American Finance Association

Options Arbitrage in Imperfect Markets Author(s): Stephen Figlewski Source: The Journal of Finance, Vol. 44, No. 5 (Dec., 1989), pp. 1289-1311 Published by: Blackwell Publishing for the American Finance Association Stable URL: http://www.jstor.org/stable/2328643 Accessed: 22/10/2009 05:25
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=black. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Blackwell Publishing and American Finance Association are collaborating with JSTOR to digitize, preserve and extend access to The Journal of Finance.

http://www.jstor.org

THE JOURNAL OF FINANCE * VOL. XLIV, NO. 5 * DECEMBER 1989

Options Arbitrage in Imperfect Markets
STEPHEN FIGLEWSKI* ABSTRACT strategy-hedging the option against Optionvaluationmodelsarebasedon an arbitrage continuouslyuntil expiration-that is only possible asset and rebalancing the underlying in a frictionlessmarket.This paper simulatesthe impact of marketimperfectionsand other problems with the "standard"arbitrage trade, including uncertain volatility, transactionscosts, indivisibilities,and rebalancingonly at discrete intervals. We find that, in an actual marketsuch as that for stock index options,the standardarbitrageis exposed to such large risk and transactionscosts that it can only establish very wide options prices. This has importantimplicationsfor price deterboundson equilibrium minationin options markets,as well as for testing of valuationmodels. AMONG ALL THEORIES IN finance, the Black-Scholes option pricing model has perhapshad the biggest impact on the real world of securities trading. Virtually all marketparticipantsare awareof the modeland use it in their decision making. Academics regularly test the model's valuation on actual market prices and typically concludethat, while not every featureis accountedfor, the modelworks very well in explaining observedoption prices.' Most option valuation models are based on an arbitrageargument.Under the assumptionsof the model, the option can be combinedwith the underlyingasset into a hedged position that is riskless for local changes in the asset's price and in time and must therefore earn the riskless interest rate. This leads to a theoreticalvalue for the option such that profitablearbitrageis ruled out. However,while virtually all options traders are awareof option pricing theory and most use it in some way, the arbitragemechanism assumed in deriving the theory cannot work in a real options market in the same way that it does in a frictionless market. The disparity between options arbitrage in theory and in practice is the subject of this paper. Some of the importantassumptionsmade in derivingthe Black-Scholes model are the following. * The price of the underlyingasset follows a logarithmicdiffusionprocess that can be written

dP/P = R dt + v dz,

(1)

where R is the drift of the price per unit time, dt denotes an infinitesimal
* Stern School of Business, New York University.The authorwouldlike to thank John Merrick, Roni Michaely,William Silber,and the referee,MarkRubinstein,for helpfulcomments. 1 Empiricalstudies of the option pricing model include Black and Scholes (1972), Galai (1977), and Macbeth and Merville (1979), among many others. Galai (1983b) provides a review of the literatureon testing option models.

1289

actual volatility. and dz represents Brownian motion. the realization of a random variable distributedas normal with mean 0 dt and variance 1 dt.Boyle and Emanuel(1980) observethat the probabilitydistributionof requiredfor empiricaltests on options. unknown volatility affects both the return and the risk in options arbitrage. at Galai (1983a) finds that there is little impact on the mean return earnedby a hedgedposition but that its varianceis increased. Volatilityestimates impliedby the option pricing modelalso are'highlyvariable. prices can make nonlocal changes from one trade to the next with no possibility of rebalancinga hedgedportfolio in between. There are no transactions costs. forminga riskless hedge. Securities are also indivisible.2 One of the biggest problems in real world options trading is determiningthe volatility of the underlyingasset's price. i. hedgereturnsis affected. Most important is the error in evaluating the fair price for the option. in theory.including Black (1976). Thus. Beckers (1981). The arbitrage position will not earn its excess return risklessly if an incorrect hedge ratio is used. In fact. Too low (high) a volatility estimate gives a model value that is also too low (high).An importantassumptionneededfor this result is that aggregation investor. so that the marketbehavesas if there were a single "representative" 'Time variationin volatility of stock prices has been discussedby a numberof authors.the realizedvolatility in a given (finite) series of prices will differ from the true value due to samplingvariation.both Brennan (1979) and Rubinstein (1976) show that continuousrebalancingis not of necessary for the Black-Scholes model to hold.and holding until option expiration. Between trading sessions.even if the true value for volatility to vary randomlyover a wide range. However. With the right comnbination security price optionpricewill be the Black-Scholesvalue and distribution investorutility function.e.and thereforeso is the methodology Interestingly. and the arbitrageby which the theoretical pricing relation is supposed to be enforced. . For example.prices do not follow a continuous diffusion when the market is closed.such a trade is complicated by the fact that the unknown volatility is also a determinantof the hedge ratio. This is not a known constant parameter. form a fully hedged position earning a return higher than the riskless rate. and for options that are not deep-in-the-moneythe price is quite sensitive to small changes in the volatility parameter.1290 The Journal of Finance * * * * time increment.the equilibrium is even if rebalancing impossible.. appear Moreover. and also the market's volatility estimate. and there are no taxes or constraints on short selling with full use of the proceeds. for 2Rebalancing discreteintervalsratherthan continuouslyhas been examinedby severalauthors.v is the volatility of proportionalprice changeper unit time. There is no limit on borrowing or lending at the same riskless interest rate. nor is there even general agreementon the best procedurefor estimating it. rebalancingcontinuously. There are no indivisibilities. Most traders would prefer to trade stock.3 is known. An investor who has a more accurate volatility estimate than the market can. On the contrary. The volatility v is known. Errors in predicting volatility lead to two kinds of errors in trading options. none of these assumptions is true of real financial markets. conditionshold. and Christie(1982).cannot actually be done with real options.as shown by Latane and Rendleman(1976) and Rubinstein(1985). Markets are "perfect"in other ways.

84 by selling one. an application of option arbitrage. and so forth. two.That is a function of the actual path taken by prices.for the most part. The researchercannot know. For some cases this is possible. short sales with full use of the proceeds. Arbitragerelationships that hold in theory are always affected by transactions costs in practice. and Leland (1985) is able to make some headway in determining theoretically the effect of proportional transactions costs. or three contracts. suppose stock index options with market exposure equal to $500.Options Arbitrage in Imperfect Markets 1291 example.4 Arbitragebounds on options prices cannot be easily computed.the total transaction cost in a particular arbitragetrade will dependon how much the position has to be rebalanced.on historical data with realistic transactions costs and rebalancing. the mathematicalproblemsraised by treating realistic market imperfectionsare too complex to be tractable theoretically.28. The researcheris limited to examining a single set of data that may not be very long and over which he or she has no control." It is obvious that the other perfect markets assumptions.Phillips and Smith (1980) documentthe costs of setting up and unwindingan options hedge and then show that these outweighthe possibleprofits uncoveredby many earlierstudies.Because of the dynamicnature of the hedgingstrategy.the indivisibility of the futures contract will lead to hedging inaccuracy. .56. absence of taxes. A second solution is to simulate trading strategies on historical option price data and to tabulate the results. because contract size is large.5 Analyzing historical data has always been the standardapproachfor testing option models. For instance.but they also find that these potential profit opportunities disappearwhen some estimate of the transactionscosts involved is considered.They find substantialdeviationsbetweenactual results and theoreticalestimates. do not hold any better in real markets than the ones we have alreadymentioned. 0. such as unlimited borrowingat the riskless interest rate. How can we find out how much options arbitrage is affected by market imperfections?One possibility would be to attempt to incorporatethe imperfections directly into our theoretical valuation models. Obviously.000. This allows one to construct only positions with hedge ratios of 0. The effect of indivisibilities is greaterwhen futures contracts are used to hedge an options position. the value of one S&P contract is about $140. so the trader cannot know in advance how large these costs will be. for example. what the true ex ante distributionfor prices was.transactions costs create bounds aroundthe theoretical price within which the market price may fall without giving rise to a profitablearbitrageopportunitylarge enough to cover the cost of exploiting it. However. but there are several problemswith it. 5 Garcia and Gould (1987) simulate the performanceof portfolio insurance. Broadly speaking. Nor can a researchertesting an option pricing model on market data know how big a deviation has to be before it representsa large enough after-cost profit to be a true "mispricing. Other difficulties 4Most published empiricaltests of option pricing models find some mispricingin the market relative to the models' prescriptions.000worth of stock are to be hedged using Standardand Poor's 500 futures contracts.For example.in roundlots of 100 shares. At current (January 1989) prices.we can computethe amount of mispricingthat arises when the wrong volatility estimate is used. of Probably the most important "imperfection" real financial markets is the existence of transactions costs. or 0.

