DELTA HEDGING WITH THE SMILE

Sami Vhmaa
*

Graduate School oI Finance and Financial Accounting
Department oI Accounting and Finance, University oI Vaasa




Abstract

This paper shows that the delta hedging perIormance oI the Black-Scholes
model can be substantially improved with a rather simple adjustment oI the
Black-Scholes delta. By utilizing the volatility smile, the Black-Scholes delta
can be adjusted to account Ior the inverse movements between volatility and
stock prices. Empirical tests in the FTSE 100 index option market show that the
smile-adjusted delta consistently outperIorms the Black-Scholes delta in terms oI
hedging perIormance.

JEL classification. G10; G13

Kevwords. options, delta hedging, volatility smile, smile-adjusted delta







*
Contacts: University oI Vaasa, Department oI Accounting and Finance, P.O. Box 700, FIN-65101
Vaasa, Finland; Tel. 358 6 324 8197; Fax: 358 6 324 8344; email: samiuwasa.Ii


DELTA HEDGING WITH THE SMILE



Abstract

This paper shows that the delta hedging perIormance oI the Black-Scholes
model can be substantially improved with a rather simple adjustment oI the
Black-Scholes delta. By utilizing the volatility smile, the Black-Scholes delta
can be adjusted to account Ior the inverse movements between volatility and
stock prices. Empirical tests in the FTSE 100 index option market show that the
smile-adjusted delta consistently outperIorms the Black-Scholes delta in terms oI
hedging perIormance.

JEL classification. G10; G13

Kevwords. options, delta hedging, volatility smile, smile-adjusted delta















2
1. INTRODUCTION

Practitioners oIten price and hedge options using the Black-Scholes (1973) model,
although many studies have shown that it does not adequately describe stock price
dynamics. In particular, the constant volatility assumption oI the Black-Scholes model
is obviously violated.
1
It has been recognized Ior a long time that volatility is time-
varying, and in addition, tends to be inversely related to stock prices |see e.g., Black
(1976a)|. One maniIestation oI the misspeciIication oI the Black-Scholes model is the
volatility smile, the variation oI implied volatilities across strike prices. Especially the
implied volatility oI index options seems to be decreasing with the strike price |see e.g.,
Rubinstein (1994) and Mayhew (1995)|, instead oI being constant as assumed by the
Black-Scholes model.
2

This paper is motivated by the inverse movements between volatility and stock
prices and its implications Ior delta hedging. Since volatility is an important determinant
oI hedge ratios, incorrect volatility assumptions may lead to incorrect deltas. II volatility
is time-varying and correlated with the underlying stock returns, the delta must control
not only Ior the direct impact oI the underlying price change on the option price, but
also Ior the indirect impact oI the simultaneous change in volatility. This has been
recently noted e.g. in Derman et al. (1996), Engle and Rosenberg (2000), Coleman et al.

1
A substantial literature has devoted to the development oI option pricing models with time-varying
volatility. The Black-Scholes constant volatility assumption is relaxed in stochastic volatility models such
as Hull and White (1987) and Heston (1993) and in deterministic volatility models by Dupire (1994),
Derman and Kani (1994), Rubinstein (1994), Duan (1995), and Heston and Nandi (2000).
2
Since implied volatility oI index options is decreasing with the strike price, the volatility curve looks
more like a smirk than a smile. ThereIore, besides volatility smile, the curve is also commonly reIerred to
as volatility smirk and volatility skew.
3
(2001), and Mixon (2002). Intuitively, an inverse relation between volatility changes
and stock returns would suggest that the Black-Scholes delta is too large.
The purpose oI this study is to empirically examine whether the delta hedging
perIormance oI the Black-Scholes model can be improved by accounting Ior the inverse
movements between volatility and the underlying stock price. Following the ideas in
Derman et al. (1996) and Coleman et al. (2001), the volatility smile is utilized to adjust
the Black-Scholes delta to account Ior the inverse relationship between volatility
changes and underlying stock returns. In this study, the resulting delta is reIerred to as
the smile-adjusted delta.
Previously, studies such as Bakshi et al. (1997, 2000), Nandi (1996, 1998),
Dumas et al. (1998), Lim and Guo (2000), Coleman et al. (2001), and Lim and Zhi
(2002) have examined option hedging under time-varying volatility. Although time-
varying volatility option pricing models clearly outperIorm the Black-Scholes model in
terms oI pricing |see e.g., Bakshi et al. (1997) and Corrado and Su (1998)|
3
, such
models do not necessarily provide better hedging perIormance than the Black-Scholes
model. For instance, Bakshi et al. (1997) show that stochastic volatility models improve
the delta hedging perIormance oI the Black-Scholes model only when out-oI-the-money
options are hedged while Dumas et al. (1998) document the Black-Scholes delta to be
more accurate than the deltas obtained Irom deterministic volatility models. Whereas
the existing literature has Iocused on the hedging perIormance oI diIIerent time-varying
volatility option pricing models, the current study takes a diIIerent approach, and
contributes to the literature by examining whether the hedging perIormance oI the

3
Time-varying volatility models outperIorm the Black-Scholes model in terms oI pricing, but are still
computationally too demanding Ior real-time option pricing and even more so Ior calculating hedge
ratios.
4
Black-Scholes model can be improved with a rather simple adjustment oI the Black-
Scholes delta.
The remainder oI the paper is organized as Iollows. The smile-adjusted delta is
presented in Section 2. In section 3, the FTSE 100 index option data used in the
empirical analysis are described. Section 4 presents the methodology applied in the
paper. Empirical Iindings are reported in Section 5. Finally, concluding remarks are
oIIered in Section 6.

