Applied Mathematics, 2012, 3, 597-605

http://dx.doi.org/10.4236/am.2012.36093 Published Online June 2012 (http://www.SciRP.org/journal/am)

Black-Scholes Option Pricing Model Modified to Admit a

Miniscule Drift Can Reproduce the Volatility Smile
Matthew C. Modisett1, James A. Powell2*
1
Financial Guard Ltd., London, UK
2
Department of Mathematics and Statistics, Utah State University, Logan, USA
Email: *jim.powell@usu.edu

Received November 2, 2011; revised April 10, 2012; accepted May 17, 2012

ABSTRACT
This paper develops a closed-form solution to an extended Black-Scholes (EBS) pricing formula which admits an im-
plied drift parameter alongside the standard implied volatility. The market volatility smiles for vanilla call options on
the S&P 500 index are recreated fitting the best volatility-drift combination in this new EBS. Using a likelihood ratio
test, the implied drift parameter is seen to be quite significant in explaining volatility smiles. The implied drift parame-
ter is sufficiently small to be undetectable via historical pricing analysis, suggesting that drift is best considered as an
implied parameter rather than a historically-fit one. An overview of option-pricing models is provided as background.

Keywords: Option Pricing; Black-Scholes; Volatility Smile

1. Introduction An option is a financial contract between a seller (or

writer) and a buyer, in which a buyer, in exchange for a
An often reiterated view of modern finance states that the
premium, acquires the right (but not the obligation) to
Black-Scholes (BS) option pricing formula “accurately
buy (for a call option) or sell (for a put option) a specific
reflects” options prices, but this view forgives a free pa-
amount of an underlying commodity at a specified price
rameter (volatility) which accommodates a considerable
(the strike price, S) during a specified period of time.
degree of variations in option prices. Not only does im-
This paper focuses on call options. In our notation, w
plied volatility vary over time, but also at any point in
represents the value (or price) of the option at a point in
time over related options, that is a group of options
time. At maturity, this value represents the final settle-
whose characteristic differ only by strike price. All other
ment that the writer pays the seller,
characteristic (underlying security, maturity) are identical.
This definition does not assume that they use the same wmaturity  max  x  S , 0  ,
volatility for pricing, but it is generally assumed that the
where x is the ending value of the underlying commodity.
same interest rate is applicable. This variation, referred to
Only European options are considered, in which there is
as the volatility smile or smirk (or sneer), reflects the
no possibility of exercise before maturity.
greater sin, since variation over time could easily be as-
The basic option pricing formula was developed in a
cribed to changing conditions in the market environment,
series of papers [2-4] (leading to a Nobel Prize for Scho-
but variations at any particular instance cannot [1]. This
les and Merton in 1997). This model is the Black-Scholes
paper addresses these latter variations by offering a slight
(BS) option pricing formula. Black and Scholes devel-
modification to the BS formula which reflects most of
oped a partial differential equation (PDE) relating w (the
the volatility smile. A (small) drift parameter is added,
option price), r (the interest rate), t (time to maturity of
and a closed-form solution, quite similar to BS, is de-
the option), x (the price of the underlying security at the
rived. Using a likelihood ratio test, this new drift pa-
current time), and price volatility, v:
rameter is determined to be significant over a sample
series of financial data. Furthermore, we argue that this w w 1 2 w
 rx  rw  v 2 x 2 2 . (1)
change may explain volatility smiles. To place this model t x 2 x
in the proper context, the paper starts with background
The volatility, v, is expressed as a fraction of the un-
concerning the volatility smile and the general classes of
derlying security (in this paper v is used for volatility
models used to explain these smiles.
rather than the more typically used σ, since the latter will
*
Corresponding author. be used to represent model variance from observed prices

