International Scholarly Research Notices

International Scholarly Research Notices / 2012 / Article

Research Article | Open Access

Volume 2012 |Article ID 673607 | https://doi.org/10.5402/2012/673607

Marco Di Francesco, "A General Gaussian Interest Rate Model Consistent with the Current Term Structure", International Scholarly Research Notices, vol. 2012, Article ID 673607, 16 pages, 2012. https://doi.org/10.5402/2012/673607

A General Gaussian Interest Rate Model Consistent with the Current Term Structure

Academic Editor: J. Abellan
Received25 Apr 2012
Accepted29 Jul 2012
Published05 Sep 2012

Abstract

We describe an extension of Gaussian interest rate models studied in literature. In our model, the instantaneous spot rate 𝑟 is the sum of several correlated stochastic processes plus a deterministic function. We assume that each of these processes has a Gaussian distribution with time-dependent volatility. The deterministic function is given by an exact fitting to observed term structure. We test the model through various numeric experiments about the goodness of fit to European swaptions prices quoted in the market. We also show some critical issues on calibration of the model to the market data after the credit crisis of 2007.

1. Introduction

A short-rate model for the term structure of interest rates is based on the assumption of a specific dynamics for the instantaneous spot-rate process 𝑟 (for the definition of 𝑟, we refer for instance to the monographs [1–3]). These models were the first approach to describe and explain the shape and the moves of the term structure of interest rates. Moreover, these models are very convenient since the dynamics of the instantaneous spot rate drives all the term structure, in the sense that both rates and prices of bonds are defined as an expectation of a functional of 𝑟. Indeed, the existence of a risk-neutral measure 𝑄 implies that the price at time 𝑡 of a contingent claim with payoff 𝐻𝑇 at time 𝑇>𝑡 is given by 𝐻𝑡=𝐸𝑡𝑒−∫𝑇𝑡𝑟(𝑠)𝑑𝑠𝐻𝑇,(1.1) where 𝐸𝑡 denotes the time 𝑡-conditional expectation under the measure 𝑄. In this paper, we describe all processes under the risk-neutral measure 𝑄 (for the definition and the existence of 𝑄, we refer to the monographs [1–4]).

A full description of the main short-rate models can be found in [1–3, 5]. Here, we only recall that these models belong to two general classes: the endogenous ones, in which the observed term structure is an output, and exogenous model, in which the observed term structure is an input. A basic strategy to transform an endogenous model to an exogenous one is the inclusion of time-dependent parameters to fit exactly the observed term structure. In fact, matching exactly the term structure is equivalent to solving a system with an infinite number of equations, and this is possible only after introducing an infinite number of parameters or, equivalently, a deterministic function of time.

In this paper, we describe a general exogenous model in which the instantaneous spot rate 𝑟 is the sum of several correlated Gaussian stochastic processes with time-dependent volatility plus a deterministic function which allows an exact fitting to the observed term structure. We derive an explicit formula for discount bond, and we recall a formula to approximate a European swaption price proposed by Schrager and Pelsser in [6], that can be used to calibrate the model to the market data. To do that, we also derive the dynamics of the 𝑟 under the 𝑇 forward measure, that is, the measure associated with the zero coupon bond maturing at time 𝑇 (see [1–4]).

In literature, a model in which the instantaneous spot rate is given by the sum of two correlated Gaussian processes plus a deterministic function is known as the G2++ model. This model with constant parameters is widely studied in [2]. Moreover, the G2++ model is equivalent to the well-known two-factor Hull-White model (see for instance [5, 7, 8]). We also note that one Gaussian process plus a deterministic function is equivalent to the well-known Hull-White model (see [5, 7, 9]). In this paper, we propose a common notation for the general case in which there are 𝑛 correlated Gaussian processes with time-dependent volatility. In order to use the same notation as in [2], we will refer to this model as the G𝑛++ model. The aim of this paper is not only to provide a common notation to unify multivariate Gaussian models with an exact fit to the observed term structure. With several numerical examples, we will show that the G1++ and G2++ models with constant parameters worked well before August 2007, the beginning of the so-called “credit-crunch’’ crisis. Then, August 2007 arrived and one of the several consequences of the liquidity and credit crisis was an explosion of the swaptions volatility, specially for swaptions with short tenor and maturity. Since August 2007, the swaptions volatility surface has become very unsmooth. At the money (ATM) swaptions volatility surface at 31/12/2006 and 31/12/2011 are, respectively, shown in Tables 1 and 2 (value in percentage).


T/Tenor 1 2 5
7 10

1 13.45 13.8 14.5 14.35 13.9
2 14.4 14.75 14.8 14.5 14.1
5 14.7 14.7 14.25 13.9
13.55
7 14.05 14
13.55 13.3 13.1
10 12.95 13 12.8
12.6 12.45
15 12
11.95 12 11.9 11.85
20 11.45 11.5 11.65 11.6
11.65


T/Tenor 1 2 5
7 10

1 59.6 49 43.9 40.5
39.2
2 59.4 46.7 38.8 36.5 35.2
5 34.5 31.4 29.3 28.5
28.1
7 29.4 27.9 26.1 25.9
26.6
10 24.7
24.3 24.3 25 26.5
15
25.8 26.4 28
2930.5
20
31.6 32.1
33.1
33.8
34

With the volatility as in Table 2, the G1++ and G2++ models with constant parameters do not work well, in the sense that it’s impossible to have a good fit to the market data. To improve the calibration to ATM swaptions volatility surface at 31/12/2011, we increase the number of factors and take a functional specification for the volatility of the processes. We test a calibration to swaptions prices proposed by Schrager and Pelsser in [6]. In our experiment, to optimize parameters, we use the differential evolution algorithm as implemented in R through the package DEoptim.

In order to make the reader familiar with the notations, we start, in Section 2, with one factor model and then, in Section 3, we describe the general 𝑛 factor model. In Section 4, we show our numerical results. Finally, in Section 5, we summarize our results and discuss future developments.

As a final remark, we note that the Gaussian distribution of the processes allows a good analytical tractability and the construction of efficient and fairly fast numerical procedures for pricing any type of payoff. On the other hand, the Gaussian distribution of the processes leads to the possibility of negative rates and so the model is hardly applicable to some concrete problems.

2. The G1++ Model

In this section, we assume that the dynamics of the instantaneous short-rate process under the risk-neutral measure 𝑄 is given by 𝑟(𝑡)=𝑥(𝑡)+𝑓(𝑡),(2.1) where the process {𝑥(𝑡)∶𝑡≥0} satisfies the stochastic differential equation 𝑥𝑑𝑥(𝑡)=âˆ’ğ‘Žğ‘¥(𝑡)𝑑𝑡+ğœŽ(𝑡)𝑑𝑊(𝑡),(0)=0,(2.2) where ğ‘Ž is a positive constant, ğœŽ(𝑡) is a deterministic function of time that is regular enough to ensure the existence and uniqueness of a solution, and 𝑊(𝑡) is a standard Brownian motion. The function 𝑓 is deterministic and is given by an exact fitting to the term structure of discount factor observed in the market. Using the same notation as in [2], we will refer to this model as the G1++ model.

