ArticlePDF Available

Algorithms. Back to basics: A new approach to the discrete dividend problem

  • Finance Press, Newport Beach, Ca
Back to Basics: a new approach to the
discrete dividend problem
Espen G. HaugJørgen HaugAlan Lewis§
May 26, 2003
1 Introduction
Stocks frequently pay dividends, which has implications for the value of options on these
stocks. For options on a large portfolio of stocks, one can approximate discrete dividend
payouts with a dividend yield and use the generalized Black-Scholes-Merton (BSM) model.
For options on one stock, this is not a viable approximation, and the discreteness of the
dividend has to be explicitly modelled.1We discuss how to properly make the necessary
We would like to thank Samuel Siren, Paul Wilmott, and participants in the Wilmott forum
( for useful comments and suggestions. All the usual disclaimers apply.
J.P. Morgan
Norwegian School of Economics and Business Administration
1An alternative to discrete cash dividend is discrete dividend yield. Implementation of discrete dividend
yield is well known and straight forward using recombining lattice models (see for instance Haug, 1997; Hull,
2000). Typically, however, at least the first dividend is known in advance with some confidence. Discrete
cash dividend models consequently seem to have been the models of choice among practitioners, despite
having to deal with a more complex modelling problem. For very long term options the predictability of
future dividends is less pronounced, and dividends should be somewhat correlated with the stock price level.
Moreover, cash dividends tend to be reduced following a significant stock price decrease. If the stock price
rallies, on the other hand, it indicates that the company is doing better than expected, which again can
It might come as a surprise to many readers that we write an entire paper about a supposedly
mundane issue—which is thoroughly treated in any decent derivatives text book (including,
but not limited to Cox and Rubinstein, 1985; Chriss, 1997; Haug, 1997; Hull, 2000; Mc-
Donald, 2003; Stoll and Whaley, 1993; Wilmott, 2000). It turns out, however, that some of
the adjustments suggested in the extant literature admit arbitrage—which is fine if all your
competitors use these models, but you know how to do the arbitrage-free adjustment.
Existing methods
Escrowed dividend model: The simplest escrowed dividend approach makes a simple
adjustment to the BSM formula. The adjustment consists of replacing the stock price S0by
the stock price minus the present value of the dividend S0ertDD, where Dis the size of
the cash dividend to be paid at time tD. Because the stock price is lowered, the approach will
typically lead to too little absolute price volatility (σtSt) in the period before the dividend is
paid. Moreover, it is just an approximation used to fit the ex-dividend price process into the
geometric Brownian motion (GBM) assumption of the BSM formula. The approach will in
general undervalue call options, and the mispricing is larger the later in the option’s lifetime
the dividend is paid. The approximation suggested by Black (1975) for American options
suffers from the same problem, as does the Roll-Geske-Whaley (RGW) model (Roll, 1977;
Geske, 1979, 1981; Whaley, 1981). The RGW model uses this approximation of the stock
price process, and applies a compound option approach to take into account the possibility
of early exercise. Not only does this yield a poor approximation in certain circumstances,
but it can open up arbitrage opportunities!
result in higher cash dividends. For very long term options then, it is likely that the discrete dividend yield
model can be a competing model to the discrete cash dividend model.
Several papers discuss the weakness of the escrowed dividend approach. In the case of
European options, suggested fixes are often based on adjustments of the volatility, in com-
bination with the escrowed dividend adjustment. We shortly discuss three such approaches,
all of which assume that the stock price can be described by a GBM:
1. An adjustment popular among practitioners is to replace the volatility σwith σ2=
SDertD, (see for instance Chriss, 1997). This approach increases the volatility relative
to the basic escrowed divided process. However, the adjustment yields too high volatil-
ity if the dividend is paid out early in the option’s lifetime. The approach typically
overprices call options in this situation.
2. A more sophisticated volatility adjustment also takes into account the timing of the
dividend (Haug and Haug, 1998; Beneder and Vorst, 2001). It replaces σwith σ2as
before, but not for the entire lifetime of the option. The idea behind the approxi-
mation is to leave volatility unchanged in the time before the dividend payment, and
to apply the volatility σ2after the dividend payment. Since the BSM model requires
one volatility as input, some sort of weight must be assigned to each of σand σ2.
This is accomplished by looking at the period after the dividend payment as a forward
volatility period (Haug and Haug, 1996). The single input volatility is then computed
where Tis the time of expiration for the option. The adjustment in the presence of
multiple dividends is described in Appendix A. This is still simply an adjustment to
parameters of the GBM price process, that ensures the adjusted price process remains
a GBM—at odds with the true ex-dividend price process. The adjustment therefore
remains an approximation. One can easily show numerically that this method performs
particularly poorly in the presence of multiple dividends.
3. Bos et al. (2003) suggest a more sophisticated volatility adjustment, described in Ap-
pendix B. Still this is just another “quick fix” to try to get around the problems with
the escrowed dividend price process. Numerical calculations show that this approach
offers quite accurate values provided the dividend is small to moderate. For very high
dividends the method performs poorly. The poor performance seemingly also occurs
for long term options with multiple dividends.
4. A slightly different way to implement the escrowed dividend process is to adjust the
stock price and strike (Bos and Vandermark, 2002). Even if this approach seems to
work better than the approximations mentioned above, it suffers from approximation
errors for large dividends just like approximation 3.
Lattice models: An alternative to the escrowed dividend approximation is to use non-
recombining lattice methods (see for instance Hull, 2000). If implemented as a binomial
tree one builds a new tree from each node on each dividend payment date. A problem with
all non-recombining lattices is that they are time consuming to evaluate. This problem is
amplified with multiple dividends. Schroder (1988) describes how to implement discrete
dividends in a recombining tree. The approach is based on the escrowed dividend process
idea, however, and the method will therefore significantly misprice options. Wilmott et al.
(1993, p. 402) indicate what seems to be a sounder approach to ensure a recombining tree
for the spot price process with a discrete dividend.
Problems and weaknesses of current approaches
In fact, the admission of arbitrage is merely the most egregious example among problems
or weaknesses in current approaches. Of course, we cannot claim to have seen every paper
written on this subject, but of those we have seen, we note one or more of the following
(i). Logical flaws. Many approaches use the idea of an escrowed dividend process, as
discussed above. The idea is to break up the stock price process into two pieces, a
risky part and an escrowed dividend part. The admirable goal is to “guarantee” that
the declared dividend will be received. The logical flaw is that the resulting stock price
process changes with the option expiration. Whatever the stock price process is, it
cannot depend upon which option you happen to be considering. The fact that this is
a logical fallacy is, to our thinking, under-appreciated.
(ii). Ill-defined stock price processes. Some treatments just don’t “get it” that a constant
dividend Dcan’t be paid at arbitrarily low stock prices. This tends to be a problem
with discussions of “non-recombining trees”. The problem is (a) the discussion “never”
mention that something must be said about this issue.2A related issue, which we have
seen in some recent models is that (b) negative stock prices are explicitly allowed.
