Riccardo Rebonato Global Head of Quantitative Research, FM, RBS Global Head of Market Risk, CBFM, RBS

Transcription

1 Why Neither Time Homogeneity nor Time Dependence Will Do: Evidence from the US$ Swaption Market Cambridge, May 2005 Riccardo Rebonato Global Head of Quantitative Research, FM, RBS Global Head of Market Risk, CBFM, RBS December 2004

2 Goals of this talk 1. Providing an effective and concise description of the whole swaption matrix 2. Explaining its changes in a more parsimonious and financially more transparent way than using PCA 3. Highlighting some intrinsic problems of time-homogenous and timedependent approaches to derivatives pricing 4. Suggesting how these limitations may be overcome in a computationally tractable manner. 2

3 Why do we want to do this? Intrinsic interest in a parsimonious description of the static and dynamic features of the swaption matrix Describing patterns and regularities using few, physically-justifiable parameters means understanding what determines and drives a certain phenomenon. (By-product: some insight into what we really need for option pricing in general ) 3

4 Our setting Local fitting to the swaption matrix using a version of the LIBOR Market Model Drivers are forward rates (about which we have good intuition) Swap rates are by-products Definition: The (j,k) entry of the swaption matrix contains the Black implied percentage volatility of the swaption expiring at the i-th maturity into a swap of length k. I condense the whole swaption matrix into 60 numbers: 0.5, 1, 3, 5, 10-year-expiry options INTO 1, 2, 3, 5, 7, 10-year-maturity swaps 4

5 The dynamics of the forward rates df i f i i j, f j dt i t, T i dz i # (eq0) df i f i i j, f j dt i t,t i dz i # (eq10) 5

6 The forward rate (y-axis) for a given probability density (x-axis) for various values of the displacement coefficient (5-year horizon, 4%-forward rate) a=0.01 a=0.5 a=2 a=5 a=

7 A Functional Form for the Instantaneous-Volatility Function Let us begin by assuming that the instantaneous volatility at time t of the forward rate expir T should be given by the expression: t T a b T t exp c T t d a b exp c d T t The quantity therefore represents the residual time to maturity of a particular forward rate the functional form chosen, the presence of a linear term together with a decaying exponential a the existence of a hump in the curve, and the asymptotic instantaneous volatility is assumed to asymptotically to a finite value, d. A few typical shapes are displayed in Fig. 3 below. 7

8 Possible shapes of the volatility function. Note how both excited (Series 5) and normal states (series 1 to 4) can be obtained. Instantaneous Volatility Curves 44.00% 39.00% Instantaneous Volatility 34.00% 29.00% 24.00% 19.00% Series1 Series2 Series3 Series4 Series % 9.00% 4.00% Time 8

9 Moving to two regimes The possibility of two regimes can be accounted for by positing: i n t, T i a t n b t n T t exp c t n T t d t n i x t, T i a t x b t x T t exp c t x T t d t x 9

10 Our immediate goal Explaining the evolution of swaption prices using a description for the instantaneous volatility the instantaneous correlation of forward rates We must establish a link between the two 10

11 The time dependence of the instantaneous volatility of the six forward rates in a 3 x 3 semi-annual swaption. The time on the x-axis indicates the time to expiry of the swaption. 24.0% 22.0% 20.0% 18.0% 16.0% 14.0% Fwd1 Fwd2 Fwd3 Fwd4 Fwd5 Fwd6 12.0% 10.0%

12 The empirical evidence main features 1. Long periods when the swaption matrix remains largely self-similar, with minor moves up or down (normal periods) 2. Rare, sudden onset of excited periods, where the shape of the swaption surface changes abruptly 3. Reversion to the normal state after a relatively short period of time 12

13 Swaption prices during the 24-May-99 to 29-June-99 period 18.00% 17.00% 16.00% 15.00% 14.00% 13.00% 12.00% 11.00% 10.00% 11-Jun x1 0.5x10 1x7 3x5 5x3 10x2 24-May-99 13

