Mathematical Finance, Vol. 7, No.

4 (October 1997), 325–348

A CONTINUITY CORRECTION FOR DISCRETE BARRIER OPTIONS
MARK BROADIE AND PAUL GLASSERMAN
Columbia Business School, New York
STEVEN KOU
Department of Statistics, University of Michigan

The payoff of a barrier option depends on whether or not a specified asset price, index, or rate
reaches a specified level during the life of the option. Most models for pricing barrier options assume
continuous monitoring of the barrier; under this assumption, the option can often be priced in closed
form. Many (if not most) real contracts with barrier provisions specify discrete monitoring instants;
there are essentially no formulas for pricing these options, and even numerical pricing is difficult. We
show, however, that discrete barrier options can be priced with remarkable accuracy using continuous
barrier formulas by applying a simple continuity correction to the barrier. The correction shifts the

barrier away from the underlying by a factor of exp(βσ
t), where β ≈ 0.5826, σ is the underlying
volatility, and t is the time between monitoring instants. The correction is justified both theoretically
and experimentally.
KEY WORDS: path-dependent options, Siegmund’s corrected diffusion approximation, level crossing
probabilities

1. THE MAIN RESULT
1.1. Introduction
A barrier option is activated (knocked in) or extinguished (knocked out) when a specified
asset price, index, or rate reaches a specified level. The simplest such options are otherwise
standard calls and puts that are knocked in or knocked out by the underlying asset itself.
Some variants tie the barrier crossing to one variable and the payoff to another; others
specify “binary” payoffs in place of the usual payoffs for calls and puts. Taken together,
these are among the most popular options with path-dependent payoffs. The knock-in and
knock-out features lower the price of an option, and may provide a payoff distribution that
better matches a hedger’s risk or a speculator’s view.
Most models of barrier options assume continuous monitoring of the barrier: a knock-in
or knock-out is presumed to occur if the barrier is breached at any instant in the life of the
option. Under this assumption, Merton (1973) obtained a formula for pricing a knock-out
call. Subsequent work on pricing continuously monitored barrier options includes Heynen
and Kat (1994a, 1994b), Kunitomo and Ikeda (1992), Rich (1994a, 1994b), and Rubinstein
and Reiner (1991). However, a sizable portion of real contracts with barrier features specify fixed times for monitoring of the barrier—typically, daily closings. One article in the
trade literature (Derivatives Week 1995a) faults existing pricing models for not addressing

Manuscript received September 1995; final revision received April 1996.
Address correspondence to the first two authors at Columbia Business School, 415 Uris Hall, New York,
NY 10027; e-mail: mbroadie@research.gsb.columbia.edu; pglasser@research.gsb.columbia.edu, or to the third
author at the Dept. of Statistics, University of Michigan, Ann Arbor, MI 48109; e-mail: Kou@stat.lsa.umich.edu.
c 1997 Blackwell Publishers, 350 Main St., Malden, MA 02148, USA, and 108 Cowley Road, Oxford,
OX4 1JF, UK.

325

326

M. BROADIE, P. GLASSERMAN, AND S. KOU

this feature. Another (Derivatives Week 1995b) discusses concerns that when monitoring
times are not specified, extraneous barrier breaches may occur in less liquid markets while
the major Western markets are closed. Moreover, numerical examples, including those in
Chance (1994), Flesaker (1992), and Kat and Verdonk (1995), indicate that there can be
substantial price differences between discrete and continuous barrier options, even under
daily monitoring of the barrier. Unfortunately, the exact pricing results available for continuous barriers do not extend to the discrete case.1 Even numerical methods using standard
lattice techniques or Monte Carlo simulation face significant difficulties in incorporating
discrete monitoring, as demonstrated in Broadie, Glasserman, and Kou (1996).
In this paper, we introduce a simple continuity correction for approximate pricing of
discrete barrier options. Our method uses formulas for the prices of continuous barrier
options but shifts the barrier to correct for discrete monitoring. The shift is determined
solely by the monitoring frequency, the asset volatility, and a constant β ≈ 0.5826. It
is therefore trivial to implement. Compared with using the unadjusted continuous price,


our formula reduces the error from O(1/ m) to o(1/ m), as the number of monitoring
points m increases. Numerical results indicate that the approximation is accurate enough to
correctly price barrier options in all but the most extreme circumstance; i.e., except when
the price of the underlying asset nearly coincides with the barrier.
Our analysis is based on the usual Black–Scholes market assumptions (Black and Scholes
1973). In particular, the asset price {St , t ≥ 0} follows the stochastic differential equation

(1.1)

dS
= ν dt + σ d Z ,
S

where Z is a standard Wiener process, ν and σ > 0 are constants, and S0 is fixed. The
term structure is flat, and we let r denote the constant, continuously compounded risk-free
interest rate. The price of a claim contingent on S is the expected present value of its cash
flows under the equivalent martingale measure, which sets ν = r in (1.1).
Let H denote the level of the barrier. An up option has H > S0 and a down option has
H < S0 ; in particular, we always assume H = S0 . The asset price reaches the barrier for
the first time at2
(1.2)

τ H = inf{t > 0 : St = H };

this is an up-crossing if S0 < H and a down-crossing if S0 > H . A knock-in call option
with maturity T and strike K pays (ST − K )+ at time T if τ H ≤ T and zero otherwise. Its
price is thus3
e−r T E[(ST − K )+ ; τ H ≤ T ],
the expectation taken with respect to the equivalent martingale measure. For a put option,
replace (ST − K )+ with (K − ST )+ , and for knock-out options replace the event {τ H ≤ T }
1 The price of a discrete barrier option can be expressed in “closed form” in terms of multivariate normal
probabilities. The dimension of the relevant multivariate normal distribution is equal to the number of monitoring
instants. This is typically too large for numerical evaluation.
2 Here and in what follows, we adopt the usual convention that the infimum of an empty set is infinity, so that
in particular τ H = ∞ if the asset prices never reach the barrier.
3 For a random variable X and an event A, the notation E[X ; A] means E[X 1 ], with 1 the indicator of A.
A
A

} is the asset price at monitoring instants. Formulas for pricing these continuously monitored barrier options are reviewed in Section 3. √ where + applies if H > S0 . 2 . A Sketch of the Argument Sections 3 and 4 and the appendices are devoted to the proof of Theorem 1. − applies if H < S0 . Suppose.1) yields d log S = (ν − 1 σ 2 ) dt + σ d Z . τ H ≤ m]. 1.1) is that Theorem 1. Let us write Si for Si t . Consider a binary knock-in option paying one dollar if the asset price reaches H at some point in [0. m.1. we should first shift the barrier away from S0 by a factor of exp(βσ results in Section 2 suggest that the resulting approximation is remarkably accurate. .2. that r − 1 σ 2 = 0. where t = ˜ ˜ T /m. 1. This result indicates that to use the continuous price as an approximation to the discrete √ t). . The price of a discrete knock-in call option is given by ˜ ˜ e−r T E[( Sm − K )+ . . i = 0. Then (1. so that under the equivalent martingale 2 √ measure log St has zero drift. K }. .5826. with ζ 2 the Riemann zeta function. i = 0. Numerical price.1 applies to binary barrier options as well. we illustrate the argument for an example. ˜ inf{n > 0 : Sn < H }. Let V (H ) be the price of the corresponding continuously monitored barrier option. THEOREM 1. A by-product of the analysis in later sections (Corollary 4. .4 Define b and c via the relations H = S0 exp(bσ T ) and 4 Ito’s lemma applied to (1. Now suppose the barrier is monitored only at times i t. The following result shows how to adjust the continuous formula to obtain a far better approximation to the discrete price. A consequence of our analysis is the obvious conclusion that the discrete price converges to the continuous price as the monitoring frequency increases. . 1. . suggesting that the continuous price may be used as a naive approximation.A CONTINUITY CORRECTION FOR DISCRETE BARRIER OPTIONS 327 with its complement {τ H > T }. so that { Si . the same modifications as before yield puts and knock-outs. T ] and if at time T it is below K . τH = ˜ S0 < H S0 > H. Define ˜ inf{n > 0 : Sn > H }.3) Vm (H ) = V (H e±βσ √ T /m 1 )+o √ m . and β = −ζ ( 1 )/ 2π ≈ 0. for purposes of illustration. Let Vm (H ) be the price of a discretely monitored knock-in or knockout down call or up put with barrier H . In general. with H > max{S0 . there are no easily computed closed-form expressions for the prices of these discrete barrier options.1. To give some insight into the result.

