Baixe o app para aproveitar ainda mais
Prévia do material em texto
impa Instituto Nacional de Matemática Pura e Aplicada Backtesting SVI Pararneterization of Implied Volatilities Author: Rafael Balestro Dias da Silva Advisor: Vinícius Viana Luiz Albani Rio de Janeiro July 2016 Acknowledgements I dedicate the present Work to my parents Ana and Mauro, girlfriend Luiza for the understanding in relation to my co Who helped me a lot With the mathematics and to my goo the presence and great debates. Also, I Would like to than Raphael, Luis, Pedro and Zeca - and to my friends at Vinc special thank to Professor Jorge Zubelli, Ph.D and to my a‹ the immeasurable support during the construction of this ii Who made it possible, to my beloved ns (f. Á tant absence, to my brother Vitor friends Gabriela and Leonardo for to my friends at Ancar Ivanhoe i Partners - Renato and Tiago. A :lvisor Vinicius Albaní, Ph.D for all master thesis. Abstract Since the stock market crash in 1987, many generalizations of Black-Scholes model have been introduced, in order to incorporate the so-called volatilíty smile effect. In this master thesis We study the numerical aspects of the stochastic Volatility inspired (SVI) parameterization of implied volatility smile. lntroduced at the end of 1990°s, this model - Which is basically a parameterized functional form - presents a nice smile adherence, Which made it popular Within market practitioners mainly by its low computational cost. We present numerical experiments using real data, evaluating how reliable Would the fitted parameters be on calculating volatilities for future dates. Key Words: Financial Options, Financial Derivatives, Volatility Smile, Black-Scholes Formula, SVI fit, Stochastic Differential Equation, Variance l\/lodelling. iii Contents Index Introduction 1 Local Volatility Surface and Dupire”s Equation 1.1 Derivation of Dupire Equation . . . . . . . . . . . . . 1.2 Local Volatility and Implied Volatility . . . . . . . . 2 SVI Parameterization: An Introduction 2.1 Roger Lee°s Moment Formula . . . . . . . . . . . . . 2.2 Static Arbitrage . . . . . . . . . . . . . . . . . . . . . 2.2.1 Calendar Spread Arbitrage . . . . . . . . . . . 2.2.2 Butterfly Arbitrage . . . . . . . . . . . . . . . 2.3 Different SVI Formulations for Implied Volatility Slices 2.3.1 The Raw SVI Parameterization . . . . . . . . 2.3.2 The natural SVI parameterization . . . . . . . 3 Parameters l\/Iodelling and Preliminary Results 3.1 The Algc›rithm . . . . . . . . . . . . . . . . . . . . . 3.2 The Implementation . . . . . . . . . . . . . . . . . . 3.3 Illustrative Results . . . . . . . . . . . . . . . . . . . 4 Backtesting SVI 4.1 Backtesting l\/Iethodology . . . . . . . . . . . . . . . . 4.2 DAX Index spot options . . . . . . . . . . . . . . . . 4.2.1 Backtest Results for DAX Index Spot Options 4.3 SPX Index spot options . . . . . . . . . . . . . . . . 4.3.1 Backtest Results for SPX Index Spot Options 4.4 PETR4 Equity Options . . . . . . . . . . . . . . . . . 4.4.1 Backtest Results for PETR4 Equity Options . 4.5 Oonfidence Intervals . . . . . . . . . . . . . . . . . . 4.5.1 Numerical Example . . . . . . . . . . . . . . . 4.6 An Alternative Approach . . . . . . . . . . . . . . . . 5 Concluding Remarks Appendix V O O O Q O O O O Q U O O O O O D 0 0 0 0 0 0 0 0 0 0 ø ø O O o o o o o Q Q o o o o o o n 0 D 0 0 O 0 0 0 0 0 0 0 O 0 O 0 0 D 0 0 0 0 0 0 0 D 0 0 0 O O o o o o o o o o o o o o o 0 0 n Q Q Q Q Q o n Q Q Q Q o 0 0 0 0 0 0 0 0 0 O 0 0 0 0 0 O O o o o o o Q o Q Q Q Q Q Q O O Q Q Q o o Q n n Q Q Q o o a n V 1 3 3 4 ©©¬I¬Í 10 11 11 15 17 17 18 18 21 21 22 24 29 30 35 37 42 42 43 47 49 Vi Bibliography OONTENTS 51 List of Figures 2.1 2.2 2.3 2.4 2.5 2.6 3.1 3.2 3.3 r'\r'\t'\r'\r'\r'\r”r'\r'\r'\r'\r“r'\r'\r'\r'\r'\r'\ oooõc:‹»|>oow›~ \l 9 1 Í Í Í Í Í Í Í Í Í 4. 4.2 Ó3<`.Yl›-lšC/~Dl\D›-\© 7 8 9 0 4.21 4.22 4.23 4.24 4.25 4.26 4.27 <;>1›-|>›-lšC.»JO0hD Generic Curve Shape of a Raw SVI Parameterization . . . . . . . . . . . . . . . 1 Íncreasing parameter a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1ncreasing parameter b . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1 reasing parameter p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Í reasing parameter m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1 reasing parameter U . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 E(J E(J E(J SPX Implied Volatility Curve at October, 13th 2015 . . . . . . . . . . . . . . . . 19 SPX Implied Volatility Curve at October, 14th 2015 . . . . . . . . . . . . . . . . 19 SPX Implied Volatility Curve at October, 15th 2015 . . . . . . . . . . . . . . . . 20 DAX Index Call Options” SVI Fitted Curve (expiring: March/ 16) . . . . . . . . 23 gíy) function for DAX Index Calls Options SVI Fit . . . . . . . . . . . . . . . . 23 DAX Index Put Options” SVI Fitted Curve (expiring: March/ 16) . . . . . . . . 23 gíy) function for DAX Index Puts Options SVI Fit . . . . . . . . . . . . . . . . 24 D>+1 Market vs SVI Implied Vols. Comparison for DAX Call Options . . . . . . 25 D>+1 Market Prices vs Model Prices Comparison for DAX Call Options . . . . . 25 D>+1 Market vs SVI Implied Vols. Comparison for DAX Put Options . . . . . . 25 D>+1 Market Prices vs Model Prices Comparison for DAX Put Options . . . . . 26 D>+3 Market vs SVI Implied Vols. Comparison for DAX Call Options . . . . . . 26 L>+3 Market Prices vs Model Prices Comparison for DAX Call Options . . . . . 26 D'+3 Market vs SVI Implied Vols. Comparison for DAX Put Options . . . . . . 27 _J+3 Market Prices vs Model Prices Comparison for DAX Put Options . . . . . 27 L'+5 Market vs SVI Implied Vols. Comparison for DAX Call Options . . . . . . 27 D>+5 Market Prices vs Model Prices Comparison for DAX Call Options . . . . . 28 _, Market vs SVI Implied Vols. Comparison for DAX Put Options . . . . . . 28 Market Prices vs Model Prices Comparison for DAX Put Options . . . . . 28 SPX 1ndex Call Options” SVI Fitted Curve (expiring: March/ 16) . . . . . . . . 29 g(y) function for SPX 1ndex Calls Options SVI Fit . . . . . . . . . . . . . . . . 29 SPX Index Put Options” SVI Fitted Curve (expiring: March/ 16) . . . . . . . . . 30 gíy) function for SPX 1ndex Puts Options SV1 Fit . . . . . . . . . . . . . . . . . 30 D'+1 Market vs SVI Irnplied Vols. Comparison for SPX Call Options . . . . . . 31 D>+1 Market Prices vs À/Iodel Prices Comparison for SPX Call Options . . . . . 31 D'+1 Market vs SVI Irnplied Vols. Comparison for SPX Put Options . . . . . . 32 D*-'fl Market Prices vs À/Iodel Prices Comparison for SPX Put Options . . . . . 32 D>+3 Market vs SVI Irnplied Vols. Comparison for SPX Call Options . . . . . . 32 D›+3 Market Prices vs À/Iodel Prices Comparison for SPX Call Options . . . . . 33 D>+3 Market vs SVI Irnplied Vols. Comparison for SPX Put Options . . . . . . 33 J J J _J-J-cn D'-'v5 vii viii LIST OF FIGURES 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 28 29 30 31 32 33 34 35 36 37 38 3 4 ›-|\›-|\›-|\›-|\›-|\›-|\›-|\›-|\ FI Q) II z) II z) Cí Q) 9 0 OO\lÓ3CI!›-|šO~DN>›-\ 9 0 1 2 3 D>+3 Market Prices vs Model Prices Comparison for SPX Put Options . D'+5 Market vs SVI Implied Vols. Comparison for SPX Call Options . . DH'-5 Market Prices vs Model Prices Comparison for SPX Call Options . D>+5 Market vs SVI Implied Vols. Comparison for SPX Put Options . . D'+5 Market Prices vs Model Prices Comparison for SPX Put Options . PETR4 Equity Call Options” SVI Fitted Curve (expiring: March/ 16) . . 35 g(y) function for PETR4 Equity Cal1s Options SVI Fit . . . . . . . . . . PETR4 Equity Put Options” SVI Fitted Curve (expiring: March/ 16) . . DH-1 Mark D>+1 Mark D'+1 Mark D*-|-1 Mark L>+3 Mark D'-I-3 Mark D>+3 Mark L>+3 Mark D'-I-5 Mark J J JJL'-I-5 Mark D>+5 Mark et et et et et et et et et '+5 Market et et D>+1 DAX Calls Prices Confidence Interval . . D'+1 DAX Call Model Prices with its Confidence Intervals . . . . . . . . gíy) function for PETR4 Equity Puts Optionsvs SVI Implied Vols. Comparison S1Prices vs Model Prices Compari vs SVI Implied Vols. Comparison Comparison for PETR4 Put OptionsPrices vs Model Prices SVI Fit . . . . . . . . . . for PETR4 Call Options . on for PETR4 Call Options for PETR4 Put Options . vs SVI Implied Vols. Comparison for PETR4 Call Options . Prices vs Model Prices Comparison for PETR4 Call Options vs SVI Implied Vols. Comparison for PETR4 Put Options . Prices vs Model Prices Comparison for PETR4 Put Options vs SVI Implied Vols. Comparison for PETR4 Call Options . Prices vs Model Prices Comparison for PETR4 Call Options vs SVI Implied Vols. Comparison for PETR4 Put Options . Prices vs Model Prices Comparison for PETR4 Put Options SPX European Calls - Interpolated Total Variance Surface . . . . . . . . SPX European Calls - Interpolated Implied Volatilities Surface . . . . . . SVI Backtest with Interpolated Volatility Surface . . . . . . . . . . . . . List of Tables 3.1 ›-|\›-|\›-I\›-l\›-|\›-|\ Cä<;fl›-|>O0hD›-\ Calibrated Parameters for the Example Curves . . . Calibrated Parameters for DAX Index Spot Options Quantitative Indexes for DAX Index Spot Options . Calibrated Parameters for SPX Index Options . . . Quantitative Indexes for SPX Index Options . . . . Calibrated Parameters for PETR4 Index Options . Quantitative Indexes for PETR4 Index Spot Options ix LIST OF TABLES Introduction The price of securities such as bonds, stocks or ƒutures cannot be predicted. Because it is subjected to different sources of risk and shocks, derivative contracts were introduced to protect market practítioners and companies from such randomness. As an example, an European call (or European put) option is a contract that gives to its holder the right to buy (or sell) a given amount of an asset at a specific maturity time for a fixed strike price. In contrast, American options can be exercised at any time until its