Gaining the alpha advantage in volatility

trading

Artur Sepp
artursepp@gmail.com

Quantitative Investment Strategies Summit

Global Derivatives Trading & Risk Management 2015
Amsterdam
May 18, 2015

1
Electronic copy available at: https://ssrn.com/abstract=3032098
Headlines

1. Present some empirical evidence for short volatility strategies and the
cyclical pattern of their P&L: alpha in good times, beta in bad times

2. Introduce a factor model with risk-aversion to explain the risk-premium

of short volatility strategies as a compensation to bear losses in bad
market regimes

3. Consider an econometric model for statistical inference of market

regimes and for optimal position sizing

4. Illustrate model applications for generating alpha from volatility strate-

gies

2
Electronic copy available at: https://ssrn.com/abstract=3032098
Short volatility strategies out-perform the benchmark
CBOE indices related to short volatility strategies (not delta-hedged)

The longest time series from 1986 including 87’ crash, 07’ GFC

• Call (BuyWrite Index): long S&P500 index and short 3m at-the-money

index call (quarterly re-balancing)

• Put (PutWrite Index): short 3m at-the-money put on S&P500

Both have limited upside but unlimited downside

Yet outperform the S&P500 index with higher Sharpe ratios

S&P500 Call Put

Return 9.4% +0.7% +2%

Volatility 17% ×0.72 ×0.70

Sharpe 0.53 × 1.5 × 1.7

3
Short volatility strategies out-perform the benchmark in
the long-term by a wide margin
S&P500 Call Put
1$ Value 8.2 × 1.4 × 2.0

16 1$ investment
S&P 500 index
12
Call Index
Put Index
8

0
Jun-86

Jun-90

Jun-94

Jun-98

Jun-02

Jun-06

Jun-10

Jun-14
4
 Out-performance of short volatility strategies is related
to the risk-premium in implied volatility
Implied volatility is a measure of volatility at which options are quoted/traded
in the market (Black-Scholes vol)
Realized volatility is a measure of price-returns volatility to replicate op-
tion pay-offs by delta-hedging
Volatility Risk-premium = Implied volatility − Realized volatility
Figure: Proxy of volatility risk-premium =
VIX at month start−Realized volatility of S&P500 daily returns in this month

Volatility Risk-Premium
20%

0%

-20% Average = 3.6%

-40% +1 StandardDeviation=12%

-60% -1 StandardDeviation=-5%
Jul-86

Jul-90

Jul-94

Jul-98

Jul-02

Jul-06

Jul-10

Jul-14
5
Straddle is the strategy for monetizing the volatility risk-
premium
Short Straddle = short at-the-money put and call options

Figure: Pay-off profile is function of absolute return: its value is sensitive

to the level of volatility
Delta-hedged straddle P&L = Volatility Risk-premium × |Straddle Vega|

10%
Straddle Pay-off
5%

0%

-5%
Price return
-10%
-10% -5% 0% 5% 10%
6
Skew risk-premium in implied volatility is more signifi-
cant than the volatility premium
Volatility skew measures how implied (at-the-money) or statistical (real-
ized) volatility changes when the index changes
Volatility skew is negative because of the leverage effect
Implied skew = (105% Call Vol − 95% Put Vol) / 5%
Realized skew is measured by the volatility beta
Figure: Proxy of skew risk-premium =
1m Implied Skew−1m Realized skew of S&P500

0.5 Skew Risk-Premium (negative)

0.0
-0.5
Average = -0.29
-1.0 -1 StandardDeviation=-0.47
-1.5 +1 StandardDeviation=-0.11
Feb-06
Feb-07
Feb-08
Feb-09
Feb-10
Feb-11
Feb-12
Feb-13
Feb-14
7
Risk-reversal is the strategy for monetizing the skew
risk-premium
Short Risk-Reversal = short put with out-of-the money strike and long
call with out-of-the money strike

Figure: Pay-off profile of 95% − 105% risk-reversal: its value is sensitive

to implied skeweness / spot-vol correlation

Delta-hedged risk-reversal P&L:

P&L = Skew Risk-premium × Risk-Reversal Vanna

10%
Risk-Reversal Pay-off
5%

0%

-5%
Price return
-10%
-10% -5% 0% 5% 10% 8
The volatility risk-premium is realized by selling options
at implied volatility net of realized volatility as costs for
delta-hedging

Can the risk-premium be source of αlpha or it is just a beta?