˜ with Wn = n Zi . P.4) is an application of the reflection principle (Karatzas and Shreve 1991. τ ≤ m).6) and treating Wm and Rm as though they were independent we get (1. The price of the option under discrete monitoring of the barrier is e−r T times √ ˜ ˜ ˜ P( Sm < K . and using the fact (to be reviewed in Section 4. max log St ≥ bσ T ) 0≤t≤T √ = P(log ST ≥ (2b − c)σ T ) = 1 − Φ(2b − c). GLASSERMAN. with t = T /m. Applying this to (1. pp. Because r − 1 σ 2 = 0. Notice that √ ˜ ˜ ˜ P(Wm > x m) = 1 − Φ(x). ˜ √ √ ˜ where Rm = Wτ − b m is the overshoot above level b m (see Figure 1. that for the option to be knocked in. . t. τ ≤ m) = P(Wm > 2(b m + Rm ) − c m).6) √ √ √ ˜ ˜ P(Wm < c m. the asset price must reach H at some time in {0. Suppose. max Sn ≥ H ) = P(Wm < c m. m t}. now. the value of the option is e−r T times (1. Expanding in a formal Taylor series with ϕ denoting the standard normal density. 2 t. where Φ is the standard normal cumulative distribution function. . BROADIE.5) √ √ P(ST < K .7) √ √ Rm ˜ P(Wm > 2(b m + Rm ) − c m) ≈ E 1 − Φ 2 b + √ m −c .4) (1. so the reflection principle yields (1. ˜ 0≤n≤m √ ˜ ˜ where τ is the first time W exceeds b m.1). i=1 where the Z i are independent standard normal random variables. Assuming continuous monitoring of the barrier. τ H ≤ T ) = P(log ST < cσ T . Equation (1. we may represent 2 the asset price at these monitoring instants as ˜ Sn = S0 eσ √ ˜ t Wn . . KOU √ K = S0 exp(cσ T ).328 M.3) that E[Rm ] → β. we get √ √ 2Rm 1 ˜ P(Wm > 2(b m + Rm ) − c m) ≈ E 1 − Φ(2b − c) − √ ϕ(2b − c) + o √ m m 2 1 ≈ 1 − Φ(2b − c) − √ E[Rm ]ϕ(2b − c) + o √ m m . AND S. The increments of the random walk W are ˜ symmetrically distributed. 79–80). .

Siegmund (1979. and identify the resulting expression as an expansion of the continuous price.1 rigorously: we find a suitable representation of the discrete price. Overshoot and reflection principle illustration.1.3. with the price of a continuously monitored barrier option √ √ knocked in at level H exp(βσ t). √ and thus agrees. 2β 1 = 1 − Φ(2b − c) − √ ϕ(2b − c) + o √ m m β 1 −c +o √ . expand it in a Taylor series. = 1−Φ 2 b+ √ m m In light of (1. evaluate the limits of the coefficients as m → ∞. 1985).7). The heuristic argument just sketched points out the key steps in proving Theorem 1. Siegmund and Yuh . Continuity corrections for level-crossing probabilities of random walks have a long history in sequential analysis and. this is P(ST < K . to a lesser extent. the necessary asymptotic independence result is given in Section 4. τ H exp(βσ √ t) 1 ≤ T) + o √ m . and this may be viewed as the rigorous counterpart of (1.5). Key references from the statistical literature are Chernoff (1965). Handling ordinary barrier options instead of just binary options requires consideration of integrals of the probabilities outlined above. The difference H eβσ t − H may be viewed as an approximation to the amount by which the asset price exceeds the barrier at the first monitoring instant at which it is above the barrier. up to o(1/ m). Evaluation of the limits of coefficients requires consideration of the joint limit of the overshoot and the barrier-crossing time.A CONTINUITY CORRECTION FOR DISCRETE BARRIER OPTIONS 2 ( b m + Rm ) − c m S b 329 Rm m c ˜ τ m t FIGURE 1. in risk theory.

This numerical procedure has an average accuracy of about 0.3). the continuous barrier price is $3.e. T ]. P. Specifically. We especially build on the work of Siegmund and Yuh. this heuristic had the smallest error of the three approximations. i.) In most of the test cases. (The factor 0. The modified corrected partial barrier approximation (3) in Table 2. with 0 < t1 < T .001 for this range of parameters..1% for H ≤ 98. we present numerical results to indicate the accuracy of the corrected barrier approximation given in equation (1. PERFORMANCE OF THE CORRECTED BARRIER APPROXIMATION Before proceeding with the main proof. i. It is worth emphasizing. AND S.06 while the discrete barrier option with daily monitoring is worth 25% more ($3.83).000 time steps.1 and later tables show that the continuous barrier price can differ from the discrete barrier price by economically significant amounts. this formula is remarkably accurate except in extreme cases with H very close to S0 .330 M.1 compares the approximation (and two related approximations to be described shortly) with the true value for different levels of the barrier H . 2. KOU (1982). an attempt to justify a refinement of (1) would involve the daunting task of carrying out the 5 In order to compute the true value to within 0. Table 2.58 was found experimentally to be the best among a few different values tried. 259 of Heynen and Kat (1994b). approximation (2) sets t1 = t and shifts the barrier in equation (9) on p. For example..001 on average.1. The prices under approximation (1) in the tables result from the corrected continuous barrier formula suggested by Theorem 1. The first monitoring point for a discrete barrier option does not occur until t. t1 ] or [t1 . that only approximation (1) is rigorously supported. Chernoff and Petkau (1986) have shown that a boundary correction improves the numerical solution of an optimal stopping problem based on a binomial approximation to a diffusion. The large number of steps required for this accuracy is quite computationally intensive. overall. The relative absolute error of this formula is less than 0. Also in the context of sequential analysis.000/sec on an Intel Pentium 133 MHz processor).58 t. It can be evaluated very fast (more than 10. Table 2. The options considered are down-and-out call options. for H = 97 in Table 2. we needed to run a modified trinomial routine using 80. options whose barriers are monitored throughout a time interval [0. GLASSERMAN. BROADIE. S0 > H . AitSahlia (1995) has analyzed a similar technique in the pricing of American options (without barriers). the corrected partial barrier formula (approximation (2)) leads to small overpricing in these cases.1. Approximation (2) uses a formula of Heynen and Kat (1994b) for continuous partial barrier options. however. It does so by adjusting the beginning of the partial barrier to t1 = 0.5 Table 2.1 is a further heuristic that attempts to compromise between the slight underpricing and overpricing observed with (1) and (2).e. for developments in risk theory see Asmussen (1989). The “true” value is determined from a trinomial procedure modified in several ways to specifically handle discrete barriers and is fully described in Broadie et al. It can be evaluated in about the same (negligible) computing time as the first approximation.1 shows results for two other approximations. and the option becomes worthless if S crosses the barrier before the option expires at time T . Recently. . while the procedure used to compute the true value to comparable accuracy is many orders of magnitude slower (approximately one hour on the same processor). T ]. suggesting an approximation that shifts the barrier in the formula for a partial barrier option monitored during [ t. and Woodroofe (1982). (1996). Although correcting the continuous barrier formula (approximation (1)) leads to small underpricing in some cases.