e. ' In an early paper.1292 The Journal of Finance are that actual volatilities change over time. Etzioni (1986)uses a simulationstrategyto examine alternativerebalancing portfolioinsurance. Thus.6This procedurehas several things to commend it. . Section II looks at other market imperfectionswhich predominantlyaffect the level of risk in a hedged option position. First. These are the use of an incorrectvolatility estimate in computing the hedge ratio and indivisibilities. and construct 250 randomlydrawnprice series and then do Monte Carlo simulations to determine the effect of some of the market imperfectionsdiscussed above. Second. parametersof the price-generating therefore. In Section V we present some results and discussion relating the analysis to longer maturity options and to other option replication strategies.We find that hedging an option with the underlying asset dynamically and rebalancingthe position once a day until expirationwould be exposed to such largetransactionscosts and risk in actual markets that it is impracticaleven for an options market maker. market prices may at times be distinctly nonlognormal. but risk increases. The simulations we look at in the paper cover the standard arbitragetrade with one-month options. Rebalancing less frequently can reduce costs. are directly comparableto those we would obtain by using the same parametersin a correctlyspecifiedtheoretical model. it is relatively easy to do. These results will show that rebalancing the hedge daily rather than continuously has a considerable impact on its risk. realized volatilities may differ considerablyfrom the ex ante values. "standard" This has important implications for price determinationin options markets. Section III analyzes transactions costs. Then we examine summary statistics on the realized returns and risk on the securities in the sample. such as portfolio insurance and the trading of actual market makers. the arbitragecan establish only very wide bounds on real option prices. The results we obtain empirically. Boyle (1977) proposes Monte Carlo simulation as a procedurefor valuing proceduresfor options. Third. and how the hedged positions are constructed. like other numerical methods. We compute the after'-costsreturns and risk on hedgedpositions using approximatelythe transactions cost structure that currentlyappliesto marketmakersand retail tradersin stock index options. In Section IV we consider the arbitragebounds that this cost structurewould imply and comparethem to typical marketbid-ask spreads. we know exactly what the true process are. as well as for testing valuation models. creating) more observations. R and v. The approach we will take in this paper is to simulate the performanceof options hedge strategies on simulatedprice data. The final section summarizesour findings in more detail.and so on.We will discuss the particularcase of portfolioinsurancein more detail below. In the next section we describe the experimental design.. our simulations can be made arbitrarilyaccurate simply by using (i. We also look at the performanceof alternative trading strategies designed to reducecosts by rebalancingless frequently. how the prices are generated. We specify values for the drift and volatility.

Table I shows summarystatistics for the constructedprice series in our sample. an at-themoney. and the mean terminal price was 101.0 percent and the volatility is taken to be the true volatility.with a standarddeviation across series of 4. and z is a randomdrawfrom a standardizednormalprobabilitydistribution. the strategy of buying the option.8Notice that. for call options with four different strike prices: 97. R = (log (1. v is the daily volatility.43. For this study we have set R and v to be the daily equivalentof an annual 15% and 0. calculated from 100 x (P24/Po . In pricingthe options. . timevarying value of the portfolio. so the typical month has 21 to 22 trading days.00930.85.1) and annualized at simple interest. For each series. 100. in doing this. correspondingseries of option prices and theoretical hedge ratios were constructedusing the Black-Scholes model. Next. 8There are between 250 and 260 trading days per year. The standarddeviation was annualizedby multiplyingby the squareroot of 260/24. rebalancing continuously.5 v = 0. since the value of the funds invested in the portfolio changes as it is rebalanced. This correspondsto a hedge period of about one month.15. and 105. Two hundredfifty price series of 25 observationseach (indexed as t = o. the riskless interest rate is assumed to be 5.15 at an annual rate. 7Note that. 0. This process for prices is implied by the assumed returns equation (1): Pt+1 = PteR+vz. (2) where R is the mean rate of price change per day."but our results will of course apply to other kinds of options as well. The mean of the annualizedpercentagerate of return was 15. earning the riskless rate implies earning a continuous cash flow at that rate on the current. Throughoutthe paper we use parametervalues that have been typical for actual options on broad-based stock indices.15))/260 = 0. .000538..until expirationwill returnexactly the riskless rate of interest.Options Arbitrage in Imperfect Markets 1293 I. The first line describesthe stock price series. Experimental Design If an option is priced at its arbitrage-based value in the market. 103. forming the hedged portfolio. each starting at the initial value of Po = 100. 24) were constructed. and carrying it...we begin by constructinga set of price series and option values to use as the basic data in the hedge tests.15/2600 = 0. We will use 260 in converting between daily and annualized parameter values. subsequent prices are computed according to equation (2). The initial price was 100 for each series. we have already departed from a world in which continuous rebalancing is possible.and a deep-out-of-the-moneyoption. This gives us an in-the-money. an out-of-the-money. The choice of parameterssuch as volatility is open. The hedge is rebalancedat most once a day. For convenience we will often refer to the underlyingasset as the "stock.52.That is.7To see how this strategy behaves with market imperfectionssuch as are present in actual options markets. respectively.

Amount 204 152 92 60 6. Table II shows the trading strategies and expected arbitrage profits that a trader would be able to earn under the perfect markets assumptions of the BlackScholes model.45. the trader would value the 100 strike price calls at 2. Knowing that the true price volatility of the underlying asset was 0. It is very interesting to see that the standard deviation of the realized volatilities across series was as high as 0.13 3. Dev. What this shows is that. Returns are annualized by multiplying by 260/24.150 Volatility Std.9 To options market makers.31.15 and riskless interest of 5. Its endof-period value averaged 2. the distribution of these realized volatilities is very close to Gaussian in our sample. the distribution of these values is highly skewed: of 250 options. The indicated arbitrage would then be to buy the underpriced calls and 'Due to the effect of the Central Limit Theorem. there is a sizable standard error of forecast in predicting what volatility will actually be experienced in daily prices over a month.05. Dev. In other words. knowing that the true volatility is 0. .31 2.1294 The Journal of Finance Table I Summary Statistics for Simulation Data Sample The table shows summary statistics from 250 simulated series of 25 daily prices each. Dev. 0. 152 ended up in the money. This suggests that the series are representative. Of course.02 4.022 Mean 101. Stock prices are generated with an annual drift of 15% and volatility of 0.10.15. the at-the-money call was priced initially at 2. with a standard deviation of 3.91 2.41 __________ Mean 4. Stock . If the market's volatility estimate was 0.72 Within each series of 25 prices. assuming that the market had mispriced these options." For example.84 0.15 only tells us that there is about 2/3 probability that a given month's volatility will be somewhere between 0.40 2.150. Dev.128 and 0. The mean of these sample volatilities in our constructed data turned out to be 0. The second part of the table summarizes the returns to buying call options "naked.75 1.52 3.172.01 2.15.52. the standard deviation of the log price relatives produces a realized volatility. by an average amount of 4. assuming volatility of 0.05 0.43 Initial Strike 97 100 103 105 Price 4. Initil Initial Price 100 _________ Final Price Std. Consider the at-the-money calls.25 0. even if the true volatility is known.85 _________ Mean 15.022.75. The remainder expired worthless. Option prices are computed from the underlying stock prices using the Black-Scholes model. 4. "Mean" and "Std Dev" figures refer to sample means and standard deviations across the 250 series.95 _________ Mean 0.05. a difference of that size between volatility estimates is considered very large. volatilities by (260/24)05. 15.0 percent. those calls would be selling for 1.52 Calls Percent Return Std.65 Final Value Std. 4.63 In the Money ____________ Number Avg. as it should.29 1.