The reason that authors can get away with (a) is that, in many computations, the
problematic region is a very low probability event. But, this is no excuse for failing
to completely define your model. The reader should be able to say similar bad things
about (b).
(iii). Only geometric Brownian motion is discussed. This is understandable in the early
literature, but not now-days. The weaknesses of GBM are well-know. Some fixes
include stochastic volatility, jumps, and other ideas. A dividend treatment needs to
accommodate these models.
(iv). Arbitrage issues. This has been mentioned above and an example is given below. All
2Except, to our knowledge (which, unlike the universe, is limited) Wilmott et al. (1993, p. 399), who
mention this problem and suggest to let the company go bankrupt if the dividend is larger than the asset
price. This approach avoids negative stock prices. Moreover, McDonald (2003, p. 352) points out this
problem, and suggests using the approach of Schroder (1988). As we have already pointed out, this method
has other flaws.
we will add here is that the source of this problem is (i) or (ii) above. Finally, to
paraphrase the physicist Sidney Coleman, who was speaking in another context, just
because a theory is dead wrong, doesn’t mean it can’t be highly accurate. In other
words, the reader will observe below many instances of highly accurate numerical
agreement between our current approach and some existing approximations that suffer
from (ii) for example. Nevertheless, we would argue that a theory that predicts negative
stock prices must be ‘dead wrong’ in Coleman’s sense.3
Arbitrage example: In the case of European options, the above techniques are ad hoc,
but get the job done (in most cases) when the corrections are properly carried out. To
give you an idea of when it really goes wrong, consider the model of choice for American
call options on stocks whose cum dividend price is a GBM. The Roll-Geske-Whaley (RGW)
model has for decades been considered a brilliant closed form solution to price American
calls on dividend paying stocks. Consider the case of an initial stock price of 100, strike 130,
risk free rate 6%, volatility 30%, one year to maturity, and an expected dividend payment of
7 in 0.9999 years. Using this input the RGW model posits a value of 4.3007. Consider now
another option, expiring just before the dividend payment, say in 0.9998 years. Since this in
effect is an American call on a non-dividend paying stock it is not optimal to exercise it before
maturity. In the absence of arbitrage the value must therefore equal the BSM price of 4.9183.
This is, however, an arbitrage opportunity! The arbitrage occurs because the RGW model
is mis-specified, in that the dynamics of the stock price process depends on the timing of the
dividend. Similar examples have been discussed by Beneder and Vorst (2001) and Frishling
(2002). This is not just an esoteric example, as several well known software systems use
the RGW model and other similar mis-specified models. In more complex situations than
described in this example the arbitrage will not necessarily be quite so obvious, and one
3Coleman (1985), speaking of Fermi’s theory of the ‘Weak force’: Phrased another way, the Fermi theory
is obviously dead wrong because it predicts infinite higher-order corrections, but it is experimentally near
perfect, because there are few experiments for which lowest-order Fermi theory is inadequate.
would need an accurate model to confidently take advantage of it. It is precisely such a
model we present in the present paper.
2 General Solution
A single dividend
You wish to value a Euro-style or American-style equity option on a stock that pays a
discrete (point-in-time) dividend at time t=tD. The simpler problem is to first specify a
price process whereby any dividends are reinvested immediately back into the security—this
is the so-called cum-dividend process St=S(t). In general, Stis not the market price of the
security, but instead is the market price of a hypothetical mutual fund that only invests in
the security. To distinguish the concepts, we will write the market price of the security at
time tas Yt, which we will sometimes call the ex-dividend process. Of course, if there are
no dividends, then Yt=Stfor all t. Even if the company pays a dividend, we can always
arrange things so that Y0=S0, which guarantees (by the law of one price) that Yt=Stfor
all t < tD.
In our treatment, we allow Stto follow a very general continuous-time stochastic process.
For example, your process might be one of the following (to keep things simple, we suppose
a world with a constant interest rate r):
Example (cum-dividend) processes
(P1). GBM: dSt=rStdt +σStdBt, where σis a constant volatility and Bis a standard
Brownian motion.
(P2). Jump-diffusion: dSt= (rλk)Stdt +σStdBt+StdJt, where dJtis a Poisson driven
jump process with mean jump arrival rate λ, and mean jump size k.
(P3). Jump-diffusion with stochastic volatility: dSt= (rλk)Stdt+σtStdBt+StdJt, where
σtfollows its own separate, possibly correlated, diffusion or jump-diffusion.
Consider an option at time t, expiring at time T, and assume for a moment that there are no
dividends so that Yt=Stfor all tT. In that case, clearly, models (P1) and (P2) are one-
factor models: the option value V(St, t) depends only upon the current state of one random
variable. Model (P3) is a two-factor model, V(St, σt, t). Obviously, “n-factor” models are
possible in principle, for arbitrary n, and our treatment will apply to those too. Note that
we leave the dependence on many parameters implicit.
What the examples have in common is that the stock price, with zero or more additional
factors, jointly form a Markov process,4in which the discounted stock price is a martin-
gale. Also, for simplicity, we will consider only time-homogenous processes; this means that
transition densities for the state variables depend only on the length of the time period in
question, and not on the beginning and end dates of the time period.
Choosing a dividend policy
Now we want to create an option formula for the case where the company declares a single
discrete dividend of size D, where the “ex-dividend date” is at time tDduring the option
holding period. We consider an unprotected Euro-style option, so that the option holder will
not receive the dividend. Since option prices depend upon the market price of the security,
for one factor models, we must now write V(Yt, t).
Note that we are being very careful with our choice of words; we have said that the company
4The Markov assumption is for simplicity. Perhaps it can be relaxed. We leave this issue open.
“declares” a dividend D. What we mean by that is that it is the company’s stated intention
to pay the amount Dif that is possible. When will it be impossible? We assume that the
company cannot pay out more equity than exists. For simplicity, we imagine a world where
there are no distortions from taxes or other frictions, so that a dollar of dividends is valued
the same as a dollar of equity. In addition, we always assume that there are no arbitrage
opportunities. In such a world, if the company pays a dividend D, the stock price at the
ex-dividend date must drop by the same amount: Y(tD) = Y(t
D)D. Our
notation is that t
Dis the time instantaneously before the ex-dividend date tD(in a world
of continuous trading). Since stock prices represent the price of a limited liability security,
we must have Y(tD)0, so we have a contradiction between these last two concepts if
D)< D.
