viktor todorov northwestern university · 2017-11-14 · viktor todorov northwestern university...
TRANSCRIPT
![Page 1: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/1.jpg)
Nonparametric Option-Implied Volatility
Viktor TodorovNorthwestern University
November 10, 2017
![Page 2: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/2.jpg)
Nonparametric Option-Implied Volatility November 2017
Outline
• Setup
• Close-to-Money Options as estimates of Volatility
• Characteristic Function based Volatility from Options
• Truncated Volatility from Options
• Feasible CLT
• Empirical Application
1
![Page 3: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/3.jpg)
Nonparametric Option-Implied Volatility November 2017
Setup
The underlying process is X and the dynamics for x = logX is given by:
xt =
∫ t
0
asds+
∫ t
0
σsdWs +
∫ t
0
∫Rxµ(ds, dx),
where
• Wt is a Brownian motion
• µ controls jumps
• all quantities are with respect to Q
Our interest: nonparametric inference for σt from options.
2
![Page 4: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/4.jpg)
Nonparametric Option-Implied Volatility November 2017
Setup
We use short-dated options on X at time t, which expire at t+ T :
OT (k) =
EQt (ek − ext+T )+, if k ≤ lnFT ,
EQt (ext+T − ek)+, if k > lnFT ,
where FT is the price at time t of a forward contract which expires at time t+ T .
3
![Page 5: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/5.jpg)
Nonparametric Option-Implied Volatility November 2017
Setup
Option prices shrink with T ↓ 0:
• OT (k) = O(√T ) for |k| <<
√T
• OT (k) = O(T ) for fixed or asymptotically increasing |k|
4
![Page 6: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/6.jpg)
Nonparametric Option-Implied Volatility November 2017
Close-to-Maturity Option Convergence
0.8 0.85 0.9 0.95 1 1.05 1.1
K/X
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1Black Scholes Implied Volatility
T = 8
T = 5
T = 2
T = 1
5
![Page 7: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/7.jpg)
Nonparametric Option-Implied Volatility November 2017
Close-to-Money Option Expansion
We have
OT (k) = f
(kT√Tσt
)√Tσt − |ekT − 1|Φ
(−|kT |√Tσt
)+Op(T ),
where
• f and Φ are the pdf and cdf of standard normal,
• kT is deterministic sequence with kT/√T = Op(1),
• the expansion works in presence of jumps in X.
6
![Page 8: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/8.jpg)
Nonparametric Option-Implied Volatility November 2017
Close-to-Money Option Expansion
0 50 100 150 200 250
time in days
0.01
0.02
0.03
0.04
0.05
0.06
0.07
7
![Page 9: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/9.jpg)
Nonparametric Option-Implied Volatility November 2017
Option Portfolios
ATM options contain nontrivial bias due to the jumps.
We will try an alternative strategy by using portfolios of options with different strikes.
Following Carr and Madan (2001):
EQt (f(XT )) = f(F ) +
∫ ∞−∞
f′′(ek)OT (k)e
kdk,
for any f ∈ C2 and where F is the futures price at time t with expiration at t+ T .
8
![Page 10: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/10.jpg)
Nonparametric Option-Implied Volatility November 2017
CF-Based Volatility
We can thus span:
EQt
(eiu(xt+T−xt)
)= 1− (u
2+ iu)
∫ ∞−∞
e(iu−1)k−iuxtOT (k)dk, u ∈ R.
In the Levy case:
1
Tlog(EQt
(eiu(xt+T−xt)
))= iuat −
u2
2σ
2t +
∫R(eiux − 1− iux)νt(dx).
9
![Page 11: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/11.jpg)
Nonparametric Option-Implied Volatility November 2017
CF-Based Volatility
Therefore
σ2t = −
2
Tu2<(
log
(1− (u
2+ iu)
∫ ∞−∞
e(iu−1)k−iuxtOT (k)dk
))−
2
u2
∫R(1− cos(ux))νt(dx),
and ∫R(1− cos(ux))νt(dx) ≤ Ct.
10
![Page 12: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/12.jpg)
Nonparametric Option-Implied Volatility November 2017
CF-Based Volatility
If we set
uT �1√T,
then
σ2t = −
2
Tu2T
<(
log
(1− (u
2T + iuT )
∫ ∞−∞
e(iuT−1)k−iuT xtOT (k)dk
))+O (T ) .
