Empirical Performance of Alternative Option Pricing Models Gurdip Bakshi, Charles Cao, Zhiwu Chen Journal of Finance, Volume 52, Issue 5 (Dec., 1997), 2003-2049. ‘Your use of the ISTOR database indicates your acceptance of ISTOR’s Tecms and Conditions of Use. A copy of ISTOR’s Terms and Conditions of Use is available at hup:/wwvrjstor-orglabouterms.himl, by contacting JSTOR at jstor-info@umich edu, or by calling ISTOR at (888)388-3574, (734)998-9101 or (PAX) (734)998.9113. No part ofa JSTOR transmission may be copied, downloaded, stored, further transmitted, ransferted, distributed, altered, or otherwise used, in any form or by any means, except: (1) one stored electronic and one paper copy of any article solely for your personal, non-commercial use, or Q) with prior written permission of ISTOR and the publisher of the article or other text Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the sereen or pfinted page of such transmission. Journal of Finance is published by Amecican Finance Association, Please contac he publisher for futher permissions eegarding the use ofthis wodk. Publisher cootact information may be obtained at thupivwrjtororgoumnalfina tn. Journal of Finance ©1997 American Finance Association ISTOR and the JSTOR logo are trademarks of JSTOR, and are Registered in the U.S, Patent and Trademark Otic. For more information on ISTOR contact jstor-nfo@umnich edu, ©2000 ISTOR hup:thvww stor orgy Fni Dec 15 09:32:22 2000Empirical Performance of Alternative Option Pricing Models GURDIP BAKSEI, CHARLES CAO, and ZHTWU CHEN* ABSTRACT Substantial progress has been made in developing mare realistic option pricing models, Empirically, however, i is net lzinwn whether and by how much each _genevalizstion improves option pricing end hedging. We fl this gap by ist deriving fn option mode! that allows volaubty, interest rates and jumps to be stochastic: sing S&P 500 options, we examine several alternative models from three perspec tives: €1) internal casetency of implied parameters'volatlity with relevant time: series data, (2) out-faample pricing, and (3) hedging. Overall, incorporating sto chastic volatility and jumps is importa for pricing and internal consistency, But for hedging, medeling stachastie volatlity alane yields the best performance Iy tHe Last Two DECADES, option pricing has witnessed an explosion of new ‘models that each relax some of the restrictive Black-Scholes (BS) (1973) assumptions. Examples include (i) the stochastic-interest-rate option modela of ‘Merton (1973) and Amin and Jarrow (1992); (i) the jump-diffusion/pure jump models of Bates (1991), Madan and Chang (1996), and Merton (1976); (ii) the. constant-elasticity-of-variance model of Cox and Ross (1976); {iv} the Mark- ovian models of Rubinstein (1994) and Ait-Sahalia and Lo (1996); (v) the stochastic-volatility models of Heston (1993), Hull and White (1987a}, Melino and Tumbull (1990, 1995), Scott (1987), Stein and Stein (1991), and Wiggins (1987); (vi) the stochastic-volatility and stochastic‘interest-rates models of Amin and Ng (1998), Bailey and Stulz (1989), Bakshi and Chen (19974,b), and Scoit (1997); and (vii) the stochastic-volatility jump-diffusion models of Bates (1996a,c), and Scott. (1997). This list is by no means exhaustive, yet already overwhelming to anyone who has to choose among the alternatives. To make matters worse, the number of possible option pricing models is virtually infinite. Note that every option pricing model has to make three basic assump- * alah ia at the University of Maryland, College Pars. Cas is at Pennaylvania State Univer- sity, University Park, Chen is at The Onio State Univeraty, Columbus. Tas paper aubsumes the Drevious one under the ttle “Option Pricing and Hedging Performance with Stochastic Volatility land Stochastic Interest Hates." We thank: Sanjiv Das, Ranjan DMelle, Jin-Chusn Duan, Helyett Geman, Erie Ghysels, Frank Hatheway, Steward Hodges, Ravidagannathan, Andrew Karolyi Bill Rracaw, Dilip Madan, Victor Ng, Louis Seat, Rens Seulz, Stephen Teylor, Siegftied Trautman, ‘lox Tviantis, lan Waite, and the anonymous referee, We gratefully acknowledge comments by sersinar participants at the 1996 Eurapesn Finance Aasaciation Meetings, 1997 Westar Finzmce Meetings, the Ghinese University of Hang Kong, the Hong Kong University of Science and ‘Technology, The Ohio State Univerity, and the University of New Orleane. Any remaining errors are our responsinilty alone. 20082004 The Journal of Finance tions: the underlying price process (the distributional assumption), the inter- est rate process, and the market price of factor risks. For each of the assump- tions, there are many possible choices. For instance, the underlying price can follow either a continuous-time or a disereto-time process. Among possible continuous-time processes, it can be Markov or non-Markov, a diffusion ot a nondiffusion, a Poisson or a non-Poiséon jump process, a mixture of jump and diffusion components with or without stochastic volatility and with or without random jumps. For the term structure of interest rates, there are similarly snany choices, While the search for that perfect option pricing model can be endless, we are tempted to ask: What do we gain from each generalized feature? Is the gain, if any, from a more realistie feature worth the additional complexity or implementational casts? Can any of the relaxed assumptions help resolve known empirical biases associated with the Black-Scholes for- mula, such as the volatility smiles (e.g., Rubinstein (1985, 1994))? As a prae- tical matter, that perfectly specified option pricing model is bound to be too complex for applications. Ultimately, it js a choice among misspecified models, made perhaps hased on (i) “which is the least misspecified?” (ii) “which results in the lowest pricing errors?” and (iif) “which achieves the hest hedging per- formance?” These empirical questions must be answered before the potential of Yecent advances in theory can be fully realized in practical applications, ‘The purpose of the present article isto fill in this gap and conduct a compre- hensive empirical study on the relative merits of competing option pricing mod- els. To this goal, we first develop in closed form an implementable option pricing model that admits stochastic volatility, stochastic interest rates, and random jumps, which will be abbreviated as the SVSI-J model. Te setup is rich enough to contain almost all the known closed form option formulas as special cases, including (i the Black-Scholes (BS) model, (i) the stochastic-interest-rate (SI) ‘model, (it) the stochastie-volatility (SV) model, (iv) the stochastie-volatility and stochastic-interest-rate (SVSI} model, and (v} the stochastic-volatility random- jump (SVJ) model. The eonstant-volatility jump-diffusion models of Bates (1991) ‘and Merton (1976) are special cases of the SVJ. Consequently, we concentrate our efforts on the SVSI-I and the five models just described. Besides the obvious normative reasons, a common motivation for these new models is the abundant empirical evidence that the benchmark BS formula exhibits strong pricing biases across both moneyness and maturity Ge, the “smile”) and that it especially underprices deep out-of-the-money puts and ealls (see Bates (19960) for an insightful review), Such evidence is clearly indicative of immplieit stock return distributions that are negatively skewed with higher kurto- sis than allowable in a BS log-normal distribution. Guided by thia implication, the A fewexiatng studies smvestigata the internal eonaietency of implied parametera (Bates (1991, 1996a,c), snd the pricing or the hedging performance (eg, Bakshi, Cao, and Chen (1997), Coo (1898), Dumas, Fleming, and Whaley (1895), Madan and Chang (1998), Nandi (1996), and ‘Rubinstein (1985), of alternative etochasti-volaility medels, Cao euidies a sachaste-velaiisy ‘model using currency options; Nendi investigates the pricing and singleinstrument-hedging performance using che S&P 500 fueures. In this axticle we addvess the empirical issues from siferent perspectives and under sltemative models.Empirical Performance of Aliernative Option Pricing Models 2005 search for alternative models has mostly focused on finding the “right” distribu- tional assumption. ‘The SV model, for instance, offers a flexible distributional structure in which the correlation hetween volatility shocks and underlying stock Tetums serves to control the level of skewness and the volatility variation coeffi- ‘cient serves to control the level of kurtosis. But, since volatility in the SV is ‘modeled as a diffusion and hence only allowed to follow a continuous sample path, its ability to intemalize enough short-term kurtosis and thus to price short-term options properly is limited (unless the variation coefficient of spot volatility is unreasonably high), The jump-diffusion models, on the other hand, assert that it is the oceasional, discontinuous jumps and crashes that cause the negative im- plicit skewness and high implicit kurtosis to exist in aption prices. The fact that stich jumps and crashes are allowed to be discontinuous aver time makes these models more flexible than the diffusion-stochastic-volatility model, in internaliz- ing the desired return distributions, especially at short time horizons, Therefore, ‘the randomjump and the stochastic-volatility features can in prineiple improve. the pricing and hedging of, respectively, short-term and relatively long-term options, The inclusion of a stochastic term structure model in an option pricing framework is, however, intended to improve the valuation and discounting of fature payoffs, rather than to enhanee the flexibility of permissible return distri- bbutions. Tans, while the stochastic-interest-rate feature is not expected ta help resolve the cross-sectional pricing biases, it should in principle improve the pricing fit across option maturity. ‘We implement every model by backing out, on each day, the spot volatility and structural parameters from the observed option prices of that day. ‘This approach is common in the existing literature (e.