This is the fundamental contradiction that every discrete dividend model must resolve. We
resolve it here by the following minimal modification to the dividend policy. We assume
that the company will indeed pay out its declared amount Dif S> D, abbreviating
D). However, in the case where S< D, we assume that the company pays some
lesser amount ∆(S) whereby 0 ∆(S)S. That is our general model, a “minimally
modified dividend policy.” In later sections, we show numerical results for two natural policy
choices, namely ∆(S) = S(liquidator), and ∆(S) = 0 (survivor). The first case allows
liquidation because the ex-dividend stock price (at least in all of the sample models P1–P3
above) would be absorbed at zero. We will assume that a zero stock price is always an
absorbing state. The second case (and, indeed any model where ∆(S)< S) allows survival
because the stock price process can then attain strictly positive values after the dividend
These choices, liquidation versus survival, sound dramatically different. In cases of financial
distress, where indeed the stock price is very low, they would be. But such cases are relatively
rare. As a practical matter, we want to stress that for most applications, the choice of
∆(S) for S < D has a negligible financial effect; the main point is that some choice must
be made to fully specify the model. There is little financial effect in most applications
because the probability that an initial stock price S0becomes as small as a declared dividend
Dis typically negligible; if this is not obvious, then a short computation with the log-
normal distribution should convince you. In any event, we will demonstrate various cases in
numerical examples.
To re-state what we have said in terms of an SDE for the security price process, our general
model is that the actual dividend paid becomes the random variable D(S), where
D(S) =
D, if S > D
∆(S)S, if SD
In (1) Dis the declared (or projected) dividend—a constant, independent of S. The func-
tional form for D(S) is any function that preserves limited liability. Then, the market price
of the security evolves, using GBM as the prototype, as the SDE:
D)idt +σYtdBt,(P1a)
where δ(ttD) is Dirac’s delta function centered at tD. The same SDE drift modification
occurs for (P2), (P3), or any other security price process you wish to model.
It’s worth stressing that the Brownian motion Btthat appears in (P1) and (P1a) have
identical realizations. You might want to picture a realization of Btfor 0 tT. Your
mental picture will ensure that Yt=Stfor all t < tDand YtD=StDD(StD). Note that Yt
is completely determined by knowledge of Stalone for all ttD(the fact that YtD=f(StD),
where fis a deterministic function, will be crucial later). What about t>tD? For those
(post ex-dividend date) times, little can be said about Ytgiven only knowledge of St(all you
can say is that Yt< Stif D > 0).
For our results to be useful, you need to be able to solve your model in the absence of
dividends. By that, we mean that you know how to find the option values and the transition
density for the stock price (and any other state variables) to evolve. Note that you need not
have these functions in so-called “closed-form”, but merely that you have some method of
obtaining them. This method may be an analytic formula, a lattice method, a Monte Carlo
procedure, a series solution, or whatever.
It’s awkward to keep placeholders for arbitrary state variables, so we will simply write
φ(S0, St, t) (the cum-dividend transition density), with the understanding that additional
state variable arguments should be inserted if your model needs them. To be explicit, the
transition density is the probability density for an initial state (stock price plus other state
variables) S0to evolve to the final state Stin a time t. This evolution occurs under the
risk-adjusted, cum-dividend process (or measure) such as the ones given under “Example
(cum-dividend) processes” above. For GBM, φ(S0, St, t) is the familiar log-normal density.
Option formulas are similarly displayed only for the one-factor model, with additional state
variables to be inserted by the reader if necessary.
With this discussion, let’s collect all of our stated assumptions in one place:
(A1) Markets are perfect (frictionless, arbitrage-free), and trading is in continuous-time.
(A2) After risk adjustment, every cum-dividend stock price St, jointly with n0 additional
factors (which are suppressed), evolve under a time-homogenous, (Markov) stochastic
process. Stis non-negative; if St= 0 is reached, it’s a trap state (absorbing). All these
statements also apply to the market price (ex-dividend process) Yt.
(A3) If a company declares (or you project) a discrete dividend D, this is promoted to a
random dividend policy D(S) in the minimal manner of (1). This causes the market
price of the stock to drop instantaneously on an ex-dividend date, as prototyped by
Our Main Result
We write VE(St, t;D, tD) for the time-tfair value of a European-style option that expires at
time T, in the presence of a discrete dividend Dpaid at time tD. The last two arguments
are the main parameters in the fully specified dividend policy {tD,D(S)}where t < tD< T .
If there is no dividend between time tand the option expiration T, we simply drop the last
two arguments and write VE(St, t). So, to be clear about notation, when you see an option
value V(·) that has only two arguments, this will be a formula that you know in the absence
of dividends, like the BSM formula. Again, the strike price X, option expiration T, and
other parameters and state variables have been suppressed for simplicity. Then, here is our
main result:
Proposition 1. Under assumptions (A1)(A3), the adoption by the company of a single
discrete dividend policy {tD,D(S)}, causes the fair value of a Euro-style option to change
from VE(S0,0) to VE(S0,0; D, tD), where
VE(S0,0; D, tD) = ertDZ
VE(S− D(S), tD)φ(S0, S, tD) dS. (2)
Proof. A very elaborate argument is:
(i) Let Sbe the cum-dividend process, and Ythe ex-dividend process as before. Assume
dividend policy D(S), paid on date tD. Then Yt=Stfor all t<tD,YtD=StD− D(StD),
and typically Yt6=Stfor t > tD. Assume the distribution function FSfor Sand that the
option pays off g(YT) at time T.
(ii) Relative to time t<T the payoff from the option is a random variable/uncertain cash
flow. The absence of arbitrage then implies that the price at time tfor this cash flow is
Let’s call this value V(Yt, t), which again is a random variable relative to any time t0< t.
For any ttDthis random variable is the price of an option on a non-dividend paying stock,
and therefore assumed known for any value of Yt.
(iii) We’re interested in the value of the option on Yat time 0. Since V(Yt, t) is simply a
random variable relative to today (time 0), we can use the exact same argument as in (ii)
above. Its value at time 0 must therefore, in the absence of arbitrage, be
ertE0{V(Yt, t)}.(3)
So far, the only result we’ve used is the (almost) equivalence of no arbitrage and the martin-
gale property of prices (Harrison and Kreps, 1979; Harrison and Pliska, 1981; Delbaen and
Schachermayer, 1995, among others), and we haven’t said anything about what specific date
ttDis (the argument would hold for t < tDtoo, but we want V(Yt, t) to be “known”).
(iv) Assuming sufficient regularity, we can write (3) in integral form wrt. a distribution
E0{V(Yt, t)}=Z
V(y, t)dFY(y;Y0,0, t).
The trouble with this integral is that we typically do not know the distribution function FY
unless t < tD—but in that case Vis unknown (orthogonal knowledge ;-).
(v) The main insight is now that by considering t=tDwe know that YtD=StD− D(StD).
This means that we do not need to know FY, since by the “Law of the Unconscious Statis-
E0{V(YtD, tD)}=Z
V(y, tD)dFY(y;Y0,0, tD)
V(s− D(s), tD)dFS(s;S0,0, tD).
In other words, at time t=tDwe know both how to compute Vand how to compute its
expectation E0{V(YtD, tD)}, and this is the only date for which this holds. With a time-
homogeneous transition density dFS(s;S0,0, tD) = φ(S0, s, tD)ds, and we arrive at (2).
An Example: Take GBM, where the dividend policy is ∆(S) = S(liquidator) for SD.