11
![Page 13: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/13.jpg)
Nonparametric Option-Implied Volatility November 2017
CF-Based Volatility
u
0 5 10 15 20 25
0.02
0.022
0.024
0.026
0.028
0.03
0.032
0.034
0.036
0.038
0.04Characteristic Function Based Volatility Estimates
12
![Page 14: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/14.jpg)
Nonparametric Option-Implied Volatility November 2017
CF-Based Volatility
In empirically realistic settings we have error due to
• presence of jumps in X
• finite number of options over a discrete grid of strikes
• observation error
• time-variation in characteristic triplet
We derive the order of magnitude of CF-based volatility estimator.
13
![Page 15: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/15.jpg)
Nonparametric Option-Implied Volatility November 2017
Observation Scheme
The available options are at time t, expiring at t+ T , and having log-strikes given by:
k ≡ k1 < k2 < · · · kN ≡ k,
with the corresponding strikes given by
K ≡ K1 < K2 < · · ·KN ≡ K.
We denote
∆i = ki − ki−1, for i = 2, ...., N ,
and assume for η ∈ (0, 1) a positive constant and deterministic ∆→ 0:
η∆ ≤ infi=2,...,N
∆i ≤ supi=2,...,N
∆i ≤ ∆.
14
![Page 16: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/16.jpg)
Nonparametric Option-Implied Volatility November 2017
Observation Scheme
Instead of observing OT (ki), we observe:
OT (ki) = OT (ki) + εi.
We assume
E(εi|F (0)
)= 0,
εi ⊥ εj, conditionally on F (0),
E(ε2i
∣∣F (0))
= OT (ki)2σ2
t,i,
where supi=1,...,N σ2t,i = Op(1).
15
![Page 17: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/17.jpg)
Nonparametric Option-Implied Volatility November 2017
CF-Based Volatility
The estimate of the conditional characteristic function is:
ft,T (u) = 1− (u2
+ iu)
N∑j=2
e(iu−1)kj−1−iuxtOT (kj−1)∆j, u ∈ R,
We denote
Rt,T (u) = −<(
ln(ft,T (u) ∨ T
)),
and then define
Vt,T (u) =2
Tu2Rt,T (u).
16
![Page 18: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/18.jpg)
Nonparametric Option-Implied Volatility November 2017
CF-Based Volatility: Rate of Convergence
Theorem 1. Suppose certain assumptions hold and in addition ∆ � Tα, K � T−β,
K � T γ for some α > 0, β > 0 and γ > 0. Let {uT}T be an F (0)t -adapted sequence
such that
u2TT
a.s.−→ u, where u is a finite nonnegative random variable.
Then, we have
Vt,T (uT )− Vt = Op
(T
1−r2∨√
∆
T 1/4
∨e−2(|k|∨k)
).
17
![Page 19: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/19.jpg)
Nonparametric Option-Implied Volatility November 2017
CF-Based Volatility
0 20 40 60 80 100 120 140 160 180
u
0.016
0.018
0.02
0.022
0.024
0.026
0.028
0.03
0.032
V (u)
18
![Page 20: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/20.jpg)
Nonparametric Option-Implied Volatility November 2017
CF-Based Volatility: Adaptive Estimation
Theoretically, any choice of uT � 1/√T will work
In practice the choice of uT is critical
From the characteristic function, the quantity that matters is: u2T × σ
2t × T
We set uT adaptively using a preliminary Truncation Volatility Estimator of σt
19
![Page 21: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/21.jpg)
Nonparametric Option-Implied Volatility November 2017
Truncated Volatility
We look at fη(x) = e−ηx2x2 for η > 0.
We have fη(x) ∼ x2, for |x| ∼ 0,
fη(x) ≤ 1η, for |x| > 1√
η.
Therefore with ηT →∞, fηT (x) can be used to separate volatility from jumps:∣∣∣∣ 1T EQt
(fηT (xt+T − xt)
)− σ2
t
∣∣∣∣ = Op
(√T ∨ ηTT
∨ 1√ηT
).
20
![Page 22: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/22.jpg)
Nonparametric Option-Implied Volatility November 2017
Truncated Volatility
Therefore, we look at ∫ ∞−∞
hη(k)OT (k)dk,
where
hη(k) = e−k−η(k−xt)
2[4η
2(k − xt)4
+ 2− 10η(k − xt)2
+ 2η(k − xt)3 − 2(k − xt)].
We have ∫∞−∞ h0(k)OT (k)dk = σ2
t +∫R x
2νt(dx) + op(1),∫∞−∞ hηT (k)OT (k)dk = σ2
t + op(1), for ηT →∞.
21
![Page 23: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/23.jpg)
Nonparametric Option-Implied Volatility November 2017
Truncated Volatility
The option-based Truncated Volatility estimator is defined by:
T V t,T (η) =1
T
N∑j=2
hη(kj−1)OT (kj−1)∆j, η ≥ 0.