¢., Bates (1996b)), partly out of the consideration that historical data reflect what happened in the past ‘whereas information implicit in option prices is forward-looking. Backing out the BS model's volatility and other model's parameters daily is indeed ad hoe since volatility in the BS and the structural parameters in the other models are assumed to be constant over time. But, as this internally inconsistent troat- ment: is how each madel is to he applied, we follow this convention so as to ensure each model an equal chance. In judging the alternative inodels, we employ three yardsticks. First, are the implied structural parameters consistent with those implicit in the relevant times-series data (e.g., the irplied-volatility time series, and the interest-rate time series)? Much of this part of the discussion is based on Bates’ (1996a,c) work where he studies the relative desirability of the SV versus the SVJ models, using stock index futures and curreney options, The reasoning is that if an option model is correctly shecified, its structural parameters implied by option prices will necessarily be consistent with those implicit in the observed time-series data, Second, out-of-sample pricing errors give a direst measure of model misspecifiea- tion. In particular, while a more complex model will generally lead to better in-sample fit, it will not necessarily perform better out of sample as any overfitting may be penalized. Third, hedging errors measure haw well a model eaptures the dynamic properties of option and underlying security prices. In other words, in-sample and out-of-sample pricing errors reflect a model's static performance,2006 ‘The Journal of Finance while hedging errors reflect the mode's dynamic performance. As shown later, these three yardsticks serve distinct purposes. Based on 88,749 S&P 500 call option prices from June 1986 to May 1991, we find that the SI and the SVSI-J models do not significantly improve the performance of the BS and the SVJ models, respectively. To keep the presen- tation manageable, we focus on the four models of distinct interest: the BS, the SV, the SVSI, and the SVJ. Our empirical investigation leads to the following overall conchisions. First, judged on internal parameter consistency, all mod- els are misspeeified, with the SVJ the least and the BS the most misspecified. ‘This conclusion is confirmed from several different angles. For example, ac- cording to the Rubinstein (1985) type. of implied-volatility graphs, the SVJ implied volatility smiles the least across moneyness levels, followed in increas- ing order by the SVSI, the SV, and the BS. Second, out-of-sample pricing errors are the highest for the BS, the second highest for the SV, and the lowest for the SVJ. Overall, stochastic volatility alone achieves the first-order pricing im- provement and typically reduces the BS pricing errors by 26 percent to 60 percent. However, our evidence also confirms the conjectures that (i) adding the random-jump feature improves the fit of short-term options and that (ii) including the SI feature enhances the pricing fit of long-term options. After hoth stochastic volatility and random jumps are modeled, the remaining pric- ing errors no longer exhibit clear aystematic biases (e.g., across moneyness). ‘Two types of hedging strategy are employed to gauge the relative hedging effectiveness, First, we examine minimum-variance hedges of option contracts that rely on the underlying asset as the single hedging instrument. As argued by Ross (1995), the need for this type of hedge may arise in contexts where a perfect delta-neutral hedge may not be feasible, either because of untraded risks or because of model misspecifications and transaction casts. In the presence of more than one source of risk, single-instrument hedges can only be partial. According to results from these type of hedges, the SV outperforms all the others, while the SVJ is second. Between the other two models, the BS hedges in-the- ‘money calls better than the SVSI, but the SVSI is better in hedging out-of-the- ‘money calls. This hedging result is surprising as one would expect the SVSI to perform at least as well as the BS, and the SVJ to do better than tho SV. Next, we implement a conventional delta-neutral hedge, in which as many hedging instruments as there are risk sources are used to make the net position completely risk-immunized (locally). For the case of the BS, this means that only the underlying stock will he employed to hedge a call. For the SV model, however, both the price risk and volatility risk affect the value of a call, implying that an SV-based delta-neutral hedge will need a position in the underlying stock and one in a second option contract. For the SVST, its delta-neutral hedge will involve a discount bond (to control for interest rate risk) in addition to the underlying stock, and a second option contract, When such internally consistent hedges are imple- mented, the hedging errors for the SV, the SVST and the SVJ are about 50 percent, t0 65 percent lower than those of the BS model, if each hedge is rebalanced daily, Furthermore, changing the hedge rebalancing frequency affects the BS model's hedging errors dramatically, while only affecting the other models’ performanceEmpirical Performance of Alternative Option Pricing Models 2007 ‘marginally. That is, after stochastic volatility is controlled for, the errors of delta-neutral hedge seem to be relatively insensitive to revision frequency.” How- ever, like in the single-instrament hedging case, once stochastic volatility is modeled, adding the SI or the random-jump feature does not enhanee hedging performance any further. Since the delta-neutral hedge for the BS doos not use a second option contract whereas it does for the other three models, this may have biased the deltarneutral hedging results against the BS model. To examine this point, we also implement the ad hoc BS delta-plus-vega neutral strategy in which the underlying stock and an option contract are used to neutralize both delta risk and vega risk (of the BS model). It turns out that in hedging out-of-the-money and at-the-money calls, this BS delta-plue-vega neutral strategy performs no worse than the other models’ delta-neutral hedges. Only in hedging deep in-the-money calls do the stochastic volatility models perform hetter than the BS delta-plus-vega neutral strategy. This is true regardless of hedge revision frequency. Overall, hedging performance is relatively insensitive to model ‘misspecification, since even ad hoc hedges can result in similar errors. The rest of the article proceeds as follows. Section I develops the option pricing models, Section TI provides a description of the S&P 500 option data. In Section IIT we present an estimation procedure, discuss the estimated param- eters, and evaluate the in-sample fit of each model. Section TV assesses the extent of each model's misspecification. Sections V and VI, respectively, present the out-of-sample pricing and the hedging results. Concluding re- marks are offered in Section VII, Proof of pricing equations and most formulas are provided in the Appendix. L. Option Pricing Models ‘The purpose of this section is to derive a closed-form jump-diffusion option pricing model that includes all those to be studied in the empirical sections as special cases, As such, it is then convenient to follow a standard practice and specify from the outset a stochastic structure under a risk-neutral probability measure. The existence of this measure is equivalent to the absence of free lunches, and it allows us to value future risky payoffs as if the economy were risk-neutral. First, under the risk-neutral measure, the underlying nondivi- dend-paying stock price Sif) and its components are, for any f, given by ait) Fay” RU) ~ duslde + (das + dehdate) a dV(t) = (8, — x,Vit)]dé + 0, (Vi)dw.it) (2) This finding is in aecord with Galas (1983) results that in any hedging scheme iti probably ‘more important ta contol for stochastic volatility than for diserete hedging (ace Hull and White (1987b) for 2 similar, simulation-based result for currency aptions.2008 The Journal of Finance Taft + JQ) ~ NAnf1 + wy] ~ 4 03, a3, @) where: RQ) is the time-t instantaneous spot interest rate; Ais the frequency of jumps per year; Vit) is the diffusion component of return variance (conditional on.na jump occurring); as{t) and «,(t) are each a standard Brownian motion, with Cov.{das(t), du, (t)) = pdt; Hi) is the percentage jump size (conditional on « jump occurring) that is lognormally, identically, and independently distributed over time, with unconditional mean 44; The standard deviation of In{1 + JO) és o.58 (is « Poissan jump counter with intensity A, that is, Pridg(é) = 1) = Ade and Prldgit) = 0} = 1 — dats yy 61k, and 0, are respectively the speed of adjustment, long-run mean, and variation. coefficient of the diffusion volatility Vit), git) and d(t) are uncarrelated with each other or with wg(t) and w(t). Under the assumed framework, the total return variance ean be decomposed ints tivo components 1 dS(t aeYe(‘sey) Vee) + Vat), a whore V(t) = (dt) Var{Jedait)] = Alu} + (e?? ~ 1) 0 + pl is the instantaneous variance of the jump component. This stock-return distributional assumption, similar to the one in Bates (1996a) for currency prices, offers a sufficiently versatile structure that can accommodate mast of the desired features. For instance, skewness in the distribution is controlled by either the correlation por the mean jump m,, whereas the amount of kurtosis is regulated by either the volatility diffusion parameter 7, or the magnitude and variability of the jump component. But the ability of the diffusion component Vit) to generate enough short-run negative skewness or ‘excess kurtosis is limited, as Vie)can only follow a continuous sample path. On the other hand, the discontinuous jump process can internalize any skewness and curtosis level even in the short run, especially when A, jy, and o,; are substantial. ‘Therefore, these two forces capture different aspects af return distributions. Nest, to ensure proper discounting of future cash flows, we adopt a single- factor term structure model of the Cox, Ingersoll, and Ross (1985) type as it requires the estimation of only three structural parameters: AR(t) = [6g — xqR(t)]dt + op (ROdoxlt), 5) See, for cxample, Bates (19564,<), Merton (1976), and Sect (2097) fora similar jump setup,Empirieal Performance of Alternative Option Pricing Models 2009 where Kg, fink, and ay ate respectively the speed of adjustment, long-run mean, and volatility coefficient of the R(t} process; and w,{t) is a standard Brownian motion, uncorrelated with any other process in the esonomy.* OF course, we ean add more factors to the term structure model and make the resulting bond price formulas more plausible, but that will also make the option pricing formula harder to implement. It is important to realize that the exogenous valuation framework given above can be derived from a general equilibrium in which the volatility risk VG), interest rate risk R(t), and jump risk J(tldg(t) are all rewarded. For instance, Bakshi and Chen (1997a) and Bates (1996a,¢) provide such examples in whieh each risk factor earns a risk premium proportional to the factor itself. ‘That is, the factor prices for V(¢) and R(@) are respectively b, Vie) and b,R(), for some constants 4, and 6,. These factor prices are implicitly reflected in equa tions (2) and (5) and adjusted through x, and ip, respectively, Therefore, factor risk premiums are not assumed to be zero in our framework. Rather, they have, been internalized in the stochastic structure, ‘Consider first. a zero-coupon hond that pays $1 in 7 periads from time f, and let. Bit, 2) be its current priee. Then, Bi, 1) = + wl [7 ne «| =expl—wir) — (IROL, (6) where _ es “I, a et) -a {is- wg) +2 inf Ie aa -e*) 6 ea) at m8 Veet 20%, “This assumption on the correlation between stock veturns and interest rates is somewhat ‘severe and likely counterfactual. To gauge the potential impact of shi assumption onthe resulting ‘option model's performance, we initially adopt the following stock price dynamics: $80 iss nde + WPT + op Tbe, with the rest of the stochastic structure remaining the same as given above. Under this mare realistic siructure, the eavariance between tack price changes and interest rate shocks is Coe (dS, ARIE ~ es, goa RSI, so bond markt ianovatians can be trenamictod to the sack market and viee verse The citained closed-form option peeing formula under this sesnario would hhave one more parameter a, than the ene presented ahoity, hut when we smplement thie slightly more general model, we fina its pricing and hedeing performance ta be indistinguishable from that ofthe SVSI model studied in this articl. For this reeson, we choaee to set ay y= 0. We ‘auld alsa make bath the rift and tho diffusion terms of Vie) «linear funstion of Af) and anit) Is such cates the stock returns, volatility and interest rates would all be corcelated with each other (at least globally), and we could stil derive the desited equity option valuation formula. But, that would again make the resulting farmula more complex while ot improving its performance.2010 ‘The Journal of Finance and .g() is the expectations operator with respect to the risk-neutral measure. For a European call option written on the stock with strike price K and term-to-expiration 7, its time-t price C(t, 7} must, by a standard argument, solve # Lage BUS Spe mai ay Sera yet (Wy aca aC +5 paar gig * (a — «eR) 5p - Ge RC a + ARCH, 7 SU +), RV) ~ Ch. S.R,VIP= 0. (7) subject to C(t + 7,0) = max(St + 7) ~ K, 0}, In the Appendix it is shown that C(t, 2) = STE, 1 S, R, V)- KBE, 7M, 5 8, RV), 8) where the risk-neutral probabilities, Ml, and [Ty, are recovered from inverting the respective characteristie functions (see Bates (1996a,c), Heston (1993), and ‘Scott (1997) for similar treatments) Te, 7 SO), RO, VO) wit [ef “MUFG, +, 8), RO, Vs @) 7 a ib z dg, (9) forj = 1,2, with the characteristic functions /; respectively given in equations (Ai0) and (ALD of the Appendix. The price of ¢ European put on the same stock can be determined from the put-call parity. ‘The option valuation model in equation (8) has several distinctive features. First, it applies to economies with stochastic interest rates, stochastic volatil- ity, and jump risk. It contains most existing models as special cases. For example, we obtain (i) the BS model by setting A = 0 and é, x, = 4,, = 0; (ii) the SI model by set and 6, a model by setting A = 0 and é = ky = ae = 0; (iv) the SYST model by setting 2 = 0; and () the SV model by letting Oy = kp = de = 0, where to derive each special ease from equation (8) one may need to apply L'Hopital's rule. The Appendix provides the exact option pricing formulas respectively for the SV, the SVSI, and the SVJ models. Second, this general model allows for a flexible correlation structure between the stock return and its volatility, as opposed to the perfect correlation assumed in, for instance, Heston (1993). Third, when ‘compared to the model in Seott (1997), the formula in equation (8) is parsimo- nious in the number of parameters; especially since it is given only as a function of identifiable variables such that all parameters can he estimated. ‘The pricing formula in equation (8) applies to European equity options. But in reality most option contracts are American in nature, While it is beyond the scope of the present article to derive a model for American options, it isEmpirical Performance of Alternative Option Pricing Models 201 nevertheless possible to capture the first-order effect of early exercise in the following manner. For options with early exercise potential, compute the Barone-Adesi and Whaley (1987) early-exercise premium, treating it as if the stock volatility and the yield-curve were time-invariant. Adding this early- exercise adjustment component to the European option price in equation (6) should result in a reasonable approximation of the corresponding American option price (e.g., Bates (1996a)). Alternatively, one can follow such a nonpara- metric approach as in AitSahalia and Lo (1996) and Broadie, Detemple, Ghysels, and Torres (1996) to price American options. ‘The closed-form option pricing formula in equation (8) makes it possible to derive comparative statics and hedge ratios analytically. [n the present context, there are three sources of stochastic variations over time, price risk St), volatility risk V(t) and interest rate risk RH). Consequently, there are three deltas: actt, 7) ate ik) = E90 a0) acta) all amt, Ay(t, aw = S(t) av” Kat, 1 Ww (iL) Ty Tt a(t K) = 800 Sp - KB 2 acai, a2) where, for g = V,Rand j= 1,2, am, 2 LY retain (13) ‘The second-order partial derivatives with respect to these variables are provided in the Appendix. These analytical expressions for the deltas form a convenient bbasis for constructing hedges such as the ones to be analyzed shortly. TI. Data Description Based on the following considerations, we use S&P 500 call option prices for ‘our empirical work. First, options written on this index are the most actively traded Buropean-style contracts. Second, the daily dividend distributions are available for the index (from the S&P 500 Information Bulletin). Furthermore, ‘S&P 600 options and options on S&P 500 futures have been the focus of many existing investigations including, among others, Bakshi, Cao, and Chen (1997), Bates (1996¢), Dumas, Fleming, and Whaley (1995), Madan and Chang (1996), ‘Nandi (1996), and Rubinstein (1994), Finally, we also use S&P 500 puts to estimate the pricing and hedging errors of all the models and find the results to be qualitatively similar. To save space, we only report the results based on the calls2012 ‘The Journal of Finance The sample period extends from June 1, 1988 throngh May 31, 1991. The intradaily bid-ask quotes for S&P 500 options are obtained from the Berkeley Option Database.* To ease computational burden, for each day in the sample, only the last reported bid-ask quote (prior to 3:00 pw Central Standard Time) ‘of each option contract is employed in the empirical tests. Note that the recorded S&P 500 index values are not the daily elosing index levels. Rather, they are the corresponding index levels at the moment when the option bid-ask quote is recorded, Thus, there is no nonsynchronous price issue here, except, that the S&P 500 index level itself may contain stale component stock prices at each point in time. ‘The data on the daily Treasury-bill bid and ask discounts with maturities up ‘tw one year are hand-collected from the Wall Street Journal and provided to us by Hyuk Choe and Steve Freund. By convention, the average of the hid and ask Treasury bill discounts is used and converted to an annualized interest rate. Since Treasury bills mature on Thursdays while index options expire on the third Friday of the month, we utilize the two Treasury-bill rates straddling an “option’s expiration date to obtain the interest rate corresponding to the op- tion’s maturity. This is done for each contract and each day in the sample, The 20-day Treasury bill rate is the surrogate for the short rate in equation (5), For European options, the spot stock price must be adjusted for discrete dividends, For each option contract with + periods to expiration from time ¢, we first obtain the present value of the daily dividends D() by computing Dit, 1) = Dent +8), aay where R(t, s)is the s~period yield-to-maturity. In the next step, we subtract the present value of future dividends from the time- index level, in order to obtain the dividend-exclusive S&P 500 spot index series that is later used as input into the option models, This procedure is repeated for all option maturities and for each day in our sample. Several exclusion filters are applied to construct the option bid-ask price data. First, option price quotes that are time-stamped later than 9:00 Put Central Standard Time are eliminated. This ensures that the spot price is recorded synchronously with its option counterpart. Second, as options with less than six days to expiration may induce liquidity-related biases, they are ‘exchided from the sample. Third, to mitigate the impact of price discreteness ‘on option valuation, price quotes lower than $% are not included, Finally, ‘quotes not satisfying the arbitrage restriction Git, 7) = max(O, S(t) ~ K, S(t) — D(t, 1) — KBit, 7)) (45) * Barly in the project we used only option transaction pric data for tke empirical wor, nt, that ata eetis much smaller, especially forthe hedging exercise. Nonetheless, the tesulls based on the ‘ransaction prices ave simila to those based on midpie bid-ase quotes,Empirical Performance of Alternative Option Pricing Models 2013 ‘Tabled Sample Properties of S&P 500 Index Options ‘The reported numbers aro respectively the average quaed hid-ak mid-point pice, the average elective bid-ask spread (ask price minus the bid-ask mispoint) which are shown fr parentheses, fad the tolal number of observations (in braces), fr each rooneyresematurty category. The ‘sample peri extend fora Juve 1, 1988 Usrough May 31, 1991 fora total of 38,749 cals. Dally informatio from the last quate (priar to 200 ptm. CST) af each option contract Led to obtain the summary atatstes.$ denotes the spot. S&P 800 index level and Kis the exercae price. OTM, ATM, and TTM denote outafthe money, at-che-maney, and inthe taney aptons, respectively Days-ta Expiration Maneyness Danian SE <60 0-180) =1e0 Subtotal ome 09 $168 438 $868 0.06) 016) (026) 42} (2330 3867 ang) oao97 $2.35 $802 sisi2 (0.09) (0.23) 038) 11943) ses) 965) ara Arm 097-100 $483 s12.79 820.7 (015) 4029) aso} (2708) (sis) 1003} 18625) 1.00-1.03 foaz siare $26.44 (0.35) (038) (oss) (2543) 1793} 931) isan) mM 103-1.06 wear $25.82 $33.00 (030) oan $0) (2255) ses} 738 4559) Lae $38.40 $18.06 ssa.12 (oat) 046) (0.50) (13% (5269) (33299) 3705) Subtotal ania, 14849) (78a) ares) are taken out of the sample. Based on this eriterion, 624 observations (approx: imately 1.3 pereent of the original sample) are eliminated and those calls are all deep in-the-money. We divide the option data into several categories according to either mon- eyness or term to expiration, Define St) ~ K as the time intrinsic value of call, A.call option is then said to be at-the- money (ATM) if its S/K € (0.97, 1.08); ‘out of-the-money (OTM) if SIK = 0.97; and in-the-money (ITM) if S/K = 1.03. 8 finer partition resulted in six moneyness eategories. By the term to expiration, an option contract can be classified as (i) short-term (<60 days}; (ii) medium: term (60-180 days); and (iii) long-term (>180 days). The proposed moneyness and maturity classifications produce 18 categories for which the empirical results will be reported. Table T describes certain sample properties of the S&P 500 call prices used in the study. Summary statistics are reported for the average bid-ask mid- point price, the average effective hid-ask spread (ie., the ask price minus the bid-ask midpoint), and the total number of observations, for each moneyness-2014 The Journal of Finance maturity category. Nate that there are a total of 38,749 call option observa- tions, with ITM and ATM options respectively taking up 47 percent and 28 percent of the total sample, and that the average call price ranges from $1.68 for short-term, deep OTM options to $58.12 for long-term, deep ITM calls. The effective bid-ask spread varies from $0.06 (for short-term deep OTM options) to ‘$0.80 (for long-term deep ITM options). IIL Structural Parameter Estimation and In-Sample Performance For the empirical work to follow, we concentrate on the four modols: the BS, the SV, the SVSI, and the SVJ.