t-statistics of 0.30 and -0.54, respectively.
Further, in column [4] where we use the 25 portfolios independently sorted on , and , ,
we similarly find that the estimate of remains significantly negative at -2.45 with a t-statistic
of -1.98, but the estimate of is insignificant with a t-statistic of 0.77. Thus, VOV risk is
robust in the presence of other risk factors and with different test assets.
In contrast, while the estimated price of VIX risk, , is significant at -7.09 (t =-1.84) in
column [1], it becomes insignificant when Fama-French-Carhart factors and other higher moment
factors are included as shown in columns [2], [3], and [4]. This suggests that VOV risk has stronger
pricing power than VIX risk.
In Panel B, we estimate the time-varying beta model of equation (15), using VIX and the
VOV as conditioning variables. The risk premium estimates are obtained for the expected return in
equation (16), which gives the estimates of premiums, , , , , , , , , and
. As with the constant beta results in Panel A, we find that is significant across all
different combinations of risk factors. In addition, as reported in column [6], , , , , and
, are significant suggesting that betas are indeed time-varying as suggested by Proposition
4. Even with all the other factors in columns [8], , , and , are still marginally
significant consistent with the time-varying factor loadings as per our theoretical model.
Overall, VOV risk is a surprisingly robust pricing factor!
5.4. Firm-level Fama-MacBeth regressions
Thus far, we have used various portfolio approaches to show that VOV risk is priced. In this
section, we further examine whether the pricing of VOV risk is robust to the firm-level analysis.
Using individual stocks as test assets may avoid potentially spurious results that could arise when
27
the test portfolios exhibit a factor structure (Lewellen, Nagel, and Shanken, 2010). Furthermore,
Lo and MacKinlay (1990) argue that portfolio sorts could lead to data snooping biases. In addition,
a stock-level analysis allows us to control for important firm characteristics.
Specifically, we run the following cross-sectional regression:
, , , , , , , , , , , ,
, , , , , , , ,
, , , , (17)
where the dependent variable is the monthly individual excess stock return. We use the 25
portfolios formed on intersection of , quintile portfolios and , quintile portfolios and,
following the methodology of Fama and French (1992), we assign each of the 25 portfolio-level
post-ranking beta estimates to the individual stocks within the portfolio at that time. Small stocks
ranked in the first decile based on NYSE breakpoints are excluded from the sample. For stock i
that belongs to portfolio p at time t, , , , , , , , , , , , , , , , , , , and
, , are the post-ranking betas from the time-series regression (15). Thus, not only do we
estimate time-varying betas, the individual stock betas can also vary over time because the
portfolio compositions can change each month. We also report the time-constant betas results using
the post-ranking betas , , , , , , and , , from the time-series regression (14).
One again, we use Newey and West (1987) standard errors with six lags.
, denotes a set of firm characteristic variables that consist of Size, B/M, RET_1,
RET_2_12, and ILLIQ. We also check whether our results are robust to existing well-known
volatility spreads that affect cross-sectional stock returns. We construct the implied-realized
volatility spread (IVOL-TVOL), which, as described in Bali and Hovakimian (2009), is defined as
the average of implied volatilities of at-the-money call and put options minus the total volatility
calculated using daily returns in the previous month; the call-put implied volatility spread (CIVOL-
28
PIVOL), which, as described in Bali and Hovakimian (2009) and Yan (2011), is defined as the at-
the-money call option implied volatility minus the at-the-money put option implied volatility.
Since we extract the volatility data from OptionMetrics Volatility Surface file as in Yan (2011), we
choose the 30-day maturity put and call options with deltas equal to -0.5 and 0.5, respectively.
Thus, , denotes a set of volatility characteristic variables that include IVOL-
TVOL, and CIVOL-PIVOL.
Table 6 presents the results from the firm-level Fama-MacBeth regressions. Panel A presents
the constant beta results while Panel B presents the conditional beta results. Column [1] of Panel
A shows that while is insignificant, is significantly negative at -3.378 (t=-5.10) and
the risk premium of VOV at the firm level is close to that estimated from the portfolio approach,
as reported in Table 5. Controlling for IVOL-TVOL and CIVOL-PIVOL, the estimate of
remains significantly negative at -3.10, with a t-statistic of -4.19, as shown in column [2]. In Panel
B, we find that remains significantly negative at -1.11(t=-1.89) controlling for firm
characteristics as shown in column [3] and at -1.24 (t=-1.98) further controlling for volatility
characteristics as shown in column [4]. In addition, consistent with previous findings, , , and
, are significant with t-statistics of 4.08 and -4.40, as reported in column [3], respectively
and 2.60 and -2.87, as reported in column [4], respectively. Thus, once again, consistent with our
theoretical model, we find evidence for time variation in the betas.
The firm characteristics - illiquidity, firm size, the past one month return, the realized
volatility spread, and the call-put implied volatility spread are all significant with signs that are
consistent with the prior literature.
Overall, the firm-level evidence confirms our results that VOV is a priced risk factor.
29
5.5. Market crash and volatility risk
In contrast to the results in Ang et al. (2006), Tables 5 and 6 show that the risk premium on
the volatility (VIX) betas is essentially zero. To better understand why VOV risk is priced but
volatility risk is not, we investigate the performance of the 25 portfolios independently sorted on
, and , during the normal periods and crash periods. Table 7 presents the Fama–
MacBeth (1973) factor premiums during the market crash period and the normal period for our
time-constant and time-varying beta models. In Panel A, Crash is equal to one if the monthly
market factor (MKT) is below -13.10%, its time-series mean minus three times standard deviation;
and zero otherwise; Normal is defined by 1-Crash. The results are similar when we define Crash
differently in Panels B (MKT is lower than the time-series mean minus twice the standard deviation)
and C (MKT is lower than the 5th percentile value).
