Abstract
We propose a general framework for modelling multiple yield curves which have emerged after the last financial crisis. In a general semimartingale setting, we provide an HJM approach to model the term structure of multiplicative spreads between FRA rates and simply compounded OIS risk-free forward rates. We derive an HJM drift and consistency condition ensuring absence of arbitrage and, in addition, we show how to construct models such that multiplicative spreads are greater than one and ordered with respect to the tenor’s length. When the driving semimartingale is an affine process, we obtain a flexible and tractable Markovian structure. Finally, we show that the proposed framework allows unifying and extending several recent approaches to multiple yield curve modelling.
Similar content being viewed by others
Notes
Due to the deterministic short rate, it is not necessary to distinguish the expectations with respect to different measures, but we explicitly indicate them for consistency of the exposition with the general setting of the following sections.
Note that for every tenor \(\delta\in\{\delta _{1},\ldots,\delta_{m}\}\), the process \(u_{i}^{\top}Y\) plays the role of the process \(Z\) appearing in (2.3).
The definition of local exponent in [39, Definition A.6] is slightly different since it is defined in terms of the exponential compensator of \((\int_{0}^{t} \mathsf{i}\beta_{s} \,dX_{s})_{t\geq0}\), which is due to the fact that complex-valued processes are considered in that paper. Note also that since Itô semimartingales are quasi-left-continuous, the derivatives of the modified Laplace cumulant process and of the ordinary Laplace cumulant process coincide (see e.g. [40, Remarks on p. 408]).
In particular, in [9] (according to the notation used therein), it holds that \(Z^{\delta}_{t}=\int_{t}^{t+\delta }g_{t}(s)\,ds\), for any tenor \(\delta>0\), with \(g_{t}(T)\) representing the spread between risky and risk-free instantaneous \(T\)-forward rates.
Note that our multiplicative spread \(S^{\delta}(t,T)\) corresponds to \(K_{L}(t,T,T+\delta)\), according to the notation adopted in [43].
Let us remark that in the presence of funding costs, absence of arbitrage is implied by the existence of an equivalent measure under which the risky assets \(S\) present in the market are (local) martingales when discounted with their corresponding funding rate \(r^{f}\). This can be embedded in the classical framework where ℚ is a risk neutral measure with the risk-free (OIS) bank account as numéraire, by treating \(e^{\int_{0}^{\cdot} (-r_{s}^{f}+r_{s}) \,ds}S\) as traded asset.
Indeed, as explained in [56, Sect. 4.2.1], bond prices are expressed in units of the respective currencies, while discount factors are simply the corresponding real numbers.
References
Ametrano, F.M., Bianchetti, M.: Everything you always wanted to know about multiple interest rate curve bootstrapping but were afraid to ask (2013). Preprint available at http://ssrn.com/abstract=2219548
Bianchetti, M.: Two curves, one price. Risk Magazine, 74–80 (2010)
Bielecki, T., Rutkowski, M.: Valuation and hedging of contracts with funding costs and collateralization. SIAM J. Financ. Math. 6, 594–655 (2015)
Brace, A., Gątarek, D., Musiela, M.: The market model of interest rate dynamics. Math. Finance 7, 127–155 (1997)
Brigo, D., Mercurio, F.: A deterministic-shift extension of analytically-tractable and time-homogeneous short-rate models. Finance Stoch. 5, 369–387 (2001)
Bru, M.F.: Wishart processes. J. Theor. Probab. 4, 725–751 (1991)
Carmona, R.A.H.: A unified approach to dynamic models for fixed income, credit and equity markets. In: Carmona, R.A., et al. (eds.) Paris–Princeton Lectures on Mathematical Finance 2004. Lecture Notes in Math., vol. 1919, pp. 1–50. Springer, Berlin (2007)