2. FROM BLACK-SCHOLES DELTA TO SMILE-AD1USTED DELTA

The Black-Scholes (1973) model gives the price and the hedge ratios oI an option as
Iunctions oI the underlying asset price, the strike price oI the option, risk-Iree interest
rate, time to maturity oI the option, and the volatility oI the underlying asset. By
deIinition, the volatility parameter is assumed to be a known constant. For a European
call option on a non-dividend-paying stock, the Black-Scholes delta is given by

|
|
.
|

\
|
o
o + +
=
c
o c
= o
T
T r K S
N
S
T r K S c
BS
) 2 / ( ) / ln( ) , , , , (
2
(1)

where ) ( c denotes the Black-Scholes (1973) call option pricing Iormula, ) ( N denotes
cumulative standard normal distribution Iunction, S is the price oI the underlying asset,
K is the strike price oI the option, o is the volatility oI the underlying asset, r is the risk-
Iree interest rate, and T denotes time to maturity.
5
An unIortunate inconsistency between the Black-Scholes model and empirical
regularities is the behavior oI volatility. The constant volatility assumption oI the Black-
Scholes model is indisputably violated in practice. Volatility appears to be time-varying,
and in addition, tends to be negatively correlated with changes in the underlying stock
price. Since volatility is a central determinant oI hedge ratios, an incorrect volatility
assumption may lead to incorrect deltas.
II volatility is time-varying and correlated with the underlying stock returns, the
delta must control not only Ior the direct impact oI the underlying price change on the
option price, but also Ior the indirect impact oI the volatility change which is correlated
with the underlying price change. Assume volatility to be a deterministic Iunction oI S,
K, and T. Then, by the chain rule, the delta oI the option is given by
4


S
c
S
c
c
c
c
c
+
c
c
= o

(2)

where o c cc is the vega oI the option. Since vega is always positive, equation (2)
shows that in the case oI a negative correlation between stock returns and volatility
changes, the delta should be smaller than the Black-Scholes delta. UnIortunately, the
term S c c in equation (2) is diIIicult to quantiIy.
Following Derman et al. (1996) and Coleman et al. (2001), the unknown
dependence oI volatility on the underlying stock price S c c can be approximated by

4
The need to adjust the Black-Scholes delta Ior the change in volatility has been recently noted in
Derman et al. (1996), Engle et al. (2000), Coleman et al. (2001), and Mixon (2002).
6
the slope oI the current volatility smile K c c .
5
Substituting K c c into equation (2)
gives the smile-adjusted delta as

K
v
K
c
S
c
BS BS SAD
c
c
+ o =
c
c
c
c
+
c
c
= o

(3)

The smile-adjusted delta utilizes the volatility smile together with the vega oI the
option to account Ior the relationship between volatility changes and underlying stock
returns. When the smile is downward sloping, the Black-Scholes delta is adjusted
downwards to account Ior the oIIsetting movement between volatility and asset price.
Note that in the special case oI a completely Ilat smile, the smile-adjusted delta becomes
equal to the Black-Scholes delta. By approximating S c c with K c c , it is assumed
that as S changes by one unit, there is a parallel shiIt oI K c c units in the volatility
smile. II the current smile is downward sloping, the approximation assumes that the
volatility smile is shiIted downwards as the price oI the underlying stock increases, and
thus, all Iixed-strike volatilities decrease by the amount deIined by the slope oI the
current smile and at-the-money volatility decreases twice as much.

3. DATA

The sample used in this study contains end-oI-day settlement prices oI the FTSE 100
index options traded on the London International Financial Futures and Options
Exchange (LIFFE). The FTSE 100 index is a capitalization-weighted index consisting

5
The local volatility models by Dupire (1994), Derman et al. (1994) and Rubinstein (1994) essentially
inIer the dependence oI volatility on the underlying asset price Irom the volatility smile.
7
oI the 100 largest companies traded on the London Stock Exchange. Both European and
American options on the FTSE 100 index are traded on the LIFFE. In this study, the
European-style options (ticker symbol ESX) are used. The market Ior the European-
style FTSE 100 index options is the most active equity options market in the United
Kingdom. The European-style FTSE 100 index options have a wide range oI strike
prices available Ior trading. The delivery months are March, June, September, and
December. In addition, the Iour nearest calendar months are always available Ior
trading. The FTSE 100 options expire on the third Friday oI the delivery month.
Dividend adjustments are incorporated via implied index Iutures, and thus, the FTSE
100 index options are priced as options on Iutures. The sample period used in this study
extends Irom January 2, 2001 to December 28, 2001. The settlement prices Ior the
FTSE 100 index options and the closing prices Ior the implied index Iutures are
obtained Irom the LIFFE.
The risk-Iree interest rate is needed Ior the calculation oI the deltas and Ior the
delta hedging experiment. In this study, the three-month LIBOR (London Interbank
OIIered Rate) rate is used as a proxy Ior the risk-Iree rate. The data on the LIBOR rates
are obtained Irom the British Bankers` Association. The three-month LIBOR rate
ranged Irom 3.78 to 5.73 during the sample period.

(insert Figure 1 about here)

8
The development oI the FTSE 100 index and the 1-month at-the-money (ATM)
implied volatility
6
during 2001 are presented in Figure 1. It can be noted that there
exists a strong inverse relationship between index level and volatility. Descriptive
statistics Ior the FTSE 100 index and 1-month ATM implied volatility series are
presented in Table I. During the sample period, the index level ranged Irom 4434 to
6335 points. The mean continuously compounded daily return on the FTSE 100 index
was very close to zero and the standard deviation oI the daily returns was 0.0136. The
standard deviation corresponds to annualized volatility oI about 21.60 , which is very
close to the mean level oI ATM implied volatility oI 21.50 . The inverse relationship
between index level and volatility observable in Figure 1 is conIirmed by the
considerable negative correlation between index returns and volatility changes reported
in Panel C oI Table I.