Copyright © 2012 SciRes. AM

598 M. C. MODISETT, J. A. POWELL

2 w of research to explain the volatility smile, and the focus

below). The coefficient of directly implies a log- has been to vary the underlying dynamic of the security
x 2
normal probability distribution for the price of the un- prices changes. This process has two goals: to reproduce
derlying security at maturity. Alternatively, Bouchaud witnessed option prices and to employ an assumption of
and Potters [5] show that the coefficient of the second security price changes that is reasonable. A number of
derivative term in (1) can arise from different assump- models have been proposed, outlined below (following
tions. Initially, the BS option pricing model was viewed Derman [6]).
to price options without reference to the distribution of Local volatility models (or deterministic volatility
the future price of the underlying, except for the free function models or implied binomial/trinomial tree mod-
volatility parameter. Currently, it is acknowledged that els) [1,7-9] used current prices to deduce future volatile-
the option price depends on the assumed or implicit dis- ities. These methods are now out of favor since the im-
tribution of underlying prices. plied future volatilities do not exhibit volatility smiles,
Recent research has been motivated, in large part, by rather much flatter patterns and lower levels of volatility.
the volatility smile (or smirk or sneer) effect, which Also, the smoothing of local volatility surfaces from dis-
arises from efforts to explain current prices. Given all the crete option prices constitutes a significant issue.
parameters for an option, except volatility, and given a Stochastic volatility models [10,11] consider both the
market price for the option, the pricing formula can be underlying security and its volatility as stochastic vari-
inverted to give the implied volatility arising from the ables. These models reproduce volatility smiles, but ini-
market price. If we calculated the implied volatility for a tially only symmetric ones. Later, asymmetric prices
series of options that only differ by strike, we can see were reproduced by allowing prices of the underlying
how volatility varies with respect to strike. Stereotypical security to be correlated with volatility. With these mod-
examples are shown in Figure 1. If the profile is sym- els, hedging (see below) is difficult, primarily because
metric, it is referred to as a “smile”, while asymmetric volatility is not an observable statistic, so that a regime
profiles are “smirks” or “sneers”. shift could not be noted at the time that a hedge transac-
In the original theory, option prices were viewed as a tion would be needed.
function of the underlying dynamic of the security price More recently innovations have varied aspects of the
changes. It followed that the volatility profile should be a model other than volatility. Jump diffusion models [3]
flat line, constant across all strikes, since the considered have been adapted [12] to the stochastic volatility envi-
options only differ by their strike prices; the pricing took ronment by relaxing the continuous assumption of price
place with the same underlying security over the same behavior. Stock modelers, for example, tend to think in
time to maturity with the same governing interest rates. downward jumps, like crashes, while energy price mod-
However, since the stock market crash of 1987, volatility elers might think of price spikes, the latter not only going
profiles have generally been not flat, rather displaying in the opposite direction but being shorter lived. Appro-
distinct smiles or smirks, as in Figure 1. That volatility priate parameters must be incorporated for the frequency,
profiles are not flat refutes the basic assumptions of the average intensity and standard deviation of intensity,
BS option formula. Recent years have witnessed a flurry along with any assumptions regarding persistency or
mean regression of the new price levels after the jump.
These models possess intuitive appeal since dynamics
such as crashes, spikes, and perhaps manic periods can
be reflected. As a drawback, jump models exaggerate
any hedging concerns associated with pure stochastic
models; while fast moving markets may be difficult to
hedge, jumping ones would seem impossible under in a
continuous framework. More recent universal volatility
models [13-17] comprise hybrids of the above.
All these models introduce new parameters, for exam-
ple one or more volatility parameters to create a smile,
various distribution parameters for a stochastic volatility
model, other parameters to describe jump or other effects.
These parameters can be divided into two types: those
parameters fitted to historical data and then fixed in the
Figure 1. Volatility profiles: Volatility versus strike price. If model and those allowed to vary moment to moment as
the profile is symmetric, it is referred to as a “smile”, while free parameters (similar to implied volatility). The for-
asymmetric profiles are “smirks” or “sneers”. mer are called historically-fitted and the latter are implied.