14 Swaption prices during the 31-Jul-2000 to 5-Sep % 14.00% 13.00% 12.00% 11.00% 10.00% 9.00% 18-Aug x1 0.5x10 1x7 3x5 5x3 10x2 31-Jul-00 14

15 Swaption prices during the 7-Sept-98 to 13-Oct-98 period (LTCM period) 20.00% 18.00% 16.00% 14.00% 12.00% 10.00% 25-Sep x1 0.5x10 1x7 3x5 5x3 10x2 07-Sep-98 15

16 Swaption prices during the 3-Jan-2001 to 8-Feb-2001 period (Fed cuts) 24.00% 22.00% 20.00% 18.00% 16.00% 14.00% 12.00% 10.00% 23-Jan x1 0.5x10 1x7 3x5 5x3 10x2 03-Jan-01 16

17 Swaption prices during the 24-Sep-2001 to 3-Oct-2001 period (post Sept 11) 35.00% 30.00% 25.00% 20.00% 15.00% 10.00% 12-Oct x1 0.5x10 1x7 3x5 5x3 10x2 24-Sep-01 17

18 Different shapes of non-excited swaption patterns (normalized to the 0.5 x 1 volatility for comparison) 130% 120% 110% 100% 90% 80% 70% 60% 50% 01-Mar Feb Jul Aug Nov-01 40% 30% 0.5x1 0.5x5 1x1 1x5 3x1 3x5 5x1 5x5 10x1 10x5 18

19 Same (non-excited) data as above organized by series instead of by expiries 130% 120% 110% 100% 90% 80% 70% 60% 01-Mar Feb Jul Aug Nov-01 50% 40% 30% 0.5x1 5x1 1x2 10x2 3x3 0.5x5 5x5 1x7 10x7 3x10 19

20 Different shapes of excited swaption patterns (normalized to the 0.5 x 1 volatility for comparison) 130% 120% 110% 100% 90% 80% 70% 60% 50% 26-Oct Oct Apr Sep-99 40% 30% 0.5x1 0.5x5 1x1 1x5 3x1 3x5 5x1 5x5 10x1 10x5 20

21 Same (excited) data organized by series instead of by expiries 130% 120% 110% 100% 90% 80% 70% 60% 26-Oct Oct Apr Sep-99 50% 40% 30% 0.5x1 5x1 1x2 10x2 3x3 0.5x5 5x5 1x7 10x7 3x10 21

22 A typical log-normal fit for a non-excited day (18-Feb-98). Implied volatility on the y-axis and swaption number on the x-axis: the first swaption is the 0.5 x 1 and the thirtieth the 10 x Model Real 22

23 The instantaneous volatility that produced the fit displayed above

24 The log-normal fit for an excited day (5-Nov-01) Model Real

25 The associated instantaneous volatility

26 The displaced-diffusion fit for 5-Nov-2001 (compare with the log-normal fit for the same day shown before) Model Real 26

27 The average absolute pricing errors for each swaption series in the log-normal and displaced-diffusion cases. 2.00% 1.80% 1.60% 1.40% 1.20% 1.00% 0.80% 0.60% 0.40% 0.20% 0.00% LOG DD 0.5x1 0.5x5 1x1 1x5 3x1 3x5 5x1 5x5 10x1 10x5 27

28 General features of the solutions The log-normal fits tend to find a humped volatility curve on normal days a decaying volatility curve on excited days the exception is the period before 09-11, where a decaying solution is found in the aftermath of excited periods a hybrid solution is found The implied instantaneous correlation is almost always found to be perfect how can we explain this? 28

29 The fit to the same data with constant (flat) instantaneous volatility and nonperfect instantaneous correlation Model Real 29

30 Typical fits: (6Feb 98) Model Real 30

31 Typical fits: 26-Jun Model Real 31

32 A hybrid instantaneous volatility curve in the aftermath (21 Dec 98) of the LTCM period

33 Checking the quality of the log-normal solution Desiderata: In a perfect world, the same volatility curve should describe the swaption matrix every day More realistically, we would hope that we can relate changes in the coefficients to changes in the swaption matrix changes in other observable financial quantities 33