* As stated before, the analysis is intended to present a complete picture of what each generalization of the benchmark BS model can really buy in terms of performance improvement. and whether each generalization produces a worthy tradeoff between benefits and costs. To get.a sense of what we should look for in any desirable alternative to the BS model, let us use the described data set to examine the extent and the direction of biases associated with the BS. To do this, we back out a BS implied volatility from each option price in the sample. Then, we equally weigh the implied volatilities of all eall options in a given moneyness-maturity category, ‘to produce an average implied volatility. The caleulations are similarly done for put options. Table Il reports the average BS implied-volatility values across six moneyness and three maturity eategories, for both calls and puts as well as for both the entire sample period and different subperiods. Clearly, regardless ‘of sample (sub)period and term to expiration, the BS implied volatility exhibits a strong U-shaped pattern (smile) as the call option goes from deep ITM to ATM and then to deep OTM or as the put option goes from deep OTM to ATM. and then to deep ITM, with the deepest ITM call-implied and the deepest OTM. put-implied volatilities taking the highest values, Furthermore, the volatility smiles are the atrongest for short-term options (both calls and puts), indicating ‘that short-term options are the most severely mispriced by the BS model and present perhaps the greatest challenge to any alternative option pricing model. For a given sample (sub)period and moneyness range, the implied volatility is downward-sloping in most cases and exhibits a slight, U-shape in some cases, as the term to expiration increases. This is again true for both calls and puts. ‘These findings of clear moneyness-related and maturity-related biases associ- ated with the BS are consistent with those in the existing literature (e.g., Bates (1996b)). Therefare, any acceptable alternative to the BS model must show an. ability to properly price non-ATM options, especially short-term OTM calls and puts. As the smile evidence is indicative of negatively-skewed implicit return distributions with excess kurtosis, a better model must. be based. on a distri- butional assumption that allows for negative skewness and excess kurtosis. In an earlier version of the article we also report the performance reeults far the SU model, Since incorporating stachaste ntorost rates does not help improve performance ralich, we amit the SI made! (im tho discuasions ta fallow. For che samme reason, we da not repar the SVS madel’s result,Empirical Performance of Alternative Option Pricing Models 2015 ‘Table IL Implied Volatility from the Black-Scholes Model ‘The implied volatility 1s obtained by inverting the Black-Scholes model separately foreach call (put) option contract. The implied volatilities of individual calle (puts) are then averaged seithin each monaynces-maturityealegory and across the days in the sample. Moneynese is determined by SK, where S denotes the spot SGP 500 index level and K ls the excrelse price. Call Optione Put Options Daye-io-Expiration Days-to-Expiration Sample Period S/R <6) 60-180 180 <6 60-1800 Sane 1988 <0 1827S «2005 9S May 991 04-097 1664639 T4O SLT 8 097-100 1695 «1776 —«17.72—16.95— ROD 1oj-103 18501895 1S IES 19.95 19.83 103-106 21402004981 21.99 2087-20-80 21062872 BMS 87? 2872 28.88 June 1988 <09s «172716851608 845 8.g0 a0 May 1989 094-097 16.21 1642169576 TAAL Oor-100 © 16ad_ G89 TO} SLT 17S1 18H lod-t03 17701788172 dBA. 103-106 1863756 daae 982d 20.29 210s «270220071876 28k DSA 22.34 Jane 1989— 180 os. 0a sr ona 1a tor 10]? tc renames)Empirical Performance of Alternative Option Pricing Models 2023 Observe that the option-implied parameters correspond to the risk-neutral distributions while those estimated from observed time-series data are for the true distributions. Thus, before making the desired comparisons, we need to separate out the true distributional parameter values from their risk-neutral ‘counterparts. For this, we rely on the general-equilibrium modela of Bakshi and Chen (1897a) and Bates (1996a,c) in which the factor risk premiums are proportional to the respective factors and consequently the processes for Vit), Rh, qt) and (@) under the true probability measure share the same stochastic, structure as their counterparts under the risk-neutral measure. Specifically, 44s Jus Ps Op, ip, and 0, are the same under either prohability: Only x, Kz, ', and j4, will change when the probability measure changes feom the risk- neutral ta its true counterpart. Let these parameters under the true probabil- ity measure be respectively denoted by &, Ra, A, and jy. According to Rates (1991), when the tisk aversion coefficient of the represontative agent is bounded within a reasonable range, the parameters of the true distributions ‘will not differ significantly from their risk-neutral counterparts. For the overall sample period from June 1988 to May 1991, the annualized daily S&P 500 returns have a mean of 12.7 percont, a volatility of 17.47 percent, a skewness of ~0.43, and a kurtosis of 6.58. The historical volatility is indeed lower than its option-implied counterparts (see Table IM]). The negative skewness and the high kurtosis are in contrast with the skewness (of zero] and kurtosis (of 3) allowed hy the log-normal distribution in the BS model. The distributional assumption of the BS is thus overwhelmingly rejected by the data. We only need to focus attention on the relative misspecification of the three models with stochastic volatility, In the rest of this subsection, we treat the volatility implied by all options in a given day as a surrogate for the unobservable true spot volatility of that day. Let us first examine the consistency of the option-implied correlation p with the sample correlation between daily returns and volatility changes of the S&P 500 index. Ifan option model is correctly specified, the average p value implied by the option prices mmust, equal its time-series counterpart estimated from the daily price and volatility changes. The row marked “Time-series estimate” in Table IV provides such estimates of p at —0.28, -0.23, and —0.27, respectively, under the SV, the SVJ, and the SVSI model. The magnitudes of these esti- mates are much lower than their option-implied counterparts (-0.64, ~0.57, and ~0.76), suggesting that for each model the correlation level implicit in Figure 1. The implied volatility graphs are based on the six-month sample period from July 1990 through December 1980. Using a8 inputs i) current day's intereat rate and S&P 500 index value and (i) previous day's implied structural paramaters, we invert enck option farmula from the market pree of «given option, to abtsin the model's impiod volatility corresponding to ‘his option eantvac. Por each model, the reported implied volatility in 8 given moneyness maturity ‘category Is the average ofall cals in that moneynoss-maturty category and aver the entire suxctonth period. BS, SV, SVSI, and SVF respectively tend forthe Black-Scholes, the stachasti- ‘volatility model, the stochastic-olatlity and seachastiesnterescrate model, and the stachastic volatility ode! with random jumps.‘The Journal of Finance 2024 ‘yours sous out, {a0'0} (00%) 90°) foo) [840] Leo] «Pog HMA (o'0) (o0'0) 0, foro) (00 o woo Pre 20s eo 20 sO GO- cho FOO a aN oo 60% wo OE orden steuany a siopmeng AEA SMOEmpirical Performance of Alternative Option Pricing Models 2025 option prices is inconsistent with the time-series relation between stock re- turns and implied volatility. Bach of the three models is hence significantly misspecified. On a relative scale, however, this departure between the average implied and the time-series estimated is the weakest for the SVJ, and the strongest for the SVSI. Based on his estimated general autoregressive condi- tional heteroskedasticity (EGARCH) specification for equity-return dynamics, ‘Nelson (1991) gives an estimate of ~0.12 for the correlation hetween stock returns and changes in the true volatility, which is closer to our time-series, estimates than to the average option-implied values of p. Next, we adopt the maximum-likelihood (MIL) method proposed by Bates (1996a) to estimate the structural parameters of Vit) and Rit) (wherever applicable) under the true probability measure. Take the volatility process as an example. Using the implied-volatility time series as inputs, maximize the log-likelihood function z max 5 In{P[la(Vie + 1))|V(e}I}, (18) Ratan oh where P{ |] denotes the transition density of the non-central x? distribution given by Plin Vie + Ag}|Vee)] AV(E)Vie + ADE}! [eve + agree $e 1 Gl, /02) +71, ~ oxpleVie + a0) + eVinie > as) where c! = (1/2R,) 03 (1 ~ e~**4), and Gt ) denotes the (statistical? Gamma function. The ML estimates of the structural parameters are reported in Table IV for the three models. Two observations are in order. First, for each model, the ML estimates of &, and @,, are statistically indistinguishable from their respective option-implied counterparts (except for the &, estimate of the SV model). The p-values for the null hypothesis of equality between the ML and the option-iinplied estimates are all in excess of 16 percent (except for the SVJ case noted). Second, the implied value of o, is, for each model, about four times its ML estimate, The volatility process implicit in option prices is therefore ‘much to0 volatile, relative to each implied-volatility time series! According to this yardstick, the three models are equally misspecified. This finding is similar to those of Bates (1996a,c) using currency and S&P 500 futures options. By replacing Vit) in equation (18) with R(@), we also obtain maximum- likelihood estimates for Qj, i, and og, and report them in Table IV for the SVSI (as it is the sole model assuming stochastic interest rates). Unlike the previous case for the volatility parameters, the ML estimate of a is similar to its option-implied counterpart, but the ML estimates of dy and ke are several times as large as their option-implied counterparts. That is, interest rate volatility implicit in option prices is consistent with the interest-rate time series, but the2026 The Journal of Finance ‘tmean-reverting speed and the long-run mean of the spot rate implicit in option prices are much lower than the spot rate time series suggests. A possible cause for this departure is the existence of a negative interest-rate risk premium, which tends to make the risk-neutral xq much lower than the true Rp. In summary, the models with stochastic volatility each rely on implausible levels of corzelation p and volatility variation «,, to rationalize the observed option prices. While the SV, the SVJ, and the SVSI are clearly misspecified (though to a lesser degree compared to the BS}.t° how will they perform in prieing and hedging options? We answer this question in the sections to follow. V. Out-of Sample Pricing Perfarmance We have shown that the in-sample fit of daily option prices is increasingly better as we extend from the BS to the SV and then to the SVJ model, even though going from the SV to the SVSI does not necessarily improve the fit much further. As one may atgue, this increasingly better fit might simply be a consequence of having an increasingly larger number of structural parame- ters. To lower the impact of this connection on inferences, we turn to examin- ing each model's out-of-sample cross-sectional pricing performance. For out- of-sample pricing, the presence of more parameters may actually cause over- fitting and have the model penalized if the extra parameters do not improve its structural fitting. For this purpose, we rely on previous day’s option prices to back out the required paremeter/volatility values and then use them as input to compute current day's model-hased option prices, Next, we subtract the model-deter- mined price from its observed counterpart, to compute both the absolute pricing ervor and the percentage pricing error. This procedure is repeated for every call and each day in the sample, to obtain the average absolute and the average percentage pricing errors and their associated standard errors, These steps axe separately followed for the BS, the SV, the SVSI, and the SVJ models, Table V reports the pricing results, where for clarity the standard errors for each estimate are omitted