”Look deep into nature, and then you will understand everything better”
- Albert Einstein

I present a new factor model with risk-aversion to explain volatility

risk-premium and P&L of vol strategies

9
Risk-aversse investors assign larger weights for negative
returns
”Individuals are risk averse if they always prefer to receive a fixed payment
to a random payment of equal expected value” - Dumas&Allaz (1996)

For valuation of derivative securities:

• The statistical measure of returns is estimated using econometric tools
• The valuation measure is assigned exponential weight with positive risk-
aversion parameter to increase impact of negative future returns

2.0 Weight for Statistical Measure

Weight for Valuation Measure

1.5
with Risk-Aversion

1.0

0.5
Future return
0.0
-5 0 5 10
The valuation measure with risk-aversion assigns higher
probability to negative returns if price-returns under the
statistical measure are fat-tailed
Risk-averse investors are ready to pay more to buy options for protecting
against negative returns
Figure: The statistical measure with zero skeweness and large excess
kurtosis of size 11.0 (daily returns for S&P500 index)
The valuation measure with risk-aversion implies large negative
skeweness even if the statistical distribution has no skeweness
Statistical Measure Valuation Measure
Implied Skeweness 0.00 -1.04
Fat-tailed PDF (Log-scale)
Statistical Measure

Valuation Measure
with Risk-Aversion

Future Return
-5 -3 -1 1 3 5 11
Model implies that the volatility risk-premium arises from:
• Fat tails of the statistical distribution of returns
• Investors’ risk-aversion
As driver for P&L:
Risk-premium = Implied risk − Realized risk
risk = volatility, skew, kurtosis
Statistical Measure
Implied Option Vol
Valuation Measure
with Risk-aversion

Skew
Premium
Volatility
Premium

Puts % Strike of Spot Calls

75% 100% 125%
12
 If volatility of strategy returns is negatively correlated to
market return, expected return on this strategy should
be higher
Classic capital asset pricing model (CAPM) links expected excess returns
on trading strategy to returns on a market factor:
Beta of Asset return to Market Factor
Expected Excess Return =
× Expected market Return

In my model for volatility strategies under the risk aversion:

Expected Excess Return = CAPM term + Risk-Premium term

With positive Risk-Premium term:

- Risk-Aversion Parameter
Risk-Premium term = × Beta of Asset Volatility to Market Return
× Expected market volatility

When the index goes down, realized vol of P&L of short index puts
increases due to negative gamma and vega so put sellers need to charge
extra premium for negative covariance between price-returns and P&L vol
13
 Empirical evidence: risk aversion leads to higher realized
returns on short volatility strategies
Modify my theoretical equation for the expected return under the risk-
aversion into a factor model:
StrategyReturn = βM arket × MarketReturn + βV olP remium × VolPremium + α

Volatility risk premium proxied by:

VolPremium = VIX − Monthly Realized Volatility Of S&P500

Using monthly returns for CBOE vol indices from 1986:

Call Put
Market Beta, βM arket 0.56 0.47
VolPremium Beta, βV olP remium 0.98 1.22
Alpha, α Insignicant
Explanatory power R2 82% 74%
Significance of coefficient ”VolPremium Beta”, βV olP remium, arises
from the risk-aversion implicit in option market prices
Under risk-neutral valuation theory, ”VolPremium Beta” must be zero
14
Factor model for portfolios of vol strategies
Aggregated portfolio alpha comes from the risk-premium arising
from investors’ risk-aversion
Residual risk comes from random changes in the risk-premium and
realized index volatility
By delta-hedging, the exposure to market return is negligible
With uncorrelated residual risks n and insignificant residual alphas, the
portfolio return with allocation wn is
X
PortfolioReturn = wn × βV olP remium,n × VolPremiumn
n=N
Representing vol premium by factor model:
VolPremiumn = Alphan − VolPremium Residual Riskn
with positive Alphan and stochastic residual risk (gamma, vega, etc)
Portfolio return is weighted sum of alphas and risks:
X X
PortfolioReturn = wn × Alphan − wn × VolPremium Residual Riskn
n=N n=N