maturity. These so-called Vanilla options are the most liquid and intuitively reflect the market expectation of the asset price future behaviour. More complex derivatives are key tools in hedging and trading strategies, however, they may be not sufficiently liquid to avoid arbitrage opportunities, i.e., they must be appropriately priced with mathematical models, taking into account the available market information. In the classical Black-Scholes model [2l, the dynamics of the instantaneous logarithmic of returns of an asset is given by an Itô process with constant coefficients, the asset”s drift and the volatility. Under no-arbitrage arguments, European call option price is given by the well-known Black-Scholes formula. In practice, volatility is not constant as we can see when we use Black-Scholes formula to identify it. It changes with strike and maturity. These volatilities are the so-called implied volatilities. The volatility smile can be seen when we plot such implied volatilities with the option”s time to maturity fixed and vary the strike price. This name comes from its shape, since volatility usually presents lower levels for strike prices close to the underlying asset price, or at-the-money strikes, and higher levels as we go far from the at-the-money. After the stock market crash in 1987, many generalizations of Black-Scholes model were proposed, where the volatility (or diffusion) coefficient is assumed stochastic. More precísely, let the asset price evolution be given by: dst : Lltstdt + mstdwt, É 2 O, where ,at is the drift, ut the variance (the square of volatility) and {l/Vi, t 2 0} is a Wiener process. One possible way to model the variance ut as a stochastic process is the following [23]: d'Ut : O¿(›St, Ut, JF T6(St, Ut, oz and Ú are continuous and well-behaved functions, T is a constant and {Zt, t 2 0} is another Wiener process, correlated with Wt. In a first observation, this assumption clearly adds ran- domness sources into the model. Also, note that setting T : 0 and ,at as a constant, we would have the standard time dependent Black-Scholes variance. The stochastic volatility models describe in a more precise way the evolution of the under- lying asset prices within a period of time. This is of particularly importance, when we want to evaluate path-dependent derivatives. However, the simulation of theses models have a higher computational cost. 1 2 Introduction A market where every contingent claim can be replicated through a portfolio of negotiable assets is called a complete market [16]. This is equivalent to the existence of a unique equiva- lent martingale measure, which gives, for example, unique European call prices. A necessary condition for this property to hold is the number of sources of uncertainty be less or equal to the number of underlying assets in a portfolio. So, a portfolio of assets modeled by stochastic volatility models fails to satisfy such condition, since the number of sources of uncertainties is in general higher than the number of assets. Local volatility models introduced in [6] and [5], can be seen as a particular class of stochastic volatility models (see [11]), where the diffusion coefficient, or local volatility surface, of the asset price dynamics is a deterministic function of the underlying asset”s current price and time. This model is interesting because it keeps the market completeness and the possibility of hedging positions through the derivatives. Ã/Ioreover, in [6] it is proved that the local volatility is the unique diffusion coefficient consistent with quoted European option prices. However, local volatility calibration is an ill-posed inverse problem, and some regularization technique must be applied. See [4, 7, 8, 21]. Even though there is a map between local and implied volatilities, in practice, its eval- uation involves a complicated tridimensional interpolation method, which is not necessarily well-behaved in terms of smoothness and continuity, since the maturity dates are discrete and sparse. Therefore, the quest for a model which is simple and easy to calibrate, that fits the market smile and satisfies non-arbitrage conditions brought the community to study some simplifica- tions of stochastic volatility models. Among them, there are the stochastic volatility inspired (SVI) models [10, 13], which consists in techniques to parametrize the implied volatility smile. It was firstly developed at Merryl Lynch in 1990 and publicly disseminated in [10]. Since it is relatively easy to implement and has interesting properties such as absence of static arbitrage, SVI models have been largely studied by the community in the past two decades. In addition, it has been shown in [12] that SVI is not arbitrary, since the large maturity limit of Heston implied volatility holds SVI non-arbitrage properties. This master thesis aims to explore the problem of modeling implied volatility smiles in order to guarantee some iinportant properties such as absence of calendar-spread arbitrage and static arbitrage, good fit to bid-ask implied volatilities and good behavior under extreme strikes and maturities. To do so, we will apply known SVI parameterization techniques over real data from liquidly traded European options and compare this fixed parameters with the realized implied volatilities for the next trading days. Structure of the thesis In Chapter 1, we present the derivation of Dupire”s partial dif- ferential equation and the inverse problem of local volatility surface calibration from quoted European option data through the implied diffusion process for the underlying asset”s price observed from the option prices for different strikes. In Chapter 2, we make a review of the- oretical results concerning SVI models. Chapter 3 is concerned with implied volatility surface modeling for real data. Some backtest techniques for market data are described in Chapter 4. In Chapter 5 we draw some concluding remarks. Chapter 1 Local Volatility Surface and Dupire”s Equation It is well-known that the the probability density of the underlying asset price atthe option”s maturity time can be inferred from European vanilla option prices [3, 19]. In [6, 5], it was shown that there exists a unique state-dependent diffusion coefficient consistent with these distributions. Such parameter, denoted by oL(S, t), is known as local volatility surface. There are at least two interesting interpretations of local volatility surface, it can be seen as an average of all possible instantaneous volatilities in a stochastic process, this was the initial intuition of practítioners. Another one, it determines how the implied volatilities evolve, since, by no-arbitrage arguments, there is a one-to-one map between both quantities. Below, we explain both concepts. 1. 1 Derivation of Dupire Equation Suppose the stock price follows the diffusion process below: clS Ê” : /lidt JF UL(Si›7Íš Soldl/Vi» É É 0» 1: where {Wt, t 2 0} is a standard Brownian Motion, ,at : rt - Dt is the risk-neutral drift, given by the risk-free rate rt minus the dividend yield Dt, and oL(St, t) is the local volatility surface. The undiscounted risk-neutral price C : C' exp (LT rtdt) of an European option with strike K and maturity T is given by: õ(S(), K, I JE [HlãX<[0, ST _ I/ (ST _ K)§0(ST, T, SQ)ClST, K with <p(ST, T ; S0) the probability density of the stock price St at time t : T (conditional to S0). It evolves according to the Fokker-Planck equation: 18222 Ô Õgo §E(ULST<PJ - ÊWSTWJ I Ô?- Now, assuming that go satisfies sufficient conditions such that we can differentiate under the integral sign, it follows that: 3 4 CHAPTER 1. LOCAL VOLATILITY SURFACE AND DUPIRE”S EQUATION /\./ ÔC OO 8? _ _ fK ÊÚ(STvTv S0)dSTv érõ Í :</>(K›TšS0J› and in T: N ÕC °° Ô Ô? - [K (ST _ T, Sg)CIlST, By the Fokker-Planck equation, the later equation is equivalent to W 1o2 2 2 ô/K <sT - K> (¿@<f›-LSTt›> - ¶<isTt›>) ist. Using some algebra tricks and integrating by parts twice, we find Dupire”s equation: aí 2k¢o%í ~ oõ (,)T:0-LT¶¬L/sL(T) . (I.I.1) Now, we can isolate the volatility term to get the so-called Dupire”s formula: ÔC ~ ÕC _ - T C - K-2 or ”( ” ( off) 2 ÕK2 which is well-defined due to the static no-arbitrage condition o2oTK, > 0 when K > 0. This is the most important conclusions in Dupire”s article In theory, the term in the right- hand of (1.1.2) can be easily evaluated with a complete set of European options prices, moreover, this represents the unique expression for local volatility surface. In what follows, we shall use equation (1.1.2) as the definition of local volati1ity surface. Note that, in practice, Dupire”s formula is highly unstable, since we must interpo1ate noisy and scarce data and differentiate it, in other words, local vc›latility surface calibration is an ill-posed inverse problem. See [4, 7]. 1.2 Local Volatility and Implied Volatility As one could guess, it is possible to express local volatility in terms of the implied volatility, con- ditional to K, T and S0, derived from the Black-Scholes formula, and denoted by ot;5(K, T ; S0). In this section, we follow [11] closely. Firstly, define the variables: T FT I S0 exp pitdt) 0 1.2. LOCAL VOLATILITY AND IMPLIED VOLATILITY 5 is the forward price of the stock at t : 0, K 92108 E <,.z(st,, y, T) z õ§t.S(sO, Fte?/, T)T is the Black-Scholes total implied variance. In terms of the above variables, the Black-Scholes formula for the future value of the option price becomes: is the log-strike, and C”Bs(FTzy›wJ = FT[^¡(d1)_@y/V(d2)lM-.r>-@~N<-ipiÉs Defining u(T) : LT (rt - Dt) dt, u(y,T) : C(FTey,T) and uL(y,T) : oL(FTe@/,T)2 (the local variance), the PDE in (1.1.1) can be rewritten as: Ôu VL Ô211 Õu _ I T . 1.2.28,. 