as they are generally very small and close to zero. ‘Three groups of results are presented to reflect differences in the parameter/ volatility values used in the madel price calculations. Pricing errors reported under the heading “All-Options-Based” ave obtained using the parameter/ volatility values implied by all of the previous day’s call options. Those under “Maturity-Based” are obtained using the parameter/volatility values implied by those previous-day calls whose maturities lie in the same category (short- term, medium-term, or long-term) as the option being priced. Pricing errors under “Moneyness-Based” are obtained using the parameterivolatility values implied by those previous-day calls whose moneyness levels lie in the same 9 See Bates (1996e) for other types af consistency tests. He also correta for measurement- error induced correlations among fitting errors acrose different contracts. To move an to our pricing and hedging exercise, we provide only the eonelstency test ust discussed, In addition, we conduct maxiraurlelihood estimations using ATM-option-implied volatilities and find the r- sulte similar to those reported im Table TV.Empirical Performance of Alternative Option Pricing Models 2027 category (OTM, ATM, or ITM) as the option. being priced. In other words, the pricing errors under “Maturity-Based” and “Moneyness-Based” respectively reflect each model's results from the “implied-parameter matrix” treatments based first on maturity and then on moneyness, We begin with the absolute and the percentage pricing errors, respectively given in Panels A and B of Table V, corresponding to *Ail-Options-Based.” Firat, both pricing error measures rank the SVJ model first, the SVSI second, the SV next, and the BS last, except that for a few categories either the SV or the SVSI performs slightly better than the others. According to both measures, the SVST does slightly better than the SVJ in pricing the deepest OTM calls, (regardless of maturity) and the long-term deepest ITM calls. The second part of the last statement may not be surprising since one would expect the long-term deep ITM calls to be the most sensitive to interest rates. But, the fact that the SVJ does not surpass the SVSI in pricing deep OTM calls is somewhat, 4 susprise heeause one would expect the opposite to be true, Second, regardless ‘of option moneyness or maturity, incorporating stochastic volatility produces by far the most important improvement over the BS model, reducing the absolute pricing errors typically by 20 percent to 70 percent. Pricing improve- ‘ment for both OTM (especially the deepest OTM) and ITM calle is particularly striking, For example, take a typical OTM call with moneyness less than 0.94 and with less than 60 days to expiration. From Table I, the average price for such a call is $1.68, When the BS is applied to value this call, the resulting absolute prieing error is, on average, $0.78 as shown in Table V, but when the SV is applied, the average error goes down to $0.42. As another example, for calls of the deepest moneyness (S/K = 1.06) and the longest term-to-expiration (greater than 180 days), their average price is $55.12, the BS gives an average pricing error of $1.87, and the SV results in an average error of $0.85. Table V, together with Figure 1, thus suggests that once stochastic volatility is modeled, adding other features will usually lead to second-order pricing improvement. ‘Third, for a given moneyness category and regardless of the pricing model, the absolute pricing errors typically increase from short- to medium- to long-term options. By the percentage pricing error measure, while the BS exhibits clear moneyness- and maturity-related biases, the other three models do not. except for short-term options. In fact, except for the deepest OTM calls as well as, short-term calls, the percentage pricing errors are all below 1 percent in ‘magnitude for the SV, the SVSI, and the SVJ A possible concern about the relatively large mispricing of short-term as well as OTM options is that the objective function in equation (17) is biased in favor of more expensive ealls (ie, long-term and ITM calls). In addition, as shown in Table I, far more. sample observations are in the more expensive, ITM catego- ries, which is also to the disadvantage of OTM options. As each estimation tries to minimize the sum of squared dollar pricing errors, these two factors must, have exaggerated the extent of poor fit for short-term and OTM options by each, candidate pricing model. This possible exaggeration, however, should not, affect the overall conclusion regarding the pricing structure of short-term and OTM options relative to others. The reason is that in both Table Il and Figure2028 ‘The Journal of Finance Table V Out-of- Sample Pricing Errors ‘For 2 given model, we compute the price ofeach option using the previous day's implied param- ‘ters and implied etack volatility. The reported ahaalute pricing ervor ia the sample sverage ofthe absolute difference between the marltet price and the model price for each call in a given rmoneyness-maturity category. The reported percentage pricing ervar isthe sample average af the market price minus the motel price, divided by the market pre. The resulis under “Ail Optians- ‘Based” are obtained using the parametars implied by all af the previous day's calls; these under “Maturity-Based” using the parameters implied by the peevious day's options af a given maturity ‘short, medium, oF long term to pice the currant day's options ofthe same matarty: those under “Moneyness-Raced” using the parameters implied by the previous day’s options of a given mon= ‘eyness (Out, At or In-tne-money, OTM, ATM, ITM) to price the euent day's options ofthe game ‘moneyness. Tho sample poviod is Sune 1988 May 1991, witha total of $8,748 call option prices. BS, 8V, SVSL, and SVJ, respectively stand for the Black-Scholes, the stoehastc-volatlity model, the elachastic-velauiity and slachaatiesmterest rate model, and the stochastie-valaility model with random jumps, Maturiy- Based Manaynent Based Daveta Espiration Daye Expiration Moneyness — —_vavete epiraton __Davetr Sapien ___DayeseEepireden ‘Sik Model 60 G08 =180 <0 Goa8a_=189 60 60180 180, Prnel A: Abele Peng nora <0 BS H0.78 $109 S159 Slon H4s $78 Boat 05 G08 SV 0424436188042 08082 aan SVSI 937 039 087 08 0400.82 0.20 a8 ose SVE 087 440459 G97 G40 05H 08300 ogt-os7 BS 07S L026 OTR LOT LIB MBS O88 SV 045 Gl GMO O08 Ona SVSI 040 940 085 0404082 aaH nD SV 438 038 0530280890 amt H OSL 097100 BS 051082085 © OSL OBL OBS a7D atk .0K SV 44804100900 408 SVS 047 041 05 0.29 042 O08 oa? at SV) 042 G40 052 Oh, 040008 oat 88 100-108 BS 052069 OBL OMS OHS OMT 50 O88 SV O41 G42 02 040A 0sa SYST 043 G42 O88 Oak 0049s sv) 040 042 O07 Oat 080? oak SL 108105 BS 761g aM OFT kT 8178 SV 045047085 oat Oat 01 Os Ons ge SS) 042 045 084 OAL 04d 080 oak ont 88 SVJ 089 04408808901 OSL 0.9904? F108 RS 082138 LST 0.918} OND TD SV 08% 049088047040 OOO SVS] 982 081 081 04804204? Odom SVJ 043043086 096 «059089 4d 04254Empirical Performance of Alternative Option Pricing Models 2029 ‘Table V—Continued AlLOptions sed Matuvny-Rased ‘Maneynese Base Daystoxpieation DaayetosEnpiation Daye-o-Expiration Moneyness —_Davetoeniraton __Daystrypiaiion _Days te Espation _ ‘S/R Model <60 60-180 =180 <0 0-100 e160 <60 60-180 =180 nel Br Percentage Pricing Brrr <0 BS 65.785 41.870 D6 86m a2 09% —44.058 25.30% 79.21% —18.59% -9.87% SV 3543-278 ~320 1s O82 1808 121 O71 SvSI -1122 138-248 ras -o4l -1982 -a85 -1.71 Sv 19% iad 248 OM 049 1259-158 01 94-097 BS 3828 -1B14 ~188 -\has 118-1553 -234 273 SV 1799-180 038, “ia2 050-920 0.13 —008 SVS 1159 -070 012 -Lo# 017-8138 0.08 008 SVE -1189 -103 008 ost 048 597 907-008 ogr1oo BS 3478 -an1 -195 0). Furthermore, if jump risk is present. as well, the position to be taken in the underlying stock must. also hedge the impact of jump risk on the target option, which is reflected in the last term of equation (21), This term is increasing in A and js, meaning that the larger the random-jump risk, the more adjustment need he made in the hedging position. Therefore, by considering an option model with jumps, ‘one makes the resulting hedging strategy also immunized against jump risk. In theory the constructed partial hedge requires continuous rebalancing to reflect the changing market conditions. In practice, only discrete rebalancing is possible. To derive a hedging effectiveness measure, suppose that portfolio rebalancing takes place at intervals of length At, As described above, at time t short the call option, go long in X,(t) shares of the stock and invest the residual, Xo(0), in an instantaneously maturing riskfree bond. The combined position is a self-financed portfolio. Next, at time t + At ealeulate the hedging error as follows H(t + At) = Xs(e)S(¢ + At) + Xoltie™ — C(e + At, r- A‘). (23) At the same time, reconstruct the self-financed portfolio, repeat the hedging error caleulation at time ¢ + 28¢, and co on, Record the hedging errors Hit + lad), for 2 = 1, ..., M = (x ~ eye. Finally, compute the average absolute hedging error as a function of rebalaneing frequeney Af: (At) ~ (MIM)Empirical Performance of Alternative Option Pricing Models 2038 fs | He + 1M0 |, and the average dollar-value hedging error: H¢At) = (UM) 2H, He + tas). Single-instrument hedging errors under the BS, the SV, the SVSI, and the SVJ models are similarly determined accounting for their modeling differ- ences. In the case of the 8V.J model, the same three terms as in equation (21) still determine the single stock position, except that the characteristic fune- tions used in the calculations should be adjusted to reflect the constant- interest-rate assumption. For the SV and the SVSI models, the jump risk- related term (the last term) does not appear and the other two terms remain. For the BS model, only the first term in equation (21) is used to determine the minimum-variance hedge. ‘To obtain the hedging results presented in Table VII, we follow the three steps below. First, estimate the set of parameter/volatility values implied by all call options of day t — 1. Next, on day ¢, use these parameter/volatility estimates and the current day’s spot index and interest rates, to construct the desired hedge as given in equation (21) or its model-specific version. Finally, calculate the hedging error as of day ¢ + 1 if the hedge is rebalanced daily or as of day t + 5 if the rebalancing takes place every five days. These steps are repeated for each option and every trading day in the sample. The average absolute and the average dollar hedging errors for each moneyness-maturity category are then reported for each model in Table VIL. Note that hedging results obtained respectively from the Maturity-Based and the Moneyness- Basod treatments are almost the same as these in Table VII and hence not reported. Based on the absolute hedging errors in Table VIT, the SV model is the best overall performer, followed by the SVJ model, and then by the SVSI. But, according to the dollar hedging errors, the SVSI performs the best among all four in hedging both OTM calls Grrespective of maturity) and long-term ITM calls. It is also clear from both Panels A and B that, regardless of hedge rebalancing frequency, the real significant improvement. by the stochastic- volatility models over the BS occurs only when OTM calls are being hedged. ‘When other categories of calls are the hedging target, the performance is in most cases virtually indistinguishable among the four models. The hedging- based ranking of the models ie thus in