(insert Table I about here)

Two exclusionary criteria are applied to the complete FTSE 100 index option
sample to construct the Iinal sample used in the empirical analysis. First, options with
Iewer than 5 or more than 120 trading days to maturity are eliminated Irom the sample.
This choice avoids any expiration-related unusual price Iluctuations and minimizes the
liquidity problems oIten aIIecting the prices oI long-term options.
7
Second, options with
absolute moneyness greater than 10 percent are eliminated. Moneyness is deIined as the

6
The 1-month ATM implied volatility series is obtained by linear interpolation between the two adjacent
maturity ATM implied volatilities. Three shortest maturity option series are used as Iollows. The shortest
and the second shortest maturity options are used until the shortest has 5 days to maturity. ThereaIter, the
second and the third shortest maturity options are used until the expiry oI the shortest maturity option.
7
The two shortest maturity contracts are usually the most actively traded ones.
9
ratio oI Iutures price to strike price Ior call options and strike price to Iutures price Ior
put options. The moneyness criterion is applied because deep out-oI-the-money and in-
the-money options tend to be thinly traded.
The Iinal sample contains 35 180 end-oI-day settlement prices on options with 5
to 120 trading days to maturity and moneyness between -10 and 10 percent. This sample
is considered to be a representative sample oI the most actively traded option contracts.
The sample is partitioned into three moneyness and two time to maturity categories. A
call option is said to be out-oI-the-money (OTM) iI the ratio oI Iutures price to strike
price is less than 0.97, at-the-money (ATM) iI the ratio is larger than 0.97 and less than
1.03, and in-the-money (ITM) iI the ratio is greater than 1.03. Similar terminology is
applied to put options by deIining moneyness as the ratio oI strike price to Iutures price.
An option is said to be short-term iI it has less than 40 trading days to expiration and
long-term otherwise. This maturity and moneyness partitioning produces 12 categories
Ior which the empirical results will be reported.

(insert Table II about here)

Summary statistics Ior the Iinal sample are reported in Table II. The reported
numbers are (i) the average end-oI-day settlement price, (ii) the average implied
volatility, and (iii) the total number oI observations in each maturity and moneyness
category. The Iinal sample contains almost equal amount oI call and put options and
also short and long maturity options are almost equally represented. For moneyness
classiIications, OTM and ATM options both account Ior 35 oI the total sample and
ITM options Ior the remaining 30 . As expected, the average option price increases
10
with moneyness and time to maturity. It can be noted that OTM puts are, on average,
much more expensive than OTM calls. The reported implied volatilities indicate the
presence oI volatility smile in the FTSE 100 index options; the implied volatility oI
OTM put options is signiIicantly higher than the implied volatility oI OTM call options.

4. METHODOLOGY

Since the FTSE 100 index options are priced as options on Iutures, Black`s (1976)
model Ior European Iutures options is applied in the empirical analysis to calculate the
deltas, vegas, and implied volatilities. In this study, the delta derived Irom the Black`s
(1976) model is reIerred to as the Black-Scholes delta. The deltas and vegas Ior each
option are derived Irom the Black`s model using the exact implied volatility Ior that
particular option as the volatility parameter. Since the volatility smile oI index options
tends to be approximately linear (see e.g. Derman (1999)), K c c is estimated by Iitting
a linear model oI volatility as a Iunction oI the strike price based on least squares
criterion.
The quantitative Iit oI the Black-Scholes and smile-adjusted deltas is tested
based on a Iirst-order Taylor series expansion with the Iollowing two regressions
speciIications, respectively

c +
(

A
c
c
| + o = A S
S
c
c (4)

11
c +
(

A |
.
|

\
|
c
o c
o c
c
+
c
c
| + o = A S
K
c
S
c
c (5)

where c is the option price, S is the price oI the underlying asset, K is the strike price oI
the option, and A denotes the Iirst diIIerence operator. An accurate delta estimate should
produce a very small negative o and | oI one, with the regression R
2
equal to one. The
accuracy oI the Black-Scholes and smile-adjusted deltas is compared simply based on
the R
2
oI the regressions; the more accurate delta should produce a higher R
2
.
In order to examine the delta hedging perIormance oI the Black-Scholes and
smile-adjusted deltas, a selI-Iinanced delta-hedged portIolio with one unit short position
in an option, o units oI the underlying asset, and B units oI a risk-Iree bond is
constructed. The value oI the portIolio H at time t is

t t t t t
c B S + o = H (6)

At the beginning oI the hedging horizon
0 0 0 0
S c B o = , and thus, 0
0
= H . The
delta hedging perIormance oI the two deltas is examined in 1-day, 5-day, and 10-day
hedging horizons using daily rebalancing oI the hedge portIolio. At each hedge-revision
time t the hedge parameter is recomputed and the position in the bond is adjusted to

) (
1 1 t t t t
rt
t
S B e B o o + =

(7)

The delta hedging error c Irom hedge-revision time t-1 to t is calculated as
12

1 1
+ o = c
t
rt
t t t t
B e c S (8)

and the total hedging error during the hedging horizon t is given as
T
H , the value oI
the portIolio at the end oI the hedging horizon

=
t
H = c = c
T
t
T t
1
(9)

The delta hedging perIormance oI the Black-Scholes and smile-adjusted deltas is
analyzed based on two commonly used error statistics, (i) mean absolute hedging error
(MAHE) and (ii) root mean squared hedging error (RMSHE).
8
The error statistics are
calculated as

MAHE

=
t
c
n
i
i
n
1
1
(10)

RMSHE

=
t
c
n
i
i
n
1
2
1
(11)




8
The usual practice in the hedging literature is to compare diIIerent models in terms oI mean absolute
errors, mean squared errors, or root mean squared errors |see e.g., Nandi (1998), Dumas et al. (1998),
Coleman et al. (2001)|. The hedging perIormance could, oI course, also be compared in terms oI the
standard deviation oI the hedging errors as in Engle et al. (2000). However, since the mean hedging error
tends to be very small, RMSHE and standard deviation essentially convey the same message.
13
Bootstrapping with 1000 resamplings is used to test whether the diIIerences in
the hedging perIormance oI the Black-Scholes and smile-adjusted deltas are statistically
signiIicant. Since the hedger`s main objective is risk minimization, the root mean
squared error is considered to be the primary evaluation criteria.