2 (ôyt t,y)+tz< >«› < > Differentiating the Black-Scholes formula in (1.2.1), we have: 82035 : _1_i+L2 8035 ÕLU2 8 2a) 2a›2 Õw ” 82035 : 1 _ ä ÕGBS Õydw 2 to Õw ” Õ2CBS ÔCB5 ÕCB5 TW : Ty + 28? (1.2.3) Now, we want to describe equation (1.2.2) in terms of the implied variance UBS. Since, for every y and T, u(y, T) = CBs(FT, y, oa), differentiating the equation above by y and T, it follows that: ÕU _ ÕGBS ÔCBS Õwíy- ôy + Ôw ay, (1.2.4) ou ô2oBS ó>2oBS ou o2oBS ou 2 ôott ouõ_e:õ_l2+2z›i¶íi+ às (ai) WT? M5) o oo o8-; = fã + ;J(T)@. (1.2.õ) The expression in (1.2.6) can be rewritten as: and by (1.2.2), the right-hand side of the above equation is equal to: Õu u E9211 do Õ? _”<T)U I 2L (59% Õy>” 6 CHAPTER 1. LOCAL VOLATILITY SURFACE AND DUPIRE”S EQUATION So, it follows that, oct-,›So¿ _ it ou oz» Õw ÔT_2 Ôy2 Ôy” Ô2 Ô Substituting the terms Õ-Q; and íw by the expressions in (1.2.4) and (1.2.5) in the PDE (1.2.2), IJ IJ we have: LCBSÔi:V_LLCBS 2_Ôi¬|_ 1 y Ôw¬|_ _L_i+L2 ai 2+yi Ôw ÔT 2 Õw Õy 2 ao Ôy 8 2o; 2w2 Ôy Ôy2 ” and simplifying: ôi_V 1_äôi+1 1 1+y2 Ôi2+Lyi ÔT_ L wôy 4 4 to w2 Õy 2Õy2 ” So, we just have to invert the above equation to isolate the local variance: 81 ÕTV : 1.2.7L 1yoa+1 1 1+¿2 o¿2+f‹3>2_a ( ” ao Õy 4 4 ao w2 Ôy 2 Ô;/2 The intuitive conclusion of this result is that the local variance is expressed as the forward Õ Black-Scholes implied volatilities distorted by the skew i. Ô Even though this work doesn”t aim to calibrate the loffal volatility surface, but the implied volatility one, this relation composes an important background to the overall understanding of the problem of pricing options using different approaches to the variance other than the constant one assumed in the Black-Scholes model. While Dupire”s result provided the discussion with a stronger theoretical background, the models which aim to interpolate and extrapolate the implied volatility surface from given market data is more straightforward and simpler to implement, with market smile adherence. Thus, the implied variance can also be expressed as a function of local volatilities. Chapter 2 SVI Parameterization: An Introduction Given the computational and theoretical difficulties related to the simulation of stochastic volatility models, it makes sense to search for simpler techniques that are easy to implement, arbitrage-free and that fit the market implied smile. One possible approach is the arbitrage-free interpolation for implied volatility surfaces, such as [9] and [14]. However, these models are not well suited to extrapolate the implied smile, and there is no closed-formula for volatilities. So, we choose the SVI model introduced in [10], which has a closed-form representation for the irnplied volatility function and guarantees the absence of static arbitrage. In this chapter we shall present the definitions of different static arbitrage conditions and how to prevent them in the formulation of the SVI model. More precísely, we shall state sufficient conditions for the existence of a non-negative martingale defined in the probability space where European call options prices are defined as the expected value of their payoffs, under the risk-neutral measure. We shall show also that the model is free of calendar-spread arbitrage, which means that the total variance must increase with the maturity. Visually, this means that if we plot several slices of volatility smiles (one for each maturity time) in the same log-strike >< implied volatility chart, the curves must not cross each other. Another important characteristic that the SVI model has is the consistency with respect to Roger Lee”s moment formula [18], which set bounds for the tail slopes of implied volatility at extreme maturities. 2.1 Roger Lee”s l\/Ioment Formula In [18], the author presented the moment formula of implied volatilities, which implies some properties of the implied volatility tail slopes at large and small-strikes, setting bounds for these slopes, which can be interpreted as arbitrage bounds. We shall review these results. Let us recall the notation K y :1Og< v and : O-BS('S[0iy7T)T7 T where the latter is the total implied variance of Black-Scholes model for a given maturity T. Let 5 be defined by B[y[/T : Roger Lee”s moment formula shows that, when calculating ÊR : limsupfi and BL : limsupfi, y-›oo y->-oo7 8 CHAPTER 2. SVI PARAMETERIZATION: AN INTRODUCTION it follows that /31.-t,5L G [0, 2]. Also, it states the following: Theorem 1 (Roger Lee”s Moment Formula for Implied Volatility). 1 s 1 %+íR-ã=sup{p:lE[S;¬+p]<oo}, and 1 5 1 _ E1-É;-§:sup{q:lfÍ.[STq]<oo}. In [18], the theorem above is proved when the asset ST has an arbitrary distribution. Theorem 1 shows that the best constant for the tails slopes depend only on the number of finite moments of the distribution of the underlying asset. Despite we shall not give the complete proof of Theorem 1 since it is not the focus of the present work, we shall demonstrate two important lemmas that help us to understand why SVI parameterization is consistent with Roger Lee”s moment formula. Define 50 = exp (- f0T utdt) as a discount factor and write the Black-Scholes option price for the strike K(y) as C(K(y)) : CBg(y,w(y)), where K(y) : FT exp(y) is the strike at the log-moneyness y. We have the following lemmas: Lemma 1. There exists y* > 0 such that for every y > y*, w(y) > \/2[@J[/T» Proof. Since CBS is strict monotone, the lemma follows from this: ÚBS(y›W(yJJ < ÚBs(yz \/ 2[y[/TJ» If a: > x*. On the left hand side of the above equation we have: um Ô(K(y)) z um s01E(sT _ K)+ z 0, y->oo K->oo which is guaranteed by the Dominated Convergence Theorem, since E(ST - K)+ < oo. By L”Hôpital rule, the right-hand limit can be written as: um o Su ×/21y1T>- õ F [‹1›<@> um @›<p<y>‹1>< @>] - rf F /2.B 1 _ 0 T _ _ _ 0 T y->oo y->oo which completes the proof. Ú Lemma 2. For any É > 2 there exists y* such that for all y < y*, w(@/) > \/ölyl/T Hence, for 5 : 2 the conclusion holds if and only if lP”(ST : 0) < 1 /2. Proof. For Õ > 2 there exists y* such that for all y < y*, l1””(51¬ < FT @Xp(@/)) < @(-\/f-(fi)[y[) _ @Xp(-3J)<Í)(-\/f+(fi)[”J[), where f is the characteristic function of the underlying asset”s price at t : T. 2.2. STATIC ARBITRAGE 9 The stated above is true since, as y -> oo the second part converges to lP”(ST : 0), and the first part approaches 1 (or 1 /2 in case S : 2 for the special case stated in the lemma). Therefore, due to the strict monotonicity of the Black-Scholes price of an European put option PBSI PT.›s(yzw(y)) = fi@1E(K(@/J - STJ* S fi‹›K(y)JP°(ST < FT @><i>(i/)) < PBs(@/z \//fly]/TJ» for all y < y*. E The stated above shows that the large and small tails slopes are not larger than 2 and, as a consequence, the implied Black-Scholes variance is linear in the log-strike y as T oo. These are characteristics that the SVI parameterization must hold. 2.2 Static Arbitrage We want to find an implied volatility surface that is free of static arbitrage, which means that there exists a non-negative martingale model such that: Õ(KzT) = E@l(5T _ KW» that is, the undiscounted European call option price at maturity can be written as the expected value of its final payoff in the risk-neutral measure Q for some process St. Definition 1. A uolatility surface ao is free of static arbitrage and only (i) is free from calendar spread arbitrage; (ii) is free of butterfly arbitrage at each single time slice. Note that, Item (iJ of Definition 1 guarantees the monotonicity of European call option prices with respect to theirs maturities, while Item (ii) guarantees the existence of a probability density. For more information about static arbitrage and implied volatilities, please see [20]. From now on this chapter shall review some theoretical aspects exposed in [13]. 2.2.1 Calendar Spread Arbitrage Consider two European call options of strike K and maturities T1 and T2 with T1 > T2. Since T1 > T2 we would intuitively expect that the price of the option with the larger maturity is higher than the other one. That happens because the more time we have ahead, the more uncertainty we suppose We will be exposed, so we would expect a higher total variance, therefore a higher price. That is why, whenever this condition is not respected, the calendar spread arbitrage is characterized. In order to have an accurate model to price options, one would suppose that this condition must be respected, that is, the implemented model must guarantee monotonicity for the call options with same strike and different maturities. That leads us to the following definition: Definition 2. A uolatility surface of is free of calendar spread arbitrage Õw _ t > Oat (yâ ) _ 7 forallyš R andt>0. 10 CHAPTER 2. SVI PARAMETERIZATION: AN INTRODUCTION This definition is motivated by the following lemma: Lemma 3. Considering a constant proportional to the underlying asset prices dividends applicable), the volatility map ao is free of calendar spread arbitrage if and only Ê_Í(v,t) 2 0, forally€Randt>0. Proof. Let (Xt)t¿0 be a martingale and L 2 0, 0 É t1 < t2. Then, the following relation is true: EKXT2 _ L)+l 2 EKXT1 _ L)+l /\/ /'\./ Now, let C1 and C2 be options with strikes K1 and K2, and maturities t1 and t2, respectively, and with the same log-strike or log-moneyness: K1 K2 <>í I í ieF1 F2 Xp v t t Then, the process {Xt}t¿0 defined by % for t 2 0 is a martingale and we have the following relation: /-\/ /-\/ ä _ @X1í>(_y)1El(Xzâz _ @X1>(v))+l 2 @X1@(_y)El(Xzz1 _ @><p(v))+l _ if the dividends are proportional. E This relation means that, for constant moneyness, option prices are non-decreasing in time to maturity. If this relation stands for Black-Scholes prices CB5(y, w(y, t)) then, w(y, t) is strictly increasing. 