Crépey, S., Grbac, Z., Nguyen, H., Skovmand, D.: A Lévy HJM multiple-curve model with application to CVA computation. Quant. Finance 15, 401–419 (2015)
Crépey, S., Grbac, Z., Nguyen, H.N.: A multiple-curve HJM model of interbank risk. Math. Financ. Econ. 6, 155–190 (2012)
Cuchiero, C., Filipović, D., Mayerhofer, E., Teichmann, J.: Affine processes on positive semidefinite matrices. Ann. Appl. Probab. 21, 397–463 (2011)
Cuchiero, C., Klein, I., Teichmann, J.: A new perspective on the fundamental theorem of asset pricing for large financial markets. Theory of Probability and its Applications. To appear. Preprint available at http://arxiv.org/abs/1412.7562
Cuchiero, C., Teichmann, J.: Path properties and regularity of affine processes on general state spaces. In: Donati-Martin, C., et al. (eds.) Séminaire de Probabilités XLV. Lecture Notes in Mathematics, vol. 2078, pp. 201–244. Springer, Berlin (2013)
Delbaen, F., Schachermayer, W.: A general version of the fundamental theorem of asset pricing. Math. Ann. 300, 463–520 (1994)
Dörsek, P., Teichmann, J.: Efficient simulation and calibration of general HJM models by splitting schemes. SIAM J. Financ. Math. 4, 575–598 (2013)
Douady, R., Jeanblanc, M.: A rating-based model for credit derivatives. Eur. Invest. Rev. 1, 17–29 (2002)
Duffie, D., Filipović, D., Schachermayer, W.: Affine processes and applications in finance. Ann. Appl. Probab. 13, 984–1053 (2003)
Eberlein, E., Kluge, W.: Calibration of Lévy term structure models. In: Fu, M., et al. (eds.) Advances in Mathematical Finance: In Honor of D.B. Madan, pp. 147–172. Birkhäuser, Basel (2007)
Eberlein, E., Koval, N.: A cross-currency Lévy market model. Quant. Finance 6, 465–480 (2006)
Filipović, D.: Consistency Problems for Heath–Jarrow–Morton Interest Rate Models. Lecture Notes in Mathematics, vol. 1760. Springer, Berlin (2001)
Filipović, D., Overbeck, L., Schmidt, T.: Dynamic CDO term structure modeling. Math. Finance 21, 53–71 (2011)
Filipović, D., Tappe, S., Teichmann, J.: Jump-diffusions in Hilbert spaces: existence, stability and numerics. Stochastics 82, 475–520 (2010)
Filipović, D., Tappe, S., Teichmann, J.: Term structure models driven by Wiener processes and Poisson measures: existence and positivity. SIAM J. Financ. Math. 1, 523–554 (2010)
Filipović, D., Trolle, A.B.: The term structure of interbank risk. J. Financ. Econ. 109, 707–733 (2013)
Fries, C.P.: Curves and term structure models: definition, calibration and application of rate curves and term structure models (2010). Preprint available at http://ssrn.com/abstract=2194907
Fujii, M., Shimada, A., Takahashi, A.: A market model of interest rates with dynamic basis spreads in the presence of collateral and multiple currencies. Wilmott 54, 61–73 (2011)
Glau, K., Grbac, Z., Kriens, D.: Martingale property of exponential semimartingales: a note on explicit conditions and applications to asset price and Libor models (2015). Preprint, TU Munich, available at http://arxiv.org/pdf/1506.08127v1.pdf
Grasselli, M., Miglietta, G.: A flexible spot multiple-curve model (2014). Preprint available at http://ssrn.com/abstract=2424242
Grbac, Z., Papapantoleon, A.: A tractable LIBOR model with default risk. Math. Financ. Econ. 7, 203–227 (2013)
Grbac, Z., Papapantoleon, A., Schoenmakers, J., Skovmand, D.: Affine LIBOR models with multiple curves: theory, examples and calibration. SIAM J. Financ. Math. 6, 984–1025 (2015)
Heath, D., Jarrow, R., Morton, A.: Bond pricing and the term structure of interest rates: a new methodology for contingent claims valuation. Econometrica 60, 77–105 (1992)
Henrard, M.: The irony in the derivatives discounting. Wilmott 30, 92–98 (2007)
Henrard, M.: The irony in the derivatives discounting part II: the crisis. Wilmott 2, 301–316 (2010)
Henrard, M.: Interest Rate Modelling in the Multi-Curve Framework. Palgrave Macmillan, Basingstoke (2014)