321 6.428 3.271 6. m = 50 barrier points roughly corresponds to daily monitoring of the barrier.0 0.184 6.0 0.308 6.977 5.0 0.1 Down-and-Out Call Option Price Results. In this case.060 2.242 6.160 2. Indeed.129 2.281 6.288 4. Table 2.30.0 0.1 0.810 5.579 5.0 0.2. Option price derivatives (e.242 6.0 0.288 4.0 0.0 0. and T = 0. the approximation is quite accurate even for m = 5.3 1.834 3.0 0.0 0.6 Option parameters: S0 = 100.A CONTINUITY CORRECTION FOR DISCRETE BARRIER OPTIONS 331 TABLE 2.2 shows that the approximations degrade slowly as the number of monitoring points decreases.0 0. the relative errors are slightly larger with higher volatility and longer time to maturity and are slightly lower with a higher strike (compared to Table 2.6). r = 0.322 6.427 3.977 5.845 3. Heynan and Kat (1994a) and Rich (1996) derive formulas for such .0 0.098 5.837 3.810 5.242 6.792 3.337 Relative error (in percent) (1) (2) (3) 0.098 5.808 5.185 6.099 5.907 4.810 5. expansions in Sections 3 and 4 to one more term.0 0. Table 2.585 5.977 5. m = 50 barrier points Corrected Corrected continuous partial Continuous barrier barrier Barrier barrier (1) (2) 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 6. and higher strike (K = 110). Next we consider a two-state example where the payoff of the option is determined by one state variable and the knock in and or knock out is determined by a second state variable— for example.0 0.0 0.0 0.288 4.428 3.171 6.1).907 4. long maturity (T = 2).0 0.322 6.1% for H ≤ 97 for approximation (1).398 3.184 6.288 4.4 shows how the deltas of the three approximations compare to the true delta (determined through extensive numerical computations). in marked contrast to the continuous price.0 0. “delta” = ∂ V /∂ S0 ) are of particular importance to practitioners for hedging and risk management. even though Theorem 1.977 5.1 −0.244 6.322 6.1 is based on the limit as m increases to infinity.810 5.585 5. Although this case is not covered by Theorem 1.0 0.184 6.0 0.1. K = 100.1 3.126 2.121 2.098 5.0 0.408 Modified corrected partial (3) True 6.10.g.0 0.322 6.1 0.281 6.907 4..888 4.306 6.836 3.184 6.977 5.098 5.277 4.0 0.189 1. Rich (1996) models options with default risk in this framework.2 −2. σ = 0. the correction can be applied to the continuous delta formula as well.283 6.0 0.281 6. Delta approximations (2) and (3) are better than (1) for H very close to S0 . options on stocks or stock indices which are knocked out if a currency reaches a barrier trade in the over-the-counter market. approximations (2) and (3) are slightly more complicated to implement than approximation (1).0 0.0 0.8 0.0 0.0 0. except when the barrier is very close to the underlying asset price. Relative absolute errors are less than 0.281 6.585 5.0 0. Table 2.0 0. The overall performance of the approximations remains excellent.584 5. Assuming 250 trading days per year.242 6.0 0.306 6.0 0.3 shows results for high asset volatility (σ = 0.0 0.907 4.429 3.0 0. Also.0 0.306 6.306 6.1 −0.0 0.

1 6.308 6.0 6.2 Down-and-Out Call Option Price Results.671 −0.171 6. P.6 3.10.0 6.084 4.323 6.7 1.0 0.0 4. but whose payoff is positive if the terminal asset price is above a strike level.392 6.0 0. The previous examples involved down-call options. THE CONTINUOUS PRICE The purpose of this section is to derive expressions for the continuous price V (H ) and its derivative.287 6.337 6.808 5.3 0.285 6. KOU TABLE 2.398 3.5 shows that the corrected continuous barrier formula is very accurate as long as the barrier is not too close to the initial price of asset 2.0 0.293 6.7 4. options that are knocked out when an asset price decreases to a barrier level. Table 2.210 6.321 0.332 M. In this and the remaining sections.489 −9.033 5.060 1.277 4.060 1.281 0.327 6.194 6.760 5.2 0.0 0.033 5. Corrected versions of the continuous partial barrier formulas in Heynen and Kat (1994b) could also be used to approximate the discrete barrier price.0 0. σ = 0.154 2.0 5.30. 3.0 0.2 85 87 89 91 93 95 97 99 6.210 6.0 5.398 3.032 0.081 0.292 0.327 6.0 0.099 5. and T = 0.2.210 0.0 0.673 6. we detail the case of the down-and-in call.1 −0.0 0. The payoff of the option is (ST − K )+ if S 2 is greater than H at all monitoring times.308 6.1 −1.688 5. varying m Corrected Corrected Modified continuous partial corrected Continuous barrier barrier partial m Barrier barrier (1) (2) (3) True 25 5 Relative error (in percent) (1) (2) (3) 85 87 89 91 93 95 97 99 6.923 6.0 0. The results indicate that the corrected barrier formula works well for these options.676 5.813 −5.113 2.0 6.0 5.116 −0. GLASSERMAN.0 0. More specifically.1 6.000 0. BROADIE. though perhaps not as well as for down calls.724 6.337 0.688 0..1).9 0.e.0 0.1 5.5 5.004 5.244 6.2 Option parameters: S0 = 100.0 6. 2. AND S. suppose that the risk-neutralized asset price processes are d S i = S i (r dt + σi d Z i ).050 6.9 3. i = 1.084 4.167 −2.338 6.171 6.688 5.0 6.2 0.8 0. two-state options when the barrier is continuously monitored.323 6.205 6. The “true” price is more difficult to determine because of the extra dimension involved.028 4.337 6.293 6.003 for this range of parameters. but the two-dimensional trinomial procedure used has an average accuracy of about 0.0 0.0 0.327 6.326 0.187 0.033 5.293 6.244 6.0 0.043 5. where Z 1 and Z 2 have a constant correlation ρ.210 6.088 4. Table 2.5 presents results for a two-asset discrete 1 barrier down-and-out call option.323 6. Table 2.688 5.284 6.1 0.195 6. . K = 100.1 2.323 4.1 0.808 5.1 0.6 gives results for up call options (not covered by Theorem 1.0 0.277 4.779 6.120 2.1 0.5 −2. r = 0. i.646 5.141 4.099 5.011 5.