2.2.2 Butterfly Arbitrage In this subsection we shall define butterfly arbitrage over a time slice of the volatility surface. Thus, we shall use w(y) in order to simplify the notation, since the time will be hold fixed. Now, recalling the Black-Scholes formula: CBs(v, w(v)) = 5(/\f(d+(v)) _ @Xr›(v)/\f(d-(v)))» f01' 81111 G R with ./\/1 the Gaussian cumulative distribution function and d1(y) : -y/ \/oJ(y) i \/w(y) /2. Now, let g : R -› R be a function defined by: g(y)z=(1-ÊÍT) -%y”2(Ê-fã) -1%”. (2.2.1) Definition 3. A volatility map y -> w(y) given for some maturity is free of butterfly arbitrage if the corresponding asset”s probability density is non-negative. Please refer to the 1 for the realtion between w(y) and the underlying asset probability density. sfíš \./ N) Lemma 4. A volatility map y -> w(y) given for some maturity is free of butterfly arbitrage if and only g(y)2 0 for all y É R and lim d+(y) : -oo. y-›oo 2.3. DIFFERENT SVI FORMULATIONS FOR II\/IPLIED VOLATILITY SLICES 11 Proof. As mentioned in Chapter 1, the probability density function <p can be inferred from a set of option prices by: 32001) Õ20Bs(vz w(v))W) I W I .zw > for any y 6 R and K : Ft exp(y). Now, differentiating the Black-Scholes formula, for any y G R we have: ( ) d-( )2</>(v) _ @Xp . It is important to notice that this density function may not always have its integral equals 1. In particular, call prices must go to 0 as y -> oo, which is equivalent to the second part of Lemma 4. E 2.3 Different SVI Formulations for Implied Volatility Slices In this section we shall present different functional forms which have been proposed to fit the implied volatility curve for a given maturity with real data. Firstly, we shall see the raw SVI parameterization, which was introduced in [10], then we will present an alternative (and equivalent) parameterization so-called natural SVI parameteri- zation. We shall not discuss the SVI-JW (Jump-Wing) parameterization since we are interested in time independent methods for backtesting in different dates. 2.3.1 The Raw SVI Parameterization Let XR _ {a, b, p, m, o} be a set of parameters. The raw SVI parameterization of total implied variance is: «nz X1.) z cz + 1» [pv - m> + do - mr + U2] . <2.õ.1> where b 2 0, [p] < 1, a G R, o > 0, and a + bo\/1- ,o2 2 0 to guarantee that w(y;XR) 2 0 for all y 6 R. It is easy to see that for a given set XR, the function y -> w(y, XR) is strictly convex. Now, we will investigate the intuitive meaning of the formulation parameters XR through the derivatives of equation to in (2.3.1):dia_ Z 1 Õa 7 ÔwÉ z z›<z.1-m>+¢<zz-m>2+‹›2. Õ dia m-y sw I z z bp»\/(v_m) +0 Õw o Õ? Z ¢<y-m>2+‹12' 12 Figure 2.1 shows a generic raw SVI curve with the set of parameters XR : (0.05, 0.15, 0.40, 0.30 0 45 CHAPTER 2. SVI PARAMETERIZATION: AN INTRODUCTION Figures 2.2-2.6 illustrate how variations on each parameter may change the shape of the volatil- ity curve E3 .Z 1:I*."1:1E1l rnpe I I I I I I I IÍL3 III I'“¬_'l I'.|'I III I"-.'I 11.15 .ta "¬~«.1\\\ |]_`| I I I I I I I-2 -1.5 -1 -11.5 11 11.5 1 1.5 2 Ger'1er1`1: 5"."I Cuwe furlm plied "."1:1I=E1liIit1_.f Exam |:1Ie "-. "-. _ . / _ Lug-F 1:1r¬w'ar1:I I'-1'I1:1r1e'1-mas Figure 2.1: Generic Curve Shape of a Raw SVI Parameterization 2.3. DIFFERENT SVI FORMULATIONS FOR II\/IPLIED VOLATILITY SLICES -:-1-:L1 EE a1:I*.f1:1al 11111 -:-'11-:|.'- EE a1:I*.f1:1at 11111 Incraaaing param atara Ú-5 1 1 1 1 1 1 1 íäifllãanaria E:-:ampla [ Gar'1ar1`1:E1‹arn1:1Ia1.-.-ithhi1;|hara [ 11.45 - - III.-4 EL35 l1I.3 III I"'¬_'l I'.|'I IÍL2 - fr; - _I__r" E 11.15- / _ |]_`| I I I I I I I -2 -1.5 -1 -IÍI.5 IJ III.5 1 1.5 2 Lag-F award I'-11 an a'1-11 aaa Figure 2.2: Increasing parameter a Incraaaing 1:1ararr1 atar |:1 Ú-5 1 1 1 1 1 1 1 [í5*."IGar1ar11:E:-:ampla | Ganafla E:1‹:arn1:1Ia1-zith higharh III.-45 - - III.-*I - - IIL35 - - IIL3 - - III I"-_'I I'_|'I IÍL2 - - lÍI.I5- - ¬¬-11___h¬______ 111,1 1 1 1 1 1 1 1 -2 -1.5 -1 -III.5 III III.5 1 1.5 2 L1:11;1-F 1:11*-.»1'ar1:I I'-sl an a'1-11 aaa Figure 2.3: Increasing parameter b CHAPTER 2. SVI PARAMETERIZATION: AN INTRODUCTION Ir1erea.air11;1 |:1arar1'1 et er 1; Ú-'45 1 1 1 1 1 1 1 í5'1'IGer1er1`eE:a5n'1|:1Ie Ger1en'e E1a5r1'11:1Ie'1-.-'ith higher 1; III.=1 - - 11.55 - - E3 11.5 //_ / _ mpedvefi 1:1 5.1 I'.l'I [1.2 \% /r 131.15 _ / _ _-' 1z|_1 1 1 1 1 1 1 1 -2 -1.5 -1 -III.5 III III.5 1 1.5 2 L1:11;1-F1:1r¬w'ar1:II'-111:1r1e'1-r1eaa Figure 2.4: Increasing parameter ,o Ir1ereaair11;| 1:1ararr1 eterm Ú-35 1 1 1 1 1 1 1 í5'1'IGer1en'e E:‹:arr1|:1Ie Generic E1a5r1'11:1Ie'1-.-'ith higherm 11.5 - - `“~. E5 Ê: r-¬_1 -'.r| .Z I:I1"1:1al rr1|:1e [1.2 - - 5.15- BRM _ 1z|_1 1 1 1 1 1 1 1 -2 -1.5 -1 -III.5 III III.5 1 1.5 2 Leg-F 1:1rw'ar1:I I'-111:1r1e'1-r1eaa Figure 2.5: Increasing parameter m As you can see in Figures 2.2, 2.3, 2.4, 2.5 and 2.6, 2.3. DIFFERENT SVI FORMULATIONS FOR II\/IPLIED VOLATILITY SLICES 15 í5"1'II5er1en`eE:-1:e1r1'11:1Ie Ger1er1eE:-=:=51r1'11:1Ie'-1.-'ithhi1;1her 1:1 [ Ir11:rea.air11;| param et er 1:1 11.5 - - `“~. \ E3 Ê: ru -'.r| rr1|:1e1:I1f1:1a1 1:1 `r~_~ xx/ 11 1 1 1 1 1 1 1 1 -2 -1.5 -1 -III.5 III III.5 1 1.5 2 L1:11;1-F1:1r¬w'ar1:II'-111:1r1e'1-r1eaa Figure 2.6: Increasing parameter o o when a increases, the level of total variance also increases, that is, there is a vertical translation of the curve; o increasing b, the slopes of the put and call wings also increase, which represents an increase of total variance for out-of-money strikes; o when ,o increases, there is a reduction of the call wing total variance; o changing m, there is a translation of the curve in the x axis; o increasing co, the curvature reduces in the neighborhood of the at-the-money strike. 2.3.2 The natural SVI parameterization The natural SVI parameterization w(v;><.1z-.z)_ 4+ É [1+Çfi(v -11)+ ¢(C(v -11) +/1)” + (1 _1>2)}z (2-3-2) is equivalent to the raw SVI by the following change of variables: / /1 _ 2 (a,b,,o,m,o) : [A+¶(1-,o2),Çl,,o,,Li p, 'O . (2.3.3) ( 2 2 C C Note that, now there are two “level variables””, A which has a straight unitary derivate, and w which doesn”t have, while in the raw formulation the only “level variable”” a has unitary derivate. CHAPTER 2. SVI PARAMETERIZATION: AN INTRODUCTION Chapter 3 Parameters Modelling and Preliminary Results The goal of this chapter is to find a stable and reliable way to calibrate the raw SVI parameters in (2.3.1) for a given set of implied volatilities observed in the market. It is well known that the least square technique for this problem is affected by the large quantity of local minima. Other problem regarding this modeling is that optimizing over a 5-dimensional set of parameters usually returns non-stable results. Therefore, the technique which will be applied is the one presented in [22] with a small mod- ification developed for us in order to make the choice of m and o more precise. This approach is particularly interesting because it simplifies the problem dividing it into one linear program- ming and one two-dimension equation optimization through a simple dimensional reduction. Moreover, the results are impressively good and the computational cost is low, if compared to other techniques such as stochastic volatility models. 3. 1 The Algorithm Let wT0t2¿ : wT be the total variance. Now, applying the following change of variable on the log-strikes: y-m íczí, o we turn the SVI parameterization into: a1T2t2¿(x) = aT + boT(,o;1: + \/ ;e2 + 1). The equation above shows that, for fixed values of m and o, the SVI curve is completely specified by a, ob and ,o. Hence, if we define the parameters: c : boT, d : ,oboT, A : aT, then we rewrite SVI formulation such that it is linear in the variables A, c and d: wT2t2t(¿1:) : A + da: + cx/ m2 + 1 17 18 CHAPTER 3. PARAMETERS MODELLING AND PRELIMINARY RESULTS It is important to note that the region where the parameters A, c and d are defined must respect the original parameters boundaries, so it is derived from these constraints. Define the parallelepiped D: D:{(A,c,d)€R3 : AGR, c20, d€R}, which is the region in R3 where these new variables are defined. Therefore, for a given set of n market observation pairs (att, óoÊ,”30”`§,z¿(,¿)), we define the cost function as: n 2 f($11›“T"ÃÊÍz1<1>)(A” CI dl _ 2 (A T dia T C\/ T 1 _ wT0Iifhl(i)) ' 1:1 So, it is only necessary to find the triplet (A*, c*, d*) for which Vf = 0 and then solve the problem P below with the corresponding triplet (a*, b*, ,o*): TL (P) 13315 2(wm.‹›.T.1›*.z›* (vz) _ wm‹z1~1‹@1<1>)2- ” 1:1 The objective function in the optimization problem above is so-called sum of the squared errors. Its square root is so-called RMSE. 3.2 The Implementation The algorithm presented above is strongly dependent on the initial guesses of m and o, i.e., it is a heuristic method. Therefore, in order to reduce such dependence, the algorithm must be tested with different initial values of m and o, in order to choose appropriately the initial guesses. More precísely, the minimization is initialized with all combinations of m and o ranging from 0.01 to 0.99, with the step size 0.01. We choose the pair that presents the lower RMSE values. This adds some computational cost to the problem but not sufficient to become a problem, since each calibration takes around 2-3 minutes to be performed. Also, a loop structure with 10 iterations has been applied for every pair of initials (m, o). The MATLAB@ code of the function created to calibrate these parameters can be found in Appendix 5. Note that, the bid-ask spread is not always well-behaved, mainly when we go far from the at-the-money strikes. To overcome this issue, we exclude the outliers, since they do not add relevant information to the analysis and usually are related to low trading volume options. 3.3 Illustrative Results Before implementing a backtest strategy for the fitted parameters, we illustrate the smile ad- herence of the SVI parameterization. The data is composed by SPX index options traded at 13th, 14th and 15th of October, 2015, with the shortest maturity time available - 21th of Oc- tober. The resulting SVI curves compared with the market implied volatilities can be found in Figures 3.1, 3.2 and 3.3. 