During normal times, when we remove the observations on the days with extreme market
downturns, Panel A of Table 7 shows that is significantly positive at 1.97 with a t-statistic
of 1.70, is significantly negative at -7.35 with a t-statistic of -1.88, and is significantly
negative at -3.29 with a t-statistic of -2.47. When we consider the time-varying beta, as shown in
column [2], , , , , , and , are significant, but and become
insignificant.
In contrast, during crash periods with extreme market downturns, as shown in columns [3]
and [4], becomes even more negative while becomes significantly negative and
becomes significantly positive. Thus, while large market declines reverse the risk-return
tradeoff for the market risk and volatility risk, the market turmoil strengthens investors’ demand
for compensation for VOV risk. This implies that volatility risk in asset pricing, as proposed by
30
Ang et al. (2006), has a negative premium during normal times but when combined with the
positive premium during crash periods, the overall premium is zero in Tables 5 and 6.
5.6. Time-varying volatility risk
In this section, we examine the risk premiums of the time-constant and the time-varying
volatility risk. Table 8 reports the performance of portfolios sorted on the pre-ranking , and
, , using the time-varying beta model of
, , , , , , ,
, , , , , . (18)
We first sort stocks into quintile portfolios based on , , from the lowest (quintile 1) to the
highest (quintile 5), and within each quintile we further sort stocks into quintile portfolios based
on , , . Portfolios are rebalanced monthly and are value weighted. In Panel A, each of the
quintile portfolios sorted on , , is then averaged over the five portfolios intersected with the
quintile portfolios sorted on , , resulting in , , quintile portfolios controlling for , .
In Panel B, similar approach yields , quintile portfolios controlling for , , . The results
show that, stocks with high , , (quintile 5) have a lower average return than stocks with more
negative , , (quintile 1) by 0.42 percent per month with a t-statistic of -2.35. Controlling for
Fama-French (1993) and Carhart (1997) four factor model, the “5-1” long-short portfolio yields a
significant alpha of -0.57 percent per month with a t-statistic of -2.87. In contrast, the “5-1” long-
short portfolio sorted on , only produces insignificant differentials in stock returns, at 0.10
percent per month with a t-statistic of 0.54. The results confirm the previous findings that while
the constant volatility risk premium has limited explanatory power, the time-varying volatility risk
(conditional on VOV) is priced in the cross-section of stock returns.
6. Conclusion
31
This paper extends Bollerslev, Tauchen, and Zhou (2009) to study the asset pricing
implications of the volatility-of-volatility (VOV), and contributes to the literature along two
dimensions. First, we establish that VOV is an important determinant of the market risk premium
by showing that VOV dictates the negative covariance between the contemporaneous market return
and market volatility and that it predicts future market returns.
Second, we develop a market-based three-factor model in which the market (MKT) risk,
market volatility (VIX) risk, and market VOV risk determine the cross-section of asset returns. We
find that VOV risk is robustly priced and that stocks with more negative VOV betas have
significantly higher future stock returns, even after we account for the exposures to the Fama-
French (1993) and Carhart(1997) factors, the market skewness and kurtosis factors, the jump risk
factor, as well as firm characteristics.
While, the unconditional VIX beta and the MKT beta are not priced in our sample, consistent
with the theoretical model, the conditional component of the factor loadings of both, VIX and the
MKT, are priced. All the three risks (MKT, VIX, VOV) in the asset pricing model are significantly
priced during normal times when we remove the observations with large market declines from our
sample. However, during crash periods, the price of VOV risk becomes even more significant while
the premiums on MKT and VIX have the wrong sign. The positive premium on VIX during market
turmoil combined with the negative premium during normal periods, results in an overall VIX
premium that is indistinguishable from zero. The results imply that market turmoil strengthens
investors’ demand for compensation for VOV risk.
32
References
Adrian, T., and Rosenberg, J., 2008, Stock returns and volatility: Pricing the short-run and long-
run components of market risk, Journal of Finance 63, 2997–3030.
Amihud, Y., 2002, Illiquidity and stock returns: Cross-section and time-series effects, Journal of
Financial Markets 5, 31—56.
Andersen, T. G., T. Bollerslev, and F. X. Diebold. 2007. Roughing it up: Including jump
components in the measurement, modeling, and forecasting of return volatility. Review of
Economics and Statistics 89:701–20.
Ang, A. R. J. Hodrick, Y. Xing, and X. Zhang, 2006. The cross-section of volatility and expected
returns. Journal of Finance 61: 25—299.
Avramov, D., and T. Chordia, 2006, Asset Pricing Models and Financial Market Anomalies.
Review of Financial Studies 19, 1001–1040.
Bakshi, G., and D. Madan. 2006. A theory of volatility spread. Management Science 52: 1945—
56.
Bakshi, G., N. Kapadia, and D. Madan, 2003. Stock return characteristics, skew laws, and the
differential pricing of individual equity options. Review of Financial Studies 16: 101—143.
Bali, T. and A. Hovakimian, 2009, Volatility spreads and expected stock returns, Management
Science, 55: 1797—1812.
Baltussen, G., S. Van Bekkum, and B. Van Der Grient, 2013. Unknown unknowns: Vol-of-vol
and the cross-section of stock returns. AFA 2013 San Diego Meetings Paper.
Bansal, R., and A. Yaron. 2004. Risks for the long run: A potential resolution of asset pricing
puzzles. Journal of Finance 59: 1481—1509.