Copyright © 2012 SciRes. AM

 M. C. MODISETT, J. A. POWELL 599

Local volatility models generally employ a host of im- reflects the difference between interest rate effects and
plied volatility parameters and few (if any) historically the missing random walk effects over the stagnant peri-
fitted variables. In contrast, stochastic volatility models ods. The implied drift effect could be positive or negative
employ many historically-fitted parameters and only one since the stochastic process during active periods (im-
implied parameter, the volatility. Jump process parame- plied by volatility) could create a forward expectancy
ters are generally historically-fitted. above or below that implied by interest rates.
As discussed above, current research tends to add new
parameters to models in an effort to better reflect various 2.1. Deriving an Extension of the Black-Scholes
aspects of price movements. While doing this, one must Model
be careful that new parameters not only improve the fit
The traditional derivation of BS begins by considering
but also that the degree of improvement warrants the new the neutral hedge equity constructed by selling a ratio
parameter. Below we will investigate a one-parameter 1
extension of the BS model which accounts for a small  w 
  of call option in a stock per share of stock held.
pricing drift. We evaluate its performance compared to  x 
the original BS model from the perspective both of re- Consider the stock and hedging options as a separate
flecting option prices and from that of parsimony: can the portfolio. For such a portfolio the net equity (E) invested
increased complexity be justified statistically by the de- is
gree of improvement? We fit both the BS and our exten-  w
sion to S&P 500 Options data from 2003-2005; the E x N ,
 wx 
tradeoff between additional complexity and improved
description is assessed using the likelihood ratio test. where N is the number of shares of stock held, and sub-
script notation for partial derivatives is adopted (that is,
2. Mathematical Model and Methods wx   x w ). Thus
How would drift come about? There are two alternatives. 
E  N x  wx1w , 
On the one hand, it may be a small trend that market par-
ticipants cannot observe. Actual calculations of the drift is the net change in portfolio value with changes in either
show it to be quite small, small enough to not be detect- stock or option price. Applying the usual arbitrage-free
able in a statistical analysis of price movements. This argument originally introduced by Black, Merton and
view of the drift is consistent with the contention that Scholes, it is required that option prices evolve over time
so that equity value experiences growth according to
market dynamics (alone) drive options prices, with neg-
risk-free interest rates. This is natural since the combina-
ligible distortion by market participants. On the other
tion of stocks and hedging options constitutes a risk-free
hand, the drift may be a trend perceived by market par-
portfolio. Mathematically,
ticipants whether or not it actually exists in the market.
This perceived drift alters market prices, but the neces-   
E  N x  wx1w  N x  wx1 w r t ,  (2)
sary drift is not small enough to be discredited by statis-
tical analyses. This view is consistent with the contention where r is the risk-free interest rate. Option price is ex-
that market perceptions drive market prices, potentially pected to vary with both time and price of underlying
independent of actual price dynamics. Whichever inter- stock; using Taylor’s theorem changes in option price
may be replaced with related changes in stock price in
pretation appeals to the reader, the mathematics is the
time,
same.
It is interesting to note that this additional drift pa- 1
w  wx x  wxx x 2  wt t
rameter may be viewed as arising from the assumption 2 (3)
that the underlying security experiences periods of vola-
tility and periods of relative stagnation. During volatile

 O t , xt , x .
2 3

periods, both random walk effects and interest rates im- The next step would be to substitute (3) into (2) and
pact option prices. However, during stagnant periods, thereby derive a differential equation for option prices.
any pricing effects due to volatility are dormant (by defi- However, to consistently neglect terms of differing order
nition) but the interest rate clock would still be running. as t  0 , there must be an assumption about how
Thus the implied forward expectation from current op- changes in stock price, x, vary with changes in time.
tion prices would conceptually be composed of two As done in the traditional BS assumption, here the as-
components: one due to the volatility effect (with interest sumption is that stock prices follow a random walk proc-
rates during volatile periods), and one due to interest ess with standard deviation vx in which the market vola-
rates (alone) during stagnant periods. The implied drift tility is v. Departing from BS, however, the random walk