46 0.35 0.04 -53.33 74.299 0. if the market'svolatility estimate were 0. the market option price.184 0.10. a reductionin the cost of obtaining funds) of $0.878 0.163 0.29 Excess Return .00 2.13 0. Regardlessof the actual course of prices duringthe 25-day hedge period.85 28. and hedge ratio and analyzes the arbitrage position that would be taken.19 -29.241 0.90 -74. As time elapsed and the stock price changed. Over time this would earn the riskless rate plus the initial option overpricingof $0.24 0.602.385 0.67 9.27 29.309 0.39 -53.18 percent.65 0. The initial cost of this position would be negative.17 0. One is that the dollar value of the arbitrage portfolio tends to be large compared to the .565 0. always computing the new hedge ratio using 0.759 0.10 0.87 16.01 4.64 0. As a percentageof the initial value of the hedge portfolio.000 0.602 0.490 0.81 19.206 0.13 0.96 Sell Buy Sell Buy No Trade Sell Buy Sell Buy Sell Buy Buy Sell No Trade Sell Buy Sell Buy Sell Buy Buy Sell No Trade Sell Buy Sell Buy Sell Buy Buy Sell No Trade Sell Buy Sell Buy 100 103 105 short the -stock using a hedge ratio of 0."earningthe riskless rate of return"would correspondto a continuous loss equalto payingthe riskless interest rate on the net amountof funds remaining in the position.269 0.95 7.06 20.00 4.69 28.144 0.203 0.19.48 1..817 0.478 0.99 12.20 instead of 0.548 0.15 0. buy stock.57 0.542 0. this would be an annualized 12.17 0.84 4.63 0.15 0.e.15.this position wouldbe rebalanced by adjusting the amount of stock sold short.17 0.351 0. 0.68 6.728 0.03 -16. and create a hedgedportfolio costing 52.170 0. Since the trade brings in cash at the beginning.20 Strategy Call Price 3.27 0.18 4.84 1.209 0.19 4.56 -29.10 0.20 0.20 0._ _ $ amt.548 0. On the other hand.05 2.Options Arbitragq in Imperfect Markets Table II 1295 is 0.435 Annual % 5.075 0.06 1.00 8.39 0.56 16.600 0.57 2.81 2.65. Each line gives the market's volatility. the arbitrageur wouldwrite calls at 2.84 Hedge Ratio 0. Trading Strike Price 97 Market's Volatility Estimate 0.07 28.00 10.62 3. meaning there would be a net cash inflow of 53.10 0.785 0.91 0.10 0. (i.39.000 0.240 0.20 0.50 percent.548 shares per call.135 0.10 0.000 0.215 0. The table shows two important propertiesof this arbitragetrade.96 12.15 Theoretical Arbitrage Profit When True Volatility The table shows the standard option arbitrage trade from the perspective of an arbitrageur who believes the true underlying volatility to be 0.41 0.50 17.52 -17.17 0.000 0.13 0.545 0.553 0.323 0. the initial mispricingof the option would yield an excess return.52 74.55 52.45 1.242 Call Stock Cost of Riskless Position -74.69 74.79 52.600.171 0.03 52.15 0.15 0.13 0.29 2.72 16.15 as the volatility estimate. The excess return would be 12.

The figuresare shown as annualized percentage rates. P is the price of the underlying asset. (3) where ER is the excess return. In every case we consider the arbitragetrade of buying a call option at the marketprice and selling the number of shares indicated by the hedge ratio.) II. The mean return is also nonzero in the sample but not statistically significant. each day the hedge ratio is recalculated and the hedged portfolio is rebalanced by buying or selling the underlyingasset. Also apparent in these figures is the fact that using an incorrect volatility estimate makes a bigger difference in the value of a call than in its hedge ratio. As a base for comparison. one from each price series. it is not the Black-Scholescase exactly. It does. In that case. (It should be understoodthat these properties come from the Black-Scholes equation-they are not producedby our simulation.daily rebalancingleads to a risky hedge whose annualized rate of return over the holding period has a standarddeviation of 6.Ct)ht-(Pt-Pt-1) - r(Ct1 - ht1Pt-1).52 percent. By far the largest impact is on the deep-out-of-the-moneycalls when the implied volatility is too low. and r is the one-day riskless interest rate.548 shares per option would yield zero excess return with a standard deviation of zero if it were possible to rebalance the position continuously.00 percent. Thereafter. for each combination of hedge strategy and strike price. The first line in Table III gives the Base Case results on the standarddeviation of excess returnsacross the 250 price series.is hardly affected at all by changes in volatility over a wide range. Market Imperfections and Risk This section begins to examine hedge strategies. in particular. since the position is rebalancedonly once a day. The sample mean and standarddeviation statistics for the excess return totals show how introducing different market imperfections into the system affects the expected return and the risk of an options hedge. This problem is less severe for the out-of-the-moneycalls. In the examplewe just described. The day's excess return is calculatedusing equation (3): ERt = (Ct . h is the hedge ratio called for by the particulartrading strategy. however.it is worth reflecting briefly . we begin by analyzing a "Base Case" without imperfections.leadingto 250 total excess return amounts. The delta for the at-the-money calls. the market price and the delta both are much too close to zero.) Before going on to look at market imperfections. or delta. The daily excess return figures are then cumulated. we do not reportthem. correspondto the typical methodologyused in empiricaltests of option pricing models.As mentionedabove. C is the call price. (Since none of the means for the strategiesexaminedin Table III was significantly different from zero. based on an annual rate of 5.it was necessary to take a position in over 50 dollars worth of securities and to be prepared to manage the position carefully for a month in orderto earn an excess return of 60 cents. Thus.1296 The Journal of Finance amount of mispricing. although buying the 100 strike calls at the theoretical value of 2.even for a large difference in volatility estimates.05 and selling short 0.