sharp contrast. with that. obtained ‘earlier based either on out-of-sample pricing or on internal consistency of a model's estimated structural parameters. ‘The finding that the SVJ does not improve over the SV's hedging perfor- mance seems somewhat surprising, especially given the SVJ’s better out-of- sample pricing performance (Table V). As discussed by Bates (1996a) in a different context, a possible explanation is as follows. In Table III, the average implied jump-intensity parameter A (under “All Options”) is 0.59 times per year, which means it takes, on average, about a year and a half for a jump of the average magnitude to occur. In Table VII, the results are obtained when each hedge is either rebalanced daily or once every five days. Clearly, during a one-day or five-day interval the chanee for a significant price jump (or fall) to occur is very small. Thus, once stochastic volatility is modeled, hedging per-2036 The Journal of Finance ‘Table VIL Single-Instrument Hedging Errors {mn chia tail hedge ofealis use ony the underying ano. the hedeing Instrument. Pararncters and ‘set volatility implied by al options ofthe previoun day are aed ta ctabish the crent das hedges ‘hich are then ligudated the flowing day ov five Jaya later. Far eath target call opin. 5 hedg ‘Sor lau of the Hiquldation day, the difernce petvonn ita marke price ad the repcating prtle alah average solic hedging omar andthe average dallas hedping error ace reported for each model and for each. moneyness maturity category, ‘The sumple pered ts June 1886-May 2804, Tn Calculating the hedging errors getarated with daily (or Seda) hedge rebalancing 15,041 (or 11,704) ‘haervations are used 8S, SV, SYST, and SVM veapocuvey sand for the Black Schole, the eochastc- ‘oletiiey del, the stochaalcvelality and stochaniesntarestrale mde, snd Ue stochasticlatliey ‘adel with random jampe 1-Day Revision ‘S:Day Rovision Moneynees Days-io- Expiration Days-to Expiration Sik Model <60 60-180 =180 <6 O18) =180 Panel A: Absolute Hedging Errors <094 BS. NA $05 $043—=NA $0.95 $088 sv. 025 0.37 057 omt Syst oz 0.39 062 za 8vd oz 0.39 0.59 074 094-097 BS. om agg 042 oro on 0.80 sv. 023082 04809 os 072 syst 033.34 0480.80 an 0.80 svi 033.88 44 oa os oe oar-10 BS 08 036 04s am ost sv. 030 03s 0427s on 073 syst 03 .88 048 a on one svl 030.36 043 at 73 076 100-108 BS. oa7 age 044 are ora a6 sv 088 a3? 04308 0.68, 0.88 syst 039 ata 045g 078 07s SvI 097.87 oss Oo 70 on 10.06 BS. on az 045 go 068 om . 039 a3 045, 0 485, on sve a40 at) 048 ama ama 0.76 Syl 0984.88 04868, 086 075 21.06 BS. 037 as 04881 085, 067 sv oss a9 045050 055, og SVsI 0360 at 04581 9.60 0.68 svi 035 asa oss 050 055, 062 formance may not be improved any further by incorporating jumps into the ‘option pricing framework (at least when the hedge is rebalanced frequently). B. Delta-Neutrat Hedges ‘Now, suppose that one can use whatever instruments it takes to create a perfect delta-neutral hedge. The need for a perfect hedge can arise in situa- tions where not only is the underlying price risk present, but also are volatility,Empirical Performance of Alternative Option Pricing Models 2037 Table VII-Continued Day Revisor Moneys __baveenpiaton Sk Model oo so-igo ima Paral 8 Bala Hedging Pros oS S00 WA 5 oot svst <300 sv “one os-o9 B83 -000-0.48 sv aan “om “118 sist ane “oo “16 Sv) ant “ome “02 asrieo Bs 005 ca 034 sv “oot cco “025 Syst noe om -a28 sv aoe oa 080 Lotsa BS a. eco 037 sv noe oo as Syst aoe om 30 Sv dos moor -033 Lowtgs BS 004 “oes 03, Sv “005 moe. M07 Syst “05 Zoot “a8 sv “a “os “8 aioe BS 005 -ome -o18 No “on: “on Svst 005 moe: a8 SW “008 =o02 “ate interest rate and jump risks. In conducting this exercise, however, we should first recognize that a perfect hedge may not be practically feasible in the presence of stochastic jump sizes (e.g,, for the SVJ and the SVSI-J models) ‘This difficulty is seen from the existing work by Bates (1996a), Cox and Ross (1976), and Merton (1976), For this reason, whenever jump risk is present, we follow Merton (1976) and only aim for a partial hedge in which diffusion risks are completely neutralized but jump risk is left uncontrolled for. We do this with the understanding that the overall impact on hedging effectiveness of not controlling for jump risk can be small or large, depending on whether the hedge is frequently rebalanced or not. Suppose again that the target is a short position in a call option with + periods to expiration and strike price K. Taking the SVSI-J model as the point of discussion, the hedger will need a position in (i) some X,(t) shares of the underlying stock (to contro! for price risk), (i) some Xp(¢) units of a period discount bond (ta cantrol for R(t) risk), and (ii) some X;¢) units of another call option with the same maturity but a different strike price K (or any option on2038 The Journal of Finance the stock with a different maturity) to control for volatility risk V(t). The timne-t value of this replieating portfolio is then Xo(t) + Xe(@S(t) + Xq(t)B, 1) + XLeiCtt, 1; K), where X(t) denotes the residual eash position. Deriving the dynamies for the replicating portfolio and comparing them with those of Cit, 5B), we find the following solution: Av(t, 5K) Xelt) = By ern Xo(t) = Aglt, 1) K)— Aglt, 7; RYXe(t) (28) 1 Xs) = agin (elt 5 R)Xelt) = Agit, 7, KY (26) Xa(t) = Ct, 15 K) ~ Xelt)S(E) ~ elt, 5 K) — XelCBE, 2), (27) where all the primitive deltas, Ag, Ay and Ay, are ag determined in equations (40)-(12). To examine the hedging effectiveness, at time t short the call option and establish the hedge as just described. After the next interval, compute the hedging error according to H(t + At) = Xge™ + Xs(E)S(t + At) + Xs(t)BUe + At, 7 — At) + Xo(t}Cit + At, r— At; R}— Cit + At, r~ ats Ky. (28) Like in the previous case, repeat this calenlation for each date ¢ and every target call in the sample to obtain a collection of hedging errors, which is then used to compute the average absolute and the average dollar-value hedging errors, both as functions of rebalancing frequency At. For the BS model, the delta-neutral hedge is the same as the previous single-instrument hedge and its hedging error measures are similarly caleu- lated as in (28), except that A = Xp(t) = Xe(t) = 0 and Xe(t) is the BS delta. ‘Thus, the BS delta-neutral hedge involves no other instrument. than the underlying stack. In the case of the SV model, set 1 = Xp(t) = 0 and let Sg and Aybe as determined in the SV model. Its delta-neutral hedge hence consists of 1 position in both the stack and the second option contract. For the SVJ model, set Xp(¢) = 0 and let As and Ay he as determined in the SVJ model. Clearly, when A ~ 0, the hedge created in equations (24)-(27) beeames the ane corre. sponding to the SVSI model In the eases of the SV, the SVSI, and the SVJ models, the hedge in equation (28) requires (j} the availability of prices far four time-matched target and hedging-instrumental options: C(t, 7, K), C(t, 7K), Clt + At, 7 ~ At; K), Ct + At, 7 ~ At; K) and (ii) the computation of Ag, Ay, and A for both the target and the instrumental option. Due to this requirement, we use as hedging instru- ments only options whose prices on both the hedge-construction day and the following liquidation day were quoted no more than 15 seconds apart from theEmpirical Performance of Alternative Option Pricing Models 2089 times when the respective prices for the target option were quoted. This constraint guarantees that the deltas for the target.and instrumental options ‘on the same day are computed based on the same spot price. The remaining sample for both this delta-neutral hedging exercise and the previous single- instrument hedging contains 15,041 matched pairs when hedging revision eceurs daily, and 11,704 matched paits when rebalancing takes placo at five-day intervals. As before, we use the current day’s spot index and interest rates, but parameterivolatility values implied by all of the previous-day's options, to deter- ‘mine the current day's hedging positions for each target call. Table VIII presents the average absolute and the average dollar hedging errors aeross the 18 maney- ness-maturity categories and for each of the four models. A striking pattern ‘emerging from this table is that, irrespective of moneyness-maturity category, the three modela with stochastic volatility have virtually identical delta-neutral hedg- ing errors. Therefore, consistent with the resulta of the previous subsection, adding jumps or stochastic interest rates to the SV model does not improve its hedging performance, at least: with respect to our eample data ‘When the hedges are revised daily, the BS delta-neutral hedging errors are usually two to three times aa high as the corresponding hedging errors for the other three models. Improvement by the stochastic-volatility models is even more evident when the hedge revision frequeney changes from daily to once every five days: the BS hedging errors increase dramatically while the other modele’ do not increase by much. This seems to suggest that the other three models perform much better than the BS. ‘The last observation perhaps raises more questions than answers. Is the hedging improvement by the three models with stochastic volatility a consequence of better model specification, or is it mostly due to the inclusion of a second option in their delta-neutral hedges? Is the fact that hedge revision frequeney does not affect the hedging effectiveness of the throe models by as much as it affects the performance of the BS a consequenee of better model specification, or is it-due to the indirect effect of the second call option on the position gamma measure? To answer the first question, we implement the so-called delta-plus-vega- neutral hedge for the BS model, in which the underlying stack and a second eall ‘option are used respectively to neutralize the sensitivity of the hedge to underly- ing price risk and volatility risk. This type of strategy is clearly inconsistent with, the BS setup, but such a treatment may in some sense give the BS a fairer chance. In particular, if the BS delta-plus-vega-neutral hedge results in hedging errors comparable to those from the delta-neutral hedges of the other three models, it will simply suggest that model misspecification may only have a secondary effect ‘on hedging. We report the average hedging errors of this BS delta-plus-vega- neutral strategy under the abbreviation “BSDV™ in ‘Table VIII. Except-for the 'TM categories, hedging performance is indistinguishable between the BS delta-plus- vega-neutral strategy and the delta-neutral strategies for the other three models. For the two ITM call option groups (with S/K > 1.03), however, incorporating stochastic volatility does improve upon the BS delta-plus-vega-neutral hedging performance. Thus, for hedging these ITM calls, more appropriate model specifi2040 The Journal of Finance ‘Table VET Delta-Neutral Hedging Errors In this table, all delte-neutral hedges of ealls use as many hedging instruments as there are sourees of risk Cexcept the Jurmp risk} assumed in a given option model. The aaly exception is the BS delte-plus-vega-neutral strategy, denoted by BSDV, waich uses the underlying aaset and a second call option to neutralize both the delta and vega risks of the target eal, ased on the Black-Scholes model. Parameters ard spot volatility implied by all options ofthe previous day are used ta establish ehe eurrant days hedges, whieh ae then liquidated the fallowing day ar five days later. For each target call option, its hedging error is, as of the Liguldation day, the difference ‘between its market pre and the replicating portflco value. The average absolute hedging error and the average dollar hedging error are reported foreach model and for each moneyness-malurily ‘category. The sample period is June 198#-May 1991. In ealeuating the hedging ervare generated with daily (ance every five days) hedge rebalancing, 16,041 (11,704) observations are used. BS, SV, SVSI, and SVJ, reepectivel, stand for the Black-Scholes, the stochastiewolanlity modal, the sochnsticwolatllty and stachastic interest-rate model, snd the stochastic-olatility model with random jumps 1-Day Revision ‘Day Revision Me DaysBxpieation Dayeto-Pxpiraton STR Mo ee Panel A: Abwolule Hedging Rerore 084 BS NA 4035 $043.