Jacod, J.: Multivariate point processes: predictable projection, Radon–Nikodým derivatives, representation of martingales. Z. Wahrscheinlichkeitstheor. Verw. Geb. 31, 235–253 (1974/1975)
Jacod, J., Protter, P.: Discretization of Processes, Stochastic Modelling and Applied Probability, vol. 67. Springer, Berlin (2012)
Jacod, J., Shiryaev, A.: Limit Theorems for Stochastic Processes, 2nd edn. Grundlehren der mathematischen Wissenschaften, vol. 288. Springer, Berlin/Heidelberg/New York (2003)
Jarrow, R., Turnbull, S.: Pricing derivatives on financial securities subject to credit risk. J. Finance 50, 53–85 (1995)
Jeanblanc, M., Yor, M., Chesney, M.: Mathematical Methods for Financial Markets. Springer, London (2009)
Kallsen, J., Krühner, P.: On a Heath–Jarrow–Morton approach for stock options. Finance Stoch. 19, 583–615 (2015)
Kallsen, J., Shiryaev, A.: The cumulant process and Esscher’s change of measure. Finance Stoch. 6, 397–428 (2002)
Keller-Ressel, M., Mayerhofer, E.: Exponential moments of affine processes. Ann. Appl. Probab. 25, 714–752 (2015)
Kenyon, C.: Post-shock short-rate pricing. Risk Magazine, 83–87 (2010)
Kijima, M., Tanaka, K., Wong, T.: A multi-quality model of interest rates. Quant. Finance 9, 133–145 (2009)
Klein, I., Schmidt, T., Teichmann, J.: When roll-overs do not qualify as numéraire: bond markets beyond short rate paradigms (2013). Preprint available at http://arxiv.org/abs/1310.0032
Kluge, W.: Time-inhomogeneous Lévy processes in interest rate and credit risk models. Ph.D. thesis, University of Freiburg (2005). available at https://www.freidok.uni-freiburg.de/data/2090/
Koval, N.: Time-inhomogeneous Lévy processes in cross-currency market models. Ph.D. thesis, University of Freiburg (2005). available at https://www.freidok.uni-freiburg.de/data/2041/
Krein, M.G., Nudelman, A.A.: The Markov Moment Problem and Extremal Problems: Ideas and Problems of P.L. Čebyšev and A.A. Markov and Their Further Development. Translations of Mathematical Monographs, vol. 50. Am. Math. Soc., Providence (1977)
Larsson, M., Ruf, J.: Convergence of local supermartingales and Novikov–Kazamaki type conditions for processes with jumps (2014). Preprint available at http://arxiv.org/abs/1411.6229
Levin, K., Zhang, J.X.: Bloomberg volatility cube. Tech. rep., Bloomberg L.P. (2014)
Mercurio, F.: Modern Libor market models: using different curves for projecting rates and discounting. Int. J. Theor. Appl. Finance 13, 113–137 (2010)
Mercurio, F.: A Libor market model with a stochastic basis. Risk Magazine, 96–101 (2013)
Mercurio, F., Xie, Z.: The basis goes stochastic. Risk Magazine, 78–83 (2012)
Moreni, N., Pallavicini, A.: Parsimonious HJM modelling for multiple yield-curve dynamics. Quant. Finance 14, 199–210 (2014)
Morini, M.: Solving the puzzle in the interest rate market. In: Bianchetti, M., Morini, M. (eds.) Interest Rate Modelling After the Financial Crisis, pp. 61–106. Risk Books, London (2013)
Morino, L., Runggaldier, W.J.: On multicurve models for the term structure. In: Dieci, R., et al. (eds.) Nonlinear Economic Dynamics and Financial Modelling: Essays in Honour of Carl Chiarella, pp. 275–290. Springer, Berlin (2014)
Musiela, M., Rutkowski, M.: Martingale Methods in Financial Modelling, 2nd edn. Springer, Berlin (2005)
Pallavicini, A., Tarenghi, M.: Interest rate modelling with multiple yield curves (2010). Preprint available at http://ssrn.com/abstract=1629688
Protter, P.E.: Stochastic Integration and Differential Equations. Stochastic Modelling and Applied Probability, vol. 21, 2nd edn. Springer, Berlin/Heidelberg/New York (2005). Version 2.1, corrected third printing
Acknowledgements
The research of the second author was partly supported by a Marie Curie Intra European Fellowship within the 7th European Community Framework Programme under grant agreement PIEF-GA-2012-332345.