Barndorff-Nielsen, O., and N. Shephard. 2004. Power and bipower variation with stochastic
volatility and jumps. Journal of Financial Econometrics 2: 1—37.
33
Bekaert, G., and G. Wu. 2000. Asymmetric volatility and risk in equity markets. Review of
Financial Studies 13: 1–42.
Black, F., 1976. Studies of stock price volatility changes. Proceedings of the 1976 Meetings of
the American Statistical Association, Business and Economical Statistics Section, 177–181
Bollerslev, T., N. Sizova, and G. Tauchen, 2012. Volatility in equilibrium: Asymmetries and
dynamic dependencies. Review of Finance 16: 31—80.
Bollerslev, T., G. Tauchen, and H. Zhou, 2009. Expected stock returns and variance risk premia.
Review of Financial Studies 22: 4463—4492.
Bollerslev, T., and V. Todorov, 2011. Tails, fears, and risk premia. Journal of Finance 66: 2165—
2211.
Britten-Jones, M., and A. Neuberger. 2000. Option prices, implied price processes, and stochastic
volatility. Journal of Finance 55:839—866.
Campbell, J. Y., and L. Hentschel, 1992, No news is good news: An asymmetric model of
changing volatility in stock returns, Journal of Financial Economics, 31, 281–318.
Campbell, J. Y., and R. J. Shiller, 1988. The dividend-price ratio and expectations of future
dividends and discount factors. Review of Financial Studies 1: 195—228.
Carr, P., and D. Madan. 1998. Towards a theory of volatility trading. In R. Jarrow (ed.),
Volatility: New Estimation Techniques for Pricing Derivatives, chap. 29, pp. 417—427. London:
Risk Books.
Carr, P., and L. Wu. 2009. Variance risk premia. Review of Financial Studies 22: 1311—1341.
Chang, B. Y., P. Christoffersen, and K. Jacobs, 2013. Market skewness risk and the cross section
of stock returns. Journal of Financial Economics 107: 46—68.
Christie, A. A., 1982. The stochastic behavior of common stock variances—Value, leverage and
interest rate effects. Journal of Financial Economics 10: 407–432.
34
Coval, J., and T. Shumway, 2001, Expected option returns, Journal of Finance 56, 983–1009.
Cremers, M., M. Halling, D. Weibaum, 2015, Aggregate jump and volatility risk in the cross-
section of stock returns, Journal of Finance 70: 577–614.
Drechsler, I., and A. Yaron, 2011. What's vol got to do with it. Review of Financial Studies 24:
1—45.
Drechsler, I., 2013. Uncertainty, time‐varying fear, and asset prices. Journal of Finance 68:
1843—1889.
Epstein, L.G., and S. E. Zin, 1989. Substitution, risk aversion, and the temporal behavior of
consumption and asset returns: A theoretical framework. Econometrica 57: 937—969.
Eraker, B. 2008. Affine general equilibrium models. Management Science 54:2068–80.
Fama, E. F., and K. R. French, 1992, The cross-section of expected stock returns. Journal of
Finance 47, 427—465.
Fama, E. F., and K. R. French, 1993, Common risk factors in the returns on stocks and bonds,
Journal of Financial Economics 33, 3—56.
Fama, E. F., and K. R. French, 2015, A five-factor asset pricing model, Journal of Financial
Economics 116, 1–22.
Fama, E. F., and J. D. MacBeth, 1973, Risk return, and equilibrium: Empirical tests, Journal of
Political Economy 71, 607–636.
Hou, K., C. Xue, and L. Zhang, 2015, Digesting anomalies: An investment approach, Review of
Financial Studies 28, 650-705.
Huang, X., and G. Tauchen. 2005. The relative contribution of jumps to total price variance.
Journal of Financial Econometrics 3:456–99.
35
Jiang, G., and Y. Tian. 2005. Model-free implied volatility and its information content. Review of
Financial Studies 18: 1305—1342.
Kelly, B., H. Jiang. 2014. Tail risk and asset prices. Review of Financial Studies 27, 2841–2871.
Linter, J. 1965. The valuation of risk assets and the selection of risky investments in stock
portfolios and capital budgets. Review of Economics and Statistics 47: 13—37.
Lewellen, J., S. Nagel, and J. Shanken. 2010. A skeptical appraisal of asset pricing tests. Journal
of Financial Economics 96: 175—194.
Maio, P., and P. Santa-Clara, 2012. Multifactor models and their consistency with the ICAPM.
Journal of Financial Economics 106: 586—613.
Merton, R. C., 1973. An intertemporal capital asset pricing model. Econometrica 41: 867—887.
Pan, J., 2002. The jump-risk premia implicit in options: Evidence from an integrated time-series
study. Journal of Financial Economics 63: 3—50.
Sharpe, W. F., 1964. Capital asset prices: A theory of market equilibrium under conditions of
risk. Journal of Finance 19: 425—442.
Wu, G., 2001. The determinants of asymmetric volatility. Review of Financial Studies 14, 837–
859.
Yan, S., 2011. Jump risk, stock returns, and slope of implied volatility smile. Journal of
Financial Economics 99: 216—233.
