Delta and Vega Hedging in the SABR and LMM-SABR Models

Delta and Vega Hedging in the SABR and LMM-SABR Models

CUTTING EDGE. INTEREST RATE DERIVATIVES Delta and vega hedging in the SABR and LMM-SABR models T Riccardo Rebonato, Andrey Pogudin and Richard d t T T T dwt (2) White examine the hedging performance of the SABR t and LMM-SABR models using real market data. As a Q T E dzT dwT T dt (3) by-product, they gain indirect evidence about how well t t specified the two models are. The results are extremely The LMM-SABR model extension by Rebonato (2007) posits the following joint dynamics for the forward rates and their instanta- encouraging in both respects neous volatilities: q M i i i i dft idt ft t eijdz j , i 1, N (4) j1 model (Hagan et al, 2002) has become a market i i t gt,Ti kt (5) The SABR standard for quoting the prices of plain vanilla options. Its main claim to being better than other models capable i M dk k i t =μ i + = + × of recovering exactly the smile surface (see, for example, the local t dt ht ∑ eijdz j , i N 1, 2 N (6) k i volatility model of Dupire, 1994, and Derman & Kani, 1996) is t j=1 its ability to provide better hedging (Hagan et al, 2002, Rebon- with: ato, 2004, and Piterbarg, 2005). M = N + N (7) The status of market standard for plain vanilla options enjoyed V F by the SABR model is similar to the status enjoyed by the Black NV and NF are the number of factors driving the forward rate and formula in the pre-smile days. The Libor market model (LMM) volatility dynamics, respectively: arose out of the desire to provide a dynamic extension of the Black gt,T a bT t exp cT t d (8) model to handle complex products, while recovering the Black i i i prices for the underlying plain vanilla instruments. The desire to ht,T T t exp T t (9) achieve a similar extension has led to the introduction of the i i i SABR-compatible extensions of the LMM by Henry-Labordere M 2 = (2007) and Rebonato (2007). ∑ eij 1 (10) This article explores how successful the delta and vega hedging j=1 suggested by the SABR and LMM-SABR models are in practice. and: Neither aspect has been quantitatively investigated (the discus- Edz dz dt , j, k 1,2,..., M (11) sion in Hagan et al, 2002, is qualitatively convincing, but little j k jk hard evidence is provided). Fortunately, a sufficiently long history Rebonato (2007) and Rebonato & White (2007) show how to deter- of SABR coefficients has by now become available to make an mine the parameters of the LMM-SABR model in order to recover empirical investigation meaningful. The accuracy and simplicity to a high degree of accuracy the SABR prices for caplets and swap- of the formulas to obtain the prices of swaptions given the LMM- tions of all strikes corresponding to a given set of market SABR SABR dynamics of the forward rates presented by Rebonato & parameters. In brief, the parameters of the functions g(.) are deter- White (2007) make the analysis of the effectiveness of vega hedg- mined by minimising the sum, H2, of the squared discrepancies: ing a practical proposition. N T 2 2 i gTˆ (12) 0 i Description of the models i ^ To make the presentation self-contained, we briefly present the where the sum over i runs over the N caplet expiries, g(Ti) denotes . equations of motion of the SABR and LMM-SABR (Rebonato, the root-mean-squared of the function g() to time Ti and: 2007) and the approximate swaption formulas in Rebonato & T Ti E i 0 t T (13) White (2007). 0 t 0 i Ti In the SABR model, the underlying (a forward or a swap rate, The initial valuesk 0 are then chosen to provide exact recovery of T T Ti f t, of expiry T) follows, under the T-terminal measure Q , the the SABR market quantities X0 : dynamics: Ti Ti 0 k0 (14) T gTˆ T T T T i dft ft t dzt (1) As for the function h(.), its parameters can be chosen from the 94 Risk December 2008 rebonato.indd 94 1/12/08 14:16:43 relationship from the available market-given SABR volatilities of N Estimation of the unobservable volatility. The swap rate and volatility STi: its volatility are the two state variables of the SABR model. How- 1/2 ever, only the first is directly observable from market values. To k T 2 2 Ti 0 i μ 2 gtht tdt (15) create a time series for the latent volatility state variable, we