Copyright © 2012 SciRes. AM

600 M. C. MODISETT, J. A. POWELL

is assumed to have a slight bias (as opposed to zero with

mean). Thus are made the following assumptions about
the expectation,  , of quantities in (2) and (3):
u  z,  0   S max e z  1, 0 .  
x  cxt ,. reflecting the (small) bias, or drift, in ran- Changing to a traveling frame of reference
 1 
dom steps, (4a)   z   c  r  v 2  to remove the lowest order z
 2 
wx x  0, reflecting market participants’ inability to
derivative gives a simple diffusion equation:
perceive bias in the underlying carries over to an inability
to perceive bias in the option price, (4b) 1 2
u  v u ,
2
x 2  v 2 x 2 t , reflecting growth of variance in the
with
random walk. (4c)
Here the parameter c may be positive or negative and
u  ,  0   S max e  1, 0 .  
reflects the expected drift or trend of underlying stock The convolution form of the general solution for the
prices. Note that for this to be a consistent approximation diffusion equation is
it is required that 1 
  2

   f   y  exp   2yv   dy,


u  ,   2πv 2
cO v , 2 2

  
2

meaning that the drift parameter will be small in com- where f   is the initial condition. Applying this form
parison to volatility, since v  1 (which dovetails with of the solution to the initial conditions above, and noting
the assumption that it is too small for market makers to that the integrand will be zero for   y gives
take into consideration). 1 
 y2 
   S e 

Multiplying (2) by wx, canceling a common factor of N, u  ,   2πv 2 y
 1 exp   2
2
 dy.
substituting (3) into (2), replacing the various changes in   2v  
stock price with their expectations in (4), dividing by t
and taking the limit t  0 leads directly to: From here, completing the square and using
z
2
w
  r  c x
w 1 2w
 rw  v 2 x 2 2 , (5) erf  z   e
t 2
dt
t x 2 x π 0

where c is the implied drift parameter. This is the Ex- gives

tended Black-Scholes (EBS) model. BS’s differential
  1
  v 2    v 2 
equation can be recovered by setting c to zero. Note that u   ,   e 2  1  erf   
this drift measures the deviation not from current prices, 2   2v 2 2 
   
but rather deviation from the interest rate applied to cur-
rent prices (the arbitrage-free estimate of future prices     
 1  erf     .
relative to current prices). 
 2v 
2
  

2.2. Solution of the Extended Black-Scholes Returning to the original coordinates, simplifying,
Equation evaluating at t = 0 and employing the identity
1  z 
We seek to solve (5) with a boundary condition at matur-   z   1  erf    , where Ф is the standard nor-
ity (t = t*) given by 2  2 

 
w x, t  t   max  x  S , 0  .
mal cumulative distribution function, gives

w  xe  ct   d    Se rt   d   ,
 
(6)
Introducing a change of variables,
x where
  t  t  , z  ln   , w  e  r u  z,  ,
S
  x  1 
ln     r  c  v 2  t 
d    
is introduced to remove the non-constant coefficients and S 2 
.
2 
exponential growth term, as well as to translate the con- vt
dition at maturity into an initial condition, giving
1 2  1  Under the assumption that c = 0 this reduces to the
u  v u zz   c  r  v 2  u z ,
2  2  original BS option pricing formula.