Then (2) for a call option becomes
CE(S0,0; D, tD) = ertDZ
CE(SD, tD)φ(S0, S, tD) dS. (4)
Note that the call price in the integrand of (2) is zero for S− D(S) = 0 (SD). In (4),
φ(S0, S, t) is simply the (no-dividend) log-normal density and CE(SD, tD) is simply the
no-dividend BSM formula with time-to-go TtD. For example, suppose S0=X= 100,
T= 1 (year), r= 0.06, σ= 0.30, and D= 7. Then, consider two cases; (i) tD= 0.01,
and (ii) tD= 0.99. We find from (4) the high precision results: (i) CE(100,0; 7,0.01) =
10.59143873835989 and (ii) CE(100,0; 7,0.99) = 11.57961536099359.
American-style options
It is well-known, and easily proved that, for an American-style call option with a discrete div-
idend, early exercise is only optimal instantaneously prior to the ex-dividend date (Merton,
1973). This result of course applies to the present model. Hence, to value an American-style
call option with a single discrete dividend, you merely replace (2) with
CA(S0,0; D, tD) = ertDZ
max{(SX)+, CE(S− D(S), tD)}φ(S0, S, tD) dS, (5)
Early exercise is never optimal unless there is a finite solution Sto SX=CE(SD, tD),
where we are assuming that X > D (a virtual certainty in practice). In this case, the reader
may want to break up the integral into two pieces, but we shall just leave it at (5).
For American-style put options, as is also well-known, it can be optimal to exercise at any
time prior to expiration, even in the absence of dividends. So, in this case, you are generally
forced to a numerical solution, evolving the stock price according to your model. This is
the well-known backward iteration. What may differ from what you are used to is that you
must allow for an instantaneous drop of D(S) on the ex-date.
One down, n1to go
With the sequence of dividends {(Di, ti)}n
i=1,t1< t2< . . . < tn, the argument leading to
formula (2) still holds. Simply repeat it iteratively, starting at time tn1by applying (2)
to the last dividend (Dn, tn). While straight forward, this procedure involves evaluating an
n-fold integral. We therefore show a simpler way to compute it in Section 4.
3 Dividend Models
It is now time to compare specific dividend models, ∆(S). We consider the two extreme
cases for ∆(S), use these to develop an inequality for an arbitrary dividend model, and then
illustrate the impact on option prices.
Liquidator: We consider first the situation where the dividend is reduced to StDwhen
StD< D, i.e., ∆(S)=∆l(S) = Sfor S < D. This is tantamount to the firm being
liquidated if the cum dividend stock price falls below the declared dividend. Although this
might seem an extreme assumption, keep in mind that for most reasonable parameter values,
this will be close to a zero-probability event. It is a simple approximation that succeeds in
ensuring a non-negative ex-dividend stock price. In this case (2) reduces to
E(S0,0; D, tD) = ertDZ
VE(SD, tD)φ(S0, S, tD) dS
+ ertDZD
VE(Sl(S), tD)φ(S0, S, tD) dS, (6)
= ertDZ
VE(SD, tD)φ(S0, S, tD) dS+VE(0, tD)ZD
φ(S0, S, tD) dS.
In this decomposition the “tail value” ertDVE(0, tD)RD
0φ(S0, S, tD) dSwill vanish for a call
option, but not for a put option.
Survivor: Consider next the situation where the dividend is canceled when StD< D, i.e.,
∆(S) = ∆s(S) = 0 for SD. This approximation also succeeds in ensuring a non-negative
ex-dividend stock price, and also allows the firm to live on with probability one after the
dividend payout. The option price is now similar to the one for the liquidator dividend, with
a slight modification to the tail value of the option contract:
E(S0,0; D, tD) = ertDZ
VE(SD, tD)φ(S0, S, tD) dS
VE(S, tD)φ(S0, S, tD) dS.(7)
From (6) and (7) we can now establish a result that should enable you to sleep better at
night if you are concerned with your choice of dividend policy ∆(·).
Corollary 1. Let CE(S0,0; D, tD)be given by (2) for a generic dividend policy ∆(S)such
that 0∆(S)S. For European call options
E(S0,0; D, tD)CE(S0,0; D, tD)Cs
E(S0,0; D, tD).
If StD< D with positive probability then additionally Cl
E(S0,0; D, tD)< Cs
E(S0,0; D, tD).
Proof. The R
D-integral is identical for all three options. Since 0 ∆(S)Sit follows that
CE(Sl(S), tD)φ(S0, S, tD) dS
CE(S∆(S), tD)φ(S0, S, tD) dS
CE(Ss(S), tD)φ(S0, S, tD) dS.
The strict inequality follows from a similar argument.
We could easily have established the same inequality for American call options, by simply
using more cumbersome notation that takes early exercise into account. The interesting
part of the result is the weak inequality. It tells us that if there’s a negligible difference in
prices between the liquidator and survivor dividend policies, then it doesn’t matter what
assumption you make about ∆(S) as long as it satisfies limited liability.
We end this section with an illustration of the relevance of the dividend model ∆(S). We
do so by illustrating the pricing implications of the two extreme dividend policies, liquidator
and survivor, in a specific case.
A financial fairy “tail”: A long-lived financial service firm, let’s call them Ye Olde
Reliable Insurance, paid a hefty dividend once a year, which they liked to declare well in
advance. Once declared, they had never missed a payment, not once in their 103-year history.
As their usual practice, they went ex-dividend every June 30 and declared their next dividend
in November. One November, with their stock approaching the $100 mark, the Ye Olde board
decided that 6% seemed fair and easily do-able, so they declared a $6 dividend for the next
June 30.
Unfortunately, during the very next month (December), the outbreak of a mysterious new
virus coupled with an 8.2 temblor centered in Newport Beach devastated both their prop-
erty/casualty and health insurance subs and their stock plummeted to the $10 range.
The CBOE dutifully opened a new option series striking at $10 with a leaps version expiring
one year later. To keep things simple, we will imagine this series expires exactly in one
year with an ex-dividend date at exactly the 1/2 year mark. Of course, there was much
speculation and uncertainty about the declared $6 dividend. The company’s only comment
was a terse press release saying “the board has spoken.”
So, the potential option buyer was faced with a contract with S= 10, X= 10, T= 1 year
and tD= 0.5. If they paid in full, then D= 6. Interest rates were at 6%.
Our “liquidator” option model postulates that the company would pay in full unless the
stock price StDat mid-year was below D= 6, in which case they would pay out all the
remaining equity, namely StD. The skeptics said “no way” and proposed a new “survivor”
option model in which the board would completely drop the dividend if StDD. The call
price in this new model was larger by a “tail value”. The big debate in the Wilmott forums
became what volatility should one use to compute these values and, in the end, of course,
no one knew. But everyone could agree that the stock price sure was volatile. So one way
to proceed was to compute the option price in both models for volatilities ranging from 80%
to 150% and the results are shown in the figure. Interestingly, the results only differed by
Figure 1: Relative tail value for a European call option
80 90 100 110 120 130 140 150
Difference in Option Values:
Survivor -Liquidator, as a Percent of Liquidator
Annualized Stock Price Volatility, Percent
3.5 to 7% even in this extreme scenario with a doubtful yield, if paid, in the 60% range.