36
Table 1: Properties of the daily factors
We report summary statistics and Spearman correlations for the daily factors, including the four factors (MKT, SMB, HML, and UMD) of Fama and French (1993) and Carhart (1997), the market variance risk premium (VRP), the VIX index, our measure of variance of market variance (VOV), the bipower covariance (multiplied by 22) between the intraday five-minute logarithmic return and the five-minute implied variance for S&P 500 index (CVRV), the market skewness factor (SKEW) and market kurtosis factor (KURT) of Chang, Christoffersen, and Jacobs (2013), the zero beta straddle factor of Coval and Shumway (2001), and the options return volatility factor (VOL) and the options return jump factor (JUMP) of Cremers, Halling, and Weinbaum (2015). ΔVIX is the first difference of VIX; ΔCVRV, ΔVOV, ΔSKEW, and ΔKURT are the residuals from fitting an ARMA(1,1) regression using VOV, SKEW, and KURT, respectively. The sample period is from January 1996 to December 2015.
Panel A: Summary statistics
MKT SMB HML UMD VRP VIX CVRV VOV SKEW KURT STR VOL JUMP
(%) (%) (%) (%) (%) (%) (%) (%) (%) (%) (%)
Mean 0.030 0.007 0.009 0.026 0.155 21.523 -0.074 0.047 -1.864 10.971 -0.346 -0.039 -0.209
Median 0.070 0.030 0.010 0.070 0.130 20.200 -0.016 0.002 -1.781 9.600 -1.320 -0.079 -0.757
Std.Dev. 1.233 0.627 0.647 0.958 0.197 9.175 0.415 0.497 0.641 5.538 6.212 1.528 4.406
Panel B: Spearman correlation
MKT SMB HML UMD VRP ΔVIX ΔCVRV ΔVOV ΔSKEW ΔKURT STR VOL JUMP
MKT 1.000
SMB 0.138 1.000
HML -0.223 -0.112 1.000
UMD -0.032 0.045 -0.135 1.000
VRP -0.229 -0.060 0.016 0.051 1.000
ΔVIX -0.786 -0.053 0.162 0.007 0.186 1.000
ΔCVRV 0.262 0.038 -0.058 0.036 -0.204 -0.239 1.000
ΔVOV -0.054 -0.008 -0.024 0.034 0.140 0.065 -0.208 1.000
ΔSKEW -0.253 -0.046 0.022 0.021 0.073 0.266 -0.041 -0.006 1.000
ΔKURT 0.324 0.044 -0.043 -0.005 -0.093 -0.321 0.070 -0.016 -0.888 1.000
STR -0.243 -0.085 0.033 -0.038 0.111 0.531 -0.140 0.065 0.128 -0.143 1.000
VOL -0.275 -0.094 -0.022 0.000 0.125 0.376 -0.115 0.055 0.098 -0.111 0.263 1.000
JUMP -0.171 -0.062 0.032 -0.035 0.090 0.421 -0.109 0.055 0.101 -0.112 0.929 -0.034 1.000
37
Table 2: CVRV, VOV, and future stock returns
This table reports the estimates of the one-period return predictability regression using the logarithmic returns on the S&P 500 index (rSPX) in excess of the risk-free rate (rf) on the lagged bipower covariance (multiplied by 22), between the intraday five-minute logarithmic return and the five-minute implied variance for S&P 500 index (CVRV), market volatility (VIX), variance of market variance (VOV), variance risk premium (VRP), market skewness (SKEW), market kurtosis (KURT), and jump risk (JUMP). The dependent variable in Panel A is the daily rSPX-rf multiplied by 22. In Panel B, the dependent variable is the monthly rSPX-rf and the independent variables are sampled at the end of the month. Robust Newey and West (1987) t-statistics are reported in parentheses. The sample period is from January 1996 to December 2015.
Panel A: Dependent variable= daily rSPX-rf (t)
[1] [2] [3] [4] [5]
Intercept -0.457 (-0.29) -0.729 (-0.46) 0.843 (0.36) 2.807 (0.99) -0.229 (-0.16)
VIX (t-1) 0.024 (0.29) -0.059 (-0.63) -0.062 (-0.61) -0.145 (-1.12) 0.012 (0.16)
VOV (t-1) 5.009 (2.10) 4.770 (2.05) 4.767 (2.05) 4.739 (1.66)
CVRV (t-1) -4.864 (-1.91) -0.674 (-0.28)
VRP (t-1)
13.136 (3.75) 13.307 (3.83) 14.847 (3.72)
SKEW (t-1) 2.076 (1.56) 2.300 (1.74)
KURT (t-1) 0.213 (1.43) 0.201 (1.33)
JUMP (t-1) 0.074 (0.70) 0.064 (0.62)
rSPX-rf (t-1) -0.061 (-3.29) -0.039 (-2.14) -0.037 (-2.05) -0.033 (-1.77) -0.061 (-3.26)
Adj. R2 0.012
0.019 0.019 0.016
0.012
Panel B: Dependent variable= monthly rSPX-rf (t)
[1] [2] [3] [4] [5]
Intercept 0.058 (0.06) -0.087 (-0.12) 0.418 (0.33) 0.825 (0.67) 0.029 (0.04)
VIX (t-1) 0.006 (0.13) -0.022 (-0.70) -0.029 (-0.87) -0.048 (-1.39) 0.008 (0.20)
VOV (t-1) 1.671 (2.73) 1.142 (2.02) 1.141 (1.92) 1.753 (1.52)
CVRV (t-1) -0.478 (-2.25) 0.050 (0.10)
VRP (t-1)
4.747 (4.14) 4.591 (3.83) 5.061 (4.97)
SKEW (t-1) -0.230 (-0.19) -0.098 (-0.08)
KURT (t-1) -0.069 (-0.56) -0.059 (-0.48)
JUMP (t-1) 0.014 (0.29) 0.006 (0.12)
rSPX-rf (t-1) 0.104 (1.39) 0.069 (1.02) 0.080 (1.17) 0.100 (1.35) 0.101 (1.34)
Adj. R2 0.008
0.059 0.050 0.052
0.003
38
Table 3: Two-way sorted portfolios on , and ,
At the end of each month, we run the following regression for each stock using daily returns: , , , , , , .