Copyright © 2012 SciRes. AM

 M. C. MODISETT, J. A. POWELL 601

3. Data Analysis and Model Competition served data, with wj being the model prediction for the
maximum likelihood parameters, wobs being the observed
In the absence of a market trend (c = 0), with x, r, S and
option price and  e being the standard deviation of the
t* assumed known, (6) gives a one-to-one relationship
errors (see [18] for details). Because the difference in
between volatility and option price. Inverting this rela-
these logs is the log of the ratio of the likelihoods, this
tionship individually for a series of options, differing
only in strike price, often gives a non-constant volatility test is called the likelihood ratio test. The resulting chi-
(the “smile” or “smirk”). However, when c is not zero, squared probability can be viewed as the probability that
how should one proceed? A direct analog to the above the more complex model is an improvement over the
procedure would require selecting pairs of options shar- simpler nested model (that is, the probability that the
ing all parameters save strike price, then solving the re- simpler model’s parameters lie outside the corresponding
sulting nonlinear system of equations for v and c. But parametric confidence interval of the more complex
which pairs should be selected? Or if all possibly pairs model). General wisdom states that parameters should
are considered, how are results to be interpreted? Are the not be added without penalty; the above procedure pe-
differences significant? This latter question, in fact, ap- nalizes the more complex model for its higher number of
plies also to the volatility smile: given the background parameters via the numerical degrees of freedom in the
variability in underlying stock prices and volatility, are chi-squared distribution.
the differences among applied volatilities really signifi- To determine c and v for each day of data in our EBS
cant? model and v for the BS model, we minimize the negative
In this paper, an alternate approach is employed fol- log likelihoods of each over the strike prices for that day.
lowing a model competition and selection perspective Furthermore, the negative log likelihood of each model is
advanced by Hilborn and Mangel in 1997 [18]. In this used in the likelihood ratio test, as described above. A
approach, model parameters for competing models are fit corresponding p value was determined from the chi-
to all available data using maximum likelihood, cogni- squared distribution with one degree of freedom, indi-
zant of the fact that there is process variability in the wit- cating the probability that the more complex model (EBS)
nessed parameters (e.g. stock prices are clearly not fixed is a significant improvement over its simpler companion
through the course of a day) as well as observational (BS) for that day.
variability (e.g. different market participants may assess
volatility using different metrics, over different periods 3.2. Data
of time, yielding different option prices). Model parame-
We use data from the Option Price Reporting Authority
ters are chosen to maximize likelihood, and the model
(OPRA) for S&P 500 index calls maturing December
which accounts for the most variability among alternative
(19), 2006. The data runs from November 5, 2003 to
models, with maximum parsimony, should be considered
November 3, 2005. Option Price Reporting Authority
“best”.
provides consolidated option information and is a com-
3.1. Statistical Methods mittee administered jointly by the five option exchanges:
AMEX, BOX, CBOE, CBOT, ISE, PCX, PHLX. Our
The EBS vis-à-vis the BS constitutes a case of nested data came indirectly through a third-party vendor. We
models, in which the simpler model is realized by a par- excluded 5 dates in which the data seemed in error: for
ticular choice of parameters in the more complex. A like- the 1325 strike in 2003, November 20, 21, 25, and De-
lihood ratio test may be used to characterize the signify- cember 8, 16 and 22; for the 1125 strike, the prices in
cance of differences between nested models. Assuming 2005 of October 24, 25 and 28 and November 2 also
both models are fit to data using maximum likelihood, seemed probably in error. In the 1150 Strike, the follow-
the statistic ing dates are excluded: 2005 October 19, 24, November
R  2  L  data model1   L  data model2   , 3. We consider only contracts whose strike is between
1125 and 1375, inclusive, and which is a multiple of 25.
is distributed according to a chi-squared distribution with This has the effect of concentrating on contracts more
number of degrees of freedom equal to the difference in likely to be liquid and traded, and to eliminate sporadic
numbers of parameters between model 1 (the simpler) or illiquid contracts. During this time:
and model 2 (the more complex, in which model 1 is 1) The S&P 500 index grew from 1051.81 to 1219.94,
nested). Here with a minimum of 1033.65 (November 20, 2003) and a
1 1 maximum of 1245.04 (August 3, 2005). The True Range
L  data model j       w j  wobs 
2
ln 2π e2  2 (daily high minus daily low) expressed as a percent of
2 2 e data
the index value varied from 0.25% to 2.14%, with an
is the negative log-likelihood of model j given the ob- average of 0.91% and a standard deviation of 0.37%.