4 Applications
To illustrate the application of the pricing formula we now specialize the option contracts as
well as the stock price process.
European call and put options: It is straight forward to derive the following put-call
Proposition 2. For a general cum-dividend price process Stand dividend policy D(S)as in
CE(S0,0; D, tD)+erT X+ ertD¯
D=PE(S0,0; D, tD) + S0,(8)
φ(S0, S, tD)(D∆(S)) dS
is the expected received dividend.
Proof. The idea of the argument is standard (Merton, 1973): Since CE(Y , T ;D, tD) + X=
(YX)++X=PE(Y, T , D, tD) + Y, these two portfolios must have the same value
today. Consider first the LHS. Its value is erT E0{(YX)++X}=CE(S0,0; D, tD) +
erT X. Consider next the RHS. Its value is similarly given by erT E0{(XY)++Y}=
PE(S0,0; D, tD)+erT E0(Y+¯
DT=PE(S0,0; D, tD) + S0+ erT E0¯
DT, where
DTis the future value (at time T) of the dividend received (a random variable). Since
DT= er(TtD)ZD
∆(S)φ(S0, S, tD) dS+ er(TtD)DZ
φ(S0, S, tD) dS
the result follows.
For the case of GBM stock price and liquidator dividend, ∆(S) = Sfor S < D, the value of
a European call option can be written explicitly as
CE(S0,0; D, tD) = ertDZ
CES0e[rσ2/2]tD+σtDxD, tD1
2x2dx, (9)
d=ln (D/S0)(rσ2/2) tD
A similar expression can be written down for the put option, but this is really not necessary
in light of (8). Tables 1 and 2 report option prices for European call options for small and
large dividends. The tables use the symbols:
BSM is the plain vanilla Black-Scholes-Merton model.
M73 is the BSM model with SertDDsubstituted for S—the escrowed dividend adjust-
ment (Merton, 1973).
Vol1 is identical to M73, but with an adjusted volatility. The volatility of the asset is
replaced with σ2=σS
SertDD(see for instance Chriss, 1997).
Vol2 is a slightly more sophisticated volatility adjustment than Vol1 (see Appendix A for
a short description of this technique).
Vol3 is the volatility adjustment suggested by Bos et al. (2003) (see Appendix B for a short
description of this adjustment).
BV adjusts the strike and stock price, to take into account the effects of the discrete dividend
payment (Bos and Vandermark, 2002).
Num is a non-recombining binomial tree with 500 time steps, and no adjustment to prevent
the event that SD < 0 (see for instance Hull, 2000, for the idea behind this method).
HHL(4) is the exact solution in (4).
Table 1: European calls with dividend of 7
(S= 100, T= 1, r= 6%, σ= 30%)
BSM Mer73 Vol1 Vol2 Vol3 BV Num HHL(4)
t X = 100
0.0001 14.7171 10.5805 11.4128 10.5806 10.5806 10.5806 10.5829 10.5806
0.5000 14.7171 10.6932 11.5001 11.1039 11.0781 11.0979 11.1079 11.1062
0.9999 14.7171 10.8031 11.5855 11.5854 11.5383 11.5887 11.5704 11.5887
X= 130
0.0001 4.9196 3.0976 3.7403 3.0977 3.0977 3.0977 3.0987 3.0977
0.5000 4.9196 3.1437 3.7701 3.4583 3.4203 3.4159 3.4368 3.4383
0.9999 4.9196 3.1889 3.7993 3.7993 3.6949 3.7263 3.7140 3.7263
X= 70
0.0001 34.9844 28.5332 28.9113 28.5332 28.5332 28.5332 28.5343 28.5332
0.5000 34.9844 28.7200 29.0832 28.9009 28.9047 28.9350 28.9218 28.9215
0.9999 34.9844 28.9016 29.2504 29.2504 29.2920 29.3257 29.3140 29.3257
Table 1 illustrates that the M73 adjustment is inaccurate, especially in the case when the
dividend is paid close to the option’s expiration. Moreover the Vol1 adjustment, often
used by practitioners, gives significantly inaccurate values when the dividend is close to
the beginning of the option’s lifetime. Both Vol2 and BV do much better at accurately
pricing the options. Vol3 yields values very close to the BV model. The non-recombining
tree (Num) and our exact solution (HHL(4)) give very similar values in all cases. However,
the non-recombining tree is not ensured to converge to the true solution (HHL(4)) in all
situations, unless the non-recombining tree is set up to prevent negative stock prices in the
nodes where SD < 0. This problem will typically be relevant only with a very high
divided, as we discussed in Section 3. For low to moderate cash dividends one can assume
that even the “naive” non-recombining tree and our exact solution agree to economically
significant accuracy.
Table 2: European calls with dividend of 50
(S= 100, T= 1, r= 6%, σ= 30%)
BSM Mer73 Vol1 Vol2 Vol3 BV Num HHL(4)
t X = 100
0.0001 14.7171 0.1282 2.9961 0.1283 0.1282 0.1283 0.1273 0.1283
0.5000 14.7171 0.1696 3.0678 1.4323 0.5755 0.8444 1.0687 1.0704
0.9999 14.7171 0.2192 3.1472 3.1469 1.1566 2.1907 2.1825 2.1908
X= 130
0.0001 4.9196 0.0094 1.3547 0.0094 0.0094 0.0094 0.0092 0.0094
0.5000 4.9196 0.0133 1.3556 0.4313 0.0947 0.1516 0.2264 0.2279
0.9999 4.9196 0.0184 1.3609 1.3607 0.2510 0.6120 0.6072 0.6120
X= 70
0.0001 34.9844 1.6510 7.0798 1.6517 1.6513 1.6514 1.6515 1.6517
0.5000 34.9844 1.9982 7.3874 4.9953 3.3697 4.2808 4.7304 4.7299
0.9999 34.9844 2.3780 7.7100 7.7096 4.9966 7.2247 7.2122 7.2248
Table 2 shows that the BV and the non-recombining tree have significant differences when
there’s a significant dividend in the middle of the option’s lifetime. The latter is closer to
the true value. The Vol3 model strongly underprices the option when the dividend is this
American call and put options: Most traded stock options are American. We now do
a numerical comparison of stock options with a single cash dividend payment. Tables 3–5
use the following models that differ from the European options considered above:
B75 is the approximation to the value of an American call on a dividend paying stock
suggested by Black (1975). This is basically the escrowed dividend method, where the
stock price in the BSM formula is replaced with the stock price minus the present value
of the dividend. To take into account the possibility of early exercise one also compute
an option value just before the dividend payment, without subtracting the dividend.
The value of the option is considered to be the maximum of these values.
RGW is the model of Roll (1977); Geske (1979); Whaley (1981). It is considered a closed
form solution for American call options on dividend paying stocks. As we already
know, the model is seriously flawed.
HHL(5) it the exact solution in (5), again using the liquidator policy.