15
To generate alpha from selling volatility, we need to
model market cycles and size positions accordingly
In good times:
• Exposure to index returns is removed by delta-hedging
• Residual risks of volatility premiums (gamma, vega) are idiosyncratic
so for diversified portfolio these are negligible
In bad times:
Residual risks become common with risk-premiums spiking down

Vol risk-premiums - compensation to bear losses in bad times

Log of Initial Investment
60% S&P 500 Index
40% Short 1m Delta-Hedged Straddle
20%
0%
-20%
-40%
-60%
Jan-05
Jan-06
Jan-07
Jan-08
Jan-09
Jan-10
Jan-11
Jan-12
Jan-13
Jan-14
Jan-15
16
Market cycles are evident in empirical data
• Periods with positive drift and low volatility
• Periods with large negative returns and high volatility
1900 S&P 500 index
1700
1500
1300
1100
900
700
Aug-00
Aug-01
Aug-02
Aug-03
Aug-04
Aug-05
Aug-06
Aug-07
Aug-08
Aug-09
Aug-10
Aug-11
Aug-12
Aug-13
Aug-14
90% Monthly volatility of daily returns
80%
70%
60%
50%
40%
30%
20%
10%
0%
Aug-00
Aug-01
Aug-02
Aug-03
Aug-04
Aug-05
Aug-06
Aug-07
Aug-08
Aug-09
Aug-10
Aug-11
Aug-12
Aug-13
Aug-14
17
 Market cycles are statistically significant with fat tails of
returns explained by regime changes in realized volatility

Left figure: QQ-plot of monthly returns on the S&P 500 from 1986
Right: QQ-plot of monthly returns normalized by the realized historic
volatility of daily returns within given month
Using statistical tests for normality of returns:
• Strong presumption against normality of returns
• Cannot reject normality of volatility-normalized returns

Monthly Return Volatity Normalized Return

Data Quantiles

Data Quantiles

Normal Quantiles Normal Quantiles

18
Kalman filter of Hamilton (1988) is applied to infer the
probabilities of the stressed regimes out-of-sample
Apply regime-switching model with normal and stressed market cy-
cles
Model statistical inference:
1. Distributions of returns in normal and stressed regimes
2. In-sample inferred probability of stressed regime from observed data
3. Out-of-sample forecast for regime probabilities using Bayesian-type
update with new market data
Probability of Stressed Regime
Model with In-Sample Forecast
Model with Out-of-Sample Forecast
100%

50%

0%
Jan-00
Jan-01
Jan-02
Jan-03
Jan-04
Jan-05
Jan-06
Jan-07
Jan-08
Jan-09
Jan-10
Jan-11
Jan-12
Jan-13
Jan-14
Jan-15
19
Regime identification and their probabilities are applied
for optimal position sizing

Maximize the log of investment value to obtain the Kelly betting rule
conditional on market regimes:
Expected Return In Normal Regime
Position In Normal Regime =
Variance of Return In Normal Regime
Expected Return In Stressed Regime
Position In Stressed Regime =
Variance of Return In Stressed Regime

Expected Return In Normal Regime is positive so be long risk-premiums

Expected Return In Stressed Regime is negative so be short risk-premiums

20
 Volatility selling using 1m delta-hedged short Straddle
on the S&P500 index - the strategy with optimal posi-
tion sizing outperforms
Daily re-hedging and weekly re-balancing
Transaction costs: 5bp per spot, 1% per implied vol
• Continuous: roll into new straddle every week
• Optimal: size position using out-of-sample forecast of regime probabilities:
No trading when forecast probability of stressed regimes is high
Continuous Optimal %
Trades # 519 173 33%
Sortino -0.39 0.98
Sharpe -0.34 0.74
40,000 Continuous
Straddle P&L Optimal Sizing
20,000