34 Time series of the a coefficient for the log-normal fits a a 0 22-Oct Mar Jul Nov

35 Time series of the b coefficient for the log-normal fits b Oct Mar Jul Nov b 35

36 Time series of the c coefficient for the log-normal fits c c Oct Mar Jul Nov-01 36

37 Time series of the d coefficient for the log-normal fits d Oct Mar Jul Nov-01 d 37

38 Do we need a displaced-diffusion solution? Does the displaced-diffusion solution do appreciably better on average? Does the displaced-diffusion solution do appreciably better during particular periods? Is it possible to associate period when the displaced-diffusion solution performs better with regimes of other economic variables? 38

39 The figure of merit for the log-normal and the displaced-diffusion fits (arbitrary units) Log Displ Nov Mar Aug Dec-01 39

40 The ratio of the figures of merit in Fig Nov Jun Dec Jul Jan Aug Feb Sep Apr

41 A scatter-plot of the figures of merit for the log-normal and displaced-diffusion fits (arbitrary units, points linked by a timeline)

42 Ratio of the log-normal to displaced diffusion figures of merit against the average level of rates and the level of the 6-month LIBOR rate (all the curves have been normalized to their initial value for ease of comparison) Dec Jul Jan Aug Mar Sep Apr Oct

43 Can we do better with displaced diffusion? What do we mean by better: Better fit Better financial story (discriminating between excited and normal states) Changes in coefficients more closely linked to changes in the empirical swaption matrix 43

44 The displaced-diffusion fit for 5-Nov-2001 (compare with the log-normal fit for the same day shown before) Model Real 44

45 The log-normal fit for 10-Sep-2001, i.e. for the day before the events of September Model Real

46 The instantaneous volatility function for the log-normal fit for 10-Sep

47 The displaced-diffusion fit for 10-Sep-2001, ie for the day before the events of September Model Real 47

48 The instantaneous volatility function for the displaced-diffusion fit for 10-Sep

49 The location of the maxima for the log-normal and displaced-diffusion fits LOG DD /01/ /05/ /09/ /01/ /05/ /09/ /01/ /05/ /09/ /01/ /05/ /09/2001

50 Time series of the a coefficient for the displaced-diffusion fits (after normalization as discussed in the text) a /01/ /04/ /07/ /10/ /01/ /04/ /07/ /10/ /01/ /04/ /07/ /10/ /01/ /04/ /07/ /10/2001

51 Time series of the b coefficient for the displaced-diffusion fits (after normalization as discussed in the text) b /09/ /01/ /05/ /09/ /01/ /05/ /09/ /01/ /05/ /09/ /01/ /05/2001

52 Time series of the c coefficient for the displaced-diffusion fits (after normalization as discussed in the text) c /10/ /01/ /04/ /07/ /10/ /01/ /04/ /07/ /10/ /01/ /04/ /07/ /10/ /01/ /04/ /07/2001

53 Time series of the d coefficient for the displaced-diffusion fits (after normalization as discussed in the text) d /01/ /04/ /07/ /10/ /01/ /04/ /07/ /10/ /01/ /04/ /07/ /10/ /01/ /04/ /07/ /10/2001

54 Remaining problems The aftermath of excited periods is still problematic No time-homogeneous (deterministic or stochastic) volatility function will do No deterministically time-dependent volatility function will do We need at least to regimes (two-state Markov chain) 54

55 Using these results to evolve the swaption matrix in the real-world measure The usual approach: PCA Poor in practice, wrong in principle Violation of static arbitrage conditions 55

56 Correlation between the model and observed changes in implied volatilities, for the whole data set (data labelled Whole ) and for the last period (data labelled Last ) Log(Last) DD(Last) Log(Whole) DD(Whole) x1 0.5x5 1x1 1x5 3x1 3x5 5x1 5x5 10x1 10x5 56