Integrating equation (2.2), we have, for each 𝑠≤𝑡, 𝑟(𝑡)=𝑥(𝑠)ğ‘’âˆ’ğ‘Ž(𝑡−𝑠)+î€œğ‘¡ğ‘ ğœŽ(𝑠)ğ‘’âˆ’ğ‘Ž(𝑡−𝑢)𝑑𝑊(𝑢)+𝑓(𝑡).(2.3)

Therefore, 𝑟(𝑡) conditional to 𝐹𝑠, the ğœŽ-field representing the information available in the market up to time 𝑠 (see for instance [1, 2, 4]) is normally distributed with mean and variance given by 𝐸𝑟(𝑡)∣𝐹𝑠=𝑥(𝑠)ğ‘’âˆ’ğ‘Ž(𝑡−𝑠)+𝑓(𝑡),VAR𝑟(𝑡)∣𝐹𝑠=î€œğ‘¡ğ‘ ğœŽ2(𝑠)𝑒−2ğ‘Ž(𝑡−𝑢)𝑑𝑢,(2.4) where 𝐸 and VAR denote the mean and the variance under the measure 𝑄, respectively. In the case ğœŽ(𝑡)=ğœŽ is a positive constant function, we have VAR𝑟(𝑡)∣𝐹𝑠=ğœŽ2𝑡𝑠𝑒−2ğ‘Ž(𝑡−𝑢)ğœŽğ‘‘ğ‘¢=22ğ‘Ž1−𝑒−2ğ‘Ž(𝑡−𝑠).(2.5) We denote by 𝑃(𝑡,𝑇) the price at time 𝑡 of a zero coupon bond maturing at time 𝑇 with unit nominal value. The existence of the risk-neutral measure 𝑄 implies that 𝑒𝑃(𝑡,𝑇)=𝐸−∫𝑇𝑡𝑟(𝑠)𝑑𝑠∣𝐹𝑡.(2.6) The model fits the current observed term structure of the discount factor if, for each maturity 𝑇, the discount factor 𝑃(0,𝑇) is equal to the one observed in the market 𝑃𝑀(0,𝑇). To calculate 𝑃(0,𝑇), we have to integrate the process 𝑟(𝑡) over the interval [0,𝑇]. Notice that, since the process 𝑥(𝑡) is normally distributed conditional on 𝐹0, then also ∫𝑇0𝑥(𝑡)𝑑𝑡 is itself normally distributed. Indeed, by Fubini’s Theorem for stochastic integral (see [3]), we have 𝑇0𝑥(𝑡)𝑑𝑡=𝑇0𝑡0ğœŽ(𝑢)ğ‘’âˆ’ğ‘Ž(𝑡−𝑢)=𝑑𝑊(𝑢)𝑑𝑡𝑇0ğœŽ(𝑢)ğ‘’ğ‘Žğ‘¢î€œğ‘‡ğ‘¢ğ‘’âˆ’ğ‘Žğ‘¡=𝑑𝑡𝑑𝑊(𝑢)𝑇0ğœŽ(𝑢)ğ‘Žî€·1âˆ’ğ‘’âˆ’ğ‘Ž(𝑇−𝑢)𝑑𝑊(𝑢).(2.7) So, the integral ∫𝑇0𝑥(𝑡)𝑑𝑡 is normally distributed with mean zero and variance VAR𝑇0𝑥(𝑡)𝑑𝑡∣𝐹0=𝑇0ğœŽ2(𝑢)ğ‘Ž21âˆ’ğ‘’âˆ’ğ‘Ž(𝑇−𝑢)2𝑑𝑢=𝑉(0,𝑇).(2.8) In the case ğœŽ(𝑡)=ğœŽ is a positive constant function, we have ğœŽğ‘‰(0,𝑇)=2ğ‘Ž2𝑇−21âˆ’ğ‘’âˆ’ğ‘Žğ‘‡ğ‘Ž+1−𝑒−2ğ‘Žğ‘‡î‚¶2ğ‘Ž.(2.9) Finally, recalling that if 𝑍 is normal random variable with mean 𝜇𝑍 and variance ğœŽ2𝑍, then 𝐸[𝑒𝑍]=𝑒𝜇𝑍+(1/2)ğœŽ2𝑍, we have 𝑒𝑃(0,𝑇)=𝐸−∫𝑇0𝑟(𝑡)𝑑𝑡𝑒=𝐸−∫𝑇0(𝑥(𝑡)+𝑓(𝑡))𝑑𝑡=𝑒−∫𝑇0𝑓(𝑡)𝑑𝑡𝐸𝑒−∫𝑇0𝑥(𝑡)𝑑𝑡=𝑒−∫𝑇0𝑓(𝑡)𝑑𝑡𝑒+(1/2)𝑉(0,𝑇).(2.10) Clearly, the model gives an exact fitting to the market term structure of the discount factors if and only if 𝑃𝑀(0,𝑇)=𝑃(0,𝑇)(2.11) for every 𝑇>0. By (2.10), we have 𝑃𝑀(0,𝑡)=𝑒−∫𝑇0𝑓(𝑡)𝑑𝑡𝑒+(1/2)𝑉(0,𝑇).(2.12) So, the model gives an exact fitting to the market term structure of the discount factors if and only if for every 𝑇>0𝑒−∫𝑇0𝑓(𝑡)𝑑𝑡=𝑃𝑀(0,𝑇)𝑒−(1/2)𝑉(0,𝑇).(2.13) To have an explicit expression of 𝑓, denoting by 𝑓𝑀(0,𝑇) the market instantaneous forward rate, that is, 𝑃𝑀(0,𝑇)=𝑒−∫𝑇0𝑓𝑀(0,𝑡)𝑑𝑡.(2.14) We can write (2.12) 𝑒−∫𝑇0𝑓𝑀(0,𝑡)𝑑𝑡=𝑒−∫𝑇0𝑓(𝑡)𝑑𝑡𝑒+(1/2)𝑉(0,𝑇).(2.15) Then the exponents on both sides of (2.15) agree, and by differentiation we get 𝑓(𝑇)=𝑓𝑀(0,𝑇)+𝑇0ğœŽ2(𝑢)ğ‘Žî€·1âˆ’ğ‘’âˆ’ğ‘Ž(𝑇−𝑢)î€¸ğ‘’âˆ’ğ‘Ž(𝑇−𝑢)𝑑𝑢(2.16) and so, in the case ğœŽ(𝑡)=ğœŽ is a positive constant function, the model perfect fits the market term structure of the discount factor if and only if for every 𝑇>0𝑓(𝑇)=ğ‘“ğ‘€ğœŽ(0,𝑇)+22ğ‘Ž21âˆ’ğ‘’âˆ’ğ‘Žğ‘‡î€¸2.(2.17)

