This action might not be possible to undo. Are you sure you want to continue?
Blomeyer and Herb Johnson Reviewed work(s): Source: The Journal of Financial and Quantitative Analysis, Vol. 23, No. 1 (Mar., 1988), pp. 1322 Published by: University of Washington School of Business Administration Stable URL: http://www.jstor.org/stable/2331020 . Accessed: 07/05/2012 09:53
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp 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 email@example.com.
University of Washington School of Business Administration is collaborating with JSTOR to digitize, preserve and extend access to The Journal of Financial and Quantitative Analysis.
LA 70803. Numerical solutions to the American put value were produced by Parkinson (1977). which was extended to handle dividends by Blomeyer (1986). earlier helpful paper the1984American StockExchange Colloquium. The SchoolofManagement. Geske. put options.NO.Louisiana State University. The valuation model for the European put op? tion was derived by Black and Scholes (1973) for the case in which the stock price follows geometric Brownian motion. the R.andE. useful discussions An version this of was presented at comments twoanonymous of JFQAreferees. Very recently. the numeri? cal approaches would presumably yield equally accurate prices. but it cannot be made arbitrarilyaccurate. Davis.23. and ex? plained why an analytic solution may be more efficient than those numerical ap? proaches in terms of the number of critical stock price calculations required for penny accuracy. * Baton Rouge. Brennan and Schwartz (1977). of Davis. and Rubinstein values. Blomeyer and Herb Johnson* Abstract This studyis an expost performance testcomparingthe accuracy of an Americanmodel to a European model forvaluing listed options. Ross. Geske and Johnson (1984) derived an ana? lytic solution to the American put value and demonstrated that their values are virtually identical to the Cox. Geske and Shastri (1985b) compared the efficiency of these numerical approaches.AND ANALYSIS JOURNAL FINANCIAL QUANTITATIVE OF VOL. As pointed out in GJ (1984) and Geske and Shastri (1985b). the closer to marketprices than are the Geske and Johnsonmodel values are significantly Black and Scholes values. Johnson (1983) developed an analytic approximation. authors University California. 1. the Geske and JohnsonAmeri? can put valuation model is compared withthe Black and Scholes European put model. CA 95616. Omberg. Options 13 . with and without dividends. However.MARCH 1988 An Put Empirical Options Examination of the Pricing of American Edward C. relative to marketprices. and Rubinstein (1979). Although Geske and Johnson (GJ) required numerical procedures to evaluate their analytic formula. Specifically. and wouldliketo acknowledge with Castanias. Introduction The development of models capable of valuing American put options has occurred only relatively recently.respectively. and Cox. In a recent paper. Ross. R. On average. and Graduate College of Business. the computational expense for this method is lower than that for numerical ap? proaches because far fewer critical stock prices need be computed for penny ac? curacy. both models undervalue. I.
are examined in this study. Omberg considered a range of parameter values so wide that a large number of the cases he considered would never be encountered in any real data samples. free ofthe nonsynchronous trading problem. This study is a comparative ex post performance evaluation of the GJ ap? proach and the Black and Scholes (BS) European model using CBOE listed put option transactions. A total of 10. to be superior to the BS model.295 put option transac? tions. it re? cently has become possible to use a huge amount of reliable market data. since the GJ technique can allow explicitly for dividends. With the advent of trading in listed options on the Chicago Board Options Exchange (CBOE). This pricing bias is reported in graphical and tabular form. Parkinson (1977) compared his numerical solutions with over-the-counter put quotations from the New York Times and found that the American put model undervalued puts by a few percentage points. for reasonable parameter values. occurring during 13 trading days chosen from the months of June through August 1978. we briefly discuss the GJ model and the effects of cash divi? dends on American put option pricing.14 Journal of Financial and Quantitative Analysis Omberg (1986) developed a method. Of course. the GJ method seems to be sufficiently accurate (see also GJ (1984). Shastri and Titman (1984) have recently reported pre? liminary evidence on the efficiency issue. we would expect the GJ approach. p. Severe data restrictions limited such early studies. More importantly. based on a suboptimal exercise policy. However. this study includes long-lived puts on dividend-paying