05 19. if the volatility is 0.85 8. simply by rebalancing discretelyinstead of continuously.we saw in Table II that.10 K = 0.Options Arbitrage in Imperfect Markets Table III 1297 Effects of Market Imperfections on Hedge Standard Deviation (Annualized Percent Standard Deviation) The table shows the standarddeviation across 250 price series of the annualized cumulativeexcess returnsto option expirationon hedgedpositions that are long the call option and short the underlyingstock. Supposethe traderuses an incorrectvolatility estimate of 0. In the second section of Table III we look at hedgingwhen the volatility of the underlyingasset is not known.05 3.46 16.64 17.88 16.10 12.15. The riskless interest rate has been assumedto be five percent.31 24.62 6.1 3. but the realized prices during the option's lifetime could have a samplevolatility of 0.34 3.70 6. he or she would have sold it for 0. If the trader were to write the option at that price and hedge it by buying the stock. For example.52 X=103 11. The Base Case uses the exact hedge ratio computedfrom the true volatility of 0.02 K = 0.82 5. .45.52 6. the true value of the 100 strike call is 2.17 0.01 Indivisibilities K = 0.38 16.we have departedmarkedlyfrom the theoretical world of Black-Scholes.18 10.the realizedvolatility over any finite time interval must equal the true volatility with probability1.10.15.15 16.26 on the meaning of these standarddeviations.25 K= 0.a standarddeviation of more than six percent makes the position quite risky. In comparison. Indivisibilitiesresults show the effect of roundingthe correct(v = 0. both the expected return and the risk of the trader's arbitrage portfolio are affected.56 11. such as in the dailypriceserieswe are considering.If pricesgeneratedby a diffusionprocesscouldbe followed continuously.68 7.20 4.10 v= v= v= v= 0.09 19.25 6.42 3. When a trader uses a volatility estimate that is not equalto the volatility that is actually experiencedduringthe option's lifetime.06 1680.15.91 16.68 4.38 11.or it might be the stock's true ex ante volatility.82 9. A perfect hedge would then lock that mispricingin as a certain loss on the trader's position. It is apparentthat. Positions are rebalanced daily.70 IncorrectVolatility v = 0.15) hedge ratio to the nearest integermultipleof K.59 3.18 6. X=97 Base Case 3.84 24.13 0.07 13.10 volatility is only 1.84 7.80 64.76 11.64 11. Incorrect Volatility results show the effect of computingthe hedge ratio with an incorrectv. (This could be due to an incorrect estimation procedure.25 11.60 below its true value.18 X= 105 16.70 14. The second problem caused by an inaccurate volatility estimate is that the 10 Note that the samplevolatility can differ from the true volatility only in discretetime.34 X= 100 6.89 3.68 K = 0.)lo The call value at a 0.0.15 0.05.80 2868.

10 would also induce a standarddeviation in the annualizedhedge return of 8. With option contracts on N units of the underlyingasset. correct hedging (i.for the out-of-the-moneyand especially the deep-out-of-the-moneycalls. Regardlessof the delta producedby the valuation model. the impact is substantial.and at-the-money calls. dependingon whether the theoretical delta is less than or greater than 0. the less accurate the hedge can be because of indivisibility). the initial "hedged" position contains only the naked call becausethe deltas are both below 0. These results suggest that. The standarddeviation is expressedhere as a percentageof the initial .5..15) would lead to (approximately)a 0. However. For the out-of-the-moneyoptions. But by the time K is 0.it is interesting that the effect of roundingthe hedge ratio to the nearest K is not very great for the out-of-the-moneyoptions even though a given K leads to a relativelylargerinaccuracyfor them due to their smaller deltas. it might be appropriateto compute the hedge ratio for out-of-the-moneyoptions using a higher volatility than what the traderexpects to prevail in the future.5. using v = 0. For the most part.18 percent. Table III shows the standard deviation of hedge returns for several values of K. incorrecthedgingwith a volatility of 0. the hedge ratio can take only values of zero or one. the standarddeviation of returns on a hedged position does not increase much comparedto the risk level that is alreadypresent in the Base Case. However. Hedging with too high a volatility estimate does not seem to increase hedge risk much at all. in trying to cope with uncertainty about the volatility. The other imperfection we consider in Table III is the indivisibility of the underlying asset. the hedge ratio cannot be set to the exact value dictated by the valuation model and the position will necessarily be slightly over or under hedged at all times.5 and remain unhedgedif it were less than 0. There is also an interesting asymmetrybetween overestimatingand underestimatingthe volatility.1298 The Journal of Finance hedge ratio used in forming the arbitrageportfolio will be wrong. depending on whether a share is sold. The impact on hedge risk is substantialfor in. When the number of options to be traded is small or the underlyingasset is like a futures contract that must be traded in large units.e. Consider a hedged position consisting of an option on a single share of stock. This is close to a strategy of hedgingonly options that are in the money and leaving out-of-the-money options unhedged.25.e. the hedge ratio can only take on values that are an integral multiple of K = 1/N. there is a sizable degradationin the effectiveness of the hedge for all but the 105 strike calls. the results bear out the expectation that the largerthe value of K (i. How much additionalrisk does this cause? The third section of Table III shows the effect on hedge risk when the set of possible hedge ratios is constrainedby the indivisibility of the underlyingasset.60 loss.5. In this case. the greater will be the risk. However. on the groundsthat it is less costly to erron the side of overestimatingthan underestimatingvolatility for these options..0). The hedger attempting to use the Black-Scholes model in this case might sell a share if the delta were greaterthan 0. The final line in Table III shows the strategyof either no hedge or a full hedge (h = 0 or 1. For the in-the-money and the at-the-money options. underestimating the volatility leads to a considerablylarger standard deviation.

The cost is $8 per contract plus 1.6. A similar fee must be paid to exercise options expiring in the money. we assume that three option contracts are traded.) The hedge position in the stock that remains on the expiration day is liquidated in the stock market. hence. Again.05 and to sell 0. There is no chargefor exercise of options finishing in the money.As before. and it is assumed that the market maker tradesoptions without having to pay the bid-ask spread. We then look at alternativestrategies designedto limit transactions costs by reducingthe numberof rebalancingtransactions. which is very small. Transactions Costs We now turn our attention to the impact of transactions costs on options arbitrage. so that one does not have additional costs to dispose of stock acquiredthrough exercise. At expiration. The complication is that the transactions costs that must be paid in hedging an options position depend on the realizedpath taken by prices. (We assume cash settlement. and they depend on the brokerage firm (whether a "discount"or a "full service" broker) as well as on the size of the trade (with discounts for largervolume).548 shares short.05 per share plus one half of the bid-ask spread.64 (per call option on one share).5% of the dollar amount of the transaction. on . We will look at two cost structures. which we assume to be 1/8.but transactionscosts unambiguously reduce the profitability of every trading strategy.Transactions costs are paid on both the initial sale and subsequentrepurchaseof the shares. The commission on a stock trade of $35 plus 0. options finishing in the money are exercised and the remainderexpire worthless. $0. The arbitragetrade is to buy the call at its theoretical value of 2. Since retail customers pay commissions that vary with quantity.one to corresponding the trading costs borne by a typical options marketmaker. III. The mean numberof stock trades across all 250 series was 24. An options market maker is assumed to pay an exchange fee of $1 per option contracttraded (i. Table IV shows the effect of transactions costs on the option hedges we have been considering.01 per underlyingshare).e. Thus.Options Arbitrage in Imperfect Markets 1299 position value. rebalancingdaily.These are representativeof the costs applying to trading in stock index options in early 1987.The market imperfections discussed in the last section induced risk but no bias towardhigheror lowerreturns. Considerthe Base Case results for the 100 strike calls. commissions are paid on both opening and closing trades. in particular.5%of the dollar amount traded. Commissionspaid by retail investors. the market maker pays $0.we begin with an analysis of the Base Case.For hedgingtransactions in the underlyingstock. the numbersin the table become very large. calls on 300 shares.are much larger than those paid by market makers..and the other to the costs that wouldbe paid by a retail traderdealingwith a discount broker.that is. The transactionscost structurealso varies considerablyamongdifferentclasses of traders.plus half of the bid-ask spreadof 1/8. and the mean total numberof shares traded was 2.