5. EMPIRICAL RESULTS

A comparison oI the average Black-Scholes and smile-adjusted deltas Ior diIIerent
moneyness categories is presented in Table III. Since the volatility smile oI index
options is usually downward sloping and the vega Ior both call and put options is
always positive, the smile-adjusted deltas are consistently lower than the corresponding
Black-Scholes deltas. On average, the diIIerence between the two deltas is about 0.05.
Since vega is largest Ior ATM options, the diIIerence between the two deltas is naturally
greatest among ATM options.

(insert Table III about here)

Derman (1999) suggests that the size oI the delta should be conditional on the
market conditions. A smaller than Black-Scholes delta should be optimal in highly
volatile, jumpy market conditions. Mixon (2002) argues that the delta should be smaller
than the Black-Scholes delta in order to be consistent with empirical regularities oI
volatility dynamics. Previously, deltas that are diIIerent Irom the Black-Scholes deltas
have been empirically documented in Bakshi et al. (2000), Coleman et al. (2001), and
Lim et al. (2002). Bakshi et al. (2000) show that the deltas produced by stochastic
14
volatility models are smaller than the Black-Scholes deltas Ior short-maturity OTM
options, but larger Ior ITM options. Coleman et al. (2001) document the deltas based on
deterministic volatility model to be consistently smaller than the Black-Scholes deltas.
Lim et al. (2002) report that the deltas based on the Derman-Kani (1994) implied tree
model are lower than the Black-Scholes deltas, while the deltas derived Irom
Jackwerth`s (1997) model tend to be higher than the Black-Scholes deltas.
The regression results based on equations (4) and (5) are reported in Table IV.
Several Ieatures can be noted Irom Table IV. Most importantly, the R
2
values under the
Black-Scholes delta are consistently lower than under the smile-adjusted delta,
indicating a better quantitative Iit oI the latter. For the Iull sample, the Black-Scholes
model can explain 96.51 oI the observed option price changes whereas the smile-
adjusted delta explains 97.10 oI the changes. In general, the quantitative Iit oI the
both deltas is relatively high Ior all moneyness and maturity categories, with R
2
`s
ranging Irom 84.22 to 98.65 under the Black-Scholes delta and Irom 87.77 to
98.92 under the smile-adjusted delta.

(insert Table IV about here)

The regression results reported in Table IV show that the diIIerence between the
quantitative Iit oI the two models is largest among short-maturity OTM options and
smallest Ior long-maturity ITM options. Both models seem to Iit ITM options better
than ATM and OTM options. The estimates oI both models are quite close to unity Ior
ATM and ITM options. However, Ior OTM options the estimates are considerably
below unity. In most cases, there are no signiIicant diIIerences in the estimates under
15
the two models, but especially Ior OTM options, the estimates under the smile-
adjusted delta are a bit closer to unity. Overall, the results in Table IV indicate a better
quantitative Iit by the smile-adjusted delta.

(insert Table V about here)

The results oI the delta hedging experiment are reported in Tables V, VI, and
VII. The reported numbers are mean absolute hedging errors (MAHE) and root mean
squared hedging errors (RMSHE) Ior all moneyness and maturity categories. In
addition, the mean diIIerence between the error statistics is reported. The hedging
results Ior the 1-day hedging horizon are given in Table V. The results indicate that the
smile-adjusted delta is a substantial improvement over the Black-Scholes delta. For the
Iull sample, the smile-adjusted delta outperIorms the Black-Scholes delta in terms oI
hedging perIormance based both on MAHE and RMSHE. The diIIerence in both error
statistics is statistically signiIicant at the 1 level. The outperIormance oI the smile-
adjusted delta is most distinct Ior short-term OTM and ATM options. For all moneyness
and maturity categories, the RMSHE under the smile-adjusted delta is consistently
lower than under the Black-Scholes delta but the MAHE criterion indicates that the
Black-Scholes delta is more eIIective Ior hedging long-term options. However, since
hedger`s main objective is risk minimization, the emphasis in the interpretation oI the
results should be given to the RMSHE.

(insert Table VI about here)

16
Table VI presents the hedging results Ior the 5-day hedging horizon. The error
statistics reported in Table VI clearly indicate that hedging under the smile-adjusted
delta is more eIIective than under the Black-Scholes delta. Regardless oI the moneyness
and maturity oI the options, the both error statistics are lower Ior the smile-adjusted
delta than Ior the Black-Scholes delta. All diIIerences reported in Table VI are
statistically signiIicant at the 1 level. In contrast to the results in Table V, the
outperIormance oI the smile-adjusted delta seems to be more apparent in the case oI
long-term options.