stocks. but. this paper shows how well the GJ approach (or any equivalent numerical approach) values puts. Very little testing of put pricing models has been reported in the literature. They reported the American put model undervalued puts by 25 to 40 percent. The degree to which the model values differ from the market prices is compared to the degree to which the options are in or out of the money and the time remaining to option maturity. As we shall show. Not only does this study use an enormous amount of data. the development ofthe Berkeley Options Data Base. 1521). Since the BS model may currently be the most widely used. In evaluating . The empirical methodology is discussed in Section III. sometimes traders believe that models determine mar? ket prices simply by being popular and frequently used. Brennan and Schwartz (1977) com? pared their numerical solutions with prices for 55 puts traded in the New York dealer market from May 1966 to May 1969. for valuing American puts. free of nonsynchronous trading problems. which explic? itly allows for the possibility of early exercise. Omberg maintained that the GJ interpolation scheme for handling dividends can produce large errors. this paper provides a useful test of this argu? ment. However. and the results are presented in Section IV. and the derivation of the faster GJ technique for valuing American puts. Section V is a sum? mary. In Section II. II. while this study examines the accuracy issue. to test market efficiency and the accuracy of the put valuation formula. which does not. Geske-Johnson Model This study uses the theoretical model developed by GJ (1984).
by subtracting the discounted value ofthe dividends from the stock price and assuming that the put can only be exercised at maturity. Equa? tion (1) is given in footnote 6. For dividends below this critical value. For the parameters used in Figure 1. where P is the American put price. As the time to maturity increases past an ex dividend date. when they are paid during the life ofthe option. a European put). The cash dividends are $1 and the stock goes ex dividend in Vi93)4. A large dividend was chosen to accentuate the effects of divi? dends. The European puts for the non-dividend case are not as deep in the money because there are no ex dividend dates on which the stock price is expected to fall. The top two plots in Figure 1 are for puts written on stocks with dividends. See GJ (1984) for more details. The BS plot for the divi? dend case has negatively sloped segments because the particular puts considered are deep in the money. III. and the bottom two plots are for puts on stocks without dividends. we use their sequential extrapolation from four exercise points to the infinite limit. page 1520 of GJ (1984). The existence of cash dividends should reduce the magnitude of the difference between the BS European put values and the GJ American put values because cash dividends reduce the probability of early exercise of the American put op? tion. we use the GJ interpolation procedure. and 6/2months. the relative (but not necessarily absolute) difference in GJ and BS values is greater for the no dividend case because dividends tend to delay exercise and therefore make the early exercise feature less valuable.23/6 [P. For dividends above some critical value. The BS European put values are adjusted for dividends. the put value rises because the stock price is expected to fall on the ex dividend day. If there are dividends during the life of the put. respectively. these European put values happen to increase monotonically with time to maturity. Px is the price of a put that can only be exercised at maturity (i. linear interpolation is performed between the value for a put with no dividend and the value for a put with the critical dividend.] . the put can be valued directly. and the loss from having to wait an extra month to exer? cise (the interest effect) dominates the gain from having more time to maturity (the volatility effect). Methodology All put option transactions occurring during the months of June through Au- . These options have maturity ranges of one to nine months and are in the money with a stock price of $40 and an exercise price of $45. which modifies both their formula and the evaluation technique for the dividends. P2 is the price of a put that can be exercised halfway to maturity or at maturity.e.Blomeyer and Johnson 15 their analytic formula.. and P3 and P4 are the prices of puts that can be exercised at three and four equally spaced points. This point was demonstrated firstby Geske and Shastri (1985a) and is illus? trated in Figure 1 where GJ and BS values are plotted for options written on stocks with and without cash dividends. As expected. PJ + 1/6 [P2 P. which yields the following approximation to their so? lution (1) P = P4 + 29/3 [P4 PJ .