-20,000

-40,000

-60,000
Jan-05
Jan-06
Jan-07
Jan-08
Jan-09
Jan-10
Jan-11
Jan-12
Jan-13
Jan-14
Jan-15
21
 Volatility trading with 1m straddles on the S&P500 in-
dex - the strategy with optimal long/short positions
Long/short am Straddle with daily re-hedging and weekly re-balancing

• Short Only: sell using out-of-sample forecast of regime probabilities

• Long-Short: buy and sell straddles using regime probabilities

Short Only Long-Short %
Trades # 173 218 26%
Sortino 0.98 1.86 90%
Sharpe 0.74 1.04 40%
80,000 Short Only
Straddle P&L
60,000 Long-Short

40,000

20,000

-20,000
Jan-05
Jan-06
Jan-07
Jan-08
Jan-09
Jan-10
Jan-11
Jan-12
Jan-13
Jan-14
Jan-15
22
 Skew selling using risk-reversals on the S&P500 index -
the strategy with optimal position sizing outperforms
1m delta-hedged short 97.5%-102.5% Risk-reversal with daily re-hedging
and weekly re-balancing

• Continuous: roll into new risk-reversal every week

• Optimal: size position using out-of-sample forecast of regime probabilities
Continuous Optimal %
Trades # 519 173 33%
Sortino 0.34 2.09 ×6
Sharpe 0.29 1.56 ×5
30,000 Risk-Reversal P&L Continuous
Optimal Sizing
20,000

10,000

-10,000
Jan-05
Jan-06
Jan-07
Jan-08
Jan-09
Jan-10
Jan-11
Jan-12
Jan-13
Jan-14
Jan-15
23
From signal generation to hedging - empirically consis-
tent delta-hedging does contribute to the alpha

Sources of alpha advantage:

1) Signal generation

2) Delta-hedge execution

If option positions are not held to maturity - vega mark-to-market risk is

important

The minimum variance delta is applied to hedge against

changes in spot and ATM vol in the least-squares sense
by minimizing the variance of delta-hedging P&L

24
Key result: for the minimum-variance delta, the choice
of any particular stochastic or local vol model is irrele-
vant if models are calibrated to same implied vol surface

In presence of the risk-aversion, implied distributions will significantly dif-

fer from fat-tailed statistical distributions (bigger implied skeweness)

Risk-neutral vol models, calibrated to market prices, produce wrong hedges

when implied dynamics are different from realized statistical dynamics

Modeling challenge is to have a model is consistent with

both the statistical dynamics of implied and realized vols

I present the factor model for volatility dynamics based on my

work with Piotr Karasinski (Sepp-Karasinski (2012)) - the beta
stochastic volatility (SV) model

25
The volatility dynamics are log-normal: the basis for
vol-of-vol (normal, log-normal, 3/2) is important for sta-
tionary model with time-independent parameters
Compute the empirical frequency of one-month implied at-the-money
(ATM) volatility proxied by the VIX index for last 20 years
Daily observations normalized to have zero mean and unit variance
Left figure: empirical frequency of the VIX - it is definitely not normal
Right figure: the frequency of the logarithm of the VIX - it does look
like the normal density, especially for the right tail!
7% Empirical frequency of 7% Empirical frequency of
6% normalized VIX 6% normalized logarithm of the VIX
5%
Frequency

Empirical 5% Empirical

Frequency
4% Standard Normal 4% Standard Normal
3% 3%
2% 2%
1% 1%
VIX Log-VIX
0% 0%
-4 -3 -2 -1 0 1 2 3 4 -4 -3 -2 -1 0 1 2 3 4
26
Empirical frequency of realized vol is also log-normal
Compute one-month realized vol of daily returns on the S&P 500 index
for each month over non-overlapping periods for last 60 years from 1954
- normalize to have zero mean and unit variance

Left figure: frequency of realized vol - it is definitely not normal

Right figure: frequency of the logarithm of realized vol - again it does

look like the normal density, specially for the right tail!
10% Frequency of Historic 1m 10% Frequency of Logarithm of
8% Volatility of S&P500 returns 8% Historic 1m Volatility of S&P500