(insert Table VII about here)

The hedging results Ior the 10-day hedging horizon are presented in Table VII.
In general, the hedging errors Ior the 10-day horizon are very similar to the errors Ior
the 1-day and 5-day horizons in Tables V and VI. Again, the smile-adjusted delta
consistently outperIorms the Black-Scholes delta in terms oI hedging perIormance,
regardless oI the moneyness and maturity oI the options. II compared to the hedging
errors Ior the 1-day hedging horizon reported in Table V, the results in Tables VI and
VII indicate that the outperIormance oI the smile-adjusted delta is more prominent Ior
longer hedging horizons.
To Iurther investigate the hedging perIormance oI the smile-adjusted delta
relative to the Black-Scholes delta, the delta hedging experiment Ior the 1-day hedging
horizon is repeated in two diIIerent market conditions, that is in stable market oI May
2001 and in highly volatile market oI September 2001. In May 2001, the FTSE 100
index as well as the implied volatility oI the index were relatively stable and the
17
correlation between the index returns and volatility changes was low, 0.16, and
statistically insigniIicant. In contrast, September 2001 was characterized by extensive
movements both in the index and in the implied volatility. In September 2001, the
correlation between the index returns and volatility changes was 0.76 and statistically
signiIicant at the 1 level. According to Derman (1999), the optimal delta should be
equal to the Black-Scholes delta during stable market conditions whereas in highly
volatile markets the optimal delta should be smaller than the Black-Scholes delta.

(insert Table VIII about here)

Table VIII presents the hedging results Ior the stable and highly volatile market
conditions. The results Ior the stable market reported in Panel A oI Table VIII seem
very similar to the results reported Ior the total sample in Table V. The RMSHE is
consistently lower under the smile-adjusted delta whereas the MAHE suggests that the
Black-Scholes delta provides better hedges long-term options. However, in most cases
the reported diIIerences are statistically insigniIicant. The hedging results Ior the highly
volatile market conditions are reported in Panel B oI Table VIII. Both error statistics
indicate considerable outperIormance oI the smile-adjusted delta. Regardless oI the
moneyness and maturity oI the options, the smile-adjusted delta always leads to smaller
hedging errors in highly volatile market conditions. All diIIerences shown in Panel B oI
Table VIII are statistically signiIicant at the 1 level.



18
6. CONCLUSIONS

The constant volatility assumption oI the Black-Scholes (1973) model is indisputably
violated in practice. Volatility appears to be time-varying, and in addition, tends to be
negatively correlated with stock returns. Still, the Black-Scholes model is the most
common Iramework Ior pricing and hedging options, mainly due to its simplicity and
tractability. This paper is motivated by the inverse movements between volatility and
stock prices and its implications Ior delta hedging. II volatility is time-varying and
correlated with the underlying stock returns, the delta must control not only Ior the
direct impact oI the underlying price change on the option price, but also Ior the indirect
impact oI simultaneous change in volatility. Intuitively, an inverse relation between
volatility and stock returns would suggest that the Black-Scholes delta is too large.
This study empirically examines whether the delta hedging perIormance oI the
Black-Scholes model can be improved by accounting Ior the inverse movements
between volatility and the underlying stock price. In particular, a rather simple
adjustment oI the Black-Scholes delta is investigated. Following the ideas in Derman et
al. (1996) and Coleman et al. (2001), the volatility smile is utilized to adjust the Black-
Scholes delta to account Ior the negative correlation between volatility changes and
stock returns.
Empirical tests in the FTSE 100 index option market show that the delta hedging
perIormance oI the Black-Scholes model can be substantially improved by adjusting the
Black-Scholes delta to account Ior the inverse movements between volatility and stock
prices. The smile-adjusted delta consistently leads to smaller hedging errors, and thus,
outperIorms the Black-Scholes delta in terms oI hedging perIormance. The
19
outperIormance oI the smile-adjusted delta is most distinct Ior short-term OTM and
ATM options and becomes more prominent as the hedging horizon lengthens.
Since short-term OTM and ATM options tend to be the most actively traded
contracts in index option markets, the Iindings reported in this paper have signiIicant
practical relevance. The practical relevance oI the results is Iurther enhanced by the Iact
that the Black-Scholes deltas, vegas, and smiles are continuously monitored by
practitioners, and thus, the smile-adjusted delta should be simple to implement Ior
practical purposes. In general, the Iindings reported in this paper have important
implications Ior risk management oI options as they indicate the importance oI the
correlation between volatility changes and stock returns Ior appropriate risk
management. Furthermore, the empirical results demonstrate that due to the inverse
relation between volatility and stock price movements, the correct delta is smaller than
the Black-Scholes delta.











20
REFERENCES

Bakshi, G., C. Cao & Z. Chen (1997). Empirical perIormance oI alternative option
pricing models. Journal of Finance. 52: 5, 203-2049.

Bakshi, G., C. Cao & Z. Chen (2000). Pricing and hedging long-term options. Journal
of Econometrics. 94, 277-318.

Black, F. (1976a). Studies oI stock price volatility changes. Proceedings of the 1976
Meetings of the American Statistical Association, 177-181.

Black, F. (1976b). The pricing oI commodity contracts. Journal of Financial
Economics. 3, 167-179.

Black, F. & M. Scholes (1973). The pricing oI options and corporate liabilities. Journal
of Political Economv. 81, 637-659.

Coleman, T. F., Y. Kim, Y. Li & A. Verma (2001). Dynamic hedging with a
deterministic local volatility Iunction model. Journal of Risk. 4: 1, 63-89.

Corrado, C. & T. Su (1998). An empirical test oI the Hull-White option pricing model.
Journal of Futures Markets. 18: 4, 363-378.

Derman, E. (1999). Regimes oI volatility. Risk. 12: 4, 55-59.
21
Derman, E. & I. Kani (1994). Riding on a smile. Risk. 7: 2, 32-39.

Derman, E., I. Kani & J. Zou (1996). The local volatility surIace: Unlocking the
inIormation in index option prices. Financial Analvsts Journal. 52, 25-36.