Gen? eral Motors (GM). trading in for the of and thecombined written theother on four volumeof options companies. Options that were out of the money by more than $5. D = $1.2 1 During werenot on werealso traded IBM stock. Using the GJ interpolation scheme to handle multiple cash dividends. These op? tions had a median time to maturity of 80 days and a maturity range of 5 to 270 days. one trading day per week was randomly chosen for evaluation.00 4. options put was aboutequal to volume IBM options of examined thisstudy tworeasons. IBM options and graphs tables. Consequently. and theequities theother companies the dominate the wouldcompletely due $73. which resulted in a total of 13 trading days.16 Journal of Financial and Quantitative Analysis 7.50 6.50 Geske-Johnson (nodividends) Black-Scholes (no dividends) 7 4 6 3 5 TIME TO MATURITY IN MONTHS IO FIGURE 1 Model Values forPutOptionsWritten Stocks with on and without Dividends S = $40.50 r Geske-Johnson (dividends) Black-Scholes (dividends) V) cr < 7. The stock price used in the BS and GJ models was the market price of the stock. 2 Our choice of a 20-weekestimation We arbitrary.295 option transactions remained for evaluation. we are able to examine long-lived options. in IBM stock traded this absolute difference between modelandmarket values. 10. X = $45. r = 10%.1 From this data set. Dividend information was obtained from Moody's Di? vidend Record.6/2 months gust 1978 on the CBOE for Avon Products (AVP). study emphasis theoptions onecompany. inclusion theIBM op? the this examines on of tions wouldplace undue Secondly. The risk-free bond equivalent yield was calculated from the Wall Street Journal quotation for a Treasury bill maturing at approximately the same time as the option expiration date. traded therangeof 3j51to in of four a rangeof $255 to $301. and Honeywell (HON) were obtained from the Berkeley Op? tions Data Base.3/2.00 6.50 5.Theseoptions thistimeperiod.During study period.00 5. a = 0. Eastman Kodak (EK).First. net of any dividends discounted at the risk-free rate from the transaction observation date to the ex dividend date. to theverylargepremia.3. Each stock re? turn volatility was calculated using stock price range data from the 20 weeks immediately preceding the week ofthe option transaction. TD = y2. have used data that periodis somewhat . The stock return standard deviation estimates were calculated using the Parkinson (1980) extreme value method with weekly stock price ranges obtained from Barron's.00 with premia of less than fiftycents were in a restricted trading status and were excluded from the sample. After fil? tering the data.
also calculated we stockreturn wouldhave been availableto the optiontrader. with the GJ values being closer. Additionally. Although there weresmalldifferences for estimates each ofthe13 trading for weeks. i. For in-the-money options. All mean values reported in Table 1 are significant at the 1 percent level. Since the mean values can be influenced strongly by a few outliers. The values For comparison. and from 10 weekspreceding the10 weeksfollowing the standard deviations usingstock priceranges for between thesetwomethods each theobservation week. Due to early exercise. median values of the pricing bias Y are plotted for two percent ranges of Z. the degree to which the option is in (Z ^ 0) or out of the money (Z < 0). IV. the differ? ences between the two model pricing biases are small because of the decreased probability of early exercise of these options.Blomeyer and Johnson 17 The ex post performance of each model is evaluated by comparing Y. The differences between with t statistics in paren? mean values is reported in the column headed by "Af theses. where (2) (3) ~~ and Z = market model r x 100% . we compare the BS European put values and the GJ American put values to market prices. average eachcompany's panywas virtually Blackfor deviations calculated Fischer estimates very was return close to thestock standard by tility 1978andreported Cox andRubinstein in (1985).theaverage volatility eachcom? company's volatility the of vola? for identical thetwomethods. the degree to which the option is in or out ofthe money.6 cents for the GJ model and 46. to market values than the BS values. to Z.e. The median pricing bias of 38.5 percent.7 cents for the BS model indicates that both models tend to undervalue put options. median values of Y are reported with mean values reported directly below these median values.295 option transactions. but contain a fixed element as well. for out-of-the-money options. on average. none of the options is in the money by more than 17. We chose to define Y in absolute rather than relative terms because the relative bias can be enormous for very low-priced puts. greater than transactions costs. the degree to which the market and model prices differ. and for in-the-money options. This is because transactions costs are not simply a percentage of the amount bought or sold. Examining the upper part of the table. Within each cell in Table 1. In the upper part of Table 1. median and mean values of Y are reported for each model for all options. the cell in the upper left-hand corner at the intersection of the row and column labeled "all options" reports the pric? ing bias for the GJ model and the BS model for all 10.. the median pricing bias for the GJ model is substantially lower than the median BS model pricing bias. Specifically. In Figure 2. the median value may be a better measure of the model pricing bias. The absolute difference gives a better indication of whether or not price deviations are economically significant. Results Using transaction-by-transaction data. July . For out-of-the-money options. the bid-ask spread does not become negligible as the price of the put gets small.