756 1.111 5.0 0.0 0.433 2.0 0.874 20.433 2.104 1.404 8.0 0.351 4.821 19.0 0.3 0.0 0.8 0.0 0.0 0.3 85 87 89 91 93 95 97 99 2.1 −2.0 0.534 12.650 12.129 2.2.0 0.346 4.451 −0.491 2.0 0.475 2.379 5.475 0. τ H = inf{t ≥ 0 : S0 e Wt ≤ H } = inf{t ≥ 0 : Wt ≤ a}.020 9.563 6.571 0.e.371 9.937 3. There are m = 50 barrier points (daily monitoring).105 2.1 1.232 15.0 0.120 16.344 4. Thus.7 5.537 0.0 0.7 0.446 14.171 2.0 0.574 18.048 9.505 10.136 1.0 0.757 1. Panel B has T = 2.505 10. σ = 0.496 2. τ H < T ].6 10.496 2.0 0.1 4.384 8.782 20.0 0.822 19. i.459 2. Panel A has σ = 0..0 12.133 16.0 19.1 −0.0 0.941 3. T .0 0.573 18.6 0.337 2. σ = 0.566 6.3.573 7. The barrier is breached when W hits a = log(H/S0 ).761 1.0 0.0 0.1 0.0 0.573 7.0 0.0 0.572 6.0.475 2.250 1.574 7.121 16.219 10.3. We define W to be a Wiener process with variance parameter σ 2 and drift µ = r − 1 σ 2 .475 2.0 0.436 0.336 2.337 2.0 0.485 2.6 Option parameters: S0 = 100 and r = 0.0 0.8 85 87 89 91 93 95 97 99 18.0 0.0 0.815 8. the other cases following with only minor modifications.414 10.446 0.136 1.591 2.020 9. T = 0.0 0.547 6.819 0. V (H ) = e−r T E[(ST − K )+ .433 2.7 0.0 0.945 7.8 −7.572 7.384 8.336 2.061 −10.0 0. with H < S0 and H < K .4 0. .496 2.491 2.657 10.494 2.361 13.020 9.1 0.1 0.020 9.10.067 1.1 0.7 −1.243 20.505 10.136 0.9 −0.566 6.1 1.114 0.0 −0.1 3.3 Down-and-Out Call Option Price Results. This allows us to represent the asset price as St = S0 exp(Wt ) and 2 then work with W rather than S.505 10.3 0.0 −0.6.576 8. and Panel C has K = 110.254 −3.384 8.0 0.578 7.033 3.227 10.383 8.433 2.136 1.446 10.450 14.496 2.0 0.803 3.499 20.638 10.0 0.2.757 1.900 3.964 1.491 2.577 18.0 18.4 −2.0 3. varying σ .A CONTINUITY CORRECTION FOR DISCRETE BARRIER OPTIONS 333 TABLE 2.0 0.821 19.394 2.0 0.183 7.1 14. T = 0.856 17.487 14.084 10. and K Corrected Corrected Modified continuous partial corrected Continuous barrier barrier partial Panel Barrier barrier (1) (2) (3) A B C True Relative error (in percent) (1) (2) (3) 85 87 89 91 93 95 97 99 10.0 16.491 2.475 2.293 4. K = 100. K = 100.553 12.0 8.135 1.

0 0.618 0.1) (3.0 0.633 0.600 0. √ 2a µ T − √ 1−Φ x − σ σ T Proof. r = 0.0 0.0 0.066 0.750 0.857 0.1 0.798 0.0 −0.0 0.853 0.633 0.2 −0.0 0.0 0.711 0.0 0.0 0.653 0.853 0. (2r/σ 2 )−1 H S0 V (H ) = BSC(H 2 /S0 ) 2 = e2µa/σ BSC(S0 e2a ) = e −r T 2µa/σ 2 e × S0 e xσ √ T ∞ √ [log(K /S0 )]/σ T √ σ T d x.1 Option parameters: S0 = 100.0 0.607 0.0 0.972 0.851 0. K = 100.600 0. (55)) formula for a barrier option.678 0. GLASSERMAN. P. The second identity restates the first because a = log(H/S0 ).6 3.917 0.618 0. this particular representation is from Boyle and Lau (1994).711 0.3 0.994 1.0 0.591 0.0 0.679 0.0 0. write the Black– .158 0.2.710 0.653 0.0 0.4 −2.30.0 0.966 0. There are m = 50 barrier points (daily monitoring). The first identity is a consequence of Merton’s (1973.0 0.798 0.988 1.594 0.7 0.904 0.653 0.073 1.0 0.750 0.0 0.10. AND S.607 0.750 0.600 0.854 0.9 −3. (3.938 0.591 0.2 11.750 0.653 0.599 0. To get the last identity.0 0.917 0. Define BSC(s) = e−r T E[(se WT − K )+ ].0 2.334 M.4 −1.710 0.678 0.594 0.0 0.600 0.710 0.798 0.752 0.591 0.3) For a down-and-in call.0 0.0 0.922 Modified corrected partial (3) True 0. eqn.921 0.633 0.1 −0. KOU TABLE 2.633 0.0 0.0 0. BROADIE.958 Relative error (in percent) (1) (2) (3) 0.1.653 0.0 0.0 0.607 0.633 0.678 0. σ = 0.0 0. the Black–Scholes price of a call option as a function of the initial asset price.618 0.2) (3.591 0.594 0.0 0.594 0.678 0.800 0. and T = 0.2 0.594 0.915 0.4 Down-and-Out Call Option Delta Results. m = 50 barrier points Corrected Corrected continuous partial Continuous barrier barrier Barrier barrier (1) (2) 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 0. Then we have PROPOSITION 3.0 −0.0 −0.618 0.618 0.988 0.797 0.607 0.607 0.

TABLE 2.545 12.726 4.887 5.621 5.442 0.112 1.051 3.0 0.4 −3.1 −0.052 3.747 5.819 12.765 5.905 12.1 −0.404 4.246 5.012 1.121 1.886 5. r = 0. The option is knocked out if S 2 hits the 1 barrier at a monitoring time.775 12.645 3.729 4. m = 50 barrier points Barrier Continuous barrier Corrected continuous barrier True Relative error (in percent) 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 5.005 4. K = 100.985 5.314 4.433 6.10.0 0.2 Option parameters: Initial assets prices are S 1 = S 2 = 100.766 5.448 5.413 4.144 8.772 5.30.5 Option parameters: S0 = 110.2.959 4. There are m = 50 barrier points (daily monitoring).005 4.1 0.0 0.447 5.187 2.542 0.0 0.616 2.581 8. ρ = 0.626 5.452 5.1 −0. K = 100.0 1.994 6.243 5.10.707 10. and T = 0.959 6.0 0. The option payoff is (ST − K )+ at time T if the option has not been knocked out.938 0.0 0.418 0.5 Two-Asset Down-and-Out Call Option Price Results.7 1.0 0.649 2.493 5.684 10.894 12. σ1 = σ2 = 0.240 11.412 4.244 5.807 0.623 3.185 2. m = 50 barrier points Barrier Continuous barrier Corrected continuous barrier True Relative error (in percent) 155 150 145 140 135 130 125 120 115 12.1 0.639 2.3 0.986 5.066 1.0 −0.A CONTINUITY CORRECTION FOR DISCRETE BARRIER OPTIONS TABLE 2. 335 .893 5.435 0.5.038 3.005 4.620 5.2. r = 0. σ = 0.729 4.158 2. and T = 0.4 0.922 4.679 2.0 0.448 11.395 10.676 2.431 11.30.6 Up-and-Out Call Option Price Results.643 3.551 8.0 0.5 0.2 0.