pro- T 0 0 i ceeded in two different ways. Rebonato (2007) shows that this model gives prices extremely First, we simply recorded the SABR market-fitted values of the similar to the SABR caplet prices for a very large range of maturi- volatility for each trading day ti. Using the notation above, this is ties and strikes. the time series {X(tiaVi)}. Rebonato & White (2007) then go on to show that the Euro- Second, we estimated a value of the volatility in a manner more pean-style swaption prices implied by a forward-rate-based LMM- consistent with the assumption that the SABR model is a correct SABR model are very closely approximated by the prices obtained specification of the market (and its parameters, therefore, should be T from a SABR model for swaptions with initial volatility 80, volatil- constant). To do this, we calibrated again the unknown volatility to T T T T ity of volatility V , exponent B and correlation + , given by: the market prices, CK(ti), of the swaptions at time ti, but this time keeping the parameters, {Viï = Giï, Siï and Wiï}, at the same values T 1 0 0 i j T i j obtained from the fit to the market prices at time tiï. Using the nota- 0 ijWi W j k0 k0 g g dt (16) tion above, the time series thus obtained is denoted by {X(t aV )}. T 0 i iï i, j Clearly, for a perfectly specified model the parameters should never change ({V} = {V }, ), and the two time series would coin- T 1 0 0 i j T i j ˆ 2 i i+1 i V 2 ijijWi W j k0 k0 g g hij t tdt (17) cide (X(t aV) = X(t aV ), ). In reality, this will never be exactly the T 0 i i i iï i 0 i, j case, and the differences between the two time series contain useful T information about the hedging performance of the SABR model. B wkk (18) N Tests of the hedging performance of the SABR model. To k 1, n j test how well specified the SABR model is as far as delta hedging is concerned, we proceed as follows. T (19) ijij N Set of tests 1. With the first set of tests, we look at some key i, j statistics of the marginal and joint distributions of the time series i with SRt = 8iwi f t and: X(tiaVi), X(tiaViï), )X(tiaVei), )X(tiaVeiï) and )f(ti) and we test for their consistency with the SABR model. N Set of tests 2. With the second, more direct, set of tests we f i W w (20) compare the realised and predicted changes in swaption prices, i i B T T SR )C K(ti; real) and )C K(ti; pred) and at the realised and predicted ^ T ^ T changes in implied volatilities, )X K(ti; real) and )X K(ti; pred). The first two sets of tests give us information about the quality 0 0 i j T i j ˆ 2 2ijijWi W j k k g g httdt of the SABR delta hedges and of the SABR vega hedges for swap- 0 0 0 (21) ij 2 tions of different strikes but same maturity. To explore the ability V T 0 to hedge across swaptions of different expiries and tails, we must These are the equations used in our study to produce swaption resort to testing a dynamic model such as the LMM-SABR that prices given market SABR parameters for caplets. links the different swaptions together. N Tests of the hedging performance of the LMM-SABR Notation model. To assess the hedging performance of the LMM-SABR For a set of trading days {ti}, i = 0, 1, 2, ... , N, we denote: model across expiries, we proceed as follows. N the set of SABR parameters {G, W and S} at time ti by {Vi}, i = N Set of tests 3. Given a history of market SABR parameters for 0, 1, 2, ... , N; caplets, {Vi}, we calibrate the volatility and volatility-of-volatility N the same set of SABR parameters at time ti augmented by the functions that appear in the Rebonato (2007) LMM-SABR ti-value of the unobservable state variable X(ti) by {Vei}, where {Vei} model to the market prices of caplets. We then calculate the = {G, W, SXti}, i = 0, 1, 2, ... , N; changes in the prices of various swaptions using the approxima- N the difference in the swap rate between day iï and day i by tions reported in equations (16) to (21) (Rebonato & White, Ƌf(ti) } f(ti ïf(tiï), i = 1, 2, ... , N; 2007). We then compare the predicted changes in swaption prices N the difference in the volatility between day iï and day i by with the observed market changes.

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    6 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us