7 1.113 0.607 0.043 0.220 -0.30 2.443 0.411 1.210 0.269 -0.142 -2. The cost structure and arbitrage strategies are described in the text.18 0. Dev.9 1.672 0.778 Strike 103 Base Case No Rebalance Rebalance Every K Days K= 2 K-= 5 Rebalance When h Changes by K K = 0.193 1.345 0.570 -0.414 0.043 0.9 3.73 0.318 0.051 0. Mean Std.380 0.2 2.420 0.8 5.67 1.8 6.25 24.21 1.300 1.0 12.532 6.175 0.040 0.51 0. Trades is the average number of days with a transaction in the stock.710 8.3 3.621 0.3 3.7 2.25 23.641 -0. Shares Trades Traded Mean Std.354 0.025 0.8 2.230 1. Excess Return Including Transaction Costs No Costs Market Maker Retail Mean Std.022 -0.214 -0.718 -3.100 0. Strike 97 Base Case No Rebalance Rebalance Every K Days K= 2 K= 5 Rebalance When h Changes by K K = 0. Returns are in dollars per option on one share.1300 The Journal of Finance Table IV Comparison of Mean Transactions Costs The table shows the mean and standard deviation across 250 series for arbitrage excess returns including transactions costs under different rebalancing strategies.465 0.618 -0.215 0.704 -1.457 0.033 0.417 0.101 -0.9 6.664 1.334 0.785 -3. Shares traded is the average total number of shares traded in hedging the option for 25 days.036 -0.216 0.67 1.175 -5.10 2.445 0.385 0.420 0.294 1.022 0.163 0.1 2.10 K = 0.684 1.10 K = 0.0 2.107 0.23 0.509 0.144 -1.148 -0.14 0.076 0.043 0.11 0.982 -2.101 0.026 -0.213 -0.687 -0.82 0.668 0.326 0.204 0.349 0.581 -3.223 1.60 1.216 0.6 2.9 2.20 1.674 -9.461 0.592 -0.720 -0.870 1.934 -3.4 2.350 -2.695 -0.004 0.157 0.037 0.801 -5.713 7.880 1.072 0.03 0.64 0.395 0.121 0.314 1.027 -0.126 0.447 0.25 24.83 0.028 0.0 12.985 1.0 12.198 -0.269 0.0 2.145 0.019 0.21 0.16 1.819 Strike 105 Base Case No Rebalance Rebalance Every K Days K= 2 K= 5 Rebalance When h Changes by K K = 0.951 1.9 2.147 0.10 K = 0.076 0.226 -0.10 K = 0.0 1.25 24.629 -4.038 0.258 1.445 0.104 -1.318 1.879 Strike 100 Base Case No Rebalance Rebalance Every K Days K= 2 K= 5 Rebalance When h Changes by K K = 0.049 0.440 0. Dev.599 6.049 0.415 0.684 0. Dev.694 -0.022 0.341 0.032 0.6 5.128 0.0 12.116 -0.57 2.149 -0.039 0.048 0.060 -0.34 1.496 -10.64 1.605 -2.64 1.391 0.633 -4.548 -9.050 0.956 .254 1.042 0.638 -5.505 -0.508 -1.607 -1.76 0.332 0.672 -10.160 0.

with a standard deviation of $0. while the standarddeviation was only slightly affected. At the oppositepole from continuous rebalancingis no rebalancingat all. This reducesthe numberof transactions to two: an opening and a closing trade.We will see this in more detail in the next table. That strategyis examinedin the second line of results for each strike price. and the total numberof shares traded is just twice the initial hedge ratio.to $1.038. The patterns exhibited by the at-the-money calls were also present in the others. The retail trader took losses far greaterthan the initial price of the option in each case. a market maker would have experienceda mean of $-0. but it remains high enough that the trade is still very unattractive. losing more than $10 on average when the option's initial value was only about $2.180 to -0. One is to rebalance less frequentlythan every day.and thereforetrading costs.307. For the 100 strike call.318. wants is an intermediatestrategythat limits both trading What an arbitrageur costs and risk. However. Including transactions costs. hedges it accordingto the theoretical hedge ratio.420. with a standarddeviationof $0. The trader takes an options position at the outset.this is obviouslyof lesser importance than the effect on the mean return.the next two columns show the sample means and standard deviations of excess returns when there are no transactions costs.160. the sample mean excess return for the 100 strike hedges was $0. A retail traderwouldhave huge costs if he or she attemptedto trade the option arbitrageaccordingto the dictates of the Black-Scholes model. Due to the variationacross price series in the amount of rebalancing. meaningthat the net impact of transactions costs was to reducemean excess returnsby (-0. the retail trader's standard deviation was also substantially higher than that borne by the market maker.314.307. Thus. but the standarddeviation has more than quadrupled. A retail trader's cost is cut by nearly a factor of ten.Options Arbitrage in Imperfect Markets 1301 average. and holds it until expirationwithout any furthertradingof the stock.even though one would expect risk to increase when the hedge is not rebalancedas often as possible.even when done by a market maker. Commissions paid by a market maker reduce the mean to $0. For comparison. The reduction in mean for the market maker varied from -0. without taking account of transactions costs. perhapsevery two days or every week. The effect of tradingcosts can then be seen in the deviations from these values.269. It suggests that trading strategiesthat economizeon the numberof transactions or the numberof shares traded might be worth pursuing. This limits the number of trades but allows for the possibility that the hedge proportions can get far out of line in between rebalancingpoints.307.269 .the dynamic hedging strategy requiredapproximatelyfive share transactions for every share in the initial hedgeposition. Two possibilities are commonly suggested. A second approachis to monitor the discrepancy between the actual and the theoretical hedge ratios .038) = $-0. a net cost of $0. It is clear from these figures that transactions costs make a substantial differencein the outcome of an options arbitrage.134 on averageinstead of the previous $0.026. the no-rebalancingstrategy increases the mean excess returnwithout transactions costs to $0. This can lead to a very costly hedge for retail traders who pay a minimum charge per stock trade no matter how few shares are involved.0.

25. The table shows the results of two such strategies of each type.e.there will be bounds aroundthe theoretical option price within which the market price may fluctuate freely. Once again. Although these results can be analyzed carefully to try to determine which strategy performedbetter for which options. Arbitrage Bounds Based on the Standard Arbitrage In a frictionless market.. but only to rebalancewhen the hedge gets too far from the correct value. The results of the previous section allow us to analyze these bounds. The strategy allows more variation in the number of trades among hedges but keeps the hedge ratio close to the theoretical value at all times. if the marketwere pricing the 100 calls on a volatility of 0.13 while the market maker believed the true value was 0. The following two lines relate to the strategy of rebalancingonly when the actual hedge ratio differs from the theoretical by at least 0.the figuresshown in Table IV can be used to compute the bid and ask prices requiredfor them to . IV. the force of arbitrage drives the price of an option exactly to its Black-Scholes value. Lines three and four for each strike examine rebalancingthe hedge every two days and once a week (i. even for a market maker in options. the transactions costs entailed by the arbitrage strategyunderlyingthe Black-Scholes model are quite large.1302 The Journal of Finance daily. more than enough to wipe out all of the profit. costs remainsubstantial and risk levels increase quickly as the frequencyof rebalancingis reduced.10 and K = 0. because the potential arbitrageprofit would be outweighedby the cost of trying to captureit.15. the retail traderhas such heavy costs that he or she wouldnot follow any of these strategies.with the latter leading to an averagenumberof shares tradedthat is about halfway between the two polar cases of rebalancingdaily and not at all. The two values for the maximumpermitted deviation. K = 0. For example. for none of them is the resulting combinationof risk and returnvery favorable. the model indicates an arbitrageprofit of $0. Table II shows that. five trading days). There is clearly no point in discussing the returns to retail tradersany further.240. If arbitrageursderive prices at which they will enter the market to buy or sell calls by calculatingthe expected cost of the standardarbitrage. However. the transactions costs involved in trying to capturethat excess return would be $0.25. respectively. This leads to frequent rebalancingin some cases and little in others. Rebalancing only when the actual hedge ratio gets too far away from the theoreticalvalue is a logicalway to reducethe amountof tradingof the underlying asset.307. Rebalancingonly every two days or every week reducesthe number of trades substantially. With costly arbitrage.10 and 0. One of the most apparentconclusions to be drawnfrom Table IV is that. were chosen because the mean number of shares traded and the average reduction in excess return were comparable to the equivalent figures for the previousstrategies.In all cases. depending on the actual stock price path.