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix A: Pricing under collateral and FRA rates
Let us here briefly review pricing under perfect collateralization for general derivatives which we then apply to the pricing of FRAs. For a more detailed discussion on general valuation with collateralization and funding costs, we refer to the growing literature on this topic, e.g. [3] and the references therein. We here follow closely [23, Sect. 2.2].
Throughout, let \((\varOmega, \mathcal{F}, (\mathcal{F}_{t})_{t\geq0}, \mathbb {P})\) be a filtered probability space, where ℙ stands for the statistical/historical probability measure. We consider OIS zero coupon bonds as basic traded instruments, which play the role of risk-free zero coupon bonds in the classical setting. In order to guarantee no arbitrage, we assume that
-
(i)
There exists an OIS (risk-free) bank account denoted by \((B_{t})_{t\geq0}\) and such that \(B_{t}=\exp(\int_{0}^{t} r_{s} \,ds)\), where \(r\) denotes the OIS short rate;
-
(ii)
There exists an equivalent probability measure ℚ such that the OIS bonds for all maturities are ℚ-martingales when denominated in units of the OIS bank account.Footnote 7
Let now \(X\) be an \(\mathcal{F}_{T}\)-measurable payoff of some derivative security. We assume here a perfect collateral agreement where 100 % of the derivative’s present value \(V_{t}\) is posted in the collateral at any time \(t < T\). The receiver of the collateral can invest it at the risk-free rate \(r\), corresponding to the OIS short rate, and has to pay an agreed collateral rate \(r^{c}_{t}\) to the poster of the collateral. Applying risk neutral pricing, we obtain for the present value of the collateralized transaction the expression
As shown in [23, Appendix A], this formula is equivalent to
Assuming that the collateral rate \(r^{c}\) corresponds to the OIS short rate \(r\), which is usually the case, we obtain the classical risk neutral valuation formula.
Since market quotes of FRAs correspond to perfectly collateralized contracts, where the collateral rate \(r^{c}\) is assumed to be the OIS short rate \(r\), the above pricing approach is applied for the definition of FRA rates. As in classical interest rate theory, the FRA rate, denoted by \(L_{t}(T,T+\delta)\), is the rate \(K\) fixed at time \(t\) such that the value of the FRA contract, whose payoff at time \(T+\delta\) is given by \(\delta(L_{T}(T,T+\delta)- K)\), is 0. Therefore, it holds that, for all \(t\in[0,T]\) and \(T\geq0\),
Hence by Bayes’ formula,
where \(\mathbb{Q}^{T+\delta}\) denotes the \((T+\delta)\)-forward measure associated with the numéraire \(B(\cdot,T+\delta)\) and density process \(\frac{d\mathbb{Q}^{T+\delta}}{d\mathbb {Q}}|_{\mathcal {F}_{t}}=\frac{B(t,T+\delta)}{B_{t} B(0,T+\delta)}\). In particular, this provides a rigorous justification for the market practice of taking the expression (2.1) as the definition of fair FRA rates.
Appendix B: Foreign exchange analogy
For simplicity of presentation, let us consider a fixed tenor \(\delta\) and define artificial “risky” bond prices \(B^{\delta}(t,T)\) at time \(t\) and maturity \(T\) for the tenor \(\delta\) by the relation, for all \(t\leq T\) and \(T\geq0\),
While the family \(\{(B(t,T))_{t\in[0,T]},T\geq0\}\) represents prices of domestic risk-free bonds (in units of the domestic currency), \(\{ (B^{\delta}(t,T))_{t\in[0,T]},T\geq0\}\) can be thought of as representing prices of zero-coupon bonds of a foreign “risky” economy, expressed in units of the foreign currency.