see Rich (1994) and Rubinstein and Reiner (1991). (3. σ . ∂a σ where (·) is the Black–Scholes delta as a function of the initial asset price with r .3). Apply this to (3. √ and make the substitution x = (y + 2a)/(σ T ) to get (3.2. and K held fixed. For a comprehensive treatment of the other cases and of connections among the various prices. AND S.336 M.3) for the first term yields the first term of . Substituting (2µ/σ 2 ) times (3. T .4) For a down-and-in call. From (3.2) we get ∂ 2µ 2 2 V (S0 ea ) = 2 e2µa/σ BSC(S0 e2a ) + e2µa/σ · 2S0 e2a (S0 e2a ). Proof. ∂ V (S0 ea ) = e−r T ∂a × √ 2µ T σ e2µa/σ 2 ∞ √ [log(K /S0 )]/σ T √ 2a µ T − √ 1−Φ x − σ σ T + 2e −r T 2µa/σ 2 × S0 e xσ e √ T ∞ √ [log(K /S0 )]/σ T S0 e xσ √ T dx √ µ T 2a ϕ x− − √ σ σ T d x. KOU Scholes price as BSC(s) = e−r T E[(se WT − K )+ ] ∞ 1 y − µT (se y − K )+ √ ϕ dy √ σ T σ T −∞ ∞ y − µT se y dy 1−Φ √ σ T log(K /s) = e−r T = e−r T using integration by parts. We now write V (S0 ea ) for V (H ) and calculate the derivative of the continuous option price with respect to a: PROPOSITION 3. GLASSERMAN. BROADIE.2) to get V (H ) = e−r T e2µa/σ ∞ 2 log(K /S0 e2a ) y − µT √ σ T 1−Φ S0 e2a e y dy. The price of a down-and-out call can be obtained by subtracting the down-and-in price from the standard Black–Scholes price. P.

and 2 µ√ t . Most of the technical details of intermediate steps are relegated to appendices. Also. 4. We use the representation ˜ Sn = S0 exp µn t + σ ˜ = S0 exp{Wn σ √ √ n t Zi i=1 t}. 4.2) and (3. Combining Propositions 3. To get the second term in (3.1. we give the main steps of the proof of Theorem 1. ˜ Wn = n Zi + i=1 µ = r − 1 σ 2 . Wn ≤ a/(σ .1 and 3.2. we obtain formulas for the option price and its derivative in expressions of the form V (S0 ea+ ) = V (S0 ea ) + ∂a V (S0 ea ) + o( ).4). MAIN STEPS OF THE PROOF In this section. This moves a out of the argument of Φ and into the lower limit of integration.3) again and differentiate to get 2S0 e 2a (S0 e ) = e 2a −r T ∞ ∂ ∂a √ [log(K /S0 )]/σ T × S0 e xσ = e √ T √ σ T dx √ 2a µ T − √ ϕ x− σ σ T √ [log(K /S0 )]/σ T σ T × S0 e xσ T ∞ 2 √ −r T √ √ 2a µ T − √ 1−Φ x − σ σ T √ σ T d x. where ∂a is short for ∂/∂a.A CONTINUITY CORRECTION FOR DISCRETE BARRIER OPTIONS 337 (3. A simple justification of this interchange of derivative and integral first makes the change √ of variable y = x − (2a/σ T ).4) we use (3. we use Sn to denote the asset price at the n th monitoring instant. The Discrete Price ˜ As in Section 1. nT /m. where the Z i are independent standard normal random variables. we let τ = inf{n ≥ 1 : ˜ = inf{n ≥ 1 : ˜ Sn ≤ H } √ ˜ t)}.1. σ t = T /m.

τ < m]. show that the higher-order terms are o(m −1/2 ).1) and ζ = − √ > 0. 4. ˜ Our approach to proving Theorem √ will be to approximate the integrand P(τ < 1.1 below builds. . and finally identify the resulting expressions with continuous formulas from Section 3. as before. θ ∈ R}.1. The option price is ˜ ˜ Vm = e−r T E[( Sm − K )+ . KOU where. X 2 . Integration by parts and a change of variables yields the following: LEMMA 4. √ where b = ζ m for some ζ < ξ . we first generalize the setting slightly. BROADIE. To make the correspondence with the notation of Section 4. as required by Lemma 4. the case ξ < 0 (corresponding to µ > 0) follows from an analysis of ξ ≥ 0. Define n Un = Xi . Define τ = inf{n ≥ 1 : Un ≥ b}. We work with a family of probability measures {Pθ .1. set √ a µ T (4. P. Wm ≥ y m) = Pθ1 (τ < m. We use notation consistent with Siegmund and Yuh (1982). AND S. Um ≤ −y m) ˜ √ with y ≥ a/(σ T ) = −ζ . evaluate the coefficients on the terms of order m −1/2 and integrate them. with ξ ≥ 0 a finite constant. .1. . a = log(H/S0 ) < 0. τ < m] = e−r T E[(S0 e Wm σ ˜ √ t − K )+ . By the symmetry of the normal distribution. the random variables {X 1 .2. as in Lemma 4.} are independent and normally distributed with mean θ and unit variance.1. ξ =− σ σ T Then √ √ ˜ P(τ < m. Wm ≤ x).1. GLASSERMAN. even after integration. under Pθ . Wm ≥ y m) for y ≥ log(K /S0 )/σ T to terms of order m −3/2 . Approximation of the Probability To approximate the integrand in Lemma 4. With the notation above Vm = e−r T ∞ √ [log(K /S0 )]/σ T √ √ √ ˜ P(τ < m. on which Theorem 4.338 M. ˜ This expectation is an integral with respect to the density in x of the distribution P(τ < ˜ ˜ m. Wm ≥ y m)σ T S0 e yσ T dy.1 ˜ √ ˜ m. Let θ1 = ξ/ m and θ0 = −ξ/ m. the random walk Un thus has drift θ . . i=1 √ √ under Pθ .