Duan, J.-C. (1995). The GARCH option pricing model. Mathematical Finance. 5: 1, 13-
32.

Dumas, B., J. Fleming & R.E. Whaley (1998). Implied volatility Iunctions: empirical
tests. Journal of Finance. 53: 6, 2059-2106.

Dupire, B. (1994). Pricing with a smile. Risk. 7: 1, 18-20.

Engle, R. F. & J. Rosenberg (2000). Testing the volatility term structure using option
hedging criteria. Journal of Derivatives. 8: 1, 10-28.

Heston, S. (1993). A closed Iorm solution Ior options with stochastic volatility with
applications to bond and currency options. Review of Financial Studies. 6, 237-
243.

Heston, S. & S. Nandi (2000). A closed-Iorm GARCH option valuation model. Review
of Financial Studies. 13: 3, 585-625.

22
Hull, J. & A. White (1987). The pricing oI options on assets with stochastic volatilities.
Journal of Finance. 42, 281-300.

Jackwerth, J. (1997). Generalized binomial trees. Journal of Derivatives. 5: 2, 7-17.

Lim, K. & X. Guo (2000). Pricing American options with stochastic volatility: Evidence
Irom S&P 500 Iutures options. Journal of Futures Markets. 20, 625-659.

Lim, K. & D. Zhi (2002). Pricing options using implied trees: evidence Irom FTSE-100
options. Journal of Futures Markets. 22: 7, 601-626.

Mayhew, S. (1995). Implied volatility. Financial Analvsts Journal. 51, 8-20.

Mixon, S. (2002). Factors explaining movements in the implied volatility surIace.
Journal of Futures Markets. 22: 10, 915-937.

Nandi, S. (1996). Pricing and hedging index options under stochastic volatility: an
empirical examination. Federal Reserve Bank of Atlanta Working Paper 96-9.

Nandi, S. (1998). How important is the correlation between returns and volatility in
stochastic volatility model? Empirical evidence Irom pricing and hedging in the
S&P 500 index options market. Journal of Banking and Finance. 22, 589-610.

Rubinstein, M. (1994). Implied binomial trees. Journal of Finance. 69: 3, 771-818.
23
Figure 1. FTSE 100 Index and At-the-Money Implied Volatility during 2001.
3500
4000
4500
5000
5500
6000
6500
2
.
1
.
2
0
0
1
2
.
2
.
2
0
0
1
7
.
3
.
2
0
0
1
6
.
4
.
2
0
0
1
1
1
.
5
.
2
0
0
1
1
3
.
6
.
2
0
0
1
1
3
.
7
.
2
0
0
1
1
5
.
8
.
2
0
0
1
1
7
.
9
.
2
0
0
1
1
8
.
1
0
.
2
0
0
1
2
0
.
1
1
.
2
0
0
1
2
1
.
1
2
.
2
0
0
1
10,00
15,00
20,00
25,00
30,00
35,00
40,00
45,00
50,00
FTSE 100 ATM Volatility
















24
Table I. Descriptive Statistics.

Descriptive statistics Ior the FTSE 100 index and 30-day at-the-money (ATM) implied
volatility. The 30-day ATM implied volatility series is obtained by linear interpolation between
the two adjacent maturity ATM option implied volatilities. The sample period extends Irom
January 2, 2001 through December 28, 2001.

Panel A: Level
FTSE 100 Index ATM Implied Volatility
Mean 5564.16 21.50
Median 5550.60 19.83
Minimum 4433.70 14.95
Maximum 6334.50 44.93
Standard Deviation 414.04 5.72
Skewness -0.14 1.45
Excess Kurtosis -0.67 1.60

Panel B: Logarithmic Iirst diIIerences
FTSE 100 Index ATM Implied Volatility
Mean -0.0007 -0.0002
Median -0.0005 -0.0006
Minimum -0.0416 -0.1443
Maximum 0.0398 0.3087
Standard Deviation 0.0136 0.0536
Skewness -0.0507 0.9058
Excess Kurtosis 0.3620 4.1659

Panel C: Correlation
FTSE 100 Index ATM Implied Volatility
FTSE 100 Index 1.00 -0.68
ATM Implied Volatility -0.68 1.00








25
Table II. FTSE 100 Index Option Sample.
The reported numbers Ior each maturity-moneyness category are (i) the average price, (ii) the
average implied volatility, and (iii) the total number oI observations. The sample period extends
Irom January 2, 2001 through December 28, 2001. OTM, ATM, and ITM reIer to out-oI-the-
money, at-the-money, and in-the-money, respectively. Option with moneyness (deIined as F/K
Ior call options and K/F Ior put options) less than 0.97 is said to be OTM, option with
moneyness larger than 0.97 but less than 1.03 ATM, and option with moneyness greater than
1.03 ITM. Short and long reIer to options with less than 40 trading days and with 40-120 trading
days to maturity, respectively.

Panel A: All Options
Moneyness Time to Maturity
Short Long Subtotal
OTM 28.21 86.91
21.94 21.20
5980 6253 12233
ATM 136.93 227.58
21.17 21.21
5981 6292 12273
ITM 394.92 455.67
21.84 21.21
5225 5449 10674
Subtotal 17186 17994 35180

Panel B: Call Options
Moneyness Time to Maturity
Short Long Subtotal
OTM 19.93 70.87
18.87 18.87
3197 3349 6546
ATM 134.97 223.35
21.14 21.17
2998 3139 6137
ITM 388.97 458.09
25.10 23.67
2442 2561 5003
Subtotal 8637 9049 17686







26
Table II. Continued.

Panel C: Put Options
Moneyness Time to Maturity
Short Long Subtotal
OTM 37.71 105.41
25.47 23.90
2783 2904 5687
ATM 138.91 229.63
21.21 21.23
2983 3153 6136
ITM 400.13 453.53
18.99 19.03
2783 2888 5671
Subtotal 8549 8945 17494






























27
Table III. Size oI the Black-Scholes and Smile-Adjusted Deltas.

The reported numbers are average Black-Scholes (BS) and smile-adjusted (SAD) deltas deIined
as S c c c / and | || | K c S c c o c o c c + c c / / / , respectively. OTM, ATM, and ITM stand Ior out-oI-
the-money, at-the-money, and in-the-money, respectively.