1-percent Thevalues brackets median "N" with in The labeled between values /valuesparentheses. Each the medianpricing bias value is calculated for 2-percent a rangeofthedegree to whichthe in this optionis "in"or "outofthe money.column column headed "AV" contains differences mean by isthe number of observations. price. The . -20 1 TABLE between Prices Model Prices3 of Differences Market and Summary and Values Y = Pmarketof Median Mean Pmodei a Median the values. the median degree of undervaluation for the GJ model is 23.00 -1.50r 1. that nearly 70 percent of our sample consists of out-of-the-money puts. F.50 -25 20 5 -5 10 0 -15 -10 PERCENT IN/OUTOF THE MONEY FIGURE 2 Median pricing bias for Black-Scholes model (*) and the Geske-Johnsonmodel (A).18 Journal of Financial and Quantitative Analysis l. For example. however. reported in parentheses are median values of the pricing bias."The symbolsplotted thisgraphare centeredwithin range. Note. expressed as a percent of the market price.00 & a o A A A -1.9 percent. where the error is relatively large. mean All values Yaresignificant of at with values Ybelow median of values Yare of reported mean of market The in are values /expressed a percentthe of as the level.
the mean value of Y for the GJ model of 36. These findings provide 3 Ourreported with studies biasesmayappearrelatively largewhencompared previous pricing of call option variance modelperformance have used an implied that estimation pricing technique variance will a an (1979)). the mean and median biases are lower for the GJ values than for the BS values. magnitude direction thereported pricing willbe influenced thechoiceof by theimplied variance weighting technique. The probability of early exercise of an American put is positively related to the degree to which the option is in the money and to the time to option maturity. the median pricing bias of 41. the model pricing biases are reported separately for options on each of the four companies. All options with maturities of less than 60 days were written on stocks that did not go ex dividend prior to the option expiration date. the and bias of tionally. By construction. the options were written on stocks with and without dividends. the size of the GJ pricing bias does appear to become a smaller proportion of the BS pricing bias as the maturity increases. For options on stocks with dividends.5 cents lower than the BS model median pricing bias. and we examine these options in the last two sections of Table 2. Examination of the reported pricing biases in this section indicates that median and mean values of F are lower for the GJ model over all maturityranges for both in. However. Also. When broken down by in and out of the money.9 cents is 82 percent of the BS value of 45 cents. respectively. For all in-the-money op? tions. there is a substantial difference between the pricing biases ofthe two mod? els. there is a substantial difference between the pricing biases of the two models.Blomeyer and Johnson 19 in-thfe-money puts must have values close to X?5. Dividends play a part in explaining these biases.3 cents and 49. the GJ bias is also proportionally smaller than the BS bias for options on stocks with? out dividends than for options on stocks with dividends. Within the matu? rity range of 60 to 88 days. and so are easier to value than the out-of-the-money puts. the size of the pricing bias is substantially lower than the size of the bias for options with longer maturities. In Section II of Table 2. which are not similarly constrained. The median values of Y are 31. as expected.3 cents for the GJ and BS models.. For in-the-money options. Consequently. and negatively related to the magnitude of cash dividends with ex dividend dates prior to the expiration ofthe puts. In the lower part of Table 1.3 The absolute difference between the GJ and BS models is largest for the in-the-money options. and all options with maturities greater than 88 days were on stocks with dividends. for options on stocks without dividends.4 is 89 percent of the BS bias. implied (e. long term to maturity options. For options with maturities of less than 45 days. MacBethandMerville technique provide smallmodelpricing bias sincethemodelis usedto estimate of itsowninput one Addi? parameters. In every case.g. there does not appear to be a systematic relationship between the time to option maturity and the magnitude of the pricing bias. We examine these conjectures in Section I of Table 2 by stratifyingthe options into four maturityranges of 45 days each and a fifthrange for options with maturities of greater than 180 days. .2 cents for the GJ model is 4. in Section III. for options with maturities greater than 45 days. the mean GJ model bias of 63. For all out-ofthe-money options. and should decrease as cash dividends are introduced.and out-ofthe-money options. the magnitude ofthe difference in pricing bias between the European put and American put models should be large for in-the-money.