3.3. ILL USTRATIVE RESULTS -:-'.1-:L1 EE e1:I1.11:1a1 11111 -:-'.1-:L1 EE 1:I1"1:1al m|:1e 5F'1{151e;1c|:1ir1r11;|ea1I1:11:1ti1:1r15e51iI:1ra1e1:I eur1Ie'»15.mar1:etmi11 '1-'1:1I5-1II11et1:1I:1er,15th Ú-15 1 1 1 1 1 1 mark et 1:1 1:15. í51.11 Fit -II" 11.111- 11.11'- - ':- 11.1E-- -5' _ "'- _III zln I'.|'I .'¬-. _ IV' _f=~ I ¬* 1 5 1,5 ñ. '-.-' _ _ -51 -“11.15 _ _ 1--1 _ '-L* _ 5:31- -11 J' z J 3. .z'_¬;. 1 .-"'-. 11.12 "'-' '-.-'-._.--: ..-' 1z|_1I] 1 1 1 1 1 1 - .III-1 -III.III2 III III.III2 III.III=1 III.IIIEi IILIII5 III.1 L1:11;1-F1:1r¬a-'ar1:II'-111:1r'1e'1-r1e55 Figure 3.1: SPX Implied Volatility Curve at October, 13th 2015 5F'1{151e1c|:1ir1r11;|ea1I1:11:1ti1:1r15e51iI:1ra1e1:I eur1Ie'»15.mar1:etmi11 '1-'1:1I5-11II1et1:1I:1er,1=1th Ú-15 1 1 1 1 1 1 1 market 1:11:15. | í5111r11 11.11' :- -:ja 11.1E- - __. '-.- "-. 11.15- - '12 -:" _ _.¬. .I"'¬. _._v:. ¬._.' .___ .I-:I -:::- .-"' :=~ E 5.15 51-1 __ -:" .-"-.-.-' .-"-. .z"'-. -:"':- 11.12- 515 - "-. -.-" -"' ¬:":- .;'_"-. .-'-. .-'-. -:§':-.-' '-._.-' '-._.-' - 1111 1 1 1 1 1 -III.III1 III III.III1 III.III2 III.III5 III.III-1 L1:11;|-F1:1r¬w'ar1:II'-.-11:1r1e'-,-r1e55 Figure 3.2: SPX Implied Volatility Curve at October, 14th 2015 11.115 11.115 11.111' 11.1111 20 CHAPTER 3. PARAMETERS MODELLING AND PRELIMINARY RESULTS 5F'1t 151 e:-t1:1ir1`r'11;|e51I 1:11:1ti1:1r15 e51it1r5te1:I eu11.1e'115.m5rketmi1:1 '111:1I5-1III1et1:1t1er,15th I I I I I I market 1:11:15. 51.11 Fit 11.2 - - 1' 11.111 - 25 «ta - ':- e5 3:' E 1 'I-F' 1E-E = 5 _ II- 1:I1"1:1al 2:' E m|:1e _ IE- _ -L» .P _.-'_ -"':___- . 11.12- _~“` - -131 . I11. 1 - 5;-__., 5. V - ¬.__..' _:.¬._ ..I...' .'... ¬_. 5... . 13 113 1 1 1 1 1 1 - .IIIEI -III.III-1 -III.III2 III III.III2 I1.III-1 III.I1E III.III5 L1:11;1-F1:1r¬w'ar1:II'-111:1r1e'1-r1e55 Figure 3.3: SPX Implied Volatility Curve at October, 15th 2015 a b m ,o o RMSE SPX calls at October, 13th 0.0872 1.3634 0.0302 0.2365 0.0200 1.18x10_2 SPX calls at October, 14th 0.0537 2.1805 0.0399 0.3435 0.0299 0.90;1:10`3 SPX calls at October, 15th 0.0463 2.6030 0.0299 0.2067 0.0200 1.60x10`2 Table 3.1: Calibrated Parameters for the Example Curves Table 3.1 presents the parameters values calibrated from the SPX data and used to find the SVI curves plotted in Figures 3.1-3.3. As one can see, the curves fit the market data well and the RMSE is low. Chapter 4 Backtesting SVI After presenting the SVI parameterization, now, we shall backtest it. More precísely, we shall verify how accurate it is as a prediction method. 4.1 Backtesting Methodology We use the following market data: o Underlying asset end of the day price; 0 Risk-free interest rate (annualized); o Evaluate the forward price corresponding to the underlying asset (based on the end of the day price and the risk-free interest rate); o Time to maturity (in years); o European call and put bid and ask prices for every strike with negotiations on the date. Using the data above, it is possible to calculate the Black-Scholes implied volatility corre- sponding to the bid and ask prices, and calibrate the raw SVI (2.3.1) parameters through the algorithm described in 3. Once the parameters are identified, we apply two different tests that verify non-arbitrage properties. The first one is to test butterfly arbitrage, where the function defined in (2.2.1) must be non-negative in R. The second one is to test Roger Lee”s extreme slope (1) conditions, where we take into consideration the largest and smallest log-strikes in the data. Now, we can calculate implied volatilities for any log-strike using the SVI function. So, we can price any option on the considered asset/maturity pair at any time we want. In what follows, we test how accurate the calibrated SVI curve is to predict option prices at future dates. If D denotes the negotiation date considered in calibration, we evaluate option prices at the dates D+1, D¬'-3 and D+5, comparing them with market prices. We plot the corresponding implied volatilities of the model prices with market bid and ask prices. The prediction is considered accurate if the model implied volatilities fall between the market bid and ask ones. These results shall be presented in the subsequent subsections. The following functions are used to measure the distance between model prices and data, i.e., how accurate the model is: 21 22 CHAPTER 4. BACKTESTING SVI o Data misfit for volatilities: ”°' 2E <0.7SVI _ O.7:l\4arket) S:1 n 2 7 \ Z <0.7:l\4arket) ' IS where n is the total number of log-strikes in the data, os]/I is the SVI implied volatility and oM“"”ke'” is the implied volatility of the mean of market bid and ask prices. o Normalized É2-error of prices: Ê (CIYSVI _ CIš\darket)2 ' IS \ É <CIš\darket)2 7 SI where CSW is the Black-Scholes option price calculated with the SVI volatility for each log-strike. ZM= o Average error over spot: [ <CISVI _ GMarket):| 77189 7 where S0 is the underlying asset spot price. 4.2 DAX Index spot options DAX Index European type options are traded daily at Eurex Exchange in EUR currency and have a high volume of trading, so it matches our requirements for the test. We took all the necessary data for call and put options on D :22-Jan-2016, and then, we fitted the curve and tested the non-arbitrage properties. The fitted parameters can be found in Table 4.1. Figures 4.1 and 4.3 show the calibrated SVI curve for DAX Index options expiring in 15- Mar-2016. Note that, the fitted curve for put options in Figures 4.3 is not free of butterfly arbitrage, since the function g(y) defined in (2.2.1), and plotted in Figure 4.4 has negative values. It can also be observed in Figure 4.3, since one can notice that the curve presents non- desirable concavity. The same does not happen with call prices, as we can see in Figures 4.1 and 4.2, where the volatility curve is convex and the function g(y) is non-negative. a I) m ,o o DAX Index Spot Call Options 0.1083 48.6448 0.1600 0.9807 0.0100 DAX Index Spot Put Options 0.2552 -0.2282 0.01 4.0272 0.0303 Table 4.1: Calibrated Parameters for DAX Index Spot Options 4.2. DAX INDEX SPOT OPTIONS II21.ë-.1I{In1:Ie:=›:1Ca1I5 5111 Fitte1:I1Cur-.-'e an III1-22-15 5 1 1 1 I E1idIm|:1. 11515 .-1»5k|m|:1. 11515 _5'1.1'I Fit Imp. 11515 III.-4 _ _ 11.4 'I-1 ¬¬. 1:' -'_-J 1.11 1-" 'Ir II.-"r".fE3E1 ¬¬.'11-1. *¬-_ H ¬'I-1. ¬\. e1:I11551 PTH.¬..`__ . 1 'I-._ ¬¬_ 11111 1:' LAJ fill:r' ¬\¬. ¬\¬. 'L "¬-._¬_ 1-¬.¬_`_H 5.25 _ “¬-_,¬ _ ¬¬-1. 'H- --._____ ¬-_ -1.¬__`¬¬-¬-Eh ¬'-1. ¬'h. ¬¬-1. '¬.|;|_ 1 1 1 1 -2.15 -III.1 -III.III5 III III.EI5 III.1 L51;1-F5r1«.-'ar11t-«15ne'1me55 Figure 4.1: DAX Index Call Options” SVI Fitted Curve (expiring: March/ 16) 5511551155151 Q- _ S- _ 1'_ _ E- _ 51- - -1- _ 3- _ 2- _ 1 _ _ _ 1 _ 1 1 1 -2.15 -11.1 -11.115 11 11.05 11.1 L51;-Farward Man e1m e55 Figure 4.2: g(y) function for DAX Index Calls Options SVI Fit II21.11G{In1:Ie><F'ut5 51.11 Fittedfiur'-.-'e un |II1-22-15 Ú 34 1 1 1 1 1 1 E1|dIm|:1 1.1515 .¿1»5k|m|:1. 11515 [132 ___ _5'1."I Fit Imp. '1'5I5 ¬¬¬. ¬"'-.¬11. _B--u -Hx 5.5 _ zm _ ¬¬¬. ¬-1. ¬'H. ¬'H.¬--1.¬¬¬¬_¬__ ¬-_. ¬--I._ ¬-._ _-_E3 1:' r~.1- 1:11:-Ê "¬»¬.1-.__ - _ "'H.¬_`_ “-1. -J-¬. "¬-¬ -2-LH1 Ç 'Hzk 1 m1:1`e1:I1.15a.t 4:1 `r«_z Cl? ._`_L_" ""‹. H ¬`¬\'H ¬`¬\Ú _ 24 - '1¬.¬__ - "1¬.¬-.___ 1*-1. ¬'-1..M- 'I-1 ¬~. ¬¬¬.11.22 - "_ |z|_;~ü 1 1 1 1 1 1 - .1 -II|.IIIE1 -IJ.UE -11.1111 -III.IJ2 III I].ü2 III.ü-11 L5g1-F5r1«.-'ar11t-1151151111555 Figure 4.3: DAX Index Put Options” SVI Fitted Curve (expiring: March/ 16) 24 CHAPTER 4. BACKTESTING SVI Di'-.Ii Puts G['¡r] Ú 5 I I I I Ú- _ ÊH U5 - 1 _ 15 l _ E 2 _ 25 _ S _ _ \ .z5.5- .L_¬¿ _ I I I 1 I I -fill III IIIE III IZIEI -III.III-4 -IILIII2 III III III2 III III4 L ug-Forward Munevn ess Figure 4.4: g(y) function for DAX Index Puts Options SVI Fit 4.2.1 Backtest Results for DAX Index Spot Options Figures 4.6, 4.10, 4.14, 4.8, 4.12 and 4.16 show the comparison between market prices and the predicted prices at the dates D + 1, D + 3 and D + 5. The predictions are evaluated With the SVI volatilities calibrated at D. Figures 4.5, 4.9, 4.13, 4.7, 4.11 and 4.15 compares market implied volatilities of bid and ask prices and the volatilities of prices predicted by the SVImodel. Even though the SVI volatilities curve do not fall perfectly inside the bid-ask spread, the predicted prices presented good adherence to the market ones. Calls D-'fl D+3 D+5 Data misfiú fm» Vols. 1.071 ~ 1.531 ~ 2.3.1«ír1 Normalized Zz-error of prices 2.31.1 _ 6.()2.1 _ 9.76.1O_2 Average error over spot 1.01.1 _ 320.1 _ 4.1O.1O_3 Puts E E D+5 Data mísfiú før VQIS. 3.251 - 3.021 - 11.76.10-2 Normalized K2-error of prices 3.40.1 _ 3.841 _ 4.87.1O_2 Average error over spot 820.1 _ 5.45.1 _ 6.93.1O_4C)C)CJ_|:C)CJC) ›I>L\Dl\DI-\oo›-II-I C)C)CJ_|:C)CJC) ›I>1\awC,Ooow›-I Table 4.2: Quantitative Indexes for DAX Index Spot Options 4.2. DAX INDEX SPOT OPTIONS E3._ m|:1`e1:I'1"I:1aI Figure 4.5: D+1 l\/Iarket vs SVI Implied Vols. Comparison for DAX Call Options 11F'ri1:55 ._ CI|:11:1 Figure 4.6: D+1 l\/Iarket Prices vs l\/Iodel Prices Comparison for DAX Call Optlons E3._ 1:I"."I:15.I mpe Figure 4.7: D+1 l\/Iarket vs SVI Implied Vols. Comparison for DAX Put Optlons EI.3Ei D+ I DAE Indexüpti 13115 Imp. '1fuI Backtest ELSE - IÍL34 - U.32 - EI.3 - I I"\.'I CCI EI.2Ei - 11.24 - Ú.22 - I I I I 4°.¡._* 'BF -31€-¿¿_*_ _ -.Ir 1*-JIE ***+ _ *àz 511 + -Ir+* -IP *ii- íülll 51fIFitt5dCu1"-na 111 I3I+1EIi1:I|111|:1.1fnI5 D+1 .-'1'-.5kIr11p. I.f1:1I5 -III- *Ê-II: ii *'- '*_*'-_ "¬-1.+*5 5*-III-áw *fäfi1 1 1 1 `*'-5. Úl ÚÚ5 Ú ÚÚ5 Úl f Ú-Ê. 1 5 D+1 D.-fi1I{Inde>cCIption5 F'rice5 Banktest IÊÚÚ - 1 1 1 1 - Log-Forward I5'I1:1n51m 555 IIIIIIIEI - EIIÍIEI - EIIIIIÍI - 400 - 2ÚIÍI - É É É 5 É5 Ê55% 5 É5 5 55. $$ É5.5 55 111- I'-1'I1:1d5IPr11:e5 <I> Ma11c5tF'11`1:55 55“úäää 5% Log-Forward I5'I1:1n51m 555 D+1Ilê-.I'I{Ir1ci1E:>cIlI.