We sort stocks into quintile portfolios based on , , from the lowest (quintile 1) to the highest (quintile 5), and independently sort stocks into quintile portfolios based on , . Portfolios are rebalanced monthly and are value weighted. In Panel A, each of the quintile portfolios sorted on , is then averaged over the five portfolios intersected with the quintile portfolios sorted on , , resulting in , quintile portfolios controlling for , . In Panel B, similar approach yields , quintile portfolios controlling for , . The column “5-1” refers to the hedge portfolio that longs portfolio 5 and shorts portfolio 1. Using the post-formation daily portfolio returns, we estimate the post-formation factor loadings for each portfolio separately from the following time-series regressions,
, , , , , , ,
, , , , , , , , , , where , , , , , , , , , , , , , , and , , = , . Using the post-formation monthly portfolio returns, we compute the excess returns and the risk-adjusted returns of each portfolio with respect to Fama-French and Carhart four factors (MKT, SMB, HML, and UMD). Robust Newey–West (1987) t-statistics are in parentheses. The sample period is from January 1996 to December 2015.
Portfolios ranking
1 2 3 4 5 5-1 t-stat
Panel A: Performance of , sorted portfolio, controlling for ,
Excess return 0.92 0.71 0.50 0.51 0.20 -0.72 (-2.25)
α-CAPM 0.16 0.10 -0.06 -0.11 -0.58 -0.74 (-2.16)
α-FF3 0.14 0.08 -0.08 -0.12 -0.58 -0.72 (-2.33)
α-FFC4 0.32 0.12 -0.08 -0.12 -0.54 -0.86 (-2.79)
Post-ranking factor loadings
, 1.34 1.09 1.01 1.07 1.35 0.01 (0.23)
, 0.07 0.02 0.00 0.01 0.08 0.01 (0.23)
, -0.03 -0.01 0.02 0.02 0.15 0.18 (2.26)
Post-ranking time-varying factor loadings
, 1.17 0.98 0.96 1.10 1.50 0.32 (4.93)
, , 0.18 0.10 0.14 0.11 -0.06 -0.24 (-1.95)
, , 2.22 1.86 4.43 -0.89 -12.52 -14.74 (-2.00)
, 0.03 0.00 0.00 0.03 0.12 0.09 (3.67)
, , 1.08 0.74 1.94 -0.29 -5.09 -6.17 (-1.90)
, 0.05 0.03 0.00 0.00 0.08 0.03 (2.08)
Pre-formation firm characteristics
Size($b) 1.46 3.39 4.03 3.75 1.81 0.35 (4.01)
B/M 1.08 0.88 0.84 0.82 1.01 -0.07 (-1.38)
RET_2_12 12.46 14.80 15.06 14.61 12.34 -0.12 (-0.10)
RET_1 2.10 1.06 0.64 0.28 1.03 -1.07 (-5.68)
ILLIQ(106) 7.81 3.31 2.96 3.31 7.54 -0.27 (-1.65)
39
Table 3 (continued.)
Portfolios ranking
1 2 3 4 5 5-1 t-stat
Panel B: Performance of , sorted portfolio, controlling for ,
Excess return 0.68 0.65 0.68 0.57 0.26 -0.42 (-1.77)
α-CAPM -0.02 0.07 0.11 -0.08 -0.57 -0.55 (-2.59)
α-FF3 0.00 0.06 0.08 -0.12 -0.60 -0.59 (-2.78)
α-FFC4 0.11 0.12 0.10 -0.11 -0.53 -0.64 (-2.58)
Post-ranking factor loadings
, 1.23 1.01 1.01 1.13 1.47 0.24 (4.49)
, 0.03 -0.01 0.00 0.03 0.13 0.10 (3.97)
, 0.05 0.03 0.00 0.00 0.08 0.03 (1.89)
Post-ranking time varying factor loadings
, 1.27 1.10 0.99 1.08 1.28 0.02 (0.26)
, , 0.26 0.00 0.06 -0.05 0.20 -0.07 (-0.42)
, , -10.89 -5.52 -1.86 7.28 6.08 16.97 (1.53)
, 0.08 0.02 0.00 0.01 0.07 -0.01 (-0.40)
, , -4.73 -2.64 -0.61 3.18 3.17 7.90 (1.64)
, -0.02 -0.01 0.02 0.02 0.15 0.17 (2.21)
Pre-formation firm characteristics
Size($b) 1.87 4.03 4.13 3.07 1.32 -0.55 (-4.33)
B/M 1.03 0.88 0.81 0.83 1.07 0.04 (0.94)
RET_2_12 11.44 15.03 15.47 14.96 12.38 0.94 (0.57)
RET_1 1.30 0.71 0.72 0.80 1.57 0.27 (1.14)
ILLIQ(106) 8.25 3.12 2.60 2.96 8.00 -0.25 (-1.41)
40
Table 4: Two-way sorted portfolios on , and control variables
This table shows performance of portfolios sorted on , , with controlling market capitalization (Size), book-to-market ratio (B/M), past 11-month returns (RET_2_12), past one-month return (RET_1), and Amihud’s illiquidity (ILLIQ), Chang, Christoffersen, and Jacobs’s (2013) market skewness risk ( , ) and market kurtosis risk ( , ), Coval and Shumway’s (2001) volatility risk with respect to the zero beta straddle factor ( , ), Cremers, Halling, and Weinbaum’s (2015) volatility risk ( , ) and jump risk ( , ), and Kelly and Jiang’s (2014) return tail risk ( , ), respectively. We sort stocks into quintile portfolios based on , , from the lowest (quintile 1) to the highest (quintile 5), and independently sort stocks into quintile portfolios based on each control variable. Portfolios are rebalanced monthly and are value weighted. Panel A presents the results for , quintile portfolios controlling for the quintile portfolios sorted on each control variable. The results for the quintile portfolios sorted on each of the competing systematic risks without and with controlling for , quintile portfolios are presented in Panel B and in Panel C, respectively. The column “5-1” refers to the hedge portfolio that longs portfolio 5 and shorts portfolio 1. Using the post-formation monthly portfolio returns, we compute the risk-adjusted returns of each portfolio with respect to Fama-French and Carhart four factors (MKT, SMB, HML, and UMD). Robust Newey–West (1987) t-statistics are in parentheses. The sample period is from January 1996 to December 2015.