Frequency
Empirical
Frequency

6% Empirical 6%
Standard Normal Standard Normal
4% 4%

2% 2%
Vol Log-Vol
0% 0%
-4 -3 -2 -1 0 1 2 3 4 -4 -3 -2 -1 0 1 2 3 4
27
 Beta SV model is factor model for changes in vol V (tn)
predicted by returns in price
"
S(tn): #
S(tn) − S(tn−1)
V (tn) − V (tn−1) = β + V (tn−1)n
S(tn−1)
iid normal residuals n are scaled by vol V (tn−1) due to log-normality
Left figure: scatter plot of daily changes in the VIX vs returns on S&P
500 for past 14 years and estimated regression model
Volatility beta β is a measure of realized skew with high R2 = 67%
Right: time series of residuals n - the regression model is stable across
different estimation periods
Volatility beta β: expected change in ATM vol predicted by price return
For return of −1%: expected change in vol = −1.08 × (−1%) = 1.08%
20% 30% Time Series of Residual Volatility
Change in VIX

Change in VIX vs Return on S&P500

15% 20%
10% y = -1.08x 10%
R² = 67%
5% 0%
0% -10%
-10% -5% -5% 0% 5% 10% -20%
-10% -30%
Dec-99
Dec-00
Jan-02
Jan-03
Jan-04
Jan-05
Jan-06
Jan-07
Jan-08
Jan-09
Jan-10
Jan-11
Jan-12
Jan-13
-15%
-20% Return % on S&P 500 28
BSM volatility and delta-hedging P&L
Vanilla options are marked using Black-Scholes-Merton (BSM) implied
vol σBSM (K) in %-strike K relative to price S(0) (any SV model implies
quadratic form for implied vols near ATM strikes so generic approach):
σBSM (K; S) = σAT M (S) + SKEW × Z(K; S)
• Z(K; S) is log-moneyness relative to current price S:
Z(K; S) = ln (K × S(0)/S )
• SKEW < 0 is inferred from spread between call and put implied vols

In practice, this form is augmented with extras for convexity and tails

Volatility P&L arises from change in volatility at fixed strike induced by

the change in spot price S → S {1 + δS}:
δσBSM (K; S) ≡ σBSM (K; S {1 + δS}) − σBSM (K; S)
= δσAT M (S) + SKEW × δZ(K; S)
• Stochastic contributor to P&L: change in ATM vol δσAT M (S):
δσAT M (S) = σAT M (S {1 + δS}) − σAT M (S)
• Deterministic contributor to P&L: change in log-moneyness relative
to skew:
δZ(K; S) = − ln(1 + δS) ≈ −δS
29
Volatility skew-beta is applied to predict the stochastic
contributor to P&L from price return
Skew-beta β (T ) is estimated (for each maturity T ) by regression of changes
in ATM vols predicted by price-return " times skew: #
(T ) (T ) (T ) (T ) S(tn) − S(tn−1)
σATM(tn) − σATM(tn−1) = β × Skew (tn−1) ×
S(tn−1)
The skew-beta is range-bound between 1 and 2 with average of
about 1.5, with weak dependence on maturity time
Figure: skew-beta of 6m ATM vol for Stoxx50 with window of 1 and 12
months - the regression has explain of about 70% of vol changes
4 Volatility Skew Beta
1m
3 12m
2

0
Jan-08

Jan-09

Jan-10

Jan-11

Jan-12

Jan-13

Jan-14

Jan-15
-1 30
Volatility skew-beta combines the skew and volatility
P&L together - positive change in ATM vol from neg-
ative return is reduced by skew
Given price return δS: S → S {1 + δS}
1) For ATM vol change
• Volatility change is predicted by regression skew-beta:
δσAT M (S) = SKEWBETA × SKEW × δS

2) For BSM vol at fixed strikes

• Log-moneyness changes by δZ(K; S) ≈ −δS
• Volatility P&L at fixed strike is change in ATM vol plus skew P&L:
δσBSM (K) ≡ δσAT M (S) − SKEW × δS
= [SKEWBETA − 1] × SKEW × δS
BSM vol at S0
BSM vol at S1, ATM vol shift
21% BSM at S1, fixed strikes
BSM vol(K)