05 2. The break-even calculation involves only the expected cost of the arbitrage trade: the risk does not enter.60 0.128 Ask Value = 4. The results of this calculation are displayed in Table V. 0.29 0.09 0. For instance.244 0.105 No Bid Negative 1.18 0.13 Negative 4.55 0.170 Value = 2. Bid Ask Strike 100-Model Base Case No Rebalance Price Vol.156 0.191 1.089 No Bid Negative 0. we saw that the arbitrage trade would entail an average total transactions cost of $0. 3.05. 0.53 0.143 Strike 105-Model Base Case No Rebalance Price Vol.144 . If the arbitrageur's strategy is to trade the option and then hedge it and hold to expiration.74 or sell it no cheaper than 2.126 0.191 1.193 3.84 1.36 0.161 Value = 0. in this case they would be willing to "buy the option on a .15 volatility. if they thoughtthe true volatility was 0.35 0.157 Value = 0. We assume that arbitrageurs face the market maker cost structure described above.28 0.46 0.241 3.172 0. Traders often think about option pricing in the market not in terms of prices but in terms of implied volatilities.176 2." For each case in Table V we show the arbitrage boundary bid and ask prices in dollars and.19 0.14 0.31 0.57 0.70 0. Consider the Base Case for the at-the-money calls. Price Vol. he or she must buy the option at no more than 1.112 3.06 0.01 4.125 0.307.Options Arbitrage in Imperfect Markets 1303 break even or to achieve any specified probability of earning a profit.176 volatility.23 0. Market Maker Cost Structure Rebalancing Strategy Breakeven Bid Strike 97-Model Base Case No Rebalance Price Vol. In Table IV.15.91 0.125 volatility and sell it on a .77 0.058 2.88 0. However. on the following line. Price Vol. With a 0.40 0.307 = $0.74 0.183 4.107 0.96 0.244 1.82 0.173 0.46 0.125 1. The width of the no-arbitrage band is (at least) 2 x 0.59 0.76 0.61.41 0.92 0. Price Vol.35 in order to expect to break even. Price Vol. risk does come into the calculation if Table V Arbitrage Bounds with Transactions Costs The table shows the arbitrage bounds on option prices and implied volatilities at which an arbitrageur would bid for and offer options and have a 50 percent or 75 percent probability of covering costs.239 75% Profit Prob.90 0.087 3.139 Strike 103-Model Base Case No Rebalance Price Vol. 1.198 4. the option's theoretical value is 2. the implied volatilities corresponding to those prices.

of course.53 and an offer of 2. so the impliedvolatility would be negative. it is a mathematical property of a logarithmic diffusion process that rebalancing continuously would require an infinite number of transactions and would involve trading an infinite number of shares.67 standarddeviations. an arbitrageurcould expect to break even bidding 1. However. as Leland (1985) observes. the price would be quoted as 17/8 bid. the bid should be no more than 0.96 and the ask at least 3. the Base Case results indicated the narrowest or almost the narrowest spread when a 75 percent profitability hurdle was imposed.193.91 and offeringat 2. to be 75 percent sure of covering costs. In all cases. are able to rebalance their positions more frequently than every day. only. For the options with different strike prices. How large a profit probability traders will require to engage in the standard arbitragewill dependon several factors. offered at 21/8.18. for a bid-ask spreadof just about ? point. which would allow them to eliminate more of the risk than this table shows. that would be a bid of 1. several significant features are visible. . In the limit. " Market makers. It is appropriateto note here that bid-ask spreads in actual options markets are much smallerthan these figures. For the at-the-money option.67 9tandarddeviations.or better. In comparingresults in this table for the different strategies and strike prices. However." We do not report the results for the alternative strategies examined in Table IV since they were not particularlyeffective. a much largerspreadthan in the Base Case. Typical spreads for these options would be 1/4 to 3/8 for the 97 strike calls. In no case would the spread indicated in Table V be close to the observedvalue if market makers only did the standard arbitrageand requiredas much as a 75 percent probabilityof making a profit on their trades. and normally less. more rebalancing also means higher transactions costs.14. about 1/8 to 1/4 for the 100 strikes.13 on the 97 call is less than the current stock price minus the present value of the exercise price. A bid price of 3.the risk involved in the arbitragetrade is so greatthat. because the effect of the lower mean cost for the other strategies was offset by their increasedvariabilityof returns. includingtheir level of risk aversion and the degreeof competition among them.we calculate the bid-ask spreadsthat wouldproducea 75 percent probabilityof coveringcosts. there is no positive bid price that would allow a 75 percent profit probability. respectively). so an arbitrageurhas a 75 percent chance of covering costs if he or she quotes a bid lower than the model value and an offer above it by an amount equalto the expected transactionscost plus 0. and about 1/8 for the 103s and 105s. For illustration. That is. The 75th percentile of the normal distribution occurs at 0. The second case examined for each strike price is no rebalancing.This is the strategy with the lowest average transactions cost but the highest standard deviation.107 and 0. For the out-of-the-money options. First. In the Base Case for the 100 strike call. the No-Rebalancebreak-even strategyleads to bid-ask spreadsthat are comparableto those observedin actual options markets.1304 The Journal of Finance arbitrageursrequiremore than a 50 percent probabilityof covering their costs. the risk of the No-Rebalance strategy is sufficiently great that the appropriatebid price violates the option's boundaryconditions.The typical marketbid-ask spreadon a onemonth at-the-money stock index call option selling at about 2 would be no more than ? point.57 (impliedvolatilities of 0.