|D| ≤ 1. τ < m] m m √ 1 2 − E θ1 [e−2ξ Rm / m Rm (1 − τ /m)−1 · 1 ϕ (η1 ). B(τ .A CONTINUITY CORRECTION FOR DISCRETE BARRIER OPTIONS 339 Let Rm = Uτ − b. y). Um < −y m) = e2ξ ζ (1 − Φ(2ζ + y + ξ )) 2 2ξ + √ E θ1 [Rm A(τ . 1 − τ /m 1 − τ /m 1 − τ /m and ζ+y ζ+y Rm + ξ 1 − τ /m.√ + ξ 1 − τ /m .2) Pθ1 (τ < m. y). τ < m] + m m 1 m 1 2 − √ E θ1 [D · 4ξ 2 Rm B(τ . ζ+y Rm ζ+y η1 ∈ √ + ξ 1 − τ /m − √ .1. τ < m] 2 m 1 2 + E θ1 [Rm (1 − τ /m)−1 · 1 ϕ (η2 ). 2 m where |C| ≤ 1. τ < m] − √ E θ1 [Rm B(τ . ξ. such that for any t. ζ+y − ξ 1 − τ /m A(τ . y). τ < m] − E θ1 [C · 4ξ 2 Rm A(τ . y ≥ 0. 4. Limits of the Coefficients In light of Lemma 4. integrating equation (4. Rm ≤ y) → G(t.1. √ (4.2. y) = √ 1 − τ /m 1 − τ /m Then for y > −ζ . Pθ1 (τ /m ≤ t. y). 1 − τ /m 1 − τ /m 1 − τ /m Proof. . ∞).3. and define THEOREM 4. ζ )B R (y). y) = Φ − √ 1 − τ /m ζ+y 1 ϕ √ + ξ 1 − τ /m . y). τ < m] m m 2ξ 1 2 E θ [R 2 B(τ . √ + ξ 1 − τ /m + √ η2 ∈ √ . To approximate the discrete price. we evaluate the integral of each of the terms in the right-hand side of (4. See Appendix A.2) in Theorem 4. τ < m]. There is a probability distribution B R on [0.2) as m → ∞. An essential tool is the following result: LEMMA 4.1 yields the discrete price Vm .

Then. we integrate the approximations to the expectations. ∞ (4. Denote by g the density of the distribution G. 1 ϕ √ 1−t = 1. ξ. REMARK. Theorem 10. where √ √ G(t.3) lim √ m→∞ [log(K /S )]/σ T 0 ∞ =β E θ1 [Rm A(τ .340 M.2. and that the distributions of τ /m and Rm can be approximated by G and B R respectively. y). we factor them into products of expectations involving just Rm and just τ . ζ ) = 1 − Φ(ζ T −1/2 − ξ T ) + e2ξ ζ Φ(−ζ T −1/2 − ξ T ). 2. τ < m]e yσ √ [log(K /S0 )]/σ T 1 ζ+y Φ −√ 1−t 0 √ T dy √ √ − ξ 1−t g(t) dt e yσ T dy.1.1. y d B R (y) = β. AND S. So. Rm ≤ y) → G(t. BROADIE. then evaluate them under the limiting marginal distributions indicated by Lemma 4. with β as defined in Theorem 1. Proof.1).11 of Siegmund (1985) we have Pθ1 (τ ≤ ζ 2 mt. ζ )B R (y). G(t. ξ. . to approximate the expectations in Theorem 4. ξ. PROPOSITION 4. τ < m]e yσ √ 1 [log(K /S0 )]/σ T √ yσ T × g(t) dt e 0 dy. ζ )B R (y). KOU as m → ∞. Moreover. y).4) lim √ m→∞ [log(K /S )]/σ T 0 ∞ =β E θ1 [Rm B(τ . The next result justifies this process. 2. and ∞ (4. Specializing to the case of normal increments and comparing with Chernoff (1965) we find that this mean is the β defined in Theorem 1. Pθ1 (τ /m ≤ t. From Lemma 10. ξ. ζ ) is the probability that a Wiener process with drift ξ and variance parameter 1 reaches ζ by time t. To approximate integrals over y of these expectations (as required by Lemma 4. Hence. as can be seen from the formula for G.55 of Siegmund (1985) gives a general expression for the mean of B R . GLASSERMAN. P.1. Let β1 = β and let β2 denote the second moment of B R . ξ ζ. We interpret this result as stating that τ /m and Rm are asymptotically independent.1. Rm ≤ y) → G(t/ζ 2 . 1)B R (y) = G(t. √ T dy √ ζ+y +ξ 1−t √ 1−t = 1.

Comparison with the Continuous Price Consider the expression given for the discrete price in Lemma 4.3. To evaluate the limits in Proposition 4.A CONTINUITY CORRECTION FOR DISCRETE BARRIER OPTIONS 341 Proof.1.5) Φ −√ 1−t 0 1 (4.6) 0 x 1 ϕ √ 1−t √ − ξ 1 − t g(t) dt = e2ξ ζ Φ(−x − ζ − ξ ). See Appendix B.1 √ √ applies because for any y ≥ log(K /S0 )/σ T we have y > log(H/S0 )/σ T = −ζ .2) in Theorem 4. An alternative probabilistic argument notes that both sides of (4.14) of Siegmund (1985) and for the left side condition on τ . Theorem 4. Proof. and τ (ξ ) is the first time W (ξ ) reaches ζ . To dispense with integrals of the last two expectations there.5) are equal to P(τ (ξ ) < (ξ ) 1. W1 < ζ − x).6) yield the corresponding density at ζ − x. 2. ϕ (ηi ) √ 1 − τ /m where η1 . t > 0.1). 4. τ < m < ∞.3. ∞ lim sup E θ1 m→∞ √ [log(K /S0 )]/σ T √ e yσ T dy. Using Proposition 4. See Appendix C. we need the following identities: LEMMA 4.1. √ x + ξ 1 − t g(t) dt = e2ξ ζ ϕ(x + ζ + ξ ). we need the following: PROPOSITION 4. The . where W (ξ ) is a unit-variance Wiener process with drift ξ .2) using the parameters in (4. 1 (4. For the right side use equation (3. substitute the expression in (4.1 and Lemma 4. Both identities can in principle be verified algebraically using the fact that 1 ζ g(t) = √ t −3/2 exp ξ ζ − 2 2π ζ2 t + ξ 2t . For i = 1. √ 1−t Proof. Both sides of (4.2.1. η2 are as in Theorem 4.1. we can handle integrals of the first five expectations on the right side of equation (4. For the integrand appearing there.4.

we thus obtain √ √ 2ξβσ T −r T ∞ 2ξ ζ Φ(−2ζ − y − ξ )S0 e yσ T dy Vm (S0 e ) = V (S0 e ) + √ e √ e m [log(K /S0 )]/σ T √ ∞ √ 1 2βσ T −r T 2ξ ζ ϕ(2ζ + y + ξ )S0 e yσ T dy + o √ − √ e . KOU result of the substitution is of the form ∞ e Vm (S0 e ) = rT a √ e [log(K /S0 )]/σ T √ √ × σ T S0 e yσ T ∞ 2µa/σ 2 √ µ T 2a 1−Φ y− √ − σ σ T dy 1 2E θ1 [Rm (ξ A(τ . u) − B(τ . P. S0 e yσ √ T dy . and invoke the boundedness of the limits in Proposition 4. τ < m] +√ m [log(K /S0 )]/σ √T √ √ 1 × σ T S0 e yσ T dy + o √ . GLASSERMAN.1). y)).1 for = 2 and the boundedness proved in Proposition 4.3. τ < m] +√ m [log(K /S0 )]/σ √T √ √ × σ T S0 e yσ T dy 1 · · · dy + m 1 + √ · · · dy. m We may replace the remaining integral with its limit as m → ∞ and preserve the validity of this expression. we get Vm (S0 ea ) = √ √ βσ T 2µ T −r T ∞ 2µa/σ 2 e V (S0 e ) − √ √ e σ m [log(K /S0 )]/σ T √ √ 2a µ T × 1−Φ y− S0 e yσ T dy − √ σ σ T √ √ µ T 2a βσ T −r T 2µa/σ 2 ∞ − √ − √ 2e e √ ϕ y− σ m σ T [log(K /S0 )]/σ T a 1 +o √ m . m m Now apply (3.1 with = 1 and then Lemma 4. BROADIE. u) − B(τ .342 M. √ e m m [log(K /S0 )]/σ T a a Substituting for ξ and ζ as in (4.2 to write this as er T Vm (S0 ea ) = er T V (S0 ea ) ∞ 1 2E θ1 [Rm (ξ A(τ . y)). Using Proposition 4. AND S.3) to the first integral.