However, to calculate 𝑃(𝑡,𝑇) for every 0<𝑡<𝑇, we point out that we donot need to compute the whole 𝑓 curve. This is very important in order to implement the model. In fact, arguing as in (2.7), we can prove that 𝑇𝑡𝑥(𝑠)𝑑𝑠=𝑥(𝑡)1âˆ’ğ‘’âˆ’ğ‘Ž(𝑇−𝑡)ğ‘Ž+î€œğ‘‡ğ‘¡ğœŽ(𝑢)ğ‘Žî€·1âˆ’ğ‘’âˆ’ğ‘Ž(𝑇−𝑢)𝑑𝑊(𝑢).(2.18) In particular, the integral ∫𝑇𝑡𝑥(𝑠)𝑑𝑠 conditional on 𝐹𝑡 is normally distributed with mean and variance given by 𝐸𝑇𝑡𝑥(𝑠)𝑑𝑠∣𝐹𝑡=𝑥(𝑡)1âˆ’ğ‘’âˆ’ğ‘Ž(𝑇−𝑡)ğ‘Žî‚¸î€œVAR𝑇𝑡𝑥(𝑠)𝑑𝑠∣𝐹𝑡=î€œğ‘‡ğ‘¡ğœŽ2(𝑢)ğ‘Ž21âˆ’ğ‘’âˆ’ğ‘Ž(𝑇−𝑢)2𝑑𝑢=𝑉(𝑡,𝑇)(2.19) and, in the case ğœŽ(𝑡)=ğœŽ is a positive constant function ğœŽğ‘‰(𝑡,𝑇)=2ğ‘Ž2𝑇−𝑡−21âˆ’ğ‘’âˆ’ğ‘Ž(𝑇−𝑡)ğ‘Ž+1−𝑒−2ğ‘Ž(𝑇−𝑡)2ğ‘Ž.(2.20) So, we have 𝑒𝑃(𝑡,𝑇)=𝐸−∫𝑇𝑡𝑟(𝑠)𝑑𝑠𝑒=𝐸−∫𝑇𝑡𝑥(𝑠)+𝑓(𝑠)𝑑𝑠=𝑒−∫𝑇𝑡𝑓(𝑠)𝑑𝑠𝐸𝑒−∫𝑇𝑡𝑥(𝑠)𝑑𝑠=𝑒−∫𝑇0𝑓(𝑠)𝑑𝑠𝑒∫𝑡0𝑓(𝑠)𝑑𝑠𝑒−((1âˆ’ğ‘’âˆ’ğ‘Ž(𝑇−𝑡))/ğ‘Ž)𝑥(𝑡)𝑒(1/2)𝑉(𝑡,𝑇)=𝑃(by(2.13))𝑀(0,𝑇)𝑃𝑀(𝑒0,𝑡)(1/2)[𝑉(𝑡,𝑇)−𝑉(0,𝑇)+𝑉(0,𝑡)]𝑒−((1âˆ’ğ‘’âˆ’ğ‘Ž(𝑇−𝑡))/ğ‘Ž)𝑥(𝑡)=𝐴(𝑡,𝑇)𝑒−𝐵(𝑡,𝑇)𝑥(𝑡),(2.21) where, 𝑃𝐴(𝑡,𝑇)=𝑀(0,𝑇)𝑃𝑀𝑒(0,𝑡)(1/2)[𝑉(𝑡,𝑇)−𝑉(0,𝑇)+𝑉(0,𝑡)],𝐵(𝑡,𝑇)=1âˆ’ğ‘’âˆ’ğ‘Ž(𝑇−𝑡)ğ‘Ž.(2.22) Expression (2.21) is very important in simulation. We note that to generate 𝑃(𝑡,𝑇), we need only the market discount curve and the process 𝑥(𝑡). From 𝑃(𝑡,𝑇), we could derive all the other rates (forward rates, swap rates, etc.).

In order to derive an explicit formula of the price of an European swaption, we need to change the probability measure as indicated by Jamshidian in [10] and more generally by German et al. in [11]. We denote by 𝑄𝑇 the 𝑇-forward measure, given by choosing a zero coupon bond with maturity 𝑇 as numerarie. By the Girsanov’s Theorem, the dynamics of the process 𝑥(𝑡) under the measure 𝑄𝑇 is given by 𝑑𝑥(𝑡)=âˆ’ğ‘Žğ‘¥(𝑡)ğ‘‘ğ‘¡âˆ’ğœŽ2(𝑡)𝐵(𝑡,𝑇)𝑑𝑡+ğœŽğ‘‘ğ‘Šğ‘‡(𝑡),(2.23) where 𝑊𝑇 is a standard Brownian motion under 𝑄𝑇 defined by 𝑑𝑊𝑇(𝑡)=𝑑𝑊(𝑡)+ğœŽ(𝑡)𝐵(𝑡,𝑇)𝑑𝑡. Then, under the measure 𝑄𝑇, the distribution of the process 𝑥(𝑡) is still Gaussian; in particular, for every 𝑠≤𝑡≤𝑇, the mean and the variance of the process are given by 𝐸𝑇𝑥(𝑡)∣𝐹𝑠=𝑥(𝑠)ğ‘’âˆ’ğ‘Ž(𝑡−𝑠)+𝑀𝑇(𝑠,𝑡),VAR𝑇𝑥(𝑡)∣𝐹𝑠=î€œğ‘¡ğ‘ ğœŽ2(𝑢)𝑒−2ğ‘Ž(𝑡−𝑢)𝑑𝑢=ğœŽ(𝑠,𝑡),(2.24) where 𝐸𝑇 and VAR𝑇 denote, respectively, the mean and the variance under 𝑄𝑇, and 𝑀𝑇(𝑡,𝑠)=ğ‘¡ğ‘ ğœŽ2(𝑢)ğ‘’âˆ’ğ‘Ž(𝑡−𝑢)1âˆ’ğ‘’âˆ’ğ‘Ž(𝑇−𝑢)ğ‘Žğ‘‘ğ‘¢.(2.25) In the case ğœŽ(𝑡)=ğœŽ is a positive constant function, we have ğ‘€ğ‘‡ğœŽ(𝑡,𝑠)=−2ğ‘Ž21âˆ’ğ‘’âˆ’ğ‘Ž(𝑡−𝑠)+ğœŽ22ğ‘Ž2î€·ğ‘’âˆ’ğ‘Ž(𝑇−𝑡)âˆ’ğ‘’âˆ’ğ‘Ž(𝑇+𝑡−2𝑠).(2.26) Now, we consider an European call swaption with maturity 𝑇, strike 𝐾, and nominal value 𝑁, which gives the holder the right to enter at time 𝑇=𝑡0 into a swap with payment dates {𝑡1,…,𝑡𝑘}, 𝑡0<𝑡1,…,𝑡𝑘, where he pays the fixed rate 𝐾 and receives the Libor rate. Jamshidian in [10] showed how a European swaption can be decomposed into a portfolio of options on zero coupon bonds, deriving an exact closed formula for any European swaption. In particular, the price 𝐸𝑆(0,𝑇,𝐾,𝑁) at time 0 of the above swaption is given by 𝐸𝑆0,𝑇,𝑡𝑘,𝐾,𝑁=𝑁𝑃(0,𝑇)ğ¸ğ‘‡âŽ¡âŽ¢âŽ¢âŽ£îƒ©1−𝑘𝑖=1𝑐𝑖𝑃𝑇,𝑡𝑖+∣𝐹0⎤⎥⎥⎦,(2.27) where 𝑐𝑖=𝐾𝜏𝑛 and 𝑐𝑘=1+𝐾𝜏𝑛, where 𝜏𝑖 denotes the year fraction from 𝑡𝑖−1 to 𝑡𝑖, 𝑖=1,…,𝑘. Then, by (2.21) we have 𝐸𝑆0,𝑇,𝑡𝑘,𝐾,𝑁=𝑁𝑃(0,𝑇)𝑅1−𝑘𝑖=1𝑐𝑖𝐴𝑇,𝑡𝑖𝑒−𝐵(𝑇,𝑡𝑖)𝑥+𝑔(𝑥)𝑑𝑥,(2.28) where 𝑔(𝑥) denotes the density function of 𝑥(𝑇) conditioned on 𝐹0 under 𝑄𝑇, that is, 1𝑔(𝑥)=√2ğœ‹ğœŽ2𝑒(0,𝑇)−(𝑥−𝑀(0,𝑇))2/2ğœŽ2(0,𝑇).(2.29) We also point out that in the case ğœŽ(𝑡)=ğœŽ is a positive constant function, we have ğœŽ2(0,𝑇)=(ğœŽ2/2ğ‘Ž)(1−𝑒−2ğ‘Žğ‘‡). Equation (2.28) may be used in order to calibrate the model to the market prices of quoted swaptions. An analytic formula alternative to (2.28) is given by 𝐸𝑆0,𝑇,𝑡𝑘=Φ−,𝐾,𝑁𝑁𝑃(0,𝑇)𝑦1−𝑘𝑖=1𝑐𝑖𝐴𝑇,𝑡𝑖𝑒−𝐵(𝑇,𝑡𝑖)𝑀(0,𝑇)𝑒𝐵2(𝑇,𝑡𝑖)(ğœŽ2(0,𝑇)/2)Φ−𝑦2,(𝑖)(2.30) where Φ is the standard normal cumulative function, 𝑦1 and 𝑦2(𝑖) are given by 𝑦1=𝑦−𝑀(0,𝑇),ğœŽ(0,𝑇)𝑦2(𝑖)=𝑦1+𝐵𝑇,ğ‘¡ğ‘–î€¸ğœŽ(0,𝑇),(2.31) where 𝑦 is the unique solution of 𝑘𝑖=1𝑐𝑖𝐴𝑇,𝑡𝑖𝑒−𝐵(𝑇,𝑡𝑖)𝑦=1.(2.32)