Table 3: American calls with dividend of 7
(D= 7, S = 100, T= 1, r= 6%, σ= 30%)
B75 RGW Num HHL(5)
t X = 100
0.0001 10.5805 10.5805 10.5829 10.5806
0.5000 10.6932 11.1971 11.6601 11.6564
0.9999 14.7162 13.9468 14.7053 14.7162
X= 130
0.0001 3.0976 3.0976 3.0987 3.0977
0.5000 3.1437 3.1586 3.4578 3.4595
0.9999 4.9189 4.3007 4.9071 4.9189
X= 70
0.0001 30.0004 30.0004 30.0000 30.0004
0.5000 32.3034 32.3365 32.4604 32.4608
0.9999 34.9839 34.7065 34.9737 34.9839
Table 3 shows that the RGW model works reasonably well when the divided is in the very
beginning of the option lifetime. The RGW model exhibits the same problems as the sim-
pler M73 or escrowed dividend method used for European options. The pricing error is
particularly large when the dividend occurs at the end of the option’s lifetime. The B75
approximation also significantly misprices options.
Table 4: American calls with dividend of 30
(D= 30, S = 100, T= 1, r= 6%, σ= 30%)
B75 RGW Num HHL(5)
t X = 100
0.0001 2.0579 2.0579 2.0574 2.0583
0.5000 9.8827 7.5202 9.9296 9.9283
0.9999 14.7162 11.4406 14.7053 14.7162
X= 130
0.0001 0.3345 0.3345 0.3322 0.3346
0.5000 1.6439 0.6742 1.7851 1.7855
0.9999 4.9189 2.4289 4.9071 4.9189
X= 70
0.0001 30.0004 30.0004 30.0000 30.0004
0.5000 32.3034 32.0762 32.3033 32.3037
0.9999 34.9839 34.1637 34.9737 34.9839
Table 5: American calls with dividend of 50
(D= 50, S = 100, T= 1, r= 6%, σ= 30%)
B75 RGW Num HHL(5)
t X = 100
0.0001 0.1282 0.1437 0.1273 0.1922
0.5000 9.8827 5.8639 9.8745 9.8828
0.9999 14.7162 9.3137 14.7053 14.7162
X= 130
0.0001 0.0094 0.0094 0.0092 0.0094
0.5000 1.6439 0.1375 0.5112 1.6492
0.9999 4.9189 1.1029 4.9071 4.9189
X= 70
0.0001 30.0004 30.0004 30.0000 30.0004
0.5000 32.3034 32.0762 32.6600 32.3034
0.9999 34.9839 34.1637 34.9737 34.9839
For very high dividend, as in Table 5, the mispricing in the RGW formula is even more
clear; the values are significantly off compared with both non-recombining tree (Num) and
our exact solution (HHL(5)). The simple B75 approximation is remarkably accurate. The
intuition behind this is naturally that a very high dividend makes it very likely to be optimal
to exercise just before the dividend date—a situation where the B75 approximation for good
reasons should be accurate.
Multiple dividend approximation
We showed in Section 2 that it is necessary to evaluate an n-fold integral when there are
multiple dividends. It is therefore useful to have a fast, accurate approximation. We now
show how to approximate the option value in the case of a call option on a stock whose
cum-dividend price follows a GBM, using the liquidator dividend policy.
First, let’s write the exact answer on date twith a sequence of ndividends prior to Tas
Cn(S, X, t, T ), where Xis the strike and Tis the expiration date. Then, the first iteration
of (4) in an exact treatment becomes
C1(S, X, tn1, T ) = er(tntn1)Z
CBSM(S1Dn, X, tn, T )φ(S, S1, tntn1) dS1,(10)
where CBSM(·) is the BSM model. This integral is quick to evaluate, just as in the single
dividend cases tabulated above. The second iteration becomes
C2(S, X, tn2, T ) = er(tn1tn2)Z
C1(S1Dn1, X, tn1, T )φ(S, S1, tn1tn2) dS1.
Notice that we now integrate not over the BSM model, but rather the option price derived
in the first iteration (10). Evaluation of (11) therefore involves a double integral. We
know, however, that C1(·) will look like an option solution and hence will have many of
the characteristics of the BSM formula. If we can effectively parametrize C1(·) with a BSM
formula then it will be quick to evaluate (11).
Some key characteristics of C1(S, X, tn1, T ) are as follows. First, it vanishes as S0.
Second, because (standard) put-call parity becomes asymptotically exact for large S,
C1(S, X, tn1, T )Ser(Ttn1)Xer(tntn1)Dn.
This suggests the BSM parametrization
C1(S, X, tn1, T )CBSM (S, Xadj, tn1, T ),(12)
where Xadj =X+Dner(tnT). The strike adjustment ensures correct large-Sbehavior.
A little experimentation will show that the approximating BSM formula just suggested is
inaccurate for Snear the money. Still, we have another degree of freedom in our ability to ad-
just the volatility in the right-hand-side of (12). By choosing σadj so that C1(S0, X, tn1, T )
CBSM(S0, Xadj, σadj, tn1, T ), where S0is the original stock price of the problem, we obtain
an accurate approximation
C1(S, X, tn1, T )CBSM (S, Xadj, σadj, tn1, T )
that often differs by less than a penny over the full range of Son (0,).
This same scheme is used at successive iterations of the exact integration. That is, the
“previous” iteration will always be fast because it uses the BSM formula. Then, after you
get the answer, you approximate that answer by a BSM formula parameterization. In that
parameterization, you choose an adjusted strike price and an adjusted volatility to fit the
large-Sbehavior and the S0value. This enables you to move on to the next iteration.
Table 6 reports call option values when there is a dividend payment of 4 in the middle of
each year. The first column shows the years to expiration for the contracts we consider. The
models Vol2, Vol3, BV, and Num are identical to the ones described earlier. HHL is our
closed form solution from Section 2 evaluated by numerical quadrature. As we have already
mentioned, this approach is computer intensive. We have therefore limited ourself to value
options with this method with up to three dividend payments. An efficient implementation
in for instance C++ will naturally make this approach viable for any practical number of
dividend payments. Non-recombining trees are even more computer intensive, especially for
multiple dividends. They also entail problems with propagation of errors when the number
of time steps is increased, so we limited ourself to compute option values for three dividends
(3 years to maturity), with 500 time steps for T= 1,2, and 1000 time steps for T= 3. The
column Appr is the approximation just described above. The two rightmost columns report
the adjusted strike and volatility used in this approximation method.