Comparing the proportional model bias for out-of-themoney options within the same maturity ranges. the mean GJ bias is reduced to 70 percent of the BS bias. options more 88days maturity written with than prior expiration All date. the mean value of Y for the GJ model was 89 percent of the BS value. bAll on that were with 60 to prior option go options lessthan days maturity writtenstocks didnot exdividend tothe onstocks went dividend tothe that ex to were date. column contains differences headed percent The column "N" number labeled isthe of observations. For options with maturities greater than 179 days. option expiration empirical support for the Geske and Shastri observation that the existence of divi? dends should decrease the relative magnitude ofthe difference between European and American put option model values. For in-the-money options. For example.20 Journal of Financial and Quantitative Analysis a Median All values mean values Vare of atthe values Yare of with values Vbelow of median significant 1reported mean in mean with The "AV" between values /values parentheses. for all options with maturities of 60 to 88 days written on stocks with dividends. we find that the mean GJ model bias decreases from 95 percent to 80 percent of the BS value. In order to obtain a consistent sample in which all options exhibit dividend effects. we examine the effects of increasing time to option maturity on the pro? portional biases of the two models by comparing the biases reported in Section III to the biases in Section I for options with maturities greater than 89 days. level. there is a substantial decrease in the proportional bias from 74 percent to .
Rubinstein (1985) has noted that the direction of the out-of-the-money pricing bias shifted in Octo? ber 1977. although the GJ values were substantially closer to market prices than were the BS values. One possible explanation for model undervaluation is that there is some small probability of a large price drop superimposed on the Brownian motion. both mod? els tended to undervalue put options. Summary and Conclusions This study has provided an ex post performance evaluation comparing the accuracy of an American model and a European model for valuing listed options. we checked the GJ values against the binomial values. Overall. we saw no evidence of the failure of the GJ interpolation scheme mentioned in Omberg (1986). In addition. and two to four cents deep in-the-money. the existence of sys? tematic pricing biases for American call options has been well documented. and the Black and Scholes model provided the European values. They also noted that the discrepancies were a decreasing function of the time to option maturity. The performance of the GJ model was superior to the BS model perfor? mance. These stud? ies have reported that systematic call option model pricing biases do exist. Black (1975) (see also Merton (1976)) reported that the BS call option pricing model undervalued out-of-the-money options and overvalued in-the-money options. We found that the discrepan? cies between the two methods were negligible for out-of-the-money puts. We there? fore conclude that the large errors found by Omberg occur only for fairly extreme parameter values. Merton also noted that the discrepancies were largest for short term to maturity options.Blomeyer and Johnson 21 47 percent. However. The superiority of the GJ model in- . a penny or two at-the-money. several previous studies have provided direct comparisons between model values and market prices for American call options. many of them requiring linear interpolation to handle the dividends. The magnitude and direction ofthe exercise price and time to maturity bias has not been consistent from study to study. V. Blo? meyer and Klemkosky (1983) examined both the BS European model and Roll (1977) American call option model and found that both models undervalued outof-the-money options while pricing other options fairly well. Although there has been little empirical investigation of American put op? tion pricing models. with the exception of out-of-the-money options with less than 45 days to maturity where the differences were slight. Nonetheless. This study has provided evidence that systematic model pricing biases exist also for American put options. MacBeth and Merville (1979) observed an exercise price bias that was exactly opposite to that reported by Black and Merton. we would caution the reader that any treatment of dividends is ultimately an approximation since we do not ordinarily know in advance precisely what the dividends will be. These results empirically support the conjecture that the superiority of the performance of the American put model over the European put model should increase with the time to option maturityand should be greatest for in-themoney options Although we computed thousands of GJ values. The Geske and Johnson model was used to generate the American values.