To conclude this section, we notice that the G1++ model is perfectly analogous to the one proposed by Hull and White in [7] and known as the Hull-White model.

3. The G𝑛++ model

We now extend the previous one factor model to the multidimensional case. In this section, we assume that the dynamics of the instantaneous short-rate process under the risk-neutral measure 𝑄 is given by 𝑟(𝑡)=𝑛𝑖=1𝑥𝑖(𝑡)+𝑓(𝑡),(3.1) where, for every 𝑖=1,…,𝑛, the process {𝑥𝑖(𝑡)∶𝑡≥0} satisfies the stochastic differential equation 𝑑𝑥𝑖(𝑡)=âˆ’ğ‘Žğ‘–ğ‘¥ğ‘–(𝑡)𝑑𝑡+ğœŽğ‘–(𝑡)𝑑𝑊𝑖𝑥(𝑡),𝑖(0)=0,(3.2) where ğ‘Žğ‘– is a positive constant, ğœŽğ‘–(𝑡) is a deterministic function of time that is regular enough to ensure the existence and uniqueness of a solution, and 𝑊𝑖(𝑡) is a standard Brownian motion. We denote by 𝜌𝑖𝑗 the instantaneous correlation between 𝑊𝑖(𝑡) and 𝑊𝑗(𝑡), for every 𝑡>0, 𝑖,𝑗=1,…𝑛,𝑖≠𝑗; we also set 𝜌𝑖𝑖=1, for every 𝑖=1,…,𝑛. We naturally assume that −1≤𝜌𝑖𝑗≤1, and the matrix (𝜌𝑖𝑗)𝑖,𝑗=1…,𝑛 is symmetric and positive definite. The function 𝑓 is deterministic and is given by an exact fitting to the term structure of discount factors observed in the market. Using the same notation as in [2], we will refer to this model as the G𝑛++ model.