`Ipti1Jr'15Im|:1."."1JIBacktest Ú-34 1 1 1 1 1 1 1 *15 E ' I 1 1 *$**¡!lI- 15 -EI.1 -IILIII5 III IILEI5 I].I 0.32 - IIL3 - - III INJ CO I HJ CU 11.22 - - íüfl 5'1fIFitt51:ICurve 515 D+1EIid|mp.'1'1:I5 IIII+1.-'11-.5I<Ir1'1|:1.".f1JI5 1 *Ii- IIC- _ _* _ -III- 1* if 11.24 _ *` *Ê _ *C- I* |]_z~ü 1 1 1 1 1 1 1 - .1 -UJIIEI -IÍI.EIE -III.I]4 -IILU2 EI IÍLU2 I].I]4 IÍLUE Log-Forward I5'I1:1r151m 555 11F'riee5 ._ CI|:11:1 Figure 4.8: D-l-1 l\/Iarket Prices vs l\/Iodel Prices Comparison for DAX Put Options E3._ m`e1:l'1"1:1eI 13' Figure 4.9: D+3 l\/Iarket vs SVI Implied Vols. Comparison for DAX Call Options 11Pn`ee5 ._ Ope Figure 4.10: D+3 l\/Iarket Prices vs l\/Iodel Prices Comparison for DAX Call Optlons 550 D+ 1 D.1i11{ Ind exüplinne F'r1`1::e5 Elaclcl est CHAPTER 4 BACKTESTING SVI 500 ~ 450 - -'-100 - 350 - Lú II 'Z I"¬-.'I |'.|'I 4:1 200- 150- 100- É â $ Ê É 3Ê¢ iÊ . *Ê Ê 5 3 * Ê Í í 111- M1:111elPr1eee <I> Me1lcelF'11`ee5 -$- -='¬`;=- *IF <`;> 1* -II? 55' -III- -$- -III- 5-“111 Leg-F er'-11'er11 M1:1ne111¬ eee |II+3 De-.Il |r'1dexOpl:i1Jr'15lm|:1. Vol Becklest Ú-4 1 1 1 1 1 1 1 1 -0.05 -0.05 -0.04 -0.02 0 0.02 0.04 0.05 0.35 - - 0.3 - - III I"'I1.'I U1 ílllü 5"1"I FiIIeE1Cur\.-'e 111 D+3 EIi1:l|m|:1. 111115 D+3.¿'-.5lclr11p.I.f1:1l5 *1 111111 +11- +** P ***+-II* *-111, +_*_¬Ir + *fr 11%* _ +_** _ 'F |:|.2 - +'11f.¡‹. - _* 'IF*ee Leg-F er'-11'er11 M1:1ne111¬ eee D+3 DMI lndexüplions F'ri1:e5 Baeklesl 1400 1 1 1 1 1 1 1 1 -015 -0.1 -0.05 0 0.05 0.1 1uuu_ 55. _ ¬'~'.?='-21° S00 Cl? 1:1 4:1 400 *II- <`>*- <`F'-I1- <`.>iI& -5+ <¬`.É>-III- - 1315112- -=Í`.='-Ir <`3›-315 <Z`>°Ii- -=í`.>-III- -=.'Í> -;`>-lI- 15-II* -.-;}¬1--31II- 13*-11%115 #1 M 1:1 11 el Prle ee -=í> M5111 el P111: ee 12uu _ É _ $$$$$$$$à _ Q* _ ill- '$- _ eg* _ -531119 <;z.~IIIl- -=:1~-III- 200 - - Leg-Fer'-Nerd M1:1r'1e111¬ ese 1 1 1 1 -0I.15 -0.1 -0.05 0 0.05 0.1 4.2. DAX INDEX SPOT OPTIONS E3._ e1:l'1"1:1eI Figure 4.11: D¬'-3 Market vs SVI Implied Vols. Comparison for DAX Put Options Figure 4.12: D+3 l\/Iarket Prices vs l\/Iodel Prices Comparison for DAX Put Options Figure 4.13: D+5 l\/Iarket vs SVI Implied Vols. Comparison for DAX Call Options 1-1111 D+3 De-11 |1'1dekOpti1:11'1el11'1|:1. '1f1:1| Beekteet Ú-34 1 1 1 1 1 1 ílllü 5'l"IFilleE1Cur\.-'B -11- _ -311-0.32 *_ 0.3 III I"'I-1.'I CO 0.25 - 0.24 - - 111- D+3 Elid|111|:1.'1"ele D+3.-'11-.el<lm|:1.".f1Jl5 + 11% -111- -III- 11€ 41- -11- -111 -IIE -211- -IP_ _* _ 11% *P-II* *$- EL2-% 1 -0.05 -0.0Ei -0.04 -0.02 0 0.02 0.04 L1:11;1-Fer-wer11 M1:11'1e-1111 eee D+3 1221.-1111-llnde>ctI.`1|:1lio1'1e P111: ee Beekleet 451:' 1 1 1 1 1 400 350 - Pfieee I'*~1.`I U1 'CI - 200- Op1:11'1 15n_ 5u¡ í e 5 4e ÊI11 Ê.. I .e*' _ I 111- M 1:1 11 el P111: ee <I> Merkel P111: ee _ 4 *Í-1 43;. _ -1° Ô 51111- 1,. 1” _ 1* Í 111 11111- 1.* _ .I L1:11;1-Fer-wer11 M1:11'1e-1111 eee D+5 De-.1l Indeküpli 13115 Imp. 'ifel Beckleet -U0 1 -0.05 -0.05 -0.04 -0.02 0 0.02 0.04 Ú-4 1 1 1 0.35 0.3 - E5._ 1:I"1"1:1e.I mpe F' I"'I-1.'I U1 ílllfl 5"1"I FiIIe121Cur\.-'e 111 D+5 EIid|1n|:1."1'ele D+5.-'11-.el<lm|:1.".f1Jl5 +11* e +e***e51í'+*_* *É -1+1 1 1 1 ` 51* -111 -111 11.2 _ *11 _e++ 11- T3111 1* -0.15 -0.1 -0.05 0 0.05 01 Leg-F er-11-erd M1:11'1e-1111 eee CHAPTER 4. BACKTESTING SVI D+5 IIII.-fi1111Íln1:le:1›:I21|:1lic1115 F'11`1::ee Elackleel IÊÚÚ - 1 1 1 1 - lPneee1+ 111- Mede ` .¿;._111 <I> Merkel F'11`eee 43.11' 1111111_ Q,-1- _ ¢_-›11- -C1 E100- *_ - ge Ç' 111 51111- 1* _ ü-11 * 111 1-1F'1-ieee ._ CI|:11:1 <I> -1)-11 131-111 <';-1111 122-111- 1.1-111 -$-111 -$-111¢-111 -='.`3-111 11';-1+ -11-111 -='¬';=-II1 400 - - 200 - - 5%1:1- __ E 1 1 1 1$ä$$$$1I1¢-1' - .15 -III.l -III.III5 III III.EI5 II|.l L 1:11;1-Fer-.-1-erd M1:11'1e-1111 eee Figure 4.14: D-1-5 l\/Iarket Prices vs l\/Iodel Prices Comparison for DAX Call Options ílllü 5111 Filte1:lCur¬~.1e 111- D+5 Elid|111|:1.'1"ele D1+5.-'11-.el<lm|:1.'1f1Jl5 0.32 - TK- 11€- * 11€-0.3- 111 - *II- 11€- 111E3._ 41- % 111 m`e1:l'1"1:1eI 1:1 iu CO *Ê -115-_ *_ _ *Ii- 13' 11.25 _ + _ -1* 11 -11 111 0.24 - - 112 1 1 1 1 1 1 -20.1 -IILIIIB -III.II|E¡ -III.III-1 -III.III2 III III.III2 III.III-4 L1:11;1-Fer-1.-er1jM1:11'1e-111¬eee Figure 4.15: D+5 l\/Iarket vs SVI Implied Vols. Comparison for DAX Put Options D+5 IIÉI.-111-llnde>cC.`1|:1lio1'1e Prieee Beekleel 45Ú 1 1 1 1 1 1 11 Medel Pneee 4% _ -=í> Merk P11'1:ee 111 5511 _ _ -C* -11-51111 _ -1 _ -11- -C1 -111 _ {} _ -111- 3 -= 21111- 3 _ 4 Ê Op1:11'1F'11`eee I'\.'I I'_I'I I: 150- 111 _ -$- 41 I 100- 115- - 41115É 511- --1113* _*.15 1 1 1 1 1 1 -“0.1 -II|.IIIE1 -IJ.II|E -11.04 -IILIJ2 III 11.02 IIL04 L1:1g|-Fer-1.-er1jM1:111e-111¬eee Figure 4.16: D+5 l\/Iarket Prices vs Model Prices Comparison for DAX Put Options 4.3. SPX INDEX SPOT OPTIONS 29 4.3 SPX Index spot options SPX Index Spot Options are negotiated at Chicago Board Options Eaíchanged and is the most liquid equity options in the world. The underlying currency is USD (American Dollar). Figures 4.17 and 4.19 show fitted SVI curves for the SPX Index options expiring in 18-Mar- 2016 and traded at 22-.Ian-2016. As one can see through Figures 4.18 and 4.20, the SVI curves for call and put options are free of butterfly arbitrage. a b m ,o o SPX Index Spot Call Options 0.1998 0.1273 0.0644 -5.5263 0.0022 SPX Index Spot Put Options 0.2381 1.2779 0.0603 0.3069 0.0200 Table 4.3: Calibrated Parameters for SPX Index Options 5F'}{lnclekCelle 5'1"l F1l:I:e1:lCu1*-.-'e 511 III1-22-IE 0 3 1 1 5'1"I Fil |11'1|:1. '1f5l Elldlm 11515 e. D- ífl-.ek|rn|:1."."5le III 28 - `¬-1. 525 E* _ 5 É ..§ 52-1 ä _ E '-¬. â É 522 _ ¬¬1. ¬¬1. ¬1-1. 02- - ¬¬-1. ¬¬-1. ¬-1. 111€ 1 1 1 1 1 - .EI-4 EI.III2 III IILEI2 EI III-4 III IIIEI III IIIE L5g-151-1.-1-er-1:1 1115115-meee Figure 4.17: SPX Index Call Options” SVI Fitted Curve (expiring: March/ 16) 5F'Í1{C3l|3G[I1.-'] 15 15 l _ 1--1 _ 12 _ 15 _ E1:15 É _ Ei J - 4 f) - 2 _ J \ _ E 1 1 1 1 1 .U4 -0.02 0 IIL02 0.0--l 0.IIIE¡ 11.05 L5g|-F51'-1.-er1jM511e-zmeee Figure 4.18: g(y) function for SPX Index Calls Options SVI Fit 30 CHAPTER 4. BACKTESTING SVI 5F'Í1{l111:lekPul5 5111 F1tte1:l 112511-'e 511 01-22-15 Ú 42 1 1 1 1 5111 Fil |111|:1. 11515 EI15l111|:1 1151 1 5 41 ' 3í.-'I'-eklrnp. '1'c1le 0 3 El II' fr1' 555 1 _ I 12111 I1I.'I5 55-1 1 _ 5 ,H Ci=- 1 É 552 1 _ E. § 55 H 1. If ,1- 525 _ .-' 5 25 `“-“' _ 0 24 I 0.2 0.15 01 -0.05 0 0 05 01 015 L5g-1511-1-51-1:1 111511e-111555 Figure 4.19: SPX Index Put Options” SVI Fitted Curve (expiring: l\/Iarch/16) 25 SPX PIIIZS G['1.1'] 20 - 15 \ - É 10 - 5 _ 5_ ííí' _ _ 1 1 1 1 1 1 1 -02 -0.15 01 -0.05 0 0 05 01 015 L55-F 51'-11-515 M511e-1111 eee Figure 4.20: g(y) function for SPX Index Puts Options SVI Fit 4.3.1 Backtest Results for SPX Index Spot Options Figures 4.22, 4.26, 4.30, 4.24, 4.28 and 4.32 show the comparison between market prices and the predicted prices at the dates D + 1, D + 3 and D + 5. The predictions are evaluated with the SVI volatilities calibrated at D. Figures 4.21, 4.25, 4.29, 4.23, 4.27 and 4.31 compares market implied volatilities of bid and ask prices and the volatilities of prices predicted by the SVI model. As we can see in the figures and in Table 4.4 the prices given by the SVI model are close to the market ones, which means that, this model can be used to predict prices. 4.3. SPX INDEX SPOT OPTIONS Calls D+1 D+3 D+5 Data misfit for vols. 1.23.11 _1 1.24.1 _1 1.53.10_2 Normalized K2-error of prices 2.17.1 _ 2.11.1 _ 1.04.10_1 Average error over spot 2.70.1 _ 2.50.1 _ 1.40.10_3 Puts E E D+5 Data misfit for vols. 7.93.1 _ 8.97.1 _ 6.05.10_2 Normalized K2-error of prices 4.65.1 _ 9.42.1 _ 8.84.10_2 Average error over spot 2.50.1 _ 4.40.1 _ 1.80.10_3Cl)CÇJ<_J_Â;LL)LL)L.)wwwlfic.o1- ÇL>Ç_>CL>_Á;CL>Ç_>Ç_>oowwbô0.11: Table 4.4: Quantitative Indexes for SPX Index Options _lII105"."lFilte1:l1I2111--.fe 111 II21+1EIi5|r11|:1.'11'5l5 1] 3 D 1.111-.el<lr11|5.".f5l5 -11- 5 25 _ -11- -31- IÍÚÉ 525 _ É É É 111* 11"É 52--1 1111* _ E -11_ H* e** 5 22 11- - 11- 5 1 * 0.2 - - 131€ 1 1 1 1 1 1 .04 0 02 0 0.02 0.04 0.05 0.05 0.1 Figure 4.21: D+1 l\/Iarket vs SVI Implied Vols. Comparison for SPX Call Options D 1 SPI l111:lek1í25ll C1|:1l15ne F'n1:e5 Beekteet BÚ 51 1 1 15 1115rkel 51:15. + 5111 Fit 10 - 50 - -C1 50 - É 5 É -15 li 5 5 -E. O 55 _ 5 115 5 20 *_ ü * 5*** 55% 15_ 111* 5 - *1.5-21=.*._*_ '-Í? ü 11°* '$- zID‹:11:1 -F1 1:1 :_ ru D_ F' :_ r~_1 ID :_ -Ii 1:1 :_ 1:1-1 5:' :_ Wee 1: Figure 4.22: D-'fl l\/Iarket Prices vs l\/Iodel Prices Comparison for SPX Call Options stes ._ mped'lie IÍI.4E CHAPTER 4. BACKTESTING SVI EL44 - EL42 -K- IIL4 - III ILIIJ CO III LIIJ ICD Ç ‹.-:I -Pl- IIL32 IÍL3 III.2E - I I íllllíl SVI Fittedüurve sf- D+1Bid|m|:|.'lI'nIs *É D+1.-'1¬-.si<Im|:|.".fIJIs l '-'P-' % $ *E _ * 1* _ * *F _ + % *Ê _ ä 1* _ i se * _ i I I I I I -IIIJII5 III IILIII5 III.1 IIL15 IIL2 IIL25 Leg-Fur'-i-'erdtvieneymess Figure 4.23: D+1 Market Vs SVI Implied Vols. Comparison for SPX Put Options cmF'riees._ On I"¬~.'I I Z _.. U1 'Z Iun_ _É 35Ú I I I I I I Ê I . 5i.fIrIt SDI] - - zsu _ * _ 6 su_ $ âäi sí' D+1 SPI Index Put Options Prices Bsektest <;› market obs ie â ü Ê sf _ *Ê I I I I I I -DDI -IILIII5 EI IILEI5 EI.i III.i5 III.2 IIL25 Leg-Forward Menew ess Figure 4.24: D-'fl Market Prices Vs Model Prices Comparison for SPX Put Options E3._ edi.fue.t mn D+3 SPI Ir'II:IexCeJ|s Options Imp. '*I'uI Beektest Ú-32 I I I I I I [LS III.2Ei III INJ CU F ru -Pl- Ú.22 IIL2 ílllfl 5".fI Fittedüuwe se D+3 Bidlmp. 'ifnls D+3.-'1¬-.si<Im|:I.".f|:|Is % ä % %** _ em _ +++* ++* _ _ _ ä * ä |]_1Ê I I I I I I - .U4 -EI.III2 IJ I].IJ2 IILIJ4 IILIIIE Ilüfl EI.1 Leg-I' ur'-.