According to this foreign exchange analogy, one is naturally led to look at the ratio
for \(t\leq T\) and \(T\geq0\), where \(B^{\delta}(t,T)\) (resp., \(B(t,T)\)) has here to be seen as the discount factor for the foreign (resp., domestic) economy.Footnote 8 Note also that \(R^{\delta}(T,T)=1\), for all \(T\geq0\). Following the presentation in [56, Sect. 4.2.1], the quantity \(R^{\delta}(t,T)\) corresponds to the forward exchange premium between the domestic and the foreign currency over the time interval \([t,T]\). Indeed, in standard foreign exchange markets, there is the following no arbitrage relation between the spot exchange rate \(Q_{t}\) (domestic price of one unit of the foreign currency) and the forward exchange rate \(F(t,T)\) (forward price in domestic currency of one unit of the foreign currency paid at time \(T\)): we have
The multiplicative spread \(S^{\delta}(t,T)\) introduced in (2.2) corresponds now to
for all \(t\leq T\) and \(T\geq0\), while the spot multiplicative spread is simply given by
for all \(T\geq0\). In particular, note that \(R^{\delta}(T,T+\delta)\) corresponds exactly to the quantity \(Q^{\delta}_{T}\) considered in Sect. 2.1 which thus has the interpretation of a foreign exchange premium over \([T,T+\delta]\).
Since Libor rates reflect the overall credit risk of the Libor panel, the exchange rate premium \(R^{\delta}(t,T+\delta)\) can be seen as a market valuation (at time \(t\)) of the riskiness of the foreign economy, i.e. of the credit and liquidity quality of the current Libor panel over the period \([t,T+\delta]\). Moreover, according to the same interpretation, the quantity
is thus an expectation of the riskiness of the future Libor panel over the future time period \([T,T+\delta]\), as seen from the market at time \(t\) (calculated as the relative riskiness of the current Libor panel over the period \([t,T+\delta]\) relative to the one over \([t,T]\)). For instance, a large value of \(S^{\delta}(t,T)\) would mean that the market anticipates a worsening of the credit quality of the Libor panel on \([T,T+\delta]\) compared to the credit quality on \([t,T]\) as seen at time \(t\).
In this sense, the multiplicative spread \(S^{\delta}(t,T)\) is a rather natural quantity to model in a multiple curve setting, because it represents the market’s expectation at time \(t\) (being computed from financial instruments traded at date \(t\)) of the credit and liquidity quality of the Libor panel over \([T,T+\delta]\).
Appendix C: Local independence and semimartingale decomposition
In this section, we let \((X,Y)\) be a general Itô semimartingale taking values in \(\mathbb{R}^{d+n}\) and denote by \(\varPsi^{X,Y}\) its local exponent and by \(\varPsi^{X}\) and \(\varPsi^{Y}\) the local exponents of \(X\) and \(Y\), respectively, and let \(\mathcal{U}^{X,Y}\) be defined as in Definition 3.2. In view of [39, Lemma A.11], the following definition is equivalent to the notion of local independence as given in [39, Definition A.10].
Definition C.1
We say that \(X\) and \(Y\) are locally independent if outside a \(d\mathbb{Q}\otimes dt\)-null set, it holds that
Following [39, Appendix A.3], let us recall the notion of semimartingale decomposition of \(Y\) relative to \(X\). We denote by \(c^{Y,X}\) and \(c^{X}\) the second local characteristic of \((Y,X)\) and \(X\), respectively, and by \(K^{Y,X}\) and \(K^{X}\) the third local characteristic of \((Y,X)\) and \(X\), respectively. Denote also by \(\mu^{Y,X}\) the jump measure of \((Y,X)\). Supposing that \(1\in\mathcal{U}^{Y}\) (i.e. \(Y\) is exponentially special, see Proposition 3.3), let
for \(i=1,\ldots,n\), where \((c^{X})^{-1}\) denotes the pseudoinverse of the matrix \(c^{X}\) and \(X^{c}\) is the continuous local martingale part of \(X\) (see [36, Proposition I.4.27]). We call \(Y^{\parallel }:=(Y^{\parallel,1},\ldots,Y^{\parallel,n})^{\top}\) the dependent part of \(Y\) relative to \(X\) and \(Y^{\perp}:=Y-Y^{\parallel}\) the independent part of \(Y\) relative to \(X\). The following lemma corresponds to [39, Lemma A.22 and Lemma A.23].