Copyright © 2012 SciRes. AM

602 M. C. MODISETT, J. A. POWELL

(See Figure 2). well have a backlog of orders in either the option or the
2) Average transaction volume per contact grew from shares, such that a market maker could reliably see that
about 332 contracts (averaged over 5 working days) to the price will move in one direction. Thus the market
peak of 3476 and ending at 2003 contracts (See Figure maker would use an index price in the option formula
3). This increase is a natural consequence of coming that differs from the current index price.
closer to maturity since the shortest dated options are the
most traded and liquid. 3.3. Analysis of Data: Results
3) Excluding the 1125 strike contract and certain days
To compare the EBS model to the no-drift BS, two fit-
due to data issues, implied volatilities varied from 7.57%
tings are performed. First each day’s closing option
to 16.05%, with an average of 10.94% and a standard
prices are fit in the BS to the best implied volatility (one
deviation of 1.23%. For each day, the difference between
volatility for all options) that maximizes the likelihood of
max and min implied volatilities (for days in which more
the model given actual option prices. Second, for each
than one contract was traded, and excluding 1125 and
day, the EBS is fit for the best combination of drift and
1150 strikes) varied from 0.26% to 6.43%, averaging
implied volatility (one combination for all options on that
1.92% with a standard deviation of 1.04%.
day) maximizing the likelihood.
4) We note that the closing price for each option and
To evaluate the results on each day, the worst and av-
the index were used. This is a potential source of incon-
erage per-day price difference between actual and model
sistency since the last trade for an option contract may
are chosen. The error is considered as a percent of the
not occur at the end of the last moment of stock trading
underlying closing index price on that day, with the re-
(applicable for the index). Our model, like BS, assumes
sults summarized in Table 1. The error is approximately
that the current price of the index is used. Option ex-
halved by the addition of an implied drift parameter. This
changes generally close before stock exchanges, so clos-
is true whether looking at the worst price of each day, or
ing option and index prices seldom occur simultaneously.
the average over all option prices. The ratio of standard
Even if the data from within a day were used to synchro-
deviation to average for the error term does not change
nize option and index prices, the result may not be as
significantly when adding a drift parameter.
market participants view the market. Market makers may
Using a likelihood ratio test for each day, the drift pa-
rameter is seen to have over a 95% chance of improving
fit in 297 out of 344 days. Insight can be gained by com-
paring this statistic with the implied drift for each corre-
sponding day. Figure 4 illustrates that the low chances of
improving the model occur on those days in which drift
values are close to zero. This is to be expected, given that
the probability of improvement depends on where the
single parameter BS volatility estimate falls relative to
the confidence interval for the drift + volatility estimate
for EBS. The EBS fails to be clearly better exactly where
Figure 2. S&P 500 index. The underlying commodity be- its parameters become indistinguishable from the no-drift
havior for the options considered, over the time period No-
vember 2003 to November 2005. Table 1. Comparison of residuals, Classic BS with and
without drift. The modified BS model with drift (called EBS)
significantly reduces variability of volatility estimates over
the classical BS model. The average variation in each day is
generally halved, while the worst fit for each day (an outlier)
is also reduced. Further reduction in variability may not be
meaningful.

Classic BS EBS
(No drift) (With drift)
Max Error over Period 0.87% 0.58%

Average Error (Each Day’s Worst) 0.27% 0.13%

Average Error (All Prices) 0.15% 0.07%

Figure 3. Daily option volume. For each day, the average
volume for option contracts in the series studied. As option Std. Dev. of Error (Each Day’s Worst) 0.14% 0.09%
contracts approach maturity, volume naturally increases
Std. Dev. of Error (All Prices) 0.11% 0.06%
since most trading is done in the nearest contract.

Copyright © 2012 SciRes. AM

 M. C. MODISETT, J. A. POWELL 603

Figure 4. Probability that EBS improves BS. Most often, the

chance is 100% that EBS improves BS. The lower probabil-
ity cases correspond to when the drift is close to zero and
the two models are less distinguishable. Figure 5. Comparison of implied volatility estimates from
EBS and BS. The EBS model’s implied volatility estimate is
BS. This supports the contention that the implied drift generally less than that of BS, but the two track each other
parameter significantly improves the model’s fit to actual reasonable well.
prices overall.