Integrating equation (3.2), we have, for each 𝑠≤𝑡, 𝑟(𝑡)=𝑛𝑖=1𝑥𝑖(𝑠)ğ‘’âˆ’ğ‘Žğ‘–(𝑡−𝑠)+î€œğ‘¡ğ‘ ğœŽğ‘–(𝑢)ğ‘’âˆ’ğ‘Žğ‘–(𝑡−𝑢)𝑑𝑊𝑖(𝑢)+𝑓(𝑡).(3.3) Therefore, 𝑟(𝑡) conditional to 𝐹𝑠 is normally distributed with mean and variance given by 𝐸𝑟(𝑡)∣𝐹𝑠=𝑛𝑖=1𝑥𝑖(𝑠)ğ‘’âˆ’ğ‘Žğ‘–(𝑡−𝑠)+𝑓(𝑡),VAR𝑟(𝑡)∣𝐹𝑠=𝑛𝑖,𝑗=1î€œğ‘¡ğ‘ ğœŽğ‘–(𝑢)ğœŽğ‘—(𝑢)ğœŒğ‘–ğ‘—ğ‘’âˆ’ğ‘Žğ‘–(𝑡−𝑢)ğ‘’âˆ’ğ‘Žğ‘—(𝑡−𝑢)𝑑𝑢,(3.4) where 𝐸 and VAR denote the mean and the variance under the measure 𝑄, respectively. In the case that for every 𝑖=1,…,𝑛,â€‰â€‰ğœŽğ‘–(𝑡)=ğœŽğ‘– is a positive constant function, we have VAR𝑟(𝑡)∣𝐹𝑠=ğœŽğ‘–ğœŽğ‘—ğ‘Žğ‘–+ğ‘Žğ‘—ğœŒğ‘–ğ‘—î€·1−𝑒−(ğ‘Žğ‘–+ğ‘Žğ‘—)(𝑡−𝑠).(3.5) The model gives an exact fitting to the currently observed term structure of the discount factors if, for each maturity 𝑇, the discount factor 𝑃(0,𝑇) is equal to the one observed in the market 𝑃𝑀(0,𝑇). To calculate 𝑃(0,𝑇), we have to integrate the process 𝑟(𝑡) over the interval [0,𝑇]. Proceeding as in (2.7), we obtain that the integral ∫𝑇0∑𝑛𝑖=1𝑥𝑖(𝑡)𝑑𝑡, conditional to 𝐹0, is normally distributed with mean zero and variance given by VAR𝑇0𝑛𝑖=1𝑥𝑖(𝑡)𝑑𝑡∣𝐹0=𝑛𝑖,𝑗=1𝜌𝑖𝑗𝑇0ğœŽğ‘–(𝑢)ğœŽğ‘—(𝑢)1âˆ’ğ‘’âˆ’ğ‘Žğ‘–(𝑇−𝑢)ğ‘Žğ‘–1âˆ’ğ‘’âˆ’ğ‘Žğ‘—(𝑇−𝑢)ğ‘Žğ‘—ğ‘‘ğ‘¢=𝑉(0,𝑇).(3.6) In the case that, for every 𝑖=1,…,𝑛, ğœŽğ‘–(𝑡)=ğœŽğ‘– is a positive constant function, we have 𝑉(0,𝑇)=𝑛𝑖,𝑗=1ğœŒğ‘–ğ‘—ğœŽğ‘–ğœŽğ‘—ğ‘Žğ‘–ğ‘Žğ‘—î‚µğ‘‡âˆ’1âˆ’ğ‘’âˆ’ğ‘Žğ‘–ğ‘‡ğ‘Žğ‘–âˆ’1âˆ’ğ‘’âˆ’ğ‘Žğ‘—ğ‘‡ğ‘Žğ‘—+1−𝑒−(ğ‘Žğ‘–+ğ‘Žğ‘—)ğ‘‡ğ‘Žğ‘–+ğ‘Žğ‘—î‚¶.(3.7) So, we have 𝑒𝑃(0,𝑇)=𝐸−∫𝑇0𝑟(𝑡)𝑑𝑡𝑒=𝐸−∫𝑇0∑𝑛𝑖=1𝑥(𝑡)+𝑓(𝑡)𝑑𝑡=𝑒−∫𝑇0𝑓(𝑡)𝑑𝑡𝐸𝑒−∫𝑇0∑𝑛𝑖=1𝑥(𝑡)𝑑𝑡=𝑒−∫𝑇0𝑓(𝑡)𝑑𝑡𝑒+(1/2)𝑉(0,𝑇).(3.8) Clearly, the model gives an exact fitting to the market term structure of the discount factors if and only if 𝑃𝑀(0,𝑇)=𝑃(0,𝑇), for every 𝑇>0, that is, 𝑒−∫𝑇0𝑓(𝑡)𝑑𝑡=𝑃𝑀(0,𝑇)𝑒−(1/2)𝑉(0,𝑇).(3.9) To have an explicit solution of 𝑓, denoting by 𝑓𝑀(0,𝑇) the market instantaneous forward rate, we can write (3.9) 𝑒−∫𝑇0𝑓𝑀(𝑡)𝑑𝑡=𝑒−∫𝑇0𝑓(𝑡)𝑑𝑡𝑒(1/2)𝑉(0,𝑇).(3.10) Then, the exponents on both sides of (3.10) agree, and by differentiation we get 𝑓(𝑇)=𝑓𝑀(0,𝑇)+𝑛𝑖,𝑗=1𝜌𝑖𝑗𝑇0ğœŽğ‘–(𝑢)ğœŽğ‘—î‚µî€·(𝑢)1âˆ’ğ‘’âˆ’ğ‘Žğ‘–(𝑇−𝑢)î€¸ğ‘’âˆ’ğ‘Žğ‘–(𝑇−𝑢)1âˆ’ğ‘’âˆ’ğ‘Žğ‘—(𝑇−𝑢)ğ‘Žğ‘—+1âˆ’ğ‘’âˆ’ğ‘Žğ‘—(𝑇−𝑢)î€¸ğ‘’âˆ’ğ‘Žğ‘—(𝑇−𝑢)1âˆ’ğ‘’âˆ’ğ‘Žğ‘–(𝑇−𝑢)ğ‘Žğ‘–î‚¶ğ‘‘ğ‘¢.(3.11) In the case, for every 𝑖=1,…,𝑛, ğœŽğ‘–(𝑡)=ğœŽğ‘– is a positive constant function, we have 𝑓(𝑇)=𝑓𝑀1(0,𝑇)+2𝑛𝑖,𝑗=1ğœŒğ‘–ğ‘—ğœŽğ‘–ğœŽğ‘—ğ‘Žğ‘–ğ‘Žğ‘—î€·1âˆ’ğ‘’âˆ’ğ‘Žğ‘–ğ‘‡î€¸î€·1âˆ’ğ‘’âˆ’ğ‘Žğ‘—ğ‘‡î€¸.(3.12) However, as in the G1++, to calculate 𝑃(𝑡,𝑇) for every 0<𝑡<𝑇, we point out that we do not need to compute the whole 𝑓 curve. This is very important in order to implement the model. In fact, we can prove that the integral ∫𝑇𝑡∑𝑛𝑖=1𝑥𝑖(𝑠)𝑑𝑠, conditional to 𝐹𝑡, is normally distributed with mean and variance given by 𝐸𝑇𝑡𝑥(𝑠)𝑑𝑠∣𝐹𝑡=𝑛𝑖=1𝑥𝑖(𝑡)1âˆ’ğ‘’âˆ’ğ‘Žğ‘–(𝑇−𝑡)ğ‘Žğ‘–î‚¸î€œVAR𝑇𝑡𝑥(𝑠)𝑑𝑠𝐹𝑡=𝑛𝑖,𝑗=1ğœŒğ‘–ğ‘—î€œğ‘‡ğ‘¡ğœŽğ‘–(𝑢)ğœŽğ‘—(𝑢)𝜌𝑖𝑗1âˆ’ğ‘’âˆ’ğ‘Žğ‘–(𝑇−𝑢)ğ‘Žğ‘–1âˆ’ğ‘’âˆ’ğ‘Žğ‘—(𝑇−𝑢)ğ‘Žğ‘—ğ‘‘ğ‘¢=𝑉(𝑡,𝑇).(3.13) In the case, for every 𝑖=1,…,𝑛, ğœŽğ‘–(𝑡)=ğœŽğ‘– is a positive constant function, we have 𝑉(𝑡,𝑇)=𝑛𝑖,𝑗=1ğœŒğ‘–ğ‘—ğœŽğ‘–ğœŽğ‘—ğ‘Žğ‘–ğ‘Žğ‘—î‚µğ‘‡âˆ’ğ‘¡âˆ’1âˆ’ğ‘’âˆ’ğ‘Žğ‘–(𝑇−𝑡)ğ‘Žğ‘–âˆ’1âˆ’ğ‘’âˆ’ğ‘Žğ‘—(𝑇−𝑡)ğ‘Žğ‘—+1−𝑒−(ğ‘Žğ‘–+ğ‘Žğ‘—)(𝑇−𝑡)ğ‘Žğ‘–+ğ‘Žğ‘—î‚¶.(3.14) So, we have 𝑒𝑃(𝑡,𝑇)=𝐸−∫𝑇𝑡𝑟(𝑠)𝑑𝑠𝑒=𝐸−∫𝑇𝑡∑𝑛𝑖=1(𝑥𝑖(𝑠)+𝑓(𝑠))𝑑𝑠=𝑒−∫𝑇𝑡𝑓(𝑠)𝑑𝑠𝐸𝑒−∫𝑇𝑡∑𝑛𝑖=1𝑥𝑖(𝑠)𝑑𝑠=𝑒−∫𝑇0𝑓(𝑠)𝑑𝑠𝑒∫𝑡0𝑓(𝑠)𝑑𝑠𝑒−∑𝑛𝑖=1((1âˆ’ğ‘’ğ‘–âˆ’ğ‘Ž(𝑇−𝑡))/ğ‘Žğ‘–)𝑥𝑖(𝑡)𝑒(1/2)𝑉(𝑡,𝑇)=𝑃(by(3.9))𝑀(0,𝑇)𝑃𝑀(𝑒0,𝑡)(1/2)[𝑉(𝑡,𝑇)−𝑉(0,𝑇)+𝑉(0,𝑡)]𝑒−∑𝑛𝑖=1((1âˆ’ğ‘’ğ‘–âˆ’ğ‘Ž(𝑇−𝑡))/ğ‘Žğ‘–)𝑥𝑖(𝑡)=𝐴(𝑡,𝑇)𝑒−∑𝑛𝑖=1𝐵𝑖(𝑡,𝑇)𝑥𝑖(𝑡),(3.15) where 𝑃𝐴(𝑡,𝑇)=𝑀(0,𝑇)𝑃𝑀𝑒(0,𝑡)(1/2)[𝑉(𝑡,𝑇)−𝑉(0,𝑇)+𝑉(0,𝑡)],𝐵𝑖(𝑡,𝑇)=1âˆ’ğ‘’âˆ’ğ‘Žğ‘–(𝑇−𝑡)ğ‘Žğ‘–.(3.16) Expression (3.15) is particularly important in simulation. We note that to generate 𝑃(𝑡,𝑇), we need only the market discount curve and the processes 𝑥𝑖(𝑡),𝑖=1,…,𝑛. From 𝑃(𝑡,𝑇), we could derive all the other rates (forward rates, swap rates, etc.).