Table 6: European calls with multiple dividends of 4
(S= 100, X= 100, r= 6%, σ= 25%, D= 4)
TNum Vol2 Vol3 BV HHL Appr Adjusted Adjusted
strike volatility
1 10.6615 10.6585 10.6530 10.6596 10.6606 10.6606 104.122 0.2467
2 15.2024 15.1780 15.1673 15.1992 15.1989 15.1996 108.499 0.2421
3 18.5798 18.5348 18.5241 18.5981 18.5984 18.5998 113.146 0.2375
4 – 21.2297 21.2304 21.3592 – 21.3644 118.081 0.2328
5 – 23.4666 23.4941 23.6868 – 23.6978 123.320 0.2282
6 – 23.3556 25.4279 25.6907 – 25.7100 128.884 0.2237
7 – 26.9661 27.1023 27.4395 – 27.4695
The approximation we suggest above (Appr) is clearly very accurate, when compared to
our exact integration (HHL). Also the non-recombining binomial implementation (Num) of
the spot process yields results very close to our exact integration. Vol2 and Vol3 seems to
give rise to significant mispricing with multiple dividends. The BV approximation seems
somewhat more accurate. However, as we already know, it significantly misprices options
when the dividend is very high. From a trader’s perspective, our approach seems to be a
clear choice—at least if you care about having a robust and accurate model that will work
in “any” situation. Remember also that our method is valid for any price process, including
stochastic volatility, jumps, and other factors that can have a significant impact on pricing
and hedging.
Exotic and real options: Several exotic options trade in the OTC equity market, and
many are embedded in warrants and other complex equity derivatives. The exact model
treatment of options on dividend paying stocks presented in this paper holds also in these
cases. Many exotic options, in particular barrier options, are known to be very sensitive to
stochastic volatility. Luckily the model described above also holds for stochastic volatility,
jumps, volatility term structure, as well as other factors that can be of vital importance
when pricing exotic options. The model we have suggested should also be relevant to real
options pricing, when the underlying asset offers known discrete payouts (of generic nature)
during the lifetime of the real option.
Appendix A
The following is a volatility adjustment that has been suggested used in combination with
the escrowed dividend model. The adjustment seems to have been discovered independently
by Haug and Haug (1998) (unpublished working paper), as well as by Beneder and Vorst
(2001). σin the BSM formula is replaced with σadj, and the stock price minus the present
value of the dividends until expiration is substituted for the stock price.
adj =
i=1 Dierti2
(t1t0) +
i=2 Dierti2
(t2t1) + · ·· +σ2(Ttn)
(tjtj1) + σ2(Ttn)
This method seems to work better than for instance the volatility adjustment discussed by
Chriss (1997), among others. However this is still simply a rough approximation, without
much of a theory behind it. For this reason, there is no guarantee for it to be accurate in all
circumstances. Any such model could be dangerous for a trader to use.
Appendix B
Bos et al. (2003) suggest the following volatility adjustment to be used in combination with
the escrowed dividend adjustment:
σ(S, X, T )2=σ2+σrπ
+ ez2
DiDjer(ti+tj)N(z2)Nz22σmin(ti, tj)
where nis the number of dividends in the option’s lifetime, s= ln(S), x= ln[(X+DT)erT ],
where DT=Pn
iDierti, and
2, z2=z1+σT
Beneder, Reimer and Ton Vorst, “Options on Dividends Paying Stocks,” in “Proceedings
of the International Conference on Mathematical Finance” Shanghai 2001.
Black, Fischer, “Fact and Fantasy In the Use of Options,” Financial Analysts Journal,
July–August 1975, pp. 36–72.
and Myron Scholes, “The Pricing of Options and Corporate Liabilities,” Journal of
Political Economy, May–June 1973, 81, 637–654.
Bos, Remco, Alexander Gairat, and Anna Shepeleva, “Dealing with Discrete Divi-
dends,” Risk, January 2003, 16 (1), 109–112.
and Stephen Vandermark, “Finesssing Fixed Dividends,” Risk, September 2002, 15
(9), 157–158.
Chriss, Neil A.,Black-Scholes and Beyond: Option Pricing Models, New York, New York:
McGraw-Hill, 1997.
Coleman, Sidney,Aspects of Symmetry: Selected Erice Lectures, Cambridge University
Press, 1985.
Cox, John C. and Mark Rubinstein,Options Markets, Englewood Cliffs, New Jersey:
Prentice-Hall, 1985.
Delbaen, Freddy and Walter Schachermayer, “The Existence of Absolutely Continuous
Local Martingale Measures,” Annals of Applied Probability, 1995, 5(4), 926–945.
Frishling, Volf, “A Discrete Question,” Risk, January 2002, 15 (1).
Geske, Robert, “A Note on an Analytic Valuation Formula for Unprotected American
Call Options on Stocks with Known Dividends,” Journal of Financial Economics, 1979,
7, 375–380.
, “Comments on Whaley’s Note,” Journal of Financial Economics, 1981, 9, 213–215.
Harrison, J. Michael and David M. Kreps, “Martingales and Arbitrage in Multiperiod
Securities Markets,” Journal of Economic Theory, 1979, 20, 381–408.
Harrison, J.M. and S. Pliska, “Martingales and Stochastic Integrals in the Theory of
Continuous Trading,” Stochastic Processes and Their Applications, 1981, 11, 215–260.
Haug, Espen Gaarder,The Complete Guide to Option Pricing Formulas, New York, New
York: McGraw-Hill, 1997.
and Jørgen Haug, “Implied Forward Volatility,” March 1996. Presented at the Third
Nordic Symposium on Contingent Claims Analysis in Finance.
and , “A New Look at Pricing Options with Time Varying Volatility,” 1998. Unpub-
lished working paper.
Hull, John,Options, Futures, and Other Derivatives, fourth ed., Englewood Cliffs, New
Jersey: Prentice-Hall, 2000.
McDonald, Robert L.,Derivatives Markets, Pearson Education, 2003.
Merton, Robert C., “Theory of Rational Option Pricing,” Bell Journal of Economics and
Management Science, Spring 1973, 4, 141–183.
Roll, Richard, “An Analytical Formula for Unprotected American Call Options on Stocks
with Known Dividends,” Journal of Financial Economics, 1977, 5, 251–258.
Schroder, Mark, “Adapting the Binomial Model to Value Options on Assets with Fixed-
cash Payouts,” Financial Analysts Journal, November–December 1988, 44 (6), 54–62.
Stoll, Hans R. and Robert E. Whaley,Futures and Options: Theory and Applications
The Current Issues in Finance, Cincinnati, Ohio: South-Western, 1993.
Whaley, Robert E., “On the Valuation of American Call Options on Stocks with Known
Dividends,” Journal of Financial Economics, 1981, 9, 207–211.
Wilmott, Paul,Paul Wilmott on Quantitative Finance, John Wiley & Sons, 2000.
, Jeff Dewynne, and Sam Howison,Option Pricing: Mathematical Models and Com-
putation, Oxford: Oxford Financial Press, 1993.