—=SCWNAS085 08 BsDVv 035038 038 028 sv a 030 16 oat S¥SI as 0m 016 os SVE am oat 016 082 oat-os7 BS om = o38 adams Spy Oda 843,19 tk sv O08 os 20) te svsl 08320 os te er ost BS 036 0860 te KB BSD 019813181 sv Or 806K SVSI 0130181801 SV 08 OMB BAB 190-103 BS os 088 Ok OTB 0H BSDY 018k .18 18D sv. O18 ld 81s SYST 018014 SMS odd Ta 108-106 BS og. 0.87 0s 85 BsDY 016 = 013, Sv, 015 0188? ata svsl0l5 013,16 a ) 2106 Bs oa a8 eK. BSpY ots 016.238, sv, os ork 8a? Tas svsl 418 01186 k vs as O14 02000 axT watsEmpirical Performance of Alternative Option Pricing Models 2041 ‘Table VUL-Continued 3-Day Revision wt Days-o- Expiration Sik Mode 60" eo-g0 89 <0 aIRO Pane! B: Dollar Hedging Error <094 BS, NA $003 $002 NA $0.19 BSpVv aor 001 00 sv oor 0.00 oo SVS aor 0.00 0.00 sva nor 0.00 oo. ogosT = BS 0.18007 00-049 oa Bspv -004 = 0.01 000-008. -.01 sv oo 0.01 aco 000, a0. syst dor 0.01 aco 0.00 00 SWI oot 0.01 am 00k 4.03 agTi0g — -BS_ 005-003 000 ~ase 0.08 BSspY ot oat or 08 008 sv oon 2.00 dor aaa 02 syst oat oan oor aaa oor sv al 2.00 oor 000, 0.02 10-102 BS 0.80.01 000-037 -o spy oo = 0.40 Dor aor 0.03 sv 001 0.00 oot 001 aa svsl-om1 0.00 oor 0.00 0.03 sys 0.01 0.00 oor -0.02 0.02 Lobi0é = «BS -004 = 0.02 003-0 0.18 Bspv aor ood 001-005 0.08 sv. oon og — -0.00 -002 ~043 syst or 900-000-003 0.02 sys or ooo aor 0.08 0.08 21.06, BS 005-003-002 ~08 0.08 SPY 002-001 010.08, 0.02 sv aol 201 0.00 0.03. -oat SvsL_ 0.01 9.00 00 = 002-000 oor sv) 001-001 aga 00a 02-0 cation matters, whereas for hedging other calls, even ad hoc hedging strategies based on the BS may do just fine, Given the performance of the BSDV in Table VIII, it is apparent that the relative insensitivity of the other three models? hedging errors to revision frequency must ke mostly due to the use of the instrumental call option. Tt is the instrumental option position that not only neutralizes the volatility risk but also dramatically reduces the remaining gamma risk in the hedge. To see. this point, take the SV delta-neutral hedge as an example. Denote the gamma (with respect to the spot price) of the target call by s(t, 7K) and that of the instrumental call by T(t, =, K), a detailed expreesion of which is provided in the Appendix. Since the position in the instrumental eall is Xc(¢) = [Ay(t, 5 Klay,2042 The Journal of Finance B)], the remaining gamma value of the SV delta-neutral hedge is given by Tyét, 15 K) - Xcftiglt, 7; K). For a typical delta-neutral hedge under stochastic volatility, the remaining gamma value is close to zero. The following are some examples based on June 3, 1988, when the spot S&P 500 was at 266,42: ‘« For hedging the ATM call with strike price 265, the position taken in the chosen instrumental option is X= 0.97 and the remaining SV gamma value in the hedge ia 0.020 - 0.97 x 0.022 = -0.001; © For hedging the ITM call with strike price 250, Xp remaining SV gamma of the hedge is 0.011 ~ 0.81 x 0.014 5 For hedging the OTM call with strike price 275, Xo = 0.97 and the remaining SV gamma of the hedge is 0.022 ~ 0.97 % 0.022 = 0.001. For a typical BSDV hedge, the remaining BS gamma is also close to zero, which explains why the BSDV hedging errors are relatively insensitive ta revision frequeney as well. Another patiern to note from Table VIII is that the BS model's dollar hedging errors are always negative, indicating that the model ovethedges each target option, whereas the dollar hedging errors of the other models are more, random and can take either sign. Therefore, the BS formula exhibits a eys- tematic hedging bias, while the others do not. ‘Comparing Tables VIL and VIII, one can see that for a given option model, the conventional delta-neutral hedge (using as many instruments as there are somees of uncertainty) performs far better than its single-instrument. counter part, for every moneyness-maturity category. This may not he surprising as the former type af hedge involves more instruments (exeept tnder the BS model)? VIL Coneluding Remarks We have developed a parsimonious option pricing model that admaits sto- chastie volatility, stochastic intorest rates, and random jumps. It is shown that this closed-form pricing formula is practically implementable, leads to useful analytical hedge ratios, and contains many known option formulas as special cases. This last feature has made it relatively straightforward to study the relative empirical performance of several models of distinct interest. Our empirical evidence indicates that. regardless of performance yardstick, taking stochastic volatility into account is of the first-order impartanee in According to Rubinstein (1985), the volatility smile pattern and the nature of pricing biases ‘aro ime-dopendont. To seo whether our conclusion may be reversed, we separately examine the pricing and the hedging performance of the models in three eubperiods: June 1988-May 1989, ‘une 1989-1600, and June 1990-May 1901. Fach eubperiod contains about 10,000 eall option ‘observations. The eubperiod results are qualitatively the aame as those, espactively, in Tables V ‘and VIL Separately, we examine the pricing and hedging error measures of each model when the structural parametars are not updated daily. Rather, we retain the structural parameter values ‘estimated from the options of th frst day of each month and thon, fr the remainder of the month, use them as input to compute the eocresponding model-based prices foreach traded option, except thot the implied spoe volatility is updated each day based on the previous day's option prices. The ‘obtained absalute prising and hedging errors fr the subperiad June 1990-May 1991 indieste that. the performance ranking of the four models also remains the same.Empirical Performance of Alternative Option Pricing Madels 2043 improving upon the BS formula. In terms of internal consisteney, the SV, the SVJ, and the SVSI are still significantly misspecified. In particular, to ratio- nalize the negative skewness and excess kurtosis implicit in option prices, each model with stochastic volatility requires highly implausible levels of volatility- return correlation and volatility variation. But, such structural misspecifiea- tions do not necessarily preclude these models from performing better other- wise. According to the out-of-sample pricing measures, adding the random- jump feature to the SV model can further improve its performance, especially in pricing short-term options; whereas modeling stochastic interest vates can enhance the fit of long-term options. With both the SVSI and the SVJ, the remaining pricing errors show the least contract specific or market-conditions- related biases. For hedging purposes, however, incorporating either the jump or the SI feature does not seem to improve the SV model's performance further ‘The SV achieves the best hedging results among all the models studied, and its, remaining hedging errors are generally quite smalt. Therefore, the three performance yardsticks employed in this article can rank a given set of models differently as they capture and reveal distinct aspects of a pricing model. Overall, our results support the claim that a model with stochastic volatility and random jumps is a better alternative to the BS formula, because the former not only performs far better but also is practically implementable. ‘The empirical issues and questions addressed in this article can also be reexamined using data from individual stock options, American-style index options, options on futures, currency and cammodity options, and so on. Even- tually, the acceptability of option pricing models with added features will be judged not only by its implomentability, its internal consistency, and its pricing and hedging performance as demonstrated in this paper, but also by its success or failure in pricing and hedging other types of options. These exten- sions are left for future research. APPENDIX Proof of the Option Pricing Formula in Equation (8). The valuation partial differential equation (PDE) in equation (7) can be rewritten as: Lye { 1 yee FO 1 a #O a apa* Ro Mus 9 Vaz t mY spay ta GY aye 1 aC ac ac 12 gr te — 26 OBR Spat (On — «agg ~ 5, ~ RC + ABGKCU, HL + inf +), RV) - CU, HL, R,V}=0, (AD) ac + fa KV SG where we have applied the transformation L(t) = In{S(e)). Inserting the con- Jectured solution in equation (8) into (A1) produces the PDEs for the risk-2044 ‘The Journal of Finance neutralized probabilities, Il, forj = 1, 2: aa, 2 1 yam, am, 14, aT, Set (a- Mery VISE + oo spay 20" a all, ap Ph aM, all, [0 = (a= pauVISg + 9 OR ape + [On ~ ka] Rt 55" = Apa, + ABQ + In + JPM, 7; L + Inf + J], R, VP ~ MG, 5 L,R,V)}=0, (A2) oy 1 ay oa aLay * 2°" oF 1 at oh aBE,A\ atl, _ all, fa - [xe ond + pov Soe HEWN GY 9 OR RY “Bea aR |” lan ~ ar + ABG(MLG, 5 E+ inft +I], R, V)— The, pL, R, Y= 0. (48) Observe that equations (82) and (Aa) are the Folkker-Planck forward equations for probability functions. This implies that fl, and fl, must indeed be valid probability functions, with values bounded between 0 and 1. These PDEs must be solved separately subject to the terminal condition: +2 «May Te +, 0)= Leese J ‘The corresponding characteristic functions for Il, and IT, will also satisfy similar PDEs: af; 1 att ay tyke (R- Ma ty V) ee pnw te thaw sh fi 7 a hy + £6. — 00. — p91 ES ate AE 4 [4 — na) = Ausf + ABQ((L + Inf + IAC, LE + Mn + J, R, V) At, 7 L, R, Vy} = 0, (A5) and afs ah 1 an of, \B + ony sey + boty 2B (4 — nV) a, 1 ap the, oh aBte, 0) Oh, af Ww pet [Oro (se nary aR) lar ar + AE gift, 7) L + Inf +d), RV) ~ fat. 5 L, RL V)}= 0. (46)Empirical Performance of Alternative Option Pricing Models 2045 with the boundary condition: ft + 7, 0; bp =e = 1, 2 (a7) Conjecture that the solution to the PDEs (AS) and (A6) is respectively given by filt, 7, S(t), RE}, Vie); $) = explu(r) + x(r)R(E) + xc) Vie) + id In[S(t)]} (AB) fill, 7 SU), RE), Vi 6) ‘= expfa(z) + y,(7)R() + ye(7)V() + id In[S(2)] ~ Inf BEE, 7))) (AD? with 1(0) = x,{0) = x,(0) = 0 and 2(0) = y,(0) = 94(0) = 0. Solving the resulting systems of differential equations and noting that. B(t + 1, 0) = 1 respectively produce the following desired characteristic functions: fe ~ Kg)(1~eF*) filt, = ex Saf inf. - ee + [ée— wall & [en eet UL + igipadta - e“%) | — ele a + A+ dood + ig InfSCOH] Big — e°* * Fey [én — walt — 7) RO HALL + wadal Cl + wieterierees — 1] — Aidpor iglid + 1) - e-$ Teed vio}, (A10) + [8 wal] | Hern iepa.r| +89 IS] ~nkBG, 9] + gar MO Rey F Ar[CL + pyletaHe i — 1) — didbugr + idtig — A -e &) [El x, + ibpo, KL — vo}. cally2046 The Journal of Finance where fan eh 205d, & > ye, — 1 FE )pa, — 16Gb + Dos, fhe fee 2oME— 1), and b= yfu, ida. — idlid — Nor. The SI, the SV, the SVSI, the SVJ models are all nested within the general formanla in equation (8). Tn the SVJ case, for instance, the partial derivatives with respect to R vanishes in equation (A1). The general solution in equations {A8}