Call Options Put Options
Moneyness BS SAD DiIIerence BS SAD DiIIerence
Full Sample 0.46 0.41 0.05 -0.50 -0.54 0.04
OTM 0.17 0.14 0.04 -0.20 -0.25 0.05
ATM 0.50 0.44 0.06 -0.48 -0.54 0.06
ITM 0.77 0.73 0.04 -0.80 -0.84 0.04























28
Table IV. Quantitative Fit oI the Black-Scholes and Smile-Adjusted Deltas.

The reported results are based on the Iollowing regression speciIications:
( ) | | c + A c c | + o = A S S c c /
( )( ) ( ) | | c + A c o c o c c + c c | + o = A S K c S c c / / /
where c, S, K, and o denote option price, underlying index level, strike price, and implied
volatility, respectively, and A is the Iirst diIIerence operator. White`s heteroscedasticity
consistent standard errors are reported in parentheses. BS and SAD denote Black-Scholes delta
and smile-adjusted delta, respectively. OTM, ATM, and ITM stand Ior out-oI-the-money, at-
the-money, and in-the-money, respectively.

BS SAD
Moneyness
Time to
Maturity
o | R
2
o | R
2

Full Sample All -0.49 0.99 0.9651 -0.58 0.99 0.9710
(0.04) (0.00) (0.04) (0.00)
Short -0.38 0.99 0.9621 -0.39 0.99 0.9684
(0.06) (0.00) (0.05) (0.00)
Long -0.60 0.99 0.9684 -0.77 0.99 0.9740
(0.05) (0.00) (0.05) (0.00)
OTM All -0.87 0.87 0.8871 -0.90 0.88 0.9096
(0.05) (0.01) (0.05) (0.01)
Short -0.85 0.80 0.8422 -0.81 0.82 0.8777
(0.08) (0.02) (0.07) (0.02)
Long -0.95 0.90 0.9148 -1.05 0.91 0.9304
(0.07) (0.01) (0.07) (0.01)
ATM All -0.80 0.97 0.9527 -0.88 0.97 0.9599
(0.07) (0.01) (0.07) (0.00)
Short -0.91 0.97 0.9431 -0.91 0.97 0.9510
(0.12) (0.01) (0.11) (0.01)
Long -0.71 0.98 0.9624 -0.85 0.98 0.9689
(0.09) (0.01) (0.08) (0.00)
ITM All -0.49 1.03 0.9858 -0.57 1.02 0.9883
(0.06) (0.00) (0.06) (0.00)
Short -0.30 1.03 0.9865 -0.30 1.03 0.9892
(0.09) (0.00) (0.08) (0.00)
Long -0.67 1.03 0.9850 -0.82 1.02 0.9873
(0.08) (0.00) (0.08) (0.00)




29
Table V. Hedging Errors Ior 1-Day Hedging Horizon.

The reported numbers Ior each maturity-moneyness category are (i) the mean absolute hedging
errors (MAHE) and (ii) the root mean squared hedging errors (RMSHE). BS and SAD denote
Black-Scholes delta and smile-adjusted delta, respectively. The reported diIIerence is the mean
diIIerence between the errors oI the two models. Bootstrapping with 1000 resamplings is used
to test the signiIicance oI the diIIerences. OTM, ATM, and ITM stand Ior out-oI-the-money, at-
the-money, and in-the-money, respectively.

MAHE RMSHE
Moneyness
Time to
Maturity
BS SAD DiIIerence BS SAD DiIIerence
Full Sample All 4.05 3.99 0.06 * 6.92 6.31 0.61 *
Short 4.03 3.79 0.24 * 7.43 6.79 0.64 *
Long 4.06 4.18 -0.12 * 6.38 5.81 0.57 *
OTM All 3.47 3.38 0.09 * 6.15 5.51 0.64 *
Short 3.19 2.97 0.22 * 6.37 5.67 0.70 *
Long 3.74 3.78 -0.04 5.93 5.35 0.58 *
ATM All 4.93 4.86 0.07 8.03 7.39 0.64 *
Short 5.31 4.98 0.33 * 8.91 8.27 0.64 *
Long 4.57 4.73 -0.16 * 7.08 6.44 0.64 *
ITM All 3.61 3.60 0.01 6.22 5.68 0.54 *
Short 3.41 3.24 0.17 * 6.49 5.87 0.62 *
Long 3.80 3.95 -0.15 5.95 5.49 0.46 *
* signiIicant at the 0.01 level
signiIicant at the 0.05 level













30
Table VI. Hedging Errors Ior 5-Day Hedging Horizon.

The reported numbers Ior each maturity-moneyness category are (i) the mean absolute hedging
errors (MAHE) and (ii) the root mean squared hedging errors (RMSHE). BS and SAD denote
Black-Scholes delta and smile-adjusted delta, respectively. The reported diIIerence is the mean
diIIerence between the errors oI the two models. Bootstrapping with 1000 resamplings is used
to test the signiIicance oI the diIIerences. OTM, ATM, and ITM stand Ior out-oI-the-money, at-
the-money, and in-the-money, respectively.