In order to derive an explicit formula of the price of an European swaption, we need to change the probability measure as indicated by Jamshidian [10] and more generally by German et al. in [11]. We denote by 𝑄𝑇 the 𝑇-forward measure, given by choosing a zero coupon bond with maturity 𝑇 as numerarie. By the Girsanov’s Theorem, for each 𝑖=1,…,𝑛, the dynamics of the process 𝑥𝑖(𝑡) under the measure 𝑄𝑇, is given by 𝑑𝑥𝑖(𝑡)=âˆ’ğ‘Žğ‘–ğ‘¥ğ‘–(𝑡)ğ‘‘ğ‘¡âˆ’ğœŽğ‘–îƒ©(𝑡)𝑛𝑗=1ğœŽğ‘—(𝑡)𝜌𝑖𝑗𝐵𝑖(𝑡,𝑇)𝑑𝑡+ğœŽğ‘–(𝑡)𝑑𝑊𝑇𝑖(𝑡),(3.17) where 𝑊𝑇𝑖 is a standard Brownian motion under 𝑄𝑇 with 𝑑𝑊𝑇𝑖𝑑𝑊𝑡𝑗=𝜌𝑖𝑗. Moreover, the explicit solution of (3.17) is, for 𝑠≤𝑡≤𝑇𝑥𝑖(𝑡)=𝑥𝑖(𝑠)ğ‘’âˆ’ğ‘Žğ‘–(𝑡−𝑠)−𝑀𝑇𝑖(𝑠,𝑡)+ğœŽğ‘–î€œğ‘¡ğ‘ ğ‘’âˆ’ğ‘Žğ‘–(𝑡−𝑢)𝑑𝑊𝑇𝑖(𝑢),(3.18) where 𝑀𝑇𝑖(𝑠,𝑡)=ğ‘¡ğ‘ ğœŽğ‘–(𝑢)𝑛𝑗=1ğœŽğ‘—(𝑢)ğ‘Žğ‘—ğœŒğ‘–ğ‘—î€·1âˆ’ğ‘’âˆ’ğ‘Žğ‘–(𝑇−𝑢)î€¸ğ‘’âˆ’ğ‘Žğ‘–(𝑡−𝑢)𝑑𝑢,(3.19) and, in the case that, for every 𝑖=1,…,𝑛, ğœŽğ‘–(𝑡)=ğœŽğ‘– is a positive constant function 𝑀𝑇𝑖(𝑠,𝑡)=ğœŽğ‘–ğ‘›î“ğ‘—=1ğœŽğ‘—ğ‘Žğ‘—ğœŒğ‘–ğ‘—î‚µ1âˆ’ğ‘’âˆ’ğ‘Žğ‘–(𝑡−𝑠)ğ‘Žğ‘–âˆ’ğ‘’âˆ’ğ‘Žğ‘—(𝑇−𝑡)âˆ’ğ‘’âˆ’ğ‘Žğ‘—ğ‘‡âˆ’ğ‘Žğ‘–ğ‘¡+(ğ‘Žğ‘–+ğ‘Žğ‘—)ğ‘ ğ‘Žğ‘–+ğ‘Žğ‘—î‚¶.(3.20) So, under 𝑄𝑇, the distribution of 𝑟(𝑡) conditional to 𝐹𝑠 is normal with mean and variance given by 𝐸𝑇𝑟(𝑡)∣𝐹𝑠=𝑛𝑖=1𝑥𝑖(𝑠)ğ‘’âˆ’ğ‘Žğ‘–(𝑡−𝑠))−𝑀𝑇𝑖(𝑠,𝑡)+𝑓(𝑡)VAR𝑇𝑟(𝑡)∣𝐹𝑠=𝑛𝑖,𝑗=1î€œğ‘¡ğ‘ ğœŽğ‘–(𝑢)ğœŒğ‘–ğ‘—ğœŽğ‘—(𝑢)ğ‘’âˆ’ğ‘Žğ‘–(𝑡−𝑢)ğ‘’âˆ’ğ‘Žğ‘—(𝑡−𝑢)𝑑𝑢.(3.21) The price 𝐸𝑆(0,𝑇,𝑡𝑘,𝐾,𝑁) at time 0 of an European call swaption with maturity 𝑇, strike 𝐾 and nominal value 𝑁 is given by 𝐸𝑆0,𝑇,𝑡𝑘,𝐾,𝑁=𝑁𝑃(0,𝑇)ğ¸ğ‘‡âŽ¡âŽ¢âŽ¢âŽ£îƒ©1−𝑘𝑗=1𝑐𝑗𝑃𝑇,𝑡𝑗+∣𝐹0⎤⎥⎥⎦,(3.22) where 𝑐𝑗=𝐾𝜏𝑛 and 𝑐𝑘=1+𝐾𝜏𝑛, where 𝜏𝑗 denotes the year fraction from 𝑡𝑗−1 to 𝑡𝑗, 𝑗=1,…,𝑘. Then, by (3.15) we have 𝐸𝑆0,𝑇,𝑡𝑘=,𝐾,𝑁𝑁𝑃(0,𝑇)𝑅𝑛1−𝑘𝑗=1𝑐𝑗𝐴𝑇,𝑡𝑗𝑒∑𝑛𝑖=1−𝐵𝑖(𝑇,𝑡𝑗)𝑥𝑖+𝑔𝑥1,…,𝑥𝑛𝑑𝑥1⋯𝑑𝑥𝑛,(3.23) where 𝑔 denotes the density function of the vector 𝑋𝑛=(𝑥1(𝑇),…,𝑥𝑛(𝑇)) conditioned on 𝐹0 under 𝑄𝑇, that is, is the 𝑛-dimensional normal density function with vector mean and covariance matrix 𝐶𝑖𝑗(𝑇) given by 𝐸𝑋𝑛∣𝐹0=−𝑀1(0,𝑇),…,−𝑀𝑛𝐶(0,𝑇)𝑖𝑗(𝑇)=𝑛𝑖,𝑗=1𝑇0ğœŽğ‘–(𝑢)ğœŽğ‘—(𝑢)ğœŒğ‘–ğ‘—ğ‘’âˆ’ğ‘Žğ‘–(𝑇−𝑢)ğ‘’âˆ’ğ‘Žğ‘—(𝑇−𝑢)𝑑𝑢(3.24) and, in the case for every 𝑖=1,…,𝑛, ğœŽğ‘–(𝑡)=ğœŽğ‘– is a positive constant function ğ¶ğ‘–ğ‘—ğœŽ(𝑇)=ğ‘–ğœŽğ‘—ğ‘Žğ‘–+ğ‘Žğ‘—ğœŒğ‘–ğ‘—î€·1−𝑒−(ğ‘Žğ‘–+ğ‘Žğ‘—)𝑇.(3.25) Since the numerical solution of the integral in (3.23) is computationally inefficient for large 𝑛, the price of European swaption could be approximated by the approach proposed by Schrager and Pelsser in [6]. As illustrated in the Appendix of [6], this method leads to a very simple analytical pricing formula of a swaption in the framework of Gaussian interest rate. In the G𝑛++ model, in the case that the strike is in at the money, we have 𝐸𝑆0,𝑇,𝑡𝑘,𝐾,𝑁⋍𝑁VOL√2𝜋𝑘𝑖=1𝜏𝑖𝑃0,𝑡𝑖≡𝑁VOL√𝑃2𝜋𝑡𝑘𝑡1,(3.26) where VOL is the approximated volatility of the swap rate given by ⎷VOL=𝑛𝑖,𝑗=1𝑇0ğœŽğ‘–(𝑢)ğœŽğ‘—(𝑢)𝜌𝑖𝑗𝐴𝑖𝐴𝑗𝑒(ğ‘Žğ‘–+ğ‘Žğ‘—)𝑢𝑑𝑢,(3.27) where, for every 𝑖=1,…,𝑛, 𝐴𝑖=ğ‘’âˆ’ğ‘Žğ‘–ğ‘‡ğ‘ƒ(0,𝑇)𝑃𝑡𝑘𝑡1âˆ’ğ‘’âˆ’ğ‘Žğ‘–ğ‘¡ğ‘˜ğ‘ƒî€·0,𝑡𝑘𝑃𝑡𝑘𝑡1−𝐾𝑘𝑗=1ğ‘’âˆ’ğ‘Žğ‘–ğ‘¡ğ‘–ğœğ‘–ğ‘ƒî€·0,𝑡𝑖𝑃𝑡𝑘𝑡1.(3.28) We finally point out that, in the case for every 𝑖=1,…,𝑛, ğœŽğ‘–(𝑡)=ğœŽğ‘– is a positive constant function, we have ⎷VOL=𝑛𝑖,𝑗=1ğœŽğ‘–ğœŽğ‘—ğœŒğ‘–ğ‘—ğ´ğ‘–ğ´ğ‘—ğ‘’(ğ‘Žğ‘–+ğ‘Žğ‘—)𝑇−1ğ‘Žğ‘–+ğ‘Žğ‘—.(3.29)