Portfolios ranking
Control variables 1 2 3 4 5 5-1 t-stat
Panel A: Performance of , sorted portfolio, controlling for each control variable
Size 0.24 0.29 0.28 0.21 -0.16 -0.40 (-1.94)
B/M 0.42 0.14 0.01 0.03 -0.26 -0.68 (-2.48)
RET_2_12 0.17 0.08 0.12 -0.03 -0.37 -0.54 (-2.08)
RET_1 0.15 0.11 -0.05 0.02 -0.47 -0.61 (-2.16)
ILLIQ 0.10 0.25 0.23 0.13 -0.32 -0.42 (-1.98)
, 0.22 0.20 0.00 0.00 -0.41 -0.63 (-2.34)
, 0.15 0.09 0.06 -0.06 -0.43 -0.58 (-2.14)
, 0.26 0.18 -0.05 -0.03 -0.41 -0.67 (-2.35)
, 0.18 0.15 -0.05 -0.12 -0.53 -0.71 (-2.40)
, 0.26 0.18 -0.07 0.00 -0.52 -0.78 (-2.75)
, 0.39 0.28 0.09 -0.06 -0.31 -0.70 (-2.37)
41
Table 4 (continued.)
Portfolios ranking
Control variables 1 2 3 4 5 5-1 t-stat
Panel B: Performance of each control variable sorted portfolio
, 0.28 0.01 0.06 -0.08 -0.31 -0.60 (-2.50)
, -0.24 0.02 0.05 0.01 0.04 0.28 (1.11)
, 0.06 0.15 0.15 -0.22 -0.28 -0.34 (-1.01)
, -0.27 -0.03 0.12 -0.02 -0.07 0.20 (0.76)
, 0.13 0.17 0.01 -0.06 -0.41 -0.55 (-1.65)
, -0.15 -0.01 0.17 0.18 0.23 0.38 (1.44)
Panel C: Performance of each control variable sorted portfolio, controlling for ,
, 0.20 0.01 0.08 -0.04 -0.25 -0.45 (-2.13)
, -0.12 -0.08 -0.01 -0.01 0.04 0.16 (0.71)
, 0.10 0.12 0.11 -0.18 -0.19 -0.29 (-0.96)
, -0.23 -0.10 0.06 -0.05 -0.04 0.19 (0.90)
, 0.10 0.13 -0.02 -0.01 -0.35 -0.44 (-1.53)
, -0.15 -0.04 0.15 0.26 0.17 0.32 (1.30)
Table 5: Fama-MacBeth regressions
This table presents the Fama–MacBeth (1973) factor premiums for the volatility-of-volatility factor (ΔVOV), with controlling for the market factor (MKT), the volatility factor (ΔVIX), Fama-French and Carhart factors (SMB, HML, and UMD), Chang, Christoffersen, and Jacobs’s (2013) market skewness factor (ΔSKEW) and market kurtosis factor (ΔKURT), and Cremers, Halling, and Weinbaum’s (2015) jump factor (JUMP). We estimate the first stage return betas using the daily full-sample post-formation value-weighted returns. Then, we regress the cross-sectional monthly portfolio returns on daily return betas from the first stage, using Fama–MacBeth (1973) cross-sectional regression. In Panel A, the testing model is , , , where , , , , and , are the post-ranking betas from the time-series regression of , , , , , , . In Panel B, the testing model is , , , , , , , , , , , , , where the post-ranking betas are from , , , , , , , , , where , , , , , , , , and , , , , , Three sets of test portfolios are considered. In columns [1], [2], and [5] through [8], the test portfolios are the 25 portfolios independently sorted on , and , . Column [3] uses the 125 portfolios independently sorted on , , , , and , . Column [4] uses the 25 portfolios independently sorted on , and , . Portfolios are rebalanced monthly and are value weighted. Robust Newey–West (1987) t-statistics that account for autocorrelations are in parentheses. The sample period is from January 1996 to December 2015.