In diesem Kapitel diskutieren wir einige Erweiterungen der prototypischen Routine des Kapitel 5. Wir lassen zunächst die homogenen Dirichlet-Randbedingungen fallen und diskutieren Probleme mit inhomogenen Dirichlet- oder Neumann-Randbedingungen sowie Differentialgleichungen, die keine Randbedingungen benötigen. Dies führt auf eine verallgemeinerte Version der Python-Routine ``matrixgenerator’’ des Kapitels 4. Dann betrachten wir rekursive Probleme, das heisst Bewertungsprobleme, für welche wir nacheinander n partielle Differentialgleichungen lösen müssen. Zum Beispiel gibt n die Anzahl der Dividendenzahlungen oder die Anzahl der Barrierebeobachtungen während der Laufzeit eines Derivats an. Weiter betrachten wir Differentialgleichungen, deren Koeffizienten nebst des Basiswerts auch noch von der Zeit abhängen. Als Anwendungen hierzu bewerten wir Asiatische Optionen im Black-Scholes Modell und Barriere Optionen im verallgemeinerten CEV Modell. Dann zeigen wir, dass man nicht nur den Preis einer Option, sondern auch deren Griechen als Lösung einer partiellen Differentialgleichung auffassen kann. Als Anwendung davon implementieren wir die Kalibrierung des im Kapitel 1 vorgestellten CEV Modells.
Wir betrachten in diesem Kapitel nochmals Amerikanische Optionen, allerdings erfolgt deren Bewertung nun nicht mehr wie im Kapitel 2 mit Hilfe von Binomialbäumen, sondern via sogenannten linearen Komplementaritätsproblemen (LKP). Ein LKP ist - salopp ausgedrückt - eine partielle Differentialungleichung. Eine Anwendung des Finite-Differenzen-Verfahrens des Kapitel 5 führt auf eine Sequenz von Matrix-LKP. Zur Lösung dieser wenden wir das Newton-Verfahren an; dies führt zum sogenannten Primal-Dual-Active Set Algorithmus, welcher typischerweise um Faktoren schneller ist als der in der Literatur üblicherweise diskutierte Projective-Successive-Overrelaxation Algorithmus. Mit Hilfe des sogenannten freien Randes kann der Investor entscheiden, ob er die Option vorzeitig ausüben möchte. Daher geben die Python-Routinen dieses Kapitels nicht nur den Preis einer Amerikanischen Option aus, sondern auch den freien Rand. Wie schon Europäische Optionen im Kapitel 6 bewerten wir auch Amerikanische Optionen unter Berücksichtigung diskreter Dividenden. Dies führt auf eine Sequenz von LKPs. Zum Abschluss dieses Kapitels betrachten und bewerten wir die einfacheren Bermuda Optionen.
In this chapter, we construct second-order (in both space and time) FD schemes for forward PDEs and PIDEs consistent with the corresponding FD schemes for the backward PDEs and PIDEs considered in previous chapters. In this context, consistency means that the option prices obtained by solving both the forward and backward equations coincide up to some tolerance. This approach is partly inspired by Andreasen and Huge (RISK, 66–71, 2011), whose authors reported a pair of consistent finite difference schemes of first-order approximation in time for an uncorrelated local stochastic volatility model. We extend their approach by constructing schemes that are of second order in both space and time and that apply to models with jumps and discrete dividends. Taking correlation into account in our approach is also not an issue.
In this paper we consider the fair pricing of single stock futures (SSFs) and the effect of dividend risk on the dividend compensation component in the pricing formulas. SSF valuation is subject to the pricing of discrete cash dividends (not percentages or dividend yields) in the underlying stock. Discrete cash dividends present modelling challenges which are not present when dividends are in the form of yields or percentages. Problems are created for market-makers and investors when the actual cash dividend is different from that predicted by analysts and used for pricing. We propose a new model for the fair price of a single stock futures contract which addresses dividend uncertainty.
We present an accurate numerical solution for the discrete Black-Scholes equation with only a few grid points. European and American option problems with deterministic discrete dividend modelled by a jump condition at the exdividend date are solved. Fourth order finite differences are employed, as well as a grid stretching in space and a Lagrange interpolation at the ex-dividend date.
Full-text available
We consider options written on assets which pay cash dividends. Dividend payments have an effect on the value of options: high dividends imply lower call premia and higher put premia. Recently, Haug et al. [13] derived an integral representation formula that can be considered the exact solution to problems of evaluating both European and American call options and European put options. For American-style put options, early exercise may be optimal at any time prior to expiration, even in the absence of dividends. In this case, numerical techniques, such as lattice approaches, are required. Discrete dividends produce discrete shift in the tree; as a result, the tree is no longer reconnecting beyond any dividend date. While methods based on non-recombining trees give consistent results, they are computationally expensive. In this contribution, we analyse binomial algorithms for the evaluation of options written on stocks which pay discrete dividends and perform some empirical experiments, comparing the results in terms of accuracy and speed. Key wordsoptions on stocks–discrete dividends–binomial lattices
Full-text available
We are concerned with the problem of pricing plain-vanilla and barrier options with cash dividends in a piecewise lognormal model. In the plain-vanilla case, we offer a method with provides thin upper and lower bounds of the exact binomial price. In the barrier case, we provide an efficient algorithm based on suitable interpolation techniques. As by-product, we provide a new method for pricing American barrier options with continuous dividends.
This paper deals with the construction of a numerical solution of the Black–Scholes equation modeling option pricing with a discrete dividend payment. This model is a partial differential equation with two variables: the underlying asset and the time to maturity, and involves the shifted Dirac delta function centered at the dividend payment date. This generalized function is suitable for approximation by means of sequences of ordinary functions. By applying a semidiscretization technique on the asset, a numerical solution is obtained and the independence of the considered sequence in a wide class of delta defining sequences is proved. From the study of the influence of the spatial step h, it follows that the difference between the numerical solution for h and h/2 is O(h2) as h⟶0. The proposed method is useful for different discrete dividend types like a dividend of present value D0, a constant yield dividend or an arbitrary underlying asset-dependent yield dividend payment. Several illustrative examples are included.
Full-text available
This contribution deals with options on assets which pay discrete dividends. We analyze some methodologies to extract information on dividends from observable option prices. Implied dividends can be computed using a modified version of the well known put-call parity relationship. This technique is straightforward, nevertheless, its use is limited to European options and, when dealing with equities, most traded options are of American-type. As an alternative, numerical inversion of pricing methods can be used. We apply different procedures to obtain implied dividends of stocks of the Italian Derivatives Market.
Full-text available
Author begins by deducing a set of restrictions on option pricing formulas from the assumption that investors prefer more to less. These restrictions are necessary conditions for a formula to be consistent with a rational pricing theory. Attention is given to the problems created when dividends are paid on the underlying common stock and when the terms of the option contract can be changed explicitly by a change in exercise price or implicitly by a shift in the investment or capital structure policy of the firm. Since the deduced restrictions are not sufficient to uniquely determine an option pricing formula, additional assumptions are introduced to examine and extend the seminal Black-Scholes theory of option pricing. Explicit formulas for pricing both call and put options as well as for warrants and the new ″down-and-out″ option are derived. Other results.
Part I: Institutional background. Introduction to futures and options. Functioning of futures and options markets. Part II: Futures. Structure of futures prices and the basics. Hedging with future contracts. Risk and rerun in future contracts. Agricultural futures. Stock index futures. Interest rate futures. Currency futures. Part III: Options. Structure of option prices. The put-call. Determinants of options prices. Hedging with options. Common stock options. Stock index options. Interest rate options. Currency options. Part IV: Finanicial Innovation.