4. Discussion
The EBS model makes a significant improvement, but
perhaps adding more parameters could provide additional
significant increases in fit. To the knowledge of the au-
thors, there is no mathematical or statistical method for
determining whether or not further parameters could be
justified. Rather, practitioners need to look at the data Figure 6. Implied drift in the EBS model (expressed as a
and its context to determine whether or not additional rate). The above annualized rates are generally small—the
parameters could be warranted. Here it is argued that the daily price movement would be below what could be de-
closeness of fit using implied volatility-drift in the EBS tected with statistical studies. The drift parameter becomes
probably is as close as the data could reasonably allow. erratic closer to maturity as a consequence of annualizing
rate on a short-term instrument; this effect is accounted for
Referring back to Table 1, the average price error for in the next figure.
an option in the new model is 0.07%, and the average
worst for each day is 0.0013%. Can we expect greater fit
than this? These price variations are similar to bid-ask
spreads on many illiquid securities. Options have no-
where near the liquidity enjoyed by the underlying index.
To attempt any greater fitting of option prices would de-
mand explanations of variations with a magnitude similar
to that of bid-ask spreads in the security. This would Figure 7. Implied drift in the EBS model (expressed as a
seem unreasonable for a pricing model that ignores such discount price). Estimated drift rate data is converted to
bid-ask spreads. The conclusion is that the fit offered by price data. The implied drift parameter evolves well, nei-
the EBS may be as close as is reasonable. ther staying too static nor too fluctuating too erratically.
When introducing implied parameters into a model,
one should observe how the parameter evolves over time in Equation (2). As the time to maturity shortens, how-
to ensure that the implied parameter is neither too stable ever, this drift rate becomes erratic, which is typical for
nor too erratic. If too stable, the implied parameter any rate that is annualized over progressively shorter
probably masks another effect more appropriate for di- time periods; minor discrepancies in price are magnified.
rect modeling. Figure 5 compares the evolution of the To account for this, in Figure 7, the drift parameter is
implied volatilities that arise from the no-drift (BS) and expressed as a price of a zero-coupon bond whose matur-
with-drift (EBS) cases. (Gaps are due to removed data ity matches that of the option. A zero-coupon bond is a
points.) The two volatilities track each other, with the bond with no interest payment. Thus, a zero-coupon
no-drift volatility being higher than the other in practi- bond makes only a single payment, at maturity. The price
cally all cases. of such a bond is given by:
Figures 6 and 7 depict the evolution of the implied Price   payment  1  r 
t

drift parameter. Figure 6 displays the parameter itself

expressed as a rate similar to interest rates. This is the c where r is the interest rate and t the time to maturity.