57 Correlation between market and PCA-implied moves in implied volatility as a function of the number of principal components x1 1x1 3x1 5x1 10x

58 Correlation as a function of the number of PC 0.5x1 0.5x2 0.5x3 0.5x5 0.5x7 0.5x10 1x1 1x2 1x3 1x5 1x7 1x10 3x1 3x2 3x3 3x5 3x7 3x

59 5x1 5x2 5x3 5x5 5x7 5x10 10x1 10x2 10x3 10x5 10x7 10x10 Average

60 How to account for the observed features a two-state Markov chain 1. Choose a simple criterion to determine whether the swaption matrix is currently in the normal or excited state 2. The instantaneous volatility function for each forward rate can be described by either of these two functional forms: i n t,t i a t n b t n T t exp c t n T t d t n i x t,t i a tx b tx T t exp c tx T t d t x 60

61 The Two Basis States Normal Excited

62 Two-State Transition Matrix between a Normal and an Excited Regime nn nx xn xx 62

63 Lower Lower Lower a_lower b_lower c_lower d_lower a_upper b_upper c_upper d_upper Probability Up Jump Probability Down Jump Sqr(ScalingUpper) Sqr(ScalingNormal)

64 A Simple Fit t = 1 year 0.35 t = 2 year t = 4 years t = 5 years t = 12 years t = 14 years

65 Making it computationally feasible Conditional on a particular path having been realized, we simply are in a Black (displaced-diffusion) / LMM situation. To calculate caplet prices, average the analytical prices associated with a large number of volatility realizations. The analytical prices are obtained by using the (displaced-diffusion) Black formula with the appropriate conditional root-mean-square volatility. The appropriate conditional root-mean-squared volatility is obtained by integrating analytically over the segments of the two volatility states. Same approach can be used to calibrate to European swaption prices. 65

66 The variance integral for a particular realization of the Markov process 0 T u 2 du n i 1 0 t 1 u 2 t du 2 u 2 du... t 1 T t n 1 u 2 du 66

67 A particular path with state transitions Normal Excited Mixed

68 Global Disclaimer for Trading Areas The contents of this document are indicative and are subject to change without notice. This document is intended for your sole use on the basis that before entering into this, and/or any related transaction, you will ensure that you fully understand the potential risks and return of this, and/or any related transaction and determine it is appropriate for you given your objectives, experience, financial and operational resources, and other relevant circumstances. You should consult with such advisors as you deem necessary to assist you in making these determinations. The Royal Bank of Scotland plc ( RBS ) will not act as your advisor or owe any fiduciary duties to you in connection with this, and/or any related transaction and no reliance may be placed on RBS for advice or recommendations of any sort. RBS makes no representations or warranties with respect to the information and disclaims all liability for any use you or your advisors make of the contents of this document. RBS and its affiliates, connected companies, employees or clients may have an interest in financial instruments of the type described in this document and/or related financial instruments. Such interest may include dealing, trading, holding, acting as market-makers in such instruments and may include providing banking, credit and other financial services to any company or issuer of securities or financial instruments referred to herein. RBS is authorised and regulated in the UK by the Financial Services Authority, in Hong Kong by the Hong Kong Monetary Authority, in Singapore by the Monetary Authority of Singapore, in Japan by the Financial Services Agency of Japan, in Australia RBS (Australia) Pty Ltd ABN is authorised and regulated by the Australian Securities and Investments Commission and acts as agent for RBS ABN and in the US, by the New York State Banking Department and the Federal Reserve Board. The financial instruments described in this document are made in compliance with an applicable exemption from the registration requirements of the US Securities Act of N.B. Trading Areas are: Government Trading; Credit Trading; Proprietary Trading & Hedging; Capital Markets Sales; Currency Options/Equity Index; Derivatives and Futures; Interest Rate Derivatives; Sales and Marketing; Futures Broking; Margin Trading; Shipping; Strategy and Agency Treasury Services (for the purposes of these disclaimers only) 68