which is exactly the first term on the right side of equation (4. each Fθ.1) √ Pθ1 (τ < m.1.1.1 for the integrals apply equally well—indeed. Thus.m−τ (− m(y + ζ ) − Rm ) . Next we analyze Y.m−τ ( m(y + ζ ) − Rm ) − Fθ1 .2) in Theorem 4.1.1. which is to say that Vm (S0 ea ) = V (S0 ea−βσ √ t 1 )+o √ m .3) produces the price of the corresponding continuously monitored barrier option. By the definition of Un . Let us write this as (A. τ < m]. we have the following corollary.n is a normal distribution with mean nθ and variance n. Notice that if we simply omit the integral from the expression in Lemma 4. 244). we may rewrite it as √ βσ T 1 Vm (S0 e ) = V (S0 e ) − √ ∂a V (S0 ea ) + o √ m m a a . p.2. √ m−τ √ Y = e−2ξ Rm / m . Um < −y m) = e(θ1 −θ0 )b Pθ0 (Um ≥ (2ζ + y) m) −E θ1 e−(θ1 −θ0 )Rm √ √ × 1 − Fθ0 .A CONTINUITY CORRECTION FOR DISCRETE BARRIER OPTIONS 343 Comparing this with Proposition 3. The steps carried out in proving Theorem 1. Omitting the integral from (3. we have √ √ Pθ1 (τ < m. where Fθ. Then X = e2ξ ζ 1 − Φ √ (2ζ + y) m − θ0 m √ m = e2ξ ζ (1 − Φ(2ζ + y + ξ )). τ < m .n is the distribution of Un under Pθ .1 holds for binary barrier options as well.1 Directly from Siegmund and Yuh (1982. APPENDIX A: PROOF OF THEOREM 4. COROLLARY 4. with much less effort—for the integrands. Um < −y m) = X − E θ1 [Y. we obtain the price of a discretely monitored binary barrier option. Theorem 1. √ √ (y + ζ ) m − Rm + ξ(m − τ )/ m 1−Φ √ m−τ √ √ −(y + ζ ) m − Rm − ξ(m − τ )/ m −Φ . Thus. concluding the proof of Theorem 1.

such that e−x = 1 − x + cx 2 . m−τ 2 m−τ 2 . GLASSERMAN. this becomes √ Y = (e−2ξ Rm / m y+ζ − 1)Φ − √ − ξ 1 − τ /m 1 − τ /m √ + (e−2ξ Rm / √ + e−2ξ Rm / m m + 1) √ Rm m−τ ϕ y+ζ + ξ 1 − τ /m √ 1 − τ /m 2 2 Rm 1 Rm 1 ϕ (η1 ) − ϕ (η2 ). KOU Taylor expansion now gives √ m Y = e−2ξ Rm / y+ζ 1−Φ √ + ξ 1 − τ /m 1 − τ /m + √ Rm m−τ ϕ 2 y+ζ 1 Rm ϕ (η1 ) + ξ 1 − τ /m + √ 2m−τ 1 − τ /m y+ζ − Φ −√ − ξ 1 − τ /m 1 − τ /m y+ζ Rm ϕ −√ −√ − ξ 1 − τ /m 1 − τ /m m−τ + 2 1 Rm ϕ (η2 ) .344 M. m−τ 2 m−τ 2 For any x ≥ 0 there is a c with |c| ≤ 1. η2 as stated in the theorem. BROADIE. 2m−τ for some η1 . AND S. We may thus introduce C and D as stated in the theorem to get √ y+ζ Y = (−2ξ Rm / m)Φ − √ − ξ 1 − τ /m 1 − τ /m 2 Rm y+ζ − ξ 1 − τ /m Φ −√ m 1 − τ /m + C · 4ξ 2 √ Rm ϕ + (2 − 2ξ Rm / m) √ m−τ 2 Rm Rm ϕ √ m m−τ + D · 4ξ 2 √ m + e−2ξ Rm / √ y+ζ + ξ 1 − τ /m 1 − τ /m y+ζ + ξ 1 − τ /m √ 1 − τ /m 2 2 Rm 1 Rm 1 ϕ (η1 ) − ϕ (η2 ). Because Φ(−x) = 1 − Φ(x) and ϕ(−x) = ϕ(x). P.

There is a finite constant M such that for all m ∞ sup 0≤t<m √ [log(K /S0 )]/σ T B(t.A CONTINUITY CORRECTION FOR DISCRETE BARRIER OPTIONS 345 With A(τ . In particular. We will also use the easily verified identity ∞ (B. We begin by recording two useful facts. for all q > 0. y) and B(τ .1). y) + √ B(τ .2) ecy ϕ(u + y) dy = ec(c+2u)/2 Φ(−b + c + u). . y) m m 2 2 √ Rm 1 Rm 1 ϕ (η1 ) − ϕ (η2 ). y) m 3 √ Rm Rm + (2 − 2ξ Rm / m) √ B(τ . From Lemma 10.1) q lim E θ1 [Rm ] = m→∞ y q d B R (y).11 of Siegmund (1985) we get convergence of all moments of the overshoot to moments of B R : (B. + e−2ξ Rm / m m−τ 2 m−τ 2 Grouping terms according to their denominators. y) + C · 4ξ 2 m A(τ . y) √ m m 1 1 2 2 + C · 4ξ 2 Rm A(τ . b We now prove LEMMA B. this becomes √ R2 Y = (−2ξ Rm / m)A(τ . APPENDIX B: UNIFORM INTEGRABILITY The main objective of this appendix is to prove Proposition 4. y) + D · 4ξ 2 √ B(τ . y)e yσ √ T dy ≤ M. E θ1 [Rm ] → β.1. y) as defined in the theorem. y) − 2ξ Rm B(τ .1. y) m m 1 2 + √ D · 4ξ 2 Rm B(τ . we get Y = −2ξ Rm 2Rm A(τ . this concludes the proof of the theorem. + e−2ξ Rm / m m−τ 2 m−τ 2 Via (A. y) m m m 2 2 √ Rm 1 Rm 1 ϕ (η1 ) − ϕ (η2 ).