18%

15%

12% Strike K%
90% 95% 100% 105% 31
 Volatility skew-beta is applied to compute minimum-
variance delta ∆ for hedging against changes in price
and price-induced changes in implied vol
A) We adjust option delta for change in BSM implied vol at fixed strikes

B) The adjustment is proportional to option vega at this strike:

∆(K, T ) = ∆BSM (K, T ) + [SKEWBETA(T ) − 1] × SKEW(T ) × VBSM (K, T )/
∆BSM (K, T ) is BSM delta for strike K and maturity T
VBSM (K, T ) is BSM vega, both evaluated at volatility skew

I classify volatility regimes using vol skew-beta for delta-adjustments:




 ∆BSM (K, T ) + SKEW(T ) × VBSM (K, T )/S, Sticky local

∆
BSM (K, T ), Sticky strike

∆(K, T ) =


 ∆BSM (K, T ) − SKEW(T ) × VBSM (K, T )/S, Sticky delta

BSM (K, T ) + O × SKEW(T ) × VBSM (K, T )/S,
∆

Empirical

Adjustment O for empirical option delta is estimated statistically from

time series
32
Empirical skew beta applied to vol trading strategies
can significantly increase performance - P&L for delta-
hedging short 6m straddle on EuroStoxx50
Back-test of monthly rolls into new straddle with maturity of 6m from
2007 to 2015: account for delta-hedge P&L and vol P&L
Delta-hedge daily using specific rules for option delta:
• StickyStrike - hedging at sticky strike vol (BSM delta),
• StickyLocalVol - using minimum variance hedge,
• Empirical - using skew-beta with estimation window of past xx days

Only empirical hedge produces positive Sharpe ratio post costs

0.2
Sharpe Ratio Post-Costs 0.12
0.1

0.0

-0.1
-0.12
-0.2

-0.3
-0.31
-0.4
StickyStrike StickyLocalVol Empirical 33
Incorporating trading signal + empirical hedging pro-
duces superior P&L for delta-hedging short 6m straddle
on EuroStoxx50
Back-test of monthly rolls into new vols with maturity of 6m from 2007
to 2015 with trading signals using inference for market regimes
Delta-hedging using empirical skew-beta outperforms significantly

Sharpe Ratio Post-Costs

1.0
0.87
0.8

0.6 0.54
0.47
0.4

0.2

0.0
StickyStrike StickyLocalVol Empirical

34
Conclusions
In 1986, American economist Minsky proposed a financial stability theory:
”A fundamental characteristic of our economy is that the financial
system swings between robustness and fragility and these swings
are an integral part of the process that generates business cycles”

Implications
• Risk-premiums observed in short volatility strategies can be inter-
preted as costs for protecting against business and market downturns
• Volatility selling strategy needs to have an optimal and dynamic posi-
tion sizing with respect to the market regime, with regime probabilities
forecasted using new market data
• Alpha comes from the ability to avoid market downturns and empiri-
cally consistent hedging
• This approach provides further applications for portfolio allocations
and risk-management
• Similar approach for credit/high yield strategies (my presentation at
Global Derivatives 2015)
35
References
Computing option delta consistently with empirical dynamics:
Sepp, A., (2014), “Empirical Calibration and Minimum-Variance Delta
Under Log-Normal Stochastic Volatility Dynamics”
http://ssrn.com/abstract=2387845

Sepp, A. (2014), “Realized and implied index skews and minimum-variance

hedging”, Global Derivatives conference in Amsterdam 2014
kodu.ut.ee/~spartak/papers/ArturSeppGlobalDerivatives2014.pdf

Beta stochastic volatility model:

Karasinski, P., Sepp, A., (2012), “Beta stochastic volatility model,” Risk,
October, 67-73
http://ssrn.com/abstract=2150614
Optimal delta-hedging strategies for discrete hedging with trans-
action costs:
Sepp, A., (2013), “When You Hedge Discretely: Optimization of Sharpe
Ratio for Delta-Hedging Strategy under Discrete Hedging and Transac-
tion Costs,” Journal of Investment Strategies 3(1), 19-59
http://ssrn.com/abstract=1865998
36
Disclaimer

All statements in this presentation are the authors personal views and not
those of Bank of America Merrill Lynch.

37