-«srd munewiess Figure 4.25: D¬'-3 Market Vs SVI Implied Vols. Comparison for SPX Call Options 4.3. SPX INDEX SPOT OPTIONS nPrices ._ Opo Figure 4.26: D-I-3 Market Prices Vs Model Prices Comparison for SPX Call Options E3._ mp`e|:I'Ifoet Figure 4.27: D+3 Market vs SVI Implied Vols. Comparison for SPX Put Options nPrices ._ Opo Figure 4.28: D+3 Market Prices Vs Model Prices Comparison for SPX Put Options EIEI SEI Í"EI EEI 5EI -Él-4:1 SEI 2EI 1D D+3 SPI Inde>:OsJI Options Prices Eiacktest im] I I I I I I EI.5 EL45 EI.4 1:! |'.II.'I lJ'I EL3 |z|_z I I I I I I I -EIS -EI.1 -EI.El5 EI ELEI5 EI.1 EI.15 EI.2 35Ú I I I I I I I qr 5"."I Fit _ _* _ SIJEI 25EI I'\I.'I II 4:1 _ |'.|'I 4:1 IIJEI 5EI -3tI-<Í`.=- -It-=l`.`=- O1+.-_.+__ G <:> market obs. _ ir Siri Fit üü e Pc _ _ O-1. -=L'> *$-_ +++T:i.;}<_> _ +ee*Í í Ê Ê É Log-f on.-àsrd rnonepness D+3 5P}{InI:Ie><PutOptionsImp. 'Ifol Bscktest -8.04 -EI.EI2 EI EI.EI2 ELEI4 EI.EIEi ELEIB EI.i -III- 4° _ *_ _ -Jr *- -3* _ -F _ -Itr + + e e * s íülll Siri Fittedüur'-.fe se D+3 Eticllrnp. trois D+3.¿'-skIrnp.I.foIs $- _ 1* *_ _ Log-For'-«sro Monepn ess pi D+3 SPI Index Put Options Prices Becktest ~f.> msritet obs 2ser* Ê 'É' O -IP _ O _ Ê_ Ê _ ' .sí ' Log-For'-Nerd Monepn ess I I I I I I I -8.15 -Ili -EI.El5 I] EI.I]5 Ili U.15 EI.2 CHAPTER 4. BACKTESTING SVI íülll Siri Fitteciüur'-.re rir D+5 Eiiciimp. trois D+5.¿'-sicImp.IroIs EI.2S - - EI.2E¡ - - _ - *_ \ _ _* ¬-_ + ¬. stes F' I"¬¬.'| -P- '¬. mped'ro = *_ *es. _ *F _oca +_* *I-*I- *#- 'if' _ 4° _EI.2 *r -315 |z|_1Ê I I I I I - .EIB -EI.III-4 -III.II|2 III ELIII2 III.III-fi IILIIIE Log-Ion.-.-srdmoneiizness Figure 4.29: D¬'-5 Market Vs SVI Implied Vols. Comparison for SPX Call Options D+5 SPI Inciexfisii Options Prices Bscictest ÍÊÚ *_ I I I I I <:> msritet obs. P se Siri Fit IEIEI- - BEI- + - +- Q O BEI- -nPrices ._ Opo <+I- -Iii-4U- ._ _ - -c-*É Égfiä 's EU- Éärs _ _ É s ts I I I I I -REIS -EI.III-4 -III.II|2 III ELIII2 III.III-fi IILIIIE Log-Ion.-.-srdmoneiizness Figure 4.30: Dr-5 Market Prices vs Model Prices Comparison for SPX Call Options D+5SPi€In|:|exPutOptionsImp.*roIEIscictest os4 , _____oosrInuzúcuwz EL42 _ se D+5EIid|mp.'lroIs D+5.fi-.skImp.¬roIs o4_ Pxfixtä* _ oss_ _ E3 Í: cú cn ._ eciiroet 4:1 Lú -Ii mn F' Lú I"'I-i.'l EL3 - -4» - EI.2EI - *Ê *Ê - -It- U.2Ei - sie - DÊÊH -HI pis É oi? um Figure 4.31: D+5 Market Vs SVI Implied Vols. Comparison for SPX Put Options 4.4. PETR4 EQUITY OPTIONS 35 i ‹í> msritet obs. Stri Fit «$- 2I]EI - IF- Iso _ s \'_'I :Ê -c Ç '-If E 15. P Ino _ 4:- -=1`;> -2+ $- 5EI É - *I-Ii* *Ê Ê*ers rir I* E I I I I-II Is oi -nos o o os oi Log-For'-«sro iirionepn ess Figure 4.32: Dr-5 Market Prices vs Model Prices Comparison for SPX Put Options 4.4 PETR4 Equity Options Petrobras PN equity options are the most liquid equity options in Brazil. They are negotiated at BMFʧB0fuespa Stock E5126/tange in BRL (Brazilian Reais). The volume of daily trading for these options are not as big as the ones in the examples above, Which can give us less accurate results. Figures 4.33 and 4.35 shovv fitted SVI curves for PETR4 Equity options expiring in 21-Mar- 2016 and traded at 5-Feb-2016. As one can see in Figures 4.34 and 4.36, both SVI curves are free of butterfly arbitrage. 0. Õ m ,0 U PETR4 Equity Call Options 0.1998 0.1273 0.0644 -5.5263 0.0022 PETR4 Equity Put Options 0.2381 1.2779 0.0603 0.3069 0.0200 Table 4.5: Calibrated Parameters for PETR4 Index Options PETFI4 Equitpúeiis Stri Fitted Cur-.fe on U2-IEI-IE I 2 I I I I I I ¬ Stri Fit Imp. trois Eiidimp. trois ir'-.sk|mp. trois 1 1 _ 1 _ Ú? É Ê§ III EI - 'É É E. É '¬¬_ _. EI Ei "`-_ “_ - THE _,_I-P"_'_-___'-I-_'-,T- ¬"\.`¬`_` _'_'____¡_,__-I-'__ "'~¬._¬`_ ___'_'__I-I-_ EI? - I I I I I I I II|.3 -IJ.2 -Ili IJ III.i 0.2 0.3 III.-4 Log-For'-I-erdiwioneymess Figure 4.33: PETR4 Equity Call Options” SVI Fitted Curve (expiring: March/16) CHAPTER 4. BACKTESTING SVI 18 i i s Pit csiis em i i Is_ I _ I~=I_ _ 12- _ Io_ _ ÉII:o S_ _ E- ._ /\ 2 _ _ E I I I I I - .EI-4 EI III2 III IILIII2 EI.III-4 IILIIIE III IIIE L op-For'-.-Isrd i\I'Ione'IIn ess Figure 4.34: g(y) function for PETR4 Equity Calls Options SVI Fit PETFI-4 Equity Puts Stri Fitted Cur-re on III2-IIII-IE Stri Fitimp trois Eiidimp trois I4 I I I I I I I 3 _ í.-'I'-skimp. trois 12- - *IO2 II- _ É D Ib- 'El É 1_ .-ff' _ CL É .z .-' os_"-~__¬¬ _ ¬-¬¬_¬_H 'J- ¬¬¬-H ¬. .-ff,-¬-'¬-I. ¬¬-I.-I¬_`_ __.--I¬_`_ __.- EI S _"“--_ --""_ _`_i`í_'_'__ _ I;|_;I' I I I I I I I I -II|.4 -IIL3 -III.2 -EI.i III III.i III.2 IIL3 III.-4 III.5 LoI;I-For'-I.-risriji-Iione'IIness Figure 4.35: PETR4 Equity Put Options” SVI Fitted Curve (expiring: March/16) PETFI-=I I=-IIIs cw] 3 I I I I I I I I /*z 25- - 2_ _ ÉII:o i5_ _ 1- _ ELE t I I I I I I I\I- .4 -III.3 -0.2 -0.1 U III.i III.2 11.3 IJ.==i III.5 Log-For\II'erdi'-Iione'IIness Figure 4.36: g(y) function for PETR4 Equity Puts Options SVI Fit 4.4. PETR4 EQUITY OPTIONS 37 4.4.1 Backtest Results for PETR4 Equity Options Figures 4.38, 4.42, 4.46, 4.40, 4.44 and 4.48 show the comparison between market prices and the predicted prices at the dates D + 1, D + 3 and D + 5. The predictions are evaluated with the SVI volatilities calibrated at D. Figures 4.37, 4.41, 4.45, 4.39, 4.43 and 4.47 compares market implied volatilities of bid and ask prices and the volatilities of prices predicted by the SVI model. As we can see in Table 4.6, the predictions are less accurate than in the previous examples, probably due to the smaller amount of data and the smaller trading volume. However the predicted prices presented skewness similar to the market ones, which is a nice feature. ¬'-Calls D¬'-1 D¬'-3 D 5 Data misfit for vols. 1.23.10_1 1.24.1i _ 1.53.10_2 Normalized Zz-error of prices 2.17.10_1 2.11.1 _ 1.04.10_1 Average error over spot 2.70.10_3 2.50.1 _ 1.40.10_3 Puts D+1 E D+5 Data misfit for vols. 7.93.10_2 8.97.1 _ 6.05.10_2 Normalized É2-error of prices 4.65.10_2 9.421 _ 8.84.10_2 Average error over spot 2.50.10_3 4.40.1 _ 1.80.10_3C)L_)C›_*:C)Ç›L_› cowwbôc.oI~I-I Table 4.6: Quantitative Indexes for PETR4 Index Spot Options EI 1 PETFI4 Eouit'-_: Ceiis Options Imp. troi Secictest íišifl Stri Fittedüurve re D+i EIII:iimp. trois D+1 .-'1¬Isi=:Imp. trois I EIEI5- - -:-'II-:|.'- -E [Ig K / .-" Ê .f ;oss_ / _ 'É É Ê' EIS - se ors ,,, + _ or *I _ USE - + I I I I I I I III 4 IJ 3 IJ 2 -III.i IJ IJ 1 IJ 2 III 3 III 4 L oo-For'-«sro i\I'Ione'IIn ess Figure 4.37: D+1 Market vs SVI Implied Vols. Comparison for PETR4 Call Options CHAPTER 4. BACKTESTING SVI EI+1 PETFI-4 Equity Coil Options Prices Eiscittest I I I I I I I tt* ‹í> msritet obs. syi ,II _ Q _ 1.4 1.2 1- _ *- O nprices F' CIO ._ Opo 1:1 is-.I _ Ir<> EL4 - - S <`HI- -=íi>+ -II- -IF os _ _ Q 'Ô I I I I I I I -“Dri -III.3 -III.2 -III.i III III.1 III.2 IIL3 III.-4 Log-ionI-'srI:imoneyrIess Figure 4.38: D-i-1 Market Prices vs Model Prices Comparison for PETR4 Call Options D+1 PETFI4 Equity Put Options Imp. troi Becictest I.-4 I I _oo sri Firiúúcuwe [ fr- D+1 Bidimp. trois EI+i .ft-skimp. trois1.3- I.2- - /' mpeotrostes = I o.s_ ,,,+* _ 'III- EI.S - - -IIE -IIE |z|_;I' I I I Ii* I I T I I -III.3 -III.2 -III.i III III.1 III.2 III.S III.4 III.5 IILE Loq-For'-ysroiyioneymess Figure 4.39: D+1 Market vs SVI Implied Vols. Comparison for PETR4 Put Options EI+i PETFI4 Equity Put Options Prio es Becktest 3 i i i i i i i i _ <> msritet oi:Is. «I Stri Fit 2.5 _ IS _ _ ä _2 S nPrices íI.I¬ S __ I Opo -O-*I- I _ S _ o.s _ I* _ I s I [IU I I I I I I I I - .S -III.2 -Ili I] IJ.1 III.2 IIL3 IJ.-4 IJ.5 III.Ei Loq-For'-yeroiyioneymess Figure 4.40: D+1 Market Prices vs Model Prices Comparison for PETR4 Put Options 4.4. PETR4 EQUITY OPTIONS ES._ eotroet Figure 4.41: D¬'-3 Market vs SVI Implied Vols. Comparison for PETR4 Call Options Figure 4.42: D+3 Market Prices vs Model Prices Comparison for PETR4 Call Options Figure 4.43: D+3 Market vs SVI Implied Vols. Comparison for PETR4 Put Options rop EI.EI5 - EI.S5 Z ¬-IJ U1 EI.S5 EI.5 onprices._ UD SS._ dtroert mpe EI.EI - EI.S EL? ELE I EI+S PETFI4 Equity Osiis Options Imp. troi Eiscktest rir D+S EIiI:iimp. trois D+3.¿'-sicimp.trois \\~\\______/Ef *II- III- qr. 5* _ O -ir -III- _oo syi Firrúúcowz L +*+'- I.S -Sus Loq-For'-«sro iyioneyn ess EI+3 PETFI4 Equity Csii Options Prices Bscictest I I I I I I I I -III.4 -III.S -EI.2 -III.1 EI III.i EI.2 EI.S III.4 IS- 1.4- 1.2- 1- III CO ELE - EI.4 - EI.2 - i i i i i i i i <> msritet obs. Stri Fit É O O III- O O 'III''O giz* *¡"`<:› O O-IE OO O-IIII-“fr I.S 5 I I I I I I I III.4 -III.S -EI.2 -III.1 EI III.i EI.2 Log-I oryrsro moneyness D+3 PETFI4 E quity P ut Options im p. troi Becittest -:- cú- 1.4- 1.2- I 1:1 CCI EI.S EI.4 o ._2ü às iII+3 Eiiciimp. trois D+3.4.skimp.trois I _oo syi Firiúúcuwz [ / "13' 'IF. 44 1* *L *_ I-*T _ _* _ EI4 .2 I I I I I I -0.1 IJ III.i EI.2 IJ.S I].4 III.5 Loq-For'-«sro iyioneyn ess CHAPTER 4. BACKTESTING SVI EI+S PETFI4 Equity Put Options Prices Bscktest 2 5 I I I I I I ‹í> msritet obs. Stri Fit 2 _ _ 1.5- Ê - O nPrices ._: *C- O Opo I_ 4. _ O É É o.s_ _ $ S I I I I I I -DELE -III.1 III III.i IIL2 IIL3 III.4 IIL5 Loq-For'-ysroiyioneimess Figure 4.44: D¬'-3 Market Prices vs Model Prices Comparison for PETR4 Put Options EI+5 PETFI4 Equity Csiis Options Imp. troi Bscktest i.S rir D+5 Eiiciimp. trois1.2 _ D+5.¿¬-.sicimp.trois _oo sri Firrúúcowz L 1.1- 1- _ *~»~_____/ E3 mpeotrost 1:1 'ss ELF- - O * 1* + _ + EI.S- _* Ip - O os _ _ _ I;|_ I I I I I I I ÍIL4 -IIL3 -IIL2 -III.i III III.1 IIL2 IIL3 III.4 Loq-For'-ysroiyioneimess Figure 4.45: D+5 Market vs SVI Implied Vols. Comparison for PETR4 Call Options EI+5 PETFI4 Equity Coil Options Prices Becictest I I I I I I I *L <> msritet oi:Is. -z;;I qr. 5"I"I Fit 1 _ _ 1.2 III _ 4* _ nprices 5:' CCI ._ Opo 1:1 izn _ _* _ O EL4 _ - + III 'S II» ELE- <;› 4° _ <:› II» c *I *Ir O .I-_; Q [IU I I I I I I I - .4 -IJ.3 -11.2 -III.i IJ Ili I].2 IIL3 III.4 Log-Ion.-.-srcimoneyness Figure 4.46: D-E5 Market Prices vs Model Prices Comparison for PETR4 Call Options 4.4. PETR4 EQUITY OPTIONS EI+5 PETFI4 Equity Put Options Imp. troi Bscictest 2 I I _oo sri Firiúúcuwe L 41- D+5Bidimp.troIs I-3 - EI+5.