Fama-MacBeth cross-sectional regressions
Panel A: Constant beta models Panel B: Time-varying beta models
Test portfolios , × , , × , , × , , × , , × , , × , , × , , × ,
× ,
[1] [2] [3] [4] [5] [6] [7] [8]
Intercept -1.23 (-1.08) 0.30 (0.23) -1.95 (-1.87) 2.42 (1.58) 1.04 (1.39) -0.62 (-0.54) 2.36 (2.19) 0.25 (0.18)
, 1.84 (1.60) 0.33 (0.24) 2.95 (2.59) -1.92 (-1.16) -0.36 (-0.45) 1.22 (1.02) -1.72 (-1.51) 0.34 (0.23)
, -7.09 (-1.84) -4.19 (-0.96) 0.44 (0.22) 5.14 (1.12) 1.31 (0.41) -4.47 (-1.12) 3.06 (0.92) -3.17 (-0.71)
, -3.35 (-2.51) -3.01 (-2.84) -1.60 (-6.51) -2.45 (-1.98) -3.27 (-2.54) -2.72 (-2.59) -3.23 (-2.87) -2.53 (-2.77)
, , 0.90 (1.82) 0.61 (1.24)
, , 0.13 (1.85) 0.14 (1.72)
, , -0.31 (-1.87) -0.32 (-1.73)
, -0.47 (-0.58) -1.36 (-1.74) 1.69 (1.40) -0.34 (-0.54) -0.41 (-0.50)
, -0.14 (-0.26) -0.39 (-0.72) 1.38 (1.54) 0.16 (0.33) -0.05 (-0.08)
, -0.85 (-0.95) -0.19 (-0.16) 0.83 (0.87) -2.29 (-2.13) -0.35 (-0.31)
, 1.79 (1.09) 0.03 (0.30) 2.19 (1.29) 1.71 (1.06) 1.88 (1.12)
, -8.09 (-0.72) -0.15 (-0.54) -16.57 (-1.31) -15.52 (-1.22) -8.56 (-0.75)
, -1.78 (-0.10) 4.24 (0.40) 11.82 (0.77) 13.46 (0.79) -7.12 (-0.37)
Adj. R2 0.70 0.67 0.39 0.61 0.55 0.72 0.63 0.66
Table 6: Firm-level Fama-MacBeth regressions
This table reports the results for the firm-level Fama-MacBeth regressions. We run the following cross-sectional regression:
, , , , , , , , , , , , , ,
, , , , , , , , , ,
where the dependent variable is the monthly individual stock returns; We use the 25 portfolios formed on intersection of , quintile portfolios and , quintile portfolios and, following the methodology of Fama and French (1992), we assign each of the 25 portfolio-level post-ranking beta estimates to the individual stocks within the portfolio at that time. Small stocks ranked in the first decile based on NYSE breakpoints are excluded. In Panel A, for stock i that belongs to portfolio p at time t, , , , , , , and , , are the post-ranking betas from the time-series regression of , , , , , , ; In Panel B,
, , , , , , , , , , , , , , , , , , and , , are from regression of , ,
, , , , , , , , , , , .
, consists of Size, B/M, RET_2_12, RET_1, and ILLIQ; , includes the implied-realized volatility spread (IVOL-TVOL) and the call-put implied volatility spread (CIVOL-PIVOL). Robust Newey and West (1987) t-statistics that account for autocorrelations are reported in parentheses. The sample period is from January 1996 to December 2015.
Fama-MacBeth regressions: individual stocks
Panel A: Constant beta models Panel B: Time-varying beta models
[1] [2] [3] [4]
Intercept -1.09 (-0.96) 0.56 (0.44) -0.01 (-0.01) 1.45 (1.15)
Log(Size($b)) -0.27 (-4.95) -0.19 (-3.02) -0.27 (-4.92) -0.19 (-2.98)
Log(B/M) 0.11 (0.96) 0.10 (0.77) 0.11 (0.95) 0.10 (0.78)
RET_2_12 0.34 (0.99) 0.20 (0.56) 0.35 (1.02) 0.21 (0.57)
RET_1 -1.70 (-2.80) -1.52 (-2.12) -1.68 (-2.77) -1.52 (-2.12)
ILLIQ(106) 0.49 (2.46) 2.91 (3.08) 0.50 (2.51) 2.93 (3.09)
, 1.96 (1.65) 0.37 (0.28) 0.93 (0.77) -0.47 (-0.35)
, 0.08 (0.02) 4.25 (0.91) 2.57 (0.66) 5.81 (1.28)
, -3.00 (-4.33) -3.10 (-4.19) -1.11 (-1.89) -1.24 (-1.98)
, , 0.23 (0.50) -0.37 (-0.73)
, , 0.20 (4.08) 0.16 (2.60)
, , -0.49 (-4.40) -0.40 (-2.87)
IVOL-TVOL 0.76 (3.72) 0.76 (3.71)
CIVOL-PIVOL 4.05 (6.63) 4.01 (6.66)
Adj. R2 0.07 0.08 0.07 0.08
No. obs 570,718 381,187 570,718 381,187
44
Table 7: Market crash and factor premiums
This table presents the Fama–MacBeth (1973) factor premiums during the market crash period and the normal period for the market factor (MKT), the volatility factor (ΔVIX), and the volatility-of-volatility factor (ΔVOV). The test portfolios are the 25 portfolios independently sorted on , and , and the portfolios are rebalanced monthly and value weighted. Crash is equal to one if the monthly market factor (MKT) is below -13.10%, its time-series mean minus three times standard deviation; and zero otherwise; Normal is defined by 1-Crash. Using the post-formation daily portfolio returns, we compute the risk exposures with respect to our constant beta model and time-varying beta model. Then we regress the cross-sectional monthly portfolio returns on betas from the first stage, using Fama–MacBeth (1973) cross-sectional regression. The constant beta model is , ,
, where the post-ranking betas are from the time-series regression of , ,
, , , , . The time-varying beta model is
, , , , , , , , , , , , , where the post-ranking betas are from , , , , , , , , , where
, , , , , , , , and , , , , , . We report the estimates of factor premiums using the Normal periods in columns [1] and [2] and using the Crash periods in columns [3] and [4]. Robust Newey and West (1987) t-statistics that account for autocorrelations are reported in parentheses. The sample period is from January 1996 to December 2015.