MAHE RMSHE
Moneyness
Time to
Maturity
BS SAD DiIIerence BS SAD DiIIerence
Full Sample All 12.42 10.05 2.37 * 18.18 15.66 2.52 *
Short 11.24 9.28 1.96 * 18.36 16.52 1.84 *
Long 13.54 10.79 2.75 * 18.01 14.79 3.22 *
OTM All 9.95 8.46 1.49 * 14.02 12.28 1.74 *
Short 8.01 6.78 1.23 * 12.81 11.50 1.31 *
Long 11.78 10.05 1.73 * 15.07 12.98 2.09 *
ATM All 17.07 12.99 4.08 * 24.63 21.00 3.63 *
Short 17.18 13.84 3.34 * 26.53 24.17 2.36 *
Long 16.97 12.19 4.78 * 22.72 17.55 5.17 *
ITM All 10.26 8.73 1.53 * 13.70 11.94 1.76 *
Short 8.64 7.30 1.34 * 12.12 10.25 1.87 *
Long 11.83 10.11 1.72 * 15.08 13.38 1.70 *
* signiIicant at the 0.01 level
signiIicant at the 0.05 level














31
Table VII. Hedging Errors Ior 10-Day Hedging Horizon.

The reported numbers Ior each maturity-moneyness category are (i) the mean absolute hedging
errors (MAHE) and (ii) the root mean squared hedging errors (RMSHE). BS and SAD denote
Black-Scholes delta and smile-adjusted delta, respectively. The reported diIIerence is the mean
diIIerence between the errors oI the two models. Bootstrapping with 1000 resamplings is used
to test the signiIicance oI the diIIerences. OTM, ATM, and ITM stand Ior out-oI-the-money, at-
the-money, and in-the-money, respectively.

MAHE RMSHE
Moneyness
Time to
Maturity
BS SAD DiIIerence BS SAD DiIIerence
Full Sample All 20.34 19.42 0.92 32.63 26.90 5.73 *
Short 19.91 18.34 1.57 * 33.22 27.24 5.98 *
Long 20.65 20.20 0.45 32.20 26.65 5.55 *
OTM All 18.92 17.52 1.40 * 32.43 25.65 6.78 *
Short 17.82 16.29 1.53 32.75 26.07 6.68 *
Long 19.73 18.43 1.30 32.18 25.33 6.85 *
ATM All 22.85 21.73 1.12 34.77 29.16 5.61 *
Short 23.13 20.98 2.15 35.41 29.32 6.09 *
Long 22.66 22.24 0.42 34.32 29.04 5.28 *
ITM All 18.44 18.39 0.05 29.58 24.93 4.65 *
Short 17.97 17.16 0.81 30.54 25.61 4.93 *
Long 18.78 19.29 -0.51 28.86 24.42 4.44 *
* signiIicant at the 0.01 level
signiIicant at the 0.05 level













32
Table VIII. Hedging Errors in Stable and in Highly Volatile Market Conditions.
The reported numbers Ior each maturity-moneyness category are (i) the mean absolute hedging
errors (MAHE) and (ii) the root mean squared hedging errors (RMSHE). BS and SAD denote
Black-Scholes delta and smile-adjusted delta, respectively. The reported diIIerence is the mean
diIIerence between the errors oI the two models. Bootstrapping with 1000 resamplings is used
to test the signiIicance oI the diIIerences. OTM, ATM, and ITM stand Ior out-oI-the-money, at-
the-money, and in-the-money, respectively.

Panel A: Stable Market, May 2001
MAHE RMSHE
Moneyness
Time to
Maturity
BS SAD DiIIerence BS SAD DiIIerence
Full Sample All 2.79 2.75 0.04 4.07 3.92 0.15
Short 2.64 2.49 0.15 * 3.96 3.69 0.27 *
Long 2.96 3.04 -0.08 4.19 4.17 0.02
OTM All 2.17 2.13 0.04 3.31 3.17 0.14
Short 1.79 1.70 0.09 2.90 2.70 0.20
Long 2.59 2.62 -0.03 3.72 3.62 0.10
ATM All 3.75 3.66 0.09 5.07 4.88 0.19
Short 3.91 3.65 0.26 * 5.25 4.88 0.37 *
Long 3.56 3.67 -0.11 4.86 4.88 -0.02
ITM All 2.36 2.35 0.01 3.44 3.36 0.08
Short 2.08 2.00 0.08 3.07 2.92 0.15
Long 2.67 2.75 -0.08 3.81 3.79 0.02

Panel B: Highly Volatile Market, September 2001
MAHE RMSHE
Moneyness
Time to
Maturity
BS SAD DiIIerence BS SAD DiIIerence
Full Sample All 12.76 10.66 2.10 * 19.48 16.70 2.78 *
Short 13.39 11.59 1.80 * 21.50 19.02 2.48 *
Long 12.02 9.56 2.46 * 16.79 13.48 3.31 *
OTM All 11.55 9.48 2.07 * 17.76 14.77 2.99 *
Short 11.73 9.90 1.83 * 19.19 16.27 2.92 *
Long 11.33 8.98 2.35 * 15.90 12.77 3.13 *
ATM All 14.58 12.34 2.24 * 21.78 19.14 2.64 *
Short 15.98 14.10 1.88 * 24.60 22.47 2.13 *
Long 12.96 10.32 2.64 * 18.01 14.37 3.64 *
ITM All 11.62 9.68 1.94 * 17.92 15.07 2.85 *
Short 11.75 10.10 1.65 * 19.39 16.63 2.76 *
Long 11.47 9.17 2.30 * 15.97 12.94 3.03 *
* signiIicant at the 0.01 level
signiIicant at the 0.05 level