The total volatility estimator is:
QV t,T ≡ T V t,T (0).
We set the cutoff level adaptively at
ηT =ηTT
1
QV t,T
,
for some deterministic sequence ηT that depends only on T .
22
![Page 24: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/24.jpg)
Nonparametric Option-Implied Volatility November 2017
Truncated and Total Volatility: Consistency
Theorem 2. Suppose certain assumptions hold and in addition ∆ � Tα, K � T−β,
K � T γ for some α > 0, β > 0 and γ > 0. If α > 12, we have
QV t,TP−→ QVt,T .
Suppose in addition that for ηT :
ηT → 0 andηTT→∞.
Then, we also have
T V t,T (ηT )P−→ Vt.
23
![Page 25: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/25.jpg)
Nonparametric Option-Implied Volatility November 2017
Adaptive CF-Based Volatility
The adaptive choice for the characteristic exponent is given by
uT =u√T
1√T V t,T (ηT )
,
where u is some positive constant.
We further denote with Avar(Vt,T (u)) an estimate of the asymptotic variance based on
εj = OT (kj)−1
2
(OT (kj−1) + OT (kj+1)
), for j = 2, ..., N − 1.
24
![Page 26: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/26.jpg)
Nonparametric Option-Implied Volatility November 2017
CF-Based Volatility: CLT
Theorem 3. Suppose certain assumptions hold and in addition ∆ � Tα, K � T−β,
K � T γ for some α > 0, β > 0 and γ > 0. If
1
2< α <
(5
2− r)∧(
1
2+ 4(β ∧ γ)
),
then
Vt,T (uT )− Vt√Avar(Vt,T (uT ))
L−→ N(0, 1).
25
![Page 27: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/27.jpg)
Nonparametric Option-Implied Volatility November 2017
Empirical Application
With A Little Help from Yang Zhang
26
![Page 28: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/28.jpg)
Nonparametric Option-Implied Volatility November 2017
Empirical Application
Option Data:
• SPX short maturity option data at market close
• period: 01/2008 - 08/2017 (weeklies start from 01/2011)
• maturity: 2 to 5 business days (based on weeklies)
• median size of option cross-section: 59 OTM options (based on weeklies)
HF Data:
• frequency: 5-minutes during work hours
• local window: trading day
27
![Page 29: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/29.jpg)
Nonparametric Option-Implied Volatility November 2017
Option vs HF Volatility Estimates
2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018
0.2
0.4
0.6
0.8
1
1.2
1.4
28
![Page 30: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/30.jpg)
Nonparametric Option-Implied Volatility November 2017
Empirical Application
Consider optimal estimator:
Vmixt = ωt × V opt
t + (1− ωt)× V hft ,
where ωt ∈ [0, 1] is optimal weight determined by asym. variance of the two estimators.
Median value of ω is: 0.85
29
![Page 31: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/31.jpg)
Nonparametric Option-Implied Volatility November 2017
HF vs Combined Volatility Estimates
2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018
0.2
0.4
0.6
0.8
1
1.2
1.4
30
![Page 32: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/32.jpg)
Nonparametric Option-Implied Volatility November 2017
Empirical Application
Gains for forecasting:
RVt+1 = α0 + α1Xt + εt+1,
where Xt is a volatility predictor from the list:
• RVt• V opt
t
• V hft
• V mixt
• QRVt
31
![Page 33: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/33.jpg)
Nonparametric Option-Implied Volatility November 2017
Forecast Performance Relative to Benchmark RV Forecast Model
Table 1: MSEPredictor
V optt V hf
t V mixt QRVt
Rolling Window 0.7445 1.0789 0.6886 0.6868
Increasing Window 0.6607 1.1438 0.6216 0.5600
Table 2: QLIKE
Predictor
V optt V hf
t V mixt QRVt
Rolling Window 0.9828 1.0439 0.8560 0.9748
Increasing Window 0.9189 1.0618 0.8088 0.8826
32
![Page 34: Viktor Todorov Northwestern University · 2017-11-14 · Viktor Todorov Northwestern University November 10, 2017. Nonparametric Option-Implied Volatility November 2017 Outline Setup](https://reader030.vdocuments.mx/reader030/viewer/2022040112/5e99a780b0f6055d1177f1cc/html5/thumbnails/34.jpg)
Nonparametric Option-Implied Volatility November 2017
Conclusion
• We propose nonparametric option-based volatility estimates
• The estimates are based on option portfolios of short-maturity options with different
strikes
• Characteristic-based Option Portfolio for high frequencies separates volatility from
jumps
• Tuning parameter selected from Option-based Truncated Volatility
• Empirical Application shows efficiency gains over HF Volatility Estimates
33