Normal periods Crash periods
[1] [2] [3] [4]
Panel A: Crash(t) =1 if MKT(t) <-13.10% (Mean-3*SD of MKT)
Intercept -1.22 (-1.07) -0.66 (-0.56) -2.68 (-6.04) 3.79 (1.16)
, 1.97 (1.70) 1.40 (1.17) -14.02 (-80.10) -20.63 (-5.45)
, -7.35 (-1.88) -4.92 (-1.21) 23.62 (2.74) 48.87 (7.35)
, -3.29 (-2.47) -2.70 (-2.57) -10.43 (-1.96) -5.34 (-4.20)
, , 0.94 (1.87) -3.84 (-4.68)
, , 0.13 (1.83) 0.41 (0.53)
, , -0.31 (-1.84) -0.98 (-0.62)
Adj. R2 0.69 0.70 0.27 0.18
45
Table 7 (continued.)
Normal periods Crash periods
[1] [2] [3] [4]
Panel B: Crash(t) =1 if MKT(t)< -8.54% (Mean- 2*SD of MKT)
Intercept -1.59 (-1.36) -1.16 (-0.98) 9.23 (1.70) 15.02 (3.01)
, 2.60 (2.17) 2.16 (1.81) -20.18 (-4.03) -25.98 (-5.80)
, -8.67 (-2.17) -6.14 (-1.54) 38.59 (1.66) 43.80 (1.90)
, -2.53 (-2.39) -2.31 (-2.47) -26.94 (-3.10) -14.41 (-2.16)
, , 1.12 (2.16) -5.57 (-2.29)
, , 0.14 (1.71) 0.04 (0.06)
, , -0.31 (-1.69) -0.40 (-0.30)
Adj. R2 0.54 0.53 0.74 0.76
Panel C: Crash(t) =1 if MKT(t)< -7.88 (5th percentile of MKT)
Intercept -1.37 (-1.20) -1.03 (-0.86) 1.43 (0.22) 7.14 (1.04)
, 2.55 (2.17) 2.19 (1.83) -11.55 (-1.82) -17.27 (-2.56)
, -8.28 (-2.06) -5.91 (-1.43) 15.42 (0.65) 22.88 (1.02)
, -2.41 (-2.28) -2.35 (-2.43) -21.15 (-2.85) -9.71 (-1.73)
, , 1.07 (2.12) -2.35 (-0.92)
, , 0.13 (1.58) 0.20 (0.40)
, , -0.29 (-1.54) -0.74 (-0.71)
Adj. R2 0.53 0.52 0.74 0.79
46
Table 8: Performance of portfolios formed on , and , ,
At the end of each month, we run the following regression for each stock using daily returns:
, , , , , , ,
, , , , , . We first sort stocks into quintile portfolios based on , , from the lowest (quintile 1) to the highest (quintile 5), and within each quintile we further sort stocks into quintile portfolios based on , , . Portfolios are rebalanced monthly and are value weighted. In Panel A, each of the quintile portfolios sorted on , , is then averaged over the five portfolios intersected with the quintile portfolios sorted on , , resulting in , , quintile portfolios controlling for , . In Panel B, similar approach yields , quintile portfolios controlling for , , . The column “5-1” refers to the hedge portfolio that longs portfolio 5 and shorts portfolio 1. Using the post-formation monthly portfolio returns, we compute the excess returns and the risk-adjusted returns of each portfolio with respect to Fama-French and Carhart four factors (MKT, SMB, HML, and UMD). Robust Newey–West (1987) t-statistics are in parentheses. The sample period is from January 1996 to December 2015.
Portfolios ranking
1 2 3 4 5 5-1 t-stat
Panel A: Performance of , , sorted portfolio, controlling for ,
Excess return 0.68 0.71 0.65 0.49 0.26 -0.42 (-2.35)
α-CAPM -0.03 0.08 0.04 -0.18 -0.56 -0.53 (-3.09)
α-FF3 -0.03 0.07 0.03 -0.18 -0.58 -0.56 (-3.01)
α-FFC4 0.05 0.11 0.06 -0.17 -0.51 -0.57 (-2.87)
Panel B: Performance of , sorted portfolio, controlling for , ,
Excess return 0.39 0.66 0.70 0.56 0.49 0.10 (0.54)
α-CAPM -0.44 0.05 0.16 -0.07 -0.34 0.09 (0.48)
α-FF3 -0.41 0.02 0.14 -0.09 -0.36 0.05 (0.27)
α-FFC4 -0.22 0.04 0.14 -0.13 -0.29 -0.07 (-0.33)
47
Figure 1. Daily market returns (rSPX) and market volatility (VIX). We plot daily market returns (rSPX) and changes in market volatility (ΔVIX) over the time period from January 1996 through December 2015.
48
Figure 2 Daily market covariance between return and variance (CVRV), and market volatility-of-volatility (VOV). We plot daily minus market covariance between return and variance (-CVRV) and market volatility-of-volatility (VOV) over the time period from January 1996 through December 2010. We partition the tick-by-tick S&P500 index options data into five-minute intervals, and then we estimate the model-free implied variance for each interval. For each day, we use the bipower variation formula on the five-minute based annualized 30-day model-free implied variance estimates within the day, resulting in our daily measure of market volatility-of-volatility (VOV). We estimate the covariance between the market return and the market variance (CVRV) using the bipower covariation formula on the five-minute based S&P500 index returns and the five-minute based annualized 30-day model-free implied variance within the day.