-'1¬-.sI<Imp.trois 1.E- _ 1.4- _ E3._ mp`eI:itrost E 1 _ _ /Í/;è o s `__- --'__`T¡_T1 1 * M 1 4 ¬.. * _ 4 EI.E _ sie - I;|_ I I I I I I ÍIL2 -III.1 III III.i EL2 EL3 EI.4 EL5 Loq-For'-ysroiyioneimess Figure 4.47: D4-5 Market vs SVI Implied Vols. Comparison for PETR4 Put Options EI+5 PETFI4 Equity Put Options Prices Becittest 2-5 I I I I I I ‹í> msritet obs. Stri Fit 2- _ I.s_ _ S nPrices ._ Opo I _ 3 _ S É EI.5 _ _ _ S * S gn I I I I I I - .2 -III.1 III ELI EL2 ELE EI.4 ELE Loq-For'-«sro iyioneyn ess Figure 4.48: D+5 Market Prices vs Model Prices Comparison for PETR4 Put Options 42 CHAPTER 4. BACKTESTING SVI 4.5 Confidence Intervals Although we are considering volatility as a deterministic function of the log-strike, we state confidence intervals, based on the bid-ask spread of option prices as follows: o From D0 data, we evaluate the modulus of the difference between observed bid-ask implied volatilities, take the maximum over strikes and denote this quantity by Õ. o Denoting the SVI curve at each log-strike lc by 25;/¡(/c), we consider 25V¡(/c) i Õ/2 upper and lower bounds. 0 We calculate Black-Scholes prices using such bounds on the volatility to define confidence intervals for the option prices. 4.5. 1 Numerical Example Now, we evaluate such confidence interval using the D+1 backtest data for DAX Index Calls 4.2.1. Õ = 0.0423 Defining siüoniicience interysi iorE1I'1'-.IE Csiis Prices on III+1 IÊÚÚ I I I I - III Iyiooeitlp Eiounosry Prices III Model Lo¬II.rerBounoeryPrices Iyisrirzet Prices IEIDEI . CCI Z II -III- -II-BIG -*HIP *II-515 -III-III- III- *III- *III SEIEI *iii - *iii- EO ED \'_'I EI:I_ E E “EL *Ik- O 1.-2* Irflr* 4EIEI * _*ils *-12- II=** * III»III I'*~I.`I Z Z III «III IIIII IIIIII III-III -IIIIII E I I I I - .15 Eli EID5 EI ELU5 ELI L oq-For'-«sro iyioneyn ess Figure 4.49: D+1 DAX Calls Prices Confidence Interval We can also determine a confidence level for our interval by performing a Kupiec test [17] by determining the number of violations, i.e., the number of market prices outside the confidence interval. On the example above,the number of violations was 0 as we can see in Figure 4.50. 4.6. AN ALTERNATIVE APPROACH 43 Model Pnc es snci Interysis in Perspectiye EEUEI , , EEIIJEI _ _ Ú? -_-_ É Isoo_ _ _E _ _, _ ã _ III? _ IIIJ É _ E Iooo " _ CD 'El Cv E I'.|'I Z .=I=-D I*-I.'I 1:! E- i '='i_ -I IIHIic ic '|II It :-1xxIII x L"ic |_›‹: 1:1 III 1:! I'_|'I 1:1 1:1 E __ Loq-For'-«sro iyioneyn ess Figure 4.50: D-i-1 DAX Call Model Prices with its Confidence Intervals 4.6 An Alternative Approach As an attempt to refine the technique proposed in Section 4.1, to forecast implied volatilities for the day D + j with maturity time T, we evaluate, by an interpolation method, implied volatilities for the day D with maturity time T - gi. So, we use the two-dimensional interpolation method proposed in [15] and proceed as follows: 0 Calculate implied volatilities smiles for several different expiring dates 15.; a set of indexes for discrete time-to-maturity quantities) with data taken from the market; o For each maturity t G It,¿,t.¿+1I and strike K we calculate Uf,,,p(K,t)t by interpolating linearly between UÊmp(K, t¿)t.¿ and 0Êmp(K, tz-+1)t,-+1, where such values were obtained with SVI. 0 So, we evaluate the surface of implied volatilities (K, t) H aÊ.mp(K, t). As an example, we calibrated the surface for SPX European Options with data taken from @Bl00mberg in À/Iay,13th. The total variance and implied volatility surfaces can be seen in Figures 4.51 and 4.52: 44 CHAPTER 4. BACKTESTING SVI kiEI eitreeiiiences EE.I'..I'. z __..- . - oos __-~" f_ .-.,.. siso Ú.E|E i pmfl oos ..~~" “- u_u 2EI5EI EI 2EIEiEI _ Tim e-t o-I'-Iiet ury SIIIIIES 'EI-' '_ 1:1 Oil OJ E ID Ê. :_ ID Iš Figure 4.51: SPX European Calls - Interpolated Total Variance Surface _-I. .' Í -" ' _Ê ois _.-“"Ê__.-É ' -.Â É _}' :_. : É:_ oI4 _.~' ' 'CI EDss ois-_ E 'E oI2 _ E Cire oii _ ro- 11': Iš oi _| _ _ zooo _._.-._ .~ “U5no2EI5EI 4 21E|EI 215III Ú U2 EÊÚÚ Ú Time-to-iyietury Strikes Figure 4.52: SPX European Calls - Interpolated Implied Volatilities Surface As an example, we took market data on May, 18th for SPX calls expiring in May, 27th (j = 9). On the figure 4.53 you can find the SVI curve fitted with ínterpolated implied volatilities for D+j on May, 13th (a,Êmp(K, 9)) as well as the bid and ask implied volatilities from market. 4.6. AN ALTERNATIVE APPROACH ES._ mp`eI:itr'oet 0.1? 0.1E 0.15 0.14 III _ Lú III _ I"'I-I.'I 1:! _ _ SPIIE Cells Im piied troistiiities Sscktesteci 'with Stri from Osiihrsted Surface -I. -I. O ____ I I _D+] S tri Fitt e ci C urye _ _ ¬ 4 Bio Imp. trois on T-I `_~_ .¿'-.skimp.trois onT-I I ¬-I. -_ -I. III + 4 _ _* _ 0.1 _ _ 0.09 _ _ 0.0E _ _ O I I I I I I -ELEIS -ELEIES -ELEI2 -ELEIIS -III.EIi -III.IIIEIS III EI.EIEIS LoI;I-For'-ysroi-Iione'III¬ess Figure 4.53: SVI Backtest with Interpolated Volatility Surface Since the Data Misfit for the volatilities was 6.47:1110_2, the single example above shows us no clear advantage in relation to the original experiment of the present work. However, we believe it°s worth looking more carefully at it on posterior works. CHAPTER 4. BACKTESTING SVI Chapter 5 Concluding Remarks In this work we intended to backtest SVI parameterization of Black-Scholes implied volatilities in order to see how accurate such model is to predict option prices. After showing that SVI satisfies important non-arbitrage properties (see [18], I20I, [11] and 1131), we presented an algorithm based on techniques from [22] to calibrate the SVI parameter- ized volatilities from market data. The algorithm included some modifications to increase the robustness of reconstructions, in order to be less dependent on the choice of the initial guess of the parameters m and 0. This provided us the necessary tools to start the backtest and perform the numerical examples presented in Chapter 4. Finding reliable option data to test our algorithms was the first difficulty we faced. We had to gather all the necessary information about bid and ask prices of European options, choosing the ones with high trading volume. Thats why we have chosen DAX (EUREX), SPX (CBOE) and PETR4 (BMF&Bovespa) vanilla options. A visual way to evaluate how accurate are the calibrated SVI curves is to observe its position in relation to market implied volatilities. Under this perspective, SVI parameters did not produced accurate predictions for all strikes, specially those far from the at-the-money, however, it predicted important features presented by the market data, such as skewness. So, this model can be advantageous for practítioners since it is computationally cheaper than stochastic volatility models. Moreover, using SVI functional forms to price options for future scenarios might be an interesting tool in portfolio optimization techniques, such as the Black-Litterman inspired one, where the user inserts returns expectations into the model. See Going further on this research, we would like to suggest the same kind of test with smaller time frames in order to find out if it is possible to better understand how the volatilities react considering marginal disturbs on other variables. 47 CHAPTER 5. CONCLUDING REMARKS Appendix l\/[ATLAB Code for Quasi-Explicit SVI Calibration function Ia, b, rho, m, sigma, rmse] = SVI_fit(bid_vo1s,ask_vo1s,T,1og_strikes) %defining variables mid_vo1s = (ask_vo1s + bid_vo1s)/2; mid_tota1_vo1 = mid_vo1s*T; = 1ength(mid_vo1s); parameters_matrix = zeros(9801,6); T1 ui0; for m_initia1 = 0.01 : 0.01 : 0.99 for sigma_initia1 = 0.01 : 0.01 : 0.99 % changing variable 1og_strikes_chg = (1og_strikes - m_initia1)/sigma_initial; mid_tota1_vo1 = mid_vo1s*T; % Solving the linear system to reach zero for the cost function. A: Z: Par In sum(1og_strikes_chg) sum(sqrt(1og_strikes_chg.^2+1)); sum(1og_strikes_chg) sum(1og_strikes_chg.^2) 1og_strikes_chg'*sqrt(1og_strikes_chg.^2 +1); sum(sqrt(1og_strikes_chg.^2 +1)) 1og_strikes_chg'*sqrt(1og_strikes_chg.^2 +1) sum(log_strikes_chg.^2 +1)]; [sum(mid_tota1_vol); mid_tota1_vo1'*1og_strikes_chg; mid_tota1_vo1'*sqrt((log_strikes_chg.^2 +1))]; ameters = A\z; a_ti1de = Parameters(1); dz C: aí bi rho Z 0°0° Z / s Parameters(2); Parameters(3); a_ti1de/T; c/(sigma_initia1*T); = d/(b*sigma_initia1*T); ow, considering a, b and rho we will minimize the problem choosing m and igma 49 50 Appendix i = 0; m_loop = m_initial; sigma_loop = sigma_initial; while i <l0 X = S3/m('X'zIl 21); obj_function = @(X) sum((a + b*(rho*(log_strikes - x(1)) + sqrt((log_strikes - x(1)).^2 + x(2)^2)) - mid_vols).^2); x = fminsearch(obj_function,[m_loop , sigma_loop]); m_loop=x(1); sigma_loop=x(2); i=i+1; end m=m_loop; sigma=sigma_loop; fitted_values = a + b*(rho*(log_strikes - m) + sqrt((log_strikes - m).^2 + sigma^2)); lqs = sum((mid_vols - fitted_values).^2); u=u+l; parameters_matrix(u,1)=a; parametersnnatrix(u,2)=b; parameters_matrix(u,3)=rho; parameters_matrix(u,4)=m; parameters_matrix(u,5)=sigma; parameters_matrix(u,6)=lqs; end end % defining m and sigma looking for the least sguare errors sum min_error = min(parameters_matrix(:,6)); n = length(parametersnnatrix); for r = l:n; if parametersJnatrix(r,6) == min_error a = parameters4natrix(r,1); b = parametersnnatrix(r,2); rho = parametersJnatrix(r,3); m = parameters¿natrix(r,4); sigma = parameters_matrix(r,5); rmse = sqrt(parameters4natrix(r,6)); end end % graphic construction for visual analysis liminf = min(log_strikes); limsup = max(log_strikes); x_axis= linspace(liminf,limsup,10000); fitted_curve = a + b*(rho*(x_axis - m) + sqrt((x_axis - m).^2 + sigma^2)); plot(log_strikes,bid_vols,'c*'); 100 101 102 103 104 105 106 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 Appendix 51 hold on plot(log_strikes,ask_vols,'g*'); plot(x_axis,fitted_curve,'r'); xlabel('Log-forward moneyness'); ylabel('Tmplied Volatilities'); legend('SVT Fit Imp. Vols','Bid
Compartilhar