4. Numerical Results: Calibration to ATM Swaptions Prices

We test formula (3.26) in the calibration to ATM swaptions prices at 31/12/2006 and at 31/12/2011. All market data are taken from Bloomberg provider. ATM swaptions volatility surface at the reference dates are, respectively, shown in Tables 1 and 2 (value in percentage).

In our experiments, to calibrate parameters, we use the differential evolution algorithm as implemented in  R  through the package DEoptim. We set the default characteristics of the optimization algorithm. In particular, we use the second strategy instead of the classical mutation in the optimization procedure; we set a number of maximum population generators allowed equals to 400, a crossover probability equals to 0.5, and a step-size equals to 0.8. The function to minimize is the sum of the square percentage difference between the market price and the model price of European swaptions with tenor and maturity as in Table 2.

4.1. G1++ with Piecewise Constant Volatility

We consider the G1++ model and we choose a piecewise constant functional specification for the volatility of the process 𝑥(𝑡): in particular, the volatility function ğœŽ(𝑡) is constant within intervals with endpoints 𝑡=0,1,3,20. We determinate previous intervals after several experiments: we try more set of possible intervals, and we find that the best set is the previous one, even increasing the number of intervals. At 31/12/2006, the best fit gives a mean percentage error of 2.5% on the swaptions price surface. Instead of this good result, at 31/12/2011, the calibrated parameters give a mean percentage error of 7.2% on the swaptions price surface. The matrix of percentage error at 31/12/2011 is shown in Table 3.


T/Tenor 1 2 5
7 10

1 0.4139 0.0907
0.3262 0.3795
0.4364
2
0.102
0.1352
0.0029
0.0242 0.06753
5 0.0769 0.02647 0.0064 0.0041 0.0036
7 0.0702 0.03631
0.0082
0.0096
0.0142
10
0.02967
0.02107 0.01227
0.00845
0.00757
15 0.01828 0.01273
0.0033
0.0071
0.0105
20
0.00632 0.00682 0.0269
0.0422
0.0741

To improve the result of calibration, we increase the number of factors.

4.2. G2++ with Constant Parameters

We now consider the G2++ model with constant parameters. At 31/12/2006, we have a very good result: the calibrated parameters give a mean percentage error of 1.6% on the swaptions price surface. At 31/12/2011, we improve the calibration result obtained in the previous section at the same reference date, but the instantaneous correlation 𝜌12 between the two Brownian motions given by the calibration output is near to −1. Anyway, the mean percentage error over the swaptions price surface is 5.2% with the percentage error matrix given by Table 4.