arbitrage involving long dated traded options.The option theoreticalvalues are higherthan those in Tables I and II. alternatively. Comparingthe standard deviations of annualized holding-periodreturns on the hedge portfolios to those shown for the Base Case in Table III reveals a distinct decrease for the longer dated options. exactly the ones we are looking at.082. How badly market imperfections impact the standard arbitrage trade depends on the amount of rebalancingrequiredto keep the portfolio hedged. with less than three months to expiration. However.the option's "gamma. 103.024 and the same price changewouldonly produce a $0. if the stock price rises from 100 to 101 with no rebalancingin between. Because delta changes. The first three lines of the table show the theoretical option values and hedge ratios at the outset. One example of an arbitrage-like strategy that attempts to replicatelonger maturity options is portfolio insurance. The only difference was that we limited the sample to 100 price series.which we will discuss in detail below. as well as the cost of the arbitrageportfolios. the arbitrage portfoliois long the option and short the underlyingasset. As before. A. if this were a one-year option. there would be "replicationerror"of more than four cents.However. we created a new set of stock price series and call option values with 75 trading days.5. the value of the "hedged" portfolio will change by $0.our 100 strike calls have a gammaof 0.Options Arbitrage in Imperfect Markets 1305 V. In other words. Options prices and hedgeratios for calls with strikeprices of 97. Each stock price series started at 100 and then followed a logarithmicrandom walk with an annualizeddrift of 15 percent and volatility of 0. We have essentially computed the most relevant results from each of the earliertables with these new data."12 Gamma is greatest for options that are close to expiration and at the money. Longer Maturity Options To see how much our results would change with longer dated options.043. For exchange-tradedoptions. gammawouldbe only 0.012 replicationerror. and other optional contracts is not uncommon. and 105 were computed from the Black-Scholes model. 100. For example. so it produces a net cash inflow (negative"cost"). Table VI displays summaryresults for the standard arbitragewith three-month calls.those close to the money. the second derivative of the option value with respect to the stock. This is partly an artifact from annualizingthe returns. . warrants. and the values of the hedge portfolios still are very large comparedto a typical amount of mispricingof the options. the deltas are closer to 0. MultiplyingN-day cumulativeexcess returns bv 260/N 12 Gamma is defined as the derivative of the hedge ratio with respect to the stock price or. This in turn is a function of how much the hedge ratio changes as the stock price moves-in other words.15. the greatest trading volume and open interest is nearly always in the contracts for which our results are representative. Other Option Maturities and Trading Strategies The results we have developedso far apply to call options that are not too far in or out of the money and have about one month to expiration. following the same proceduresas above.

particularly for the outof-the-money options.307 below the Black-Scholes price for a one-month option.76 0. For example.55 2.65.322 0.422 0. if daily excess returns are serially independent.73.724 -66.07 103 2. Table VI bears this out. an arbitrageur would bid no more than 0.Bid) Breakeven 75% Profit Probability 97 5. However. Transactions costs for longer holding periods.49 0. Strike Price Theoretical Call Value Hedge Ratio Arbitrage Portfolio Cost Hedge Portfolio Base Case Percent Standard Deviation Transactions Costs for Market Maker (in $) Mean Cost Std.345' 0. can be expected to cumulate.62 1.1306 The Journal of Finance Table VI Summary of Results for Three-Month Options The table summarizes results from applying the procedures reported in the previous tables to a new sample consisting of 100 price series of 75 trading days each.78 1.431 0.585 -54.391 0. See the text and earlier tables for a full description. tripling the number of days in each series should only increase the standard deviation of the cumulative total by the square root of three. three-month calls eventually become one-month calls. the lower gammas for three-month options may yield smaller trading costs per day than for one-month options.52 and 0. the standard deviations of the annualized holding-period returns reported in Table VI would appear to drop by a factor of about 1/1.34 -0.91 0.346 -32.373 0.438 -41. It does seem that the arbitrage is considerably less risky per day for three-month than for one-month options. We might therefore expect the total transactions cost for the standard arbitrage to rise monotonically with option maturity.431 for a three-month call.86 1.24 105 1. At the outset.82 3. The comparable figures to have a 75 percent chance of covering costs would be 0.78 0. on the other hand. Dev.255 0. of Hedge Portfolio Arbitrage Band (Ask .24 to annualize them multiplies their sample standard deviation by the same factor. to break even on the at-the-money call. . However.29 100 3.30 -0.82 1. The standard deviations of hedge returns including costs increase also. However. respectively. but he or she would only pay the model prices less 0. even if the risk per day of the arbitrage portfolios for one-month and three-month options were the same. since the standard arbitrage requires the position to be held until expiration.16 -0. Thus. The combination of increased mean hedging cost and increased standard deviation leads to substantially wider arbitrage bands for three-month than for one-month options.84 1.70 -0.32 3.406 0. The increase in mean transactions costs ranges from about 30 percent for the 97 strike calls up to 117 percent for the 105s. the annualized standard deviations for three-month calls are substantially lower than can be accounted for in this way.

First. rather than with the underlying asset. he or she buys on one implied volatility. the important thing is how the market it will price options in the immediate future. ."3 Under the right circumstances. Normally. but not necessarily by setting up the standard arbitrage against the underlying asset. anticipating that he or she will sell it again fairly quickly at his or her offer price. and one that can potentially be hedged with other options. Rather. As with other trading strategies. as well. Second. but also for its gamma. traders will take limited positions and carefully weigh the expected profit against the risk. More precisely. and tries to sell as soon as possible on a higher implied volatility. with the result that considerable netting out of market exposure may occur. The supply of arbitrage services to a real market will not be perfectly elastic. a market maker's entire portfolio of options on a given asset is aggregated. B. The trader has less interest in the true volatility of the underlying asset than in the option's future implied volatility. Indeed. Market Making and Price Determination in the Options Market Since a rational options market maker who based his or her trading on the standard arbitrage strategy or any of the variants described in Table IV would insist on much wider bid-ask spreads than are observed in the marketplace. regardless of whether this is a very good estimate of how volatile the stock price 1 Figlewski (1988) examines the impact of incomplete arbitrage in the market for NYSE index options. a reasonable conclusion is that actual market makers should probably not follow these strategies. This trading strategy embodies important deviations from the model of market making implied by the Black-Scholes and other models based on the standard arbitrage. A typical market maker does not buy an options contract with the expectation that it will be held in inventory and hedged until expiration. An option may be hedged with another option.Options Arbitrage in Imperfect Markets 1307 The results in Table VI indicate that the standard arbitrage trade becomes less risky when longer dated options are involved. Each of these represents exposure to a type of risk. trading so as to profit from a theoretical mispricing of the option relative to its underlying asset will not be the dominant strategy. options prices may be allowed to deviate very far from their model values without inducing a large amount of arbitrage trading to push them back into line. The resulting option position is evaluated not only for its delta. when a trader takes on an options position with the expectation that will be unwound quickly in the market. However. hedges the position. since arbitrage is not riskless. or with a related futures contract. observation reveals that they do not. Mispricing of options relative to the underlying index is found to be associated with the use of an alternative (risky) trading strategy involving NYSE index futures. Options positions that are not turned over immediately are hedged. as it is in a frictionless world. he or she buys it at his or her bid price. trading costs increase with option maturity. and probably its theta (rate of time decay) and its kappa or "vega" (sensitivity to volatility movements). leading to wider arbitrage bounds.