Copyright © 2012 SciRes. AM

604 M. C. MODISETT, J. A. POWELL

Given a fixed t, a change in price implies a change in is comparable to that of options. Furthermore, the small
yield. For smaller t, the effect on yield is magnified. size of the daily drift parameter indicates that it would be
When t is very small, say a couple of weeks or less, yield unlikely to be uncovered during historical analysis.
is quite sensitive to price changes, and the effect from The new drift parameter may be interpreted in one of
day to day can look quite distorted, as in the right end of two ways: it may be an actual trend that is too small for
Figure 6. This price representation displays a variability market participants to observe; or it may be a small trend
that dampens as time approaches maturity. This confirms perceived by market participants and driving their prices,
the contention that the drift reflects price movements whether or not this drift actually exists in the market.
(either perceived by market participants or latent in the These interpretations represent two opposing views of
market) since over shorter horizons, lower price drifts pricing theory: either security prices arise strictly from
appear. underlying market dynamics, or arise from perceptions of
While yield is typically express as a percent return per market players. It is interesting that historical testing
year, the implied drift parameter in Figure 7 can be ex- cannot be used to resolve this debate, since the EBS is an
pressed per day, by taking the difference of the price option pricing model that can explain prices as much as
from 1.00 and dividing it by the number of days to ma- justifiable, while accommodating either assumed cause.
turity. This is the average daily drift implied by market
prices. Over our sample, this varied from −0.0042% to
REFERENCES
0.0204%, with an average of 0.0033% and standard de-
viation of 0.0028%. These are exceptionally small num- [1] B. Dupire, “Pricing with a Smile,” Risk, Vol. 7, No. 1,
bers which could never be confirmed from historical 1994, pp. 18-20.
analysis. This conclusion is consistent with a result from [2] F. Black and M. S. Scholes, “The Pricing of Options and
Brownian motion: that over short periods any drift ef- Corporate Liabilities,” Journal of Political Economy, Vol.
81, No. 3, 1973, pp. 637-654. doi:10.1086/260062
fects are completely dominated by the noise or random
effects (See [19] p. 656 for further discussion). Using [3] R. Merton, “Option Pricing When Underlying Stock Re-
turns Are Discontinuous,” Journal of Financial Econo-
drift as an implied parameter implicitly allows it to vary mics, Vol. 3, No. 1-2, 1976, pp. 125-144.
moment to moment (or, in our case, day to day), and thus doi:10.1016/0304-405X(76)90022-2
it would be difficult if not impossible to detect. Perhaps [4] F. Black, “The Pricing of Commodity Contracts,” Journal
if any particular drift were valid over a longer period of of Financial Economics, Vol. 3, 1976, pp. 167-179.
time (say until maturity), then there might be a chance to doi:10.1016/0304-405X(76)90024-6
confirm it with a historical study, but this is not the case. [5] J.-P. Bouchaud and M. Potters, “Back to Basics: Histori-
Given the momentary nature of an implied parameter, cal Option Pricing Revisited,” Philosophical Transac-
there is no real hope of confirming the validity of any tions of the Royal Society, Vol. 357, No. 1735, 1999, pp.
particular drift measurement. Nonetheless, the effect over 2019-2028.
time until maturity is significant and affects option pric- [6] E. Derman, “Regimes of Volatility,” Risk, Vol. 4, 1999,
ing. pp. 55-59.
[7] E. Derman and I. Kani, “Riding on a Smile,” Risk, Vol. 7,
5. Conclusions No. 2, 1994, pp. 32-39.
[8] M. Rubinstein, “Implied Binomial Trees,” Journal of
In this paper, the Black-Scholes option pricing model is Finance, Vol. 49, No. 3, 1994, pp. 771-818.
modified to have a free drift parameter alongside volatil- [9] L. Andersen and R. Brotherton-Ratcliffe, “The Equity
ity. Similar to the original BS work, the ensuing differen- Option Volatility Smile: An Implicit Finite-Difference
tial equations are developed, and a closed-form solution Approach,” Journal of Computational Finance, Vol. 1,
for this model is derived, with the resulting model re- No. 2, 1997, pp. 5-38.
ferred to as the Extended Black-Scholes (EBS). Rather [10] J. C. Hull and A. White, “An Analysis of the Bias in Op-
than fitting the new drift parameter historically, it is tion Pricing Caused by a Stochastic Volatility,” Advances
treated as an implied parameter, like volatility. Using in Futures and Options Research, Vol. 3, 1988, pp. 29-
61.
historical data, each day’s closing volatility smile is
best-fit to an implied volatility-drift combination. A like- [11] S. L. Heston, “A Closed-Form Solution for Options with
Stochastic Volatility with Applications to Bond and Cur-
lihood ratio test indicates that the implied drift parameter
rency Options,” Review of Financial Studies, Vol. 6, No.
significantly improves the fit of the EBS model as com- 2, 1993, pp. 327-343. doi:10.1093/rfs/6.2.327
pared with the BS model. It is argued that any better fit
[12] D. S. Bates, “Jumps and Stochastic Volatility: Exchange
may not be justified, since the remaining unexplained Rate Processes Implicit in Deutsche Mark Options,” Re-
differences between EBS and option prices are very view of Financial Studies, Vol. 9, No. 1, 1996, pp. 69-
similar to bid-asked spreads for securities whose liquidity 107. doi:10.1093/rfs/9.1.69

Copyright © 2012 SciRes. AM

 M. C. MODISETT, J. A. POWELL 605

[13] B. Dupire, “A Unified Theory of Volatility,” Working Vega Hedging of Exotic Options,” Conference Presenta-
Paper, 1996. tion at Global Derivatives and Risk Management, Juan-
[14] A. Lipton and W. McGhee, “Universal Barriers,” Risk, les-Pins, 26 June 2002.
Vol. 15, No. 5, 2002, pp. 81-85. [17] D. Brigo and F. Mercurio, “A Mixed-Up Smile,” Risk,
[15] M. Britten-Jones and A. Neuberger, “Option Prices, Im- Vol. 13, No. 9, 2000, pp. 123-126.
plied Prices Processes, and Stochastic Volatility,” Journal [18] R. Hillborn and M. Mangel, “The Ecological Detective:
of Finance, Vol. 55, No. 2, 2000, pp. 839-866. Confronting Models with Data,” Princeton University
doi:10.1111/0022-1082.00228 Press, Princeton, 1997.
[16] G. Blacher, “A New Approach for Designing and Cali- [19] R. L. McDonald, “Derivative Markets,” 2nd Edition,
brating Stochastic Volatility Models for Optimal Delta- Addison Wesley, New York, 2006.

Copyright © 2012 SciRes. AM