You might also like

• The Subtle Art of Not Giving a F*ck: A Counterintuitive Approach to Living a Good Life

  

  From Everand

  The Subtle Art of Not Giving a F*ck: A Counterintuitive Approach to Living a Good Life

  Mark Manson

  Rating: 4 out of 5 stars

  4/5 (5807)
• The Gifts of Imperfection: Let Go of Who You Think You're Supposed to Be and Embrace Who You Are

  

  From Everand

  The Gifts of Imperfection: Let Go of Who You Think You're Supposed to Be and Embrace Who You Are

  Brene Brown

  Rating: 4 out of 5 stars

  4/5 (1091)
• Never Split the Difference: Negotiating As If Your Life Depended On It

  

  From Everand

  Never Split the Difference: Negotiating As If Your Life Depended On It

  Chris Voss

  Rating: 4.5 out of 5 stars

  4.5/5 (843)
• • Magazines
  • Podcasts
  • Sheet music
• Principles: Life and Work

  

  From Everand

  Principles: Life and Work

  Ray Dalio

  Rating: 4 out of 5 stars

  4/5 (608)
• The Glass Castle: A Memoir

  

  From Everand

  The Glass Castle: A Memoir

  Jeannette Walls

  Rating: 4.5 out of 5 stars

  4.5/5 (1716)
• Grit: The Power of Passion and Perseverance

  

  From Everand

  Grit: The Power of Passion and Perseverance

  Angela Duckworth

  Rating: 4 out of 5 stars

  4/5 (590)
• Sing, Unburied, Sing: A Novel

  

  From Everand

  Sing, Unburied, Sing: A Novel

  Jesmyn Ward

  Rating: 4 out of 5 stars

  4/5 (1104)
• Hidden Figures: The American Dream and the Untold Story of the Black Women Mathematicians Who Helped Win the Space Race

  

  From Everand

  Hidden Figures: The American Dream and the Untold Story of the Black Women Mathematicians Who Helped Win the Space Race

  Margot Lee Shetterly

  Rating: 4 out of 5 stars

  4/5 (897)
• Shoe Dog: A Memoir by the Creator of Nike

  

  From Everand

  Shoe Dog: A Memoir by the Creator of Nike

  Phil Knight

  Rating: 4.5 out of 5 stars

  4.5/5 (537)
• The Perks of Being a Wallflower

  

  From Everand

  The Perks of Being a Wallflower

  Stephen Chbosky

  Rating: 4.5 out of 5 stars

  4.5/5 (2104)
• The Hard Thing About Hard Things: Building a Business When There Are No Easy Answers

  

  From Everand

  The Hard Thing About Hard Things: Building a Business When There Are No Easy Answers

  Ben Horowitz

  Rating: 4.5 out of 5 stars

  4.5/5 (346)
• Bad Feminist: Essays

  

  From Everand

  Bad Feminist: Essays

  Roxane Gay

  Rating: 4 out of 5 stars

  4/5 (1016)
• Elon Musk: Tesla, SpaceX, and the Quest for a Fantastic Future

  

  From Everand

  Elon Musk: Tesla, SpaceX, and the Quest for a Fantastic Future

  Ashlee Vance

  Rating: 4.5 out of 5 stars

  4.5/5 (474)
• Her Body and Other Parties: Stories

  

  From Everand

  Her Body and Other Parties: Stories

  Carmen Maria Machado

  Rating: 4 out of 5 stars

  4/5 (821)
• The Outsider: A Novel

  

  From Everand

  The Outsider: A Novel

  Stephen King

  Rating: 4 out of 5 stars

  4/5 (1850)
• The Emperor of All Maladies: A Biography of Cancer

  

  From Everand

  The Emperor of All Maladies: A Biography of Cancer

  Siddhartha Mukherjee

  Rating: 4.5 out of 5 stars

  4.5/5 (271)
• The Sympathizer: A Novel (Pulitzer Prize for Fiction)

  

  From Everand

  The Sympathizer: A Novel (Pulitzer Prize for Fiction)

  Viet Thanh Nguyen

  Rating: 4.5 out of 5 stars

  4.5/5 (122)
• Angela's Ashes: A Memoir

  

  From Everand

  Angela's Ashes: A Memoir

  Frank McCourt

  Rating: 4.5 out of 5 stars

  4.5/5 (440)
• Brooklyn: A Novel

  

  From Everand

  Brooklyn: A Novel

  Colm Toibin

  Rating: 3.5 out of 5 stars

  3.5/5 (1946)
• The Little Book of Hygge: Danish Secrets to Happy Living

  

  From Everand

  The Little Book of Hygge: Danish Secrets to Happy Living

  Meik Wiking

  Rating: 3.5 out of 5 stars

  3.5/5 (401)
• The World Is Flat 3.0: A Brief History of the Twenty-first Century

  

  From Everand

  The World Is Flat 3.0: A Brief History of the Twenty-first Century

  Thomas L. Friedman

  Rating: 3.5 out of 5 stars

  3.5/5 (2259)
• A Man Called Ove: A Novel

  

  From Everand

  A Man Called Ove: A Novel

  Fredrik Backman

  Rating: 4.5 out of 5 stars

  4.5/5 (4610)
• The Art of Racing in the Rain: A Novel

  

  From Everand

  The Art of Racing in the Rain: A Novel

  Garth Stein

  Rating: 4 out of 5 stars

  4/5 (4203)
• A Tree Grows in Brooklyn

  

  From Everand

  A Tree Grows in Brooklyn

  Betty Smith

  Rating: 4.5 out of 5 stars

  4.5/5 (1929)
• Steve Jobs

  

  From Everand

  Steve Jobs

  Walter Isaacson

  Rating: 4.5 out of 5 stars

  4.5/5 (806)
• The Yellow House: A Memoir (2019 National Book Award Winner)

  

  From Everand

  The Yellow House: A Memoir (2019 National Book Award Winner)

  Sarah M. Broom

  Rating: 4 out of 5 stars

  4/5 (98)
• Devil in the Grove: Thurgood Marshall, the Groveland Boys, and the Dawn of a New America

  

  From Everand

  Devil in the Grove: Thurgood Marshall, the Groveland Boys, and the Dawn of a New America

  Gilbert King

  Rating: 4.5 out of 5 stars

  4.5/5 (266)
• The Woman in Cabin 10

  

  From Everand

  The Woman in Cabin 10

  Ruth Ware

  Rating: 3.5 out of 5 stars

  3.5/5 (2521)
• Yes Please

  

  From Everand

  Yes Please

  Amy Poehler

  Rating: 4 out of 5 stars

  4/5 (1898)
• A Heartbreaking Work Of Staggering Genius: A Memoir Based on a True Story

  

  From Everand

  A Heartbreaking Work Of Staggering Genius: A Memoir Based on a True Story

  Dave Eggers

  Rating: 3.5 out of 5 stars

  3.5/5 (231)
• Team of Rivals: The Political Genius of Abraham Lincoln

  

  From Everand

  Team of Rivals: The Political Genius of Abraham Lincoln

  Doris Kearns Goodwin

  Rating: 4.5 out of 5 stars

  4.5/5 (234)
• Fear: Trump in the White House

  

  From Everand

  Fear: Trump in the White House

  Bob Woodward

  Rating: 3.5 out of 5 stars

  3.5/5 (738)
• Wolf Hall: A Novel

  

  From Everand

  Wolf Hall: A Novel

  Hilary Mantel

  Rating: 4 out of 5 stars

  4/5 (3811)
• John Adams

  

  From Everand

  John Adams

  David McCullough

  Rating: 4.5 out of 5 stars

  4.5/5 (2409)
• On Fire: The (Burning) Case for a Green New Deal

  

  From Everand

  On Fire: The (Burning) Case for a Green New Deal

  Naomi Klein

  Rating: 4 out of 5 stars

  4/5 (74)
• The Light Between Oceans: A Novel

  

  From Everand

  The Light Between Oceans: A Novel

  M.L. Stedman

  Rating: 4.5 out of 5 stars

  4.5/5 (789)
• The Unwinding: An Inner History of the New America

  

  From Everand

  The Unwinding: An Inner History of the New America

  George Packer

  Rating: 4 out of 5 stars

  4/5 (45)
• Manhattan Beach: A Novel

  

  From Everand

  Manhattan Beach: A Novel

  Jennifer Egan

  Rating: 3.5 out of 5 stars

  3.5/5 (792)
• The Constant Gardener: A Novel

  

  From Everand

  The Constant Gardener: A Novel

  John le Carré

  Rating: 3.5 out of 5 stars

  3.5/5 (104)
• Rise of ISIS: A Threat We Can't Ignore

  

  From Everand

  Rise of ISIS: A Threat We Can't Ignore

  Jay Sekulow

  Rating: 3.5 out of 5 stars

  3.5/5 (137)
• Little Women

  

  From Everand

  Little Women

  Louisa May Alcott

  Rating: 4 out of 5 stars

  4/5 (104)
• SSRN Id3512994

  

  Document34 pages

  SSRN Id3512994

  Mohit Gupta

  No ratings yet
• Mystic

  

  Document218 pages

  Mystic

  Mohit Gupta

  No ratings yet
• TT18 Dissertation Vu 0

  

  Document90 pages

  TT18 Dissertation Vu 0

  Mohit Gupta

  No ratings yet
• Michael Kapler

  

  Document11 pages

  Michael Kapler

  Mohit Gupta

  No ratings yet
• Gs Am Factor Fidelity Webinar 112916

  

  Document27 pages

  Gs Am Factor Fidelity Webinar 112916

  Mohit Gupta

  No ratings yet
• Alex Gillula SeniorDesignPaper - OptimalHedge

  

  Document46 pages

  Alex Gillula SeniorDesignPaper - OptimalHedge

  Mohit Gupta

  No ratings yet
• Skinny Math 051613 Geo Brownian Motion

  

  Document10 pages

  Skinny Math 051613 Geo Brownian Motion

  Mohit Gupta

  No ratings yet
• Sept 13 2006 DDE Links

  

  Document14 pages

  Sept 13 2006 DDE Links

  Mohit Gupta

  No ratings yet