T/Tenor125710

10.0010.10970.02070.071660.1703
20.1910.037810.024370.054090.129
50.70750.007070.0310.003360.04999
70.0102820.059590.07120.0337570.02974
100.081950.094050.069670.034180.02683
150.06060.062570.036580.000620.059657
200.025820.019960.005320.031790.06581

4.3. G2++ with Constant Piecewise Volatility

We consider the G2++ model and we choose a piecewise constant functional specification for both the volatility of the processes 𝑥1(𝑡) and 𝑥2(𝑡): in particular, the volatility functions ğœŽ1(𝑡) and ğœŽ2(𝑡) are constant within intervals with endpoints 𝑡=0,1,2,20. As for the G1++ model, we arrive to determinate the previous intervals after several experiments: we try more set of possible intervals and we find that the best set is the previous one. We make experiments only at 31/12/2011, since we obtained a nonimproved result at 31/12/2006 in the previous section. The calibrated parameters give a mean percentage error of 3.4% on the swaptions price surface. This is the best result that we obtain through all our experiments at 31/12/2011. The relative error matrix is shown in Table 5.


T/Tenor1 2 5
7 10

1 0.04 0.0067 0.0105 0.048050.04482
2 0.01650.08404 0.12790.035140.03316
5 0.05272 0.05583 0.04804 0.01913 0.00367
7 0.0125 0.0032 0.00730.01530.01216
10 0.08680.05430.0250.02350.0083
15 0.0694 0.0358 0.00169
0.0109 0.0367
20
0.0276 0.01
0.0448
0.051 0.0559

4.4. G2++ with a Functional Instantaneous Correlation

Since the calibrated instantaneous correlation 𝜌12 in the G2++ model with constant parameters is near to −1, we set a functional specification for 𝜌12𝜌12(𝑡)=ğœŒâˆž+1âˆ’ğœŒâˆžî€¸ğ‘’âˆ’ğœ†ğ‘¡,(4.1) where 𝜆 is a positive constant and −1â‰¤ğœŒâˆžâ‰¤1. The interpretation of these two parameters is straightforward: ğœŒâˆž is the long-term average instantaneous correlation whereas 𝜆 is the rate of convergence of 𝜌12(𝑡) to ğœŒâˆž. The other parameters are constant functions. Unfortunately, among the best calibrated parameters, that give a mean percentage error of 5.2%, ğœŒâˆž is closer to −1 as aspected, but 𝜆 is quite large (near to 15). This does not improve the calibration obtained with constant instantaneous correlation.

4.5. G3++ with Constant Parameters

We finally consider the G3++ model with constant parameters. We substantially do not improve the result obtained with the G2++ model with constant parameters. In fact, the mean percentage error still remained near to 5.2%, and the instantaneous correlation 𝜌23 between the second and the third factor is near to −1.

5. Conclusions

In this paper we describe a general exogenous model in which the instantaneous spot rate 𝑟 is the sum of correlated Gaussian stochastic processes with time-dependent volatility plus a deterministic function given by an exact fitting to the observed term structure. We also give a general notation independent of the number of factors and analogous with the notation used in literature. In our numerical experiments, we consider an increasing number of factors from one to three and different functional specifications for the volatility of factors and one also for the correlation. To calibrate the model to market data, we use an accurate and fast formula to approximate a European swaptions price proposed by Schrager and Pelsser in [6]. The algorithm used for calibration is the differential evolution as implemented in the R package DEoptim, whereas the function to minimize is the sum of the square percentage difference between the market price and the model price of swaptions. At 31/12/2006 we obtain a good fit to the market data. The G2++ model with constant parameters gives a mean percentage error of 1.6% on the swaptions price surface. Then, August 2007 arrived and the world will never be the same. We test the calibration at 31/12/2011. We obtain the best optimization with the two factors model with a piecewise constant functional specification for both the volatility functions. We get a mean percentage error on the swaptions price surface of 3.4%. We also note that setting all the parameters constant and increasing the number of the factors, we do not improve the performance of the calibration. Further investigation could be direct to different directions: we could set a different functional specification for the volatility of the factors, instead of constant or piecewise constant as in this paper; we could use a different optimization technic to calibrate the parameters instead of differential evolution algorithm, or, starting from the output of the differential evolution algorithm; we could try to increase the performance of the calibration using a deterministic optimization algorithm; finally, we could try to calibrate the parameters to other market data instead of swaptions prices.

Acknowledgments

The author wishes to thank professor Andrea Pascucci and prof. Sergio Polidoro for several helpful conversations and for some useful remarks that improved this paper. He also wishes to thank his wife Giulia for some useful remarks that correct and improve my English language. Finally, he thanks his colleagues Lorenzo and Martina that allowed him to write this paper.

References

  1. T. Bjork, Arbitrage Theory in Continuous Time, Oxford University Press, 1998.
  2. D. Brigo and F. Mercurio, Interest Rate Models—Theory and Practice, Springer Finance, Springer, Berlin, Germany, 2nd edition, 2006. View at: Zentralblatt MATH
  3. D. Filipović, Term-Structure Models, Springer Finance, Springer, Berlin, Germany, 2009. View at: Publisher Site | Zentralblatt MATH
  4. A. Pascucci, PDE and martingale methods in option pricing, vol. 2 of Bocconi & Springer Series, Springer, Milan, Italy, 2011. View at: Publisher Site | Zentralblatt MATH
  5. J. Hull and A. White, Options, Futures and Other Derivatives, Prentice Hall, Upper Saddle River, NJ, USA, 3rd edition, 1997.
  6. D. F. Schrager and A. A. J. Pelsser, “Pricing swaptions and coupon bond options in affine term structure models,” Mathematical Finance, vol. 16, no. 4, pp. 673–694, 2006. View at: Publisher Site | Google Scholar | Zentralblatt MATH
  7. J. Hull and A. White, “Pricing interest rate derivative securities,” Review of Financial Studies, vol. 3, no. 4, pp. 573–592, 1990. View at: Publisher Site | Google Scholar
  8. J. Hull and A. White, “Numerical procedures for implementing term structure models ii: two-factor models,” The Journal of Derivatives, vol. 2, no. 2, pp. 37–48, 1994. View at: Google Scholar
  9. J. Hull and A. White, “Numerical procedures for implementing term structure models i: single-factor models,” The Journal of Derivatives, vol. 2, no. 1, pp. 7–16, 1994. View at: Google Scholar
  10. F. Jamshidian, “An exact bond option formula,” The Journal of Finance, vol. 44, no. 1, pp. 205–209, 1989. View at: Google Scholar
  11. H. German, N. EL KAROUI, and J. C. Rochet, “Changes of numeraire, changes of probability measure and pricing of options,” Journal of Applied Probability, vol. 32, pp. 443–458, 1995. View at: Google Scholar

Copyright © 2012 Marco Di Francesco. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.


More related articles

 PDF Download Citation Citation
 Download other formatsMore
 Order printed copiesOrder
Views4653
Downloads759
Citations

Related articles

We are committed to sharing findings related to COVID-19 as quickly as possible. We will be providing unlimited waivers of publication charges for accepted research articles as well as case reports and case series related to COVID-19. Review articles are excluded from this waiver policy. Sign up here as a reviewer to help fast-track new submissions.