as we have mentloned.1308 The Journal of Finance will actually be over the option's lifetime. Several things account for this apparent discrepancy. 14 Brennan and Schwartz (1988) have taken a first step in modeling the short run optimal trading behavior of an arbitrageur with limited capital in a stock index futures market. in the context of portfolio insurance. Thus. so they naturally appear much smaller. at approximately one tenth of the cost. . C. if he or she expects that the market will continue to overprice such options for a while and he or she will be able to earn a quick profit by reselling it at his or her ask price. Finally. Thus. portfolio insurance programs try to replicate much longer dated options than we have been examining. and more specifically to the amount by which it may be mispriced. which reduces replication error. This is what is relevant for pricing traded options contracts in the market and evaluating a market maker's arbitrage strategy. the replication error of $199 thousand found by Etzioni seems insignificant relative to the $100 million portfolio being insured. Option Replication in Portfolio Insurance Our results appear to contrast markedly with those from studies of portfolio insurance. it is important to recognize that we are comparing the costs and replication errors of options arbitrage to the option's price. Etzioni's results considerably understate what it would have cost to replicate the put with stock transactions. while the Black-Scholes model may give a great deal of guidance about how one option should be priced to be consistent with other options on the same stock. Their approach offers a useful starting point for analyzing the more complex option market maker's problem. even though it is pretty large compared with the $750 thousand total value of the put options being replicated. a well-known application of the same kind of option replication we are examining. However.1 thousand. Etzioni (1986).14 All of these factors imply that. they are often set up to reduce the problem of high gamma near maturity. A second factor is that. uses a methodology similar to ours with daily rebalancing and finds that a portfolio insurance program replicating a one-year. at-the-money put option on a $100 million stock portfolio would cost $745. the force of arbitrage driving options prices to their theoretical values relative to the underlying asset based on the market's best estimate of its true volatility is severely blunted. Thus. for example. the cost and risk of the program are expressed relative to the total value of the insured portfolio. by targeting a final payoff pattern that is smooth rather than kinked at the strike price. there is no inconsistency between our results and those from portfolio insurance studies. Thus. Indivisibilities are also not a problem when such a large portfolio is being hedged. Moreover. it is perfectly appropriate for a market maker to buy an option that he or she believes is currently overpriced relative to its long run value. A major one is that portfolio insurance transactions are generally done in stock index futures contracts rather than the underlying stocks.5 thousand and would have a replication error of only $199.

increase risk in a hedged position but do not have a large effect on expected return. we suggest that market makers and others engaged in arbitrage of exchange traded options follow different strategies.10 (per share) or better. The tradeoff appears to be slightly more favorable for a strategy of rebalancing only as the hedge ratio moves far enough away from its correct value. However. the impact is asymmetrical. The following are the major conclusions indicated by our results. we have adopted a simulation approach that allows us to derive accurate answers for specific values of the underlying parameters of the market system. However. If the hedge ratio can be set to the nearest 0. they are prohibitive. Partly based on direct observation. . the impact of indivisibilities is limited. are precise but not completely general.25 increased the standard deviation of the hedge return by more than half. to look at cases that accurately reflect the realities of exchange traded stock index options contracts. * The normal transactions costs for the standard arbitrage induce arbitrage bounds around the theoretical option value that are substantially wider than the bid-ask spreads that are observed in practice. For a retail trader. Quick turnover also implies that traders will be more interested in forecasting the option's implied volatility for the immediate future than the true volatility of the underlying asset. A "yes or no" hedge (i. limiting possible hedge ratios to multiples of 0. Since these questions are sufficiently complex that a general theoretical treatment is infeasible. but this is not a riskless strategy. Conclusions In this paper we have addressed a number of issues involved in applying arbitragebased option valuation models to actual. Strategies for reducing transactions costs by rebalancing the hedged position less frequently do not help much: there is a substantial reduction in cost only at the expense of a substantial increase in risk. Hedges involving futures contracts that are relatively large will be most affected. * The volatility of the underlying asset is an extremely important determinant of option value. the impact on hedging accuracy is relatively slight. For them. except for far out-of-the-money options. imperfect. but sampling error makes the ex post volatility in daily closing prices hard to predict accurately even when the true underlying volatility is known. though quantitatively different. Our results.Options Arbitrage in Imperfect Markets 1309 VI. as far as is possible.e. They try to achieve quick turnover. We would anticipate that other parameter values would lead to qualitatively similar. therefore. which make it impossible to achieve exactly the right hedge ratio. markets. h = 0 or 1) is highly risky. thus reducing costs. * Indivisibilities. results. * Transactions costs to do the standard arbitrage trade upon which the BlackScholes model is based are large. even for a market maker. rather than rebalancing only after a fixed number of days. We have tried.. so that it is substantially worse to underestimate volatility than to overestimate it. except for out-of-the-money options. Mistakes in forecasting volatility cause both option values and hedge ratios to be wrong.

the standardarbitragecited in the literatureas the basis of valuationmodelsbecomes a weak force to drive actual option prices towardtheir theoretical values. 167-197. Heath. Financial Analysts Journal 44-54. 323-338. 53-68. 1980. We do not currentlyhave a model of option pricing in a market populatedby short-term. Journal of Fir. Figlewski. The pricing of contingent claims in discrete time models. Arbitrage in stock index futures. 1981. and Myron Scholes. Brenner. A survey of empirical tests of option pricing models. Journal of Business 50. Options: A Monte Carlo approach. 59-62. Journal of Finance 34. inarbitrageurswho engage almost exclusively in non-"standard".1983b. Fischer.1310 The Journal of Finance * The standardarbitragewith longer dated options is exposed to less risk per day than with the one-month calls we examined in the bulk of the paper.: Option Pricing (D. The standard arbitrageeliminates price expectations and risk aversion from option pricing in a frictionless market. 1986. UCLA Graduate School of Management.) Under these conditions.there is certainly room for these and many other factors to have an influence. 1988. Journal of Business 56. 1979. Dan. However. Working Paper.(Nor have we coveredall majorimperfections. Gould. MA). Arbitrage-based pricing of stock index options. Boyle. 259-282. to name a few. One general conclusion suggested by this research is that. Phelim. B. REFERENCES Beckers. Black. 1983a. Garcia. 1977. Andrew.nonlognormal price paths. Review of Research in Futures Markets 7. Studies of stock price volatility changes. Christie. Rebalance disciplines for portfolio insurance. . Discretely adjusted option hedges. Standard deviations implied in option prices as predictors of future stock price variability. Journal of Finance 27. Galai. Stephen. 399-418. Journal of Banking and Finance 5. An empirical study of portfolio insurance. Journal of Financial Economics 4. and taxes. . 1987. 1972. C. 1976. Tests of market efficiency of the Chicago Board Options Exchange. However. the impact of market imperfections is also large. The stochastic behavior of common stock variances. 1988. arbitrage-likestrategies. 1977. 1982. and F. and David Emanuel. Journal of Portfolio Management 13. The valuation of option contracts and a test of market efficiency. Lexington.the total transactions cost to maintain a hedgedposition through option expiration increases with option maturity. 363-382.within the wide bounds on prices that are all that can be established by the standardarbitragein an actual. completely hedged. Journal of Financial Economics 8. Eytan.and probablylargerthan many researchershave realized.. in Proceedings of the 1976 Meetings of the American Statistical Association. The components of the return from hedging options against stock. J. S.ancial Economics 10. C. Michael J.. ed. 407-432. Etzioni. Business and Economics Statistics Section. in M.imperfectoptions market. so that the arbitrage bounds on the marketprice become wider. 45-54. while empirical researchhas shown that option valuation theory plays a very important role in determiningprices in real options markets. Brennan. 250-270. and Eduardo Schwartz.having left out marginrequirements.

1976. Macbeth. Journal of Finance 40. Henry and Richard Rendleman. 455-480. The valuation of uncertain income streams and the pricing of options. 1976. 179-201. . 1980. Leland. 1173-1186. Journal of Finance 31. Mark. Standard deviations of stock price ratios implied in option prices. 369-382. . 1978. An empirical examination of the Black-Scholes call option pricing model. Options pricing and replication with transactions costs. 1985. Trading costs for listed options: The implications for market efficiency.Options Arbitrage in Imperfect Markets 1311 Latane. Phillips. Journal of Financial Economics 8. James and Larry Merville. Journal of Finance 34. Rubinstein. Journal of Finance 40. Hayne. 1283-1301. Bell Journal of Economics 7. Susan and Clifford Smith. Nonparametric tests of alternative option pricing models using all reported trades and quotes on the 30 most active CBOE option classes from August 23. 407-425. 1976 through August 31. 1979. 1985.

Sign up to vote on this title
UsefulNot useful