4) are nonnegative. y)e yσ √ √ T dy T dy.2). Using first the definition of B(t. P. Just as in Lemma B.1. we may therefore integrate with respect to the limiting distribution to get 1 β 0 ∞ √ [log(K /S0 )]/σ T √ 1 1−t ϕ √ √ ζ+y + ξ 1 − t e yσ T dy g(t) dt. √ 1−t Again interchanging the order of integration yields (4.3). y)e yσ √ [log(K /S0 )]/σ T ∞ T dy √ ζ+y dy + ξ 1 − t/m e yσ T √ 1 − t/m 1 − t/m [log(K /S0 )]/σ T √ 1 √ 2ζ = exp σ T 1 − t/m σ T 1 − t/m + 2ξ 1 − t/m + √ 2 1 − t/m √ ζ log(K /S0 ) + σ T 1 − t/m + ξ 1 − t/m + √ ×Φ − √ √ 1 − t/m σ T 1 − t/m = √ ϕ √ . APPENDIX C: PROOF OF PROPOSITION 4.1) the entire expression inside the expectation is uniformly integrable. Because all quantities inside the integral and expectation on the left side of (4. But then by (B. and this is bounded for 0 ≤ t < m. This proves uniform integrability which.4). together with two applications of Tonelli’s Theorem yields (4. by Tonelli’s Theorem we may interchange the integral and expectation to get ∞ √ [log(K /S0 )]/σ T E θ1 [Rm B(τ . AND S.3) is similar. The proof of (4. To evaluate the limit as m → ∞. y)e yσ √ T dy [log(K /S0 )]/σ T is uniformly bounded for 0 ≤ t < m.1. we get ∞ √ B(t. first considering (4.1.2 We begin with the following bound on the integral inside the expectation: .346 M. τ < m . By Lemma B. y). GLASSERMAN. BROADIE. τ < m]e yσ ∞ = E θ1 R m √ [log(K /S0 )]/σ T B(τ . We proceed with the proof of Proposition 4. KOU Proof. y) and then (B.4). we integrate to verify that ∞ √ A(t. the integral inside the expectation is uniformly bounded.

ν √ Apply this identity to the integral in the lemma by setting b = log(K /S0 )/(σ T ). by the argument of Example 2. √ √ m 1 − t/m This results in a bound of the form in the lemma. . 2. M2 .2 will be complete once we verify that √ lim sup E θ1 [exp(M3 Rm / m)] is finite as m → ∞. and all u ≥ 0. with Rm . Since 1 − Φ(x) ≤ x e−x /2 ≤ e−x for all x > 0. c = √ √ √ σ T . and u = ζ (1 − t/m)−1/2 + ξ 1 − t/m + di . Proof. 19 of Woodroofe (1982) we have 2 Pθ1 (Rm > u) = ≤ ∞ n=1 ∞ Pθ1 (τ ≥ n. 347 There are finite constants M1 . d2 ∈ 0. The proof of Proposition 4.2 on p.0 d1 ∈ − √ √ m 1 − t/m Rm .A CONTINUITY CORRECTION FOR DISCRETE BARRIER OPTIONS LEMMA C. τ ≥ n] n=1 ∞ = e−u E θ1 [e−(b−Un−1 ) . For any ν > 0. integration by parts shows that ∞ b 1 c 1c ecy ϕ (u + νy) dy = − ecb ϕ(u + νb) − 2 exp ν ν 2ν c + 2u ν Φ −νb + c +u . 2. τ ≥ n] n=1 −u ≡ M4 e .1. for all m 1 Proof. and M3 for which √ ∞ sup 0≤t<m √ [log(K /S0 )]/σ T √ e yσ T ϕ (ηi ) √ dy ≤ M1 + M2 e M3 Rm / m 1 − t/m for i = 1.2. For this we need LEMMA C. ν = 1/ 1 − t/m. There is a finite constant M4 such that Pθ1 (Rm > u) ≤ M4 e−u . Un > b + u) E θ1 [e−(b+u−Un−1 ) . i = 1.

CHERNOFF. Political Economy. and H. and H. BROADIE. H. 25–59. P. 36. F. 46–49. (1979): “Corrected Diffusion Approximations in Certain Random Walk Problems...” J. KAT.. D.” Scand.. (1989): “Risk Theory in a Markovian Environment. (1965): “Sequential Tests for the Mean of a Normal Distribution IV.. 7. SIEGMUND. M.” Doctoral dissertation. 637–659. 8. J. 141–183. Department of Finance. KOU (1996): “Connecting Discrete and Continuous PathDependent Options. RISK. Department of Operations Research. H. N. PETKAU (1986): “Numerical Solutions for Bayes Sequential Decision Problems. 4. RUBINSTEIN.. 7 (June). P. RICH.. SCHOLES (1973): “The Pricing of Options and Corporate Liabilities. (March) 18. GLASSERMAN.. RICH. BOYLE. GLASSERMAN. Res.” J. HEYNEN. 1. and S.. KAT (1994b): “Partial Barrier Options.. 7. 28–35. (1973): “Theory of Rational Option Pricing. Champaign.. LAU (1994): “Bumping Up Against the Barrier with the Binomial Method. KOU It follows now that √ E θ1 [e M3 Rm / m ∞ ] ≤ M4 √ m −r e M3 r/ e dr. I. Derivatives. 17. 53–56. Correction (1995).” Ann. (1992): “The Design and Valuation of Capped Stock Index Options. B.” J..” RISK. Columbia Business School. FLESAKER. IKEDA (1992): “Pricing Options with Curved Boundaries. SHREVE (1991): Brownian Motion and Stochastic Calculus. Econ. 7.” Bell J. . AND S.” Adv. Financial Engineering. to appear in Finance and Stochastics. M. D. BLACK. Financial Res. Derivatives Res.” working paper. KARATZAS. C. HEYNEN.” SIAM J. H. Futures Opt. P. (1996): “The Valuation of Black-Scholes Options Subject to Intertemporal Default Risk. and M. 561–584. Stat. 8 (February).” May 29. H. (1994): “The Mathematical Foundations of Barrier Option Pricing Theory.. C. 46–59. Probab. Sci. New York: Springer-Verlag.. S. Derivatives Week (1995b): “Liquidity Standards to be Drawn up for Barrier Options. CA. and A. D. 2. 267–312. BROADIE.. KUNITOMO. Stanford. MERTON. 0 which remains bounded as m → ∞. and L. 2. 2. and S. and S. M. R.348 M. 275–298.” J. 6–14. Stanford University. and E. ASMUSSEN. Actuarial J. 11. and M. REFERENCES AITSAHLIA. Comput. Statist. (1994): “The Pricing and Hedging of Limited Exercise Caps and Spreads. Derivatives Week (1995a): “Discrete Path-Dependent Options. P.. CHERNOFF. Mgmt. University of Illinois. F. R. R. M.” working paper. New York. 701–719.. (1995): “Optimal Stopping and Weak Convergence Methods for Some Problems in Financial Economics. C. REINER (1991): “Breaking Down the Barriers. M.. 3 (September/December). 253–274.” April 10. Finance. CHANCE. 16. 4 (September). VERDONK (1995): “Tree Surgery.” Adv. 69–100.” RISK. 81. D. Appl. KAT (1994a): “Crossing Barriers. 55–68.” Rev. Math.” Math. IL.” RISK. Sci.

WOODROOFE.-S. M. Wahrsch. . (1982): Nonlinear Renewal Theory in Sequential Analysis. D. SIEGMUND.A CONTINUITY CORRECTION FOR DISCRETE BARRIER OPTIONS 349 SIEGMUND. YUH (1982): “Brownian Approximations for First Passage Probabilities. verw.” Z. 239–248. (1985): Sequential Analysis: Tests and Confidence Intervals. New York: SpringerVerlag. D. and Y. 59. Philadelphia: Society for Industrial and Applied Mathematics. Gebiete..