Merton. Efficiency American Blomeyer.455-480.. "OptionPricing of cial Economics. R. Rubinstein." In "Tests of Market E. J.1 (forthcoming). Analysis. andH. "An Analytic of Approximation theAmerican Price." Journal Business.1980)." "The American Option Geske. 61-72.32 Brennan.. Exchange. 50 Put. of Approach. ExercisePolicies. Pricing Model.207-219.. Heathand Co." AdPut E. J."An Empirical J."Manuscript. Schwartz. Scholes. "Fact and Fantasy theUse of Options.. 124-144. of Journal Finan? are whenUnderlying StockReturns Discontinuous. Rubinstein. Klemkosky.61-65. Valua? of A _"Valuation Option by Approximation: Comparison Alternative 20 tion Analysis. Johnson. Merville. "OptionPricing: American of for the The Value Method Estimating WalkProblem: Extreme _"The Random 53 Variance oftheDisplacement." Journal Business."MarketEfficiency CostlyTransactions: Examination the K. Omberg. Liabilities. andE. Hall NJ: Markets. 101-121. Call Examination oftheBlack-Scholes Option MacBeth. (March1985b). and S. August 1976 through 40 1985)."Journal Political of and Black. of Models Using all Reported M. The M. of Techniques. 1511-1524. 1977). (June ofFinancial for Call Options. Options Englewood A Cox.36-41..S. 1173-1186. Prentice (1985). M." FinancialAnalysts 1975)." Journal Financialand 18 Quantitative Analysis. (Jan.9 (June 1985a). of 39 (Dec. MA: D.21-36."Journal FinancialEconomics. "The Pricing Options Corporate of 81 1973). andM. Tests of Alternative Rubinstein. (Jan. and R." Journal Banking Geske. 1984). Black.andM. Approximation theAmerican PriceforOptions Stockswith Blomeyer. Put ValuedAnalytically. (Oct. Ross. Johnson."Journal Financialand Quantitative Put for H. "An Analytic 21 and Quantitative Dividends. 1979).R. Lexington of Put "The Valuation American Options.229-233."Journal Finan? "OptionPricing: Simplified cial Economics.449-462. "The Valuationof American OptionswithExponential and vancesinFutures Options Research. (March1983).F. of (May 1977).22 Journal of Financial and Quantitative Analysis creases with the time to option maturity and is greatest for in-the-money options.." Journal Finance. of on American Call Options Stockswith for Valuation Formula Unprotected Roll. and K. (June August31. 7 Journal Finance. "An Analytic 5 Known Dividends... UCLA (1984)." R."Journal Finance. Parkinson. ChicagoBoardOptions . Economy.C. and a researcher or trader generally will be handicapped severely by using the Black and Scholes model in lieu of the more sophisticated Geske and Johnson approach. Shastri. 141-148."Journal 1986). Titman. (Nov.-March 3 1976)." Journal of with An Shastri. (July-Aug. (May/June on Put for E.1977). and Fi? Puts. M. Cox. and Option PricingTheory Applications. 1979). The performance difference between the GJ and BS models is substantial.45-71. ofFinance."The EarlyExerciseof American of nance. Lexington.. Cliffs. Brenner.251-258.34 (Dec. Books(1983).637-659. We also found that the existence of cash dividends decreases the difference be? tween American and European option model values. "Nonparametric OptionPricing Tradesand Quoteson the30 Most ActiveCBOE OptionClasses from 23.. andL. (Jan. andM. References 31 in Journal.R.229-263. ed. F.1978.