Lemma C.2
Let \((X,Y)\) be an \(\mathbb{R}^{d\times n}\)-valued Itô semimartingale and suppose that \(1\in\mathcal{U}^{Y}\). Then the following hold:
-
(i)
\(Y\mapsto Y^{\parallel}\) is a projection, in the sense that \((Y^{\parallel})^{\parallel}=Y^{\parallel}\).
-
(ii)
If \(Z\) is an Itô semimartingale locally independent of \(X\), then \((Z+Y)^{\parallel}=Y^{\parallel}\).
-
(iii)
\(\exp(Y^{\parallel,i})\) is a local martingale, for all \(i=1,\ldots,n\).
-
(iv)
\(Y^{\perp}\) and \((Y^{\parallel},X)\) are locally independent semimartingales.
Appendix D: Proofs of the results of Sect. 4
Proof of Proposition 4.4
Let us fix any \(i\in\{0,1,\ldots,m\}\). By the same argument as in [19, Corollary 5.12], it can be shown that \(\kappa_{4}(H^{\lambda}_{m+1}) \subseteq H^{\lambda,0}_{1} \) and \(\kappa_{j}^{i}(H^{\lambda}_{m+1}) \subseteq H^{\lambda,0}_{1} \) for \(j=1,2\). In the sequel, \(C\) always denotes a positive constant which can vary from line to line. The following estimates can be derived similarly as in the proof of [22, Proposition 3.2] (to which we refer the reader for more details), by using Assumption 4.3, the Hölder inequality and [22, Theorem 2.1].
For \(\kappa_{3}^{i}(h)\) and \(\kappa_{5}^{i}(h)\), we have for all \(h \in H_{m+1}^{\lambda}\), \(\xi\in\mathbb{R}^{d}\) and \(s \in\mathbb{R}_{+}\) that
and for all \(\xi\in\mathbb{R}^{d}\), \(\hat{\xi}\in\mathbb{R}^{n}\) and \(s \in\mathbb {R}_{+}\) that
Together with (4.9) and (4.10), this shows that we have \(\lim_{s \to\infty}\kappa_{3}^{i}(h)(s)=0\) and \(\lim_{s \to\infty}\kappa_{5}^{i}(h)(s)=0\). Moreover, we have
In view of the form of \(\partial_{s} \kappa^{i}_{3}(h)\) and \(\partial_{s} \kappa _{5}(h)\) as given in (4.11), (4.12), this implies that \(\kappa_{j}^{i}(H^{\lambda}_{m+1}) \subseteq H^{\lambda,0}_{1}\) for \(j=3,5\). We have thus shown that
For \(h_{1}, h_{2} \in H^{\lambda}_{m+1}\) we obtain
Furthermore, due to (4.11) and (4.12), we can estimate
where
We get for all \(\xi\in\mathbb{R}^{d}\), \(s \in\mathbb{R}_{+}\) that
Therefore,
Moreover, for every \(s \in\mathbb{R}_{+}\), we obtain
Hence,
Moreover,
Finally, we have
Summing up, we have shown that there exist constants \(Q_{i}>0\) such that condition (4.13) is satisfied for all \(h_{1},h_{2} \in H_{m+1}^{\lambda}\). □
Proof of Proposition 4.10
For every \((\omega,\omega')\in\widetilde{\varOmega}\) and \(t\geq0\), define the process \(Y^{\perp}_{t}(\omega,\omega') := y_{0} + J_{t}(\omega')\), for some starting value \(y_{0}\in\mathbb{R}_{+}\). Clearly, \(Y^{\perp}\) is a pure jump \((\mathcal{G}_{t})\)-adapted process. In order to prove the existence of a probability measure \(\widetilde {\mathbb{Q}}\) such that the jump measure of \(Y^{\perp}\) with respect to the two filtrations \((\mathcal{G}_{t})_{t\geq0}\) and \((\overline {\mathcal{G}}_{t})_{t\geq 0}\) is given by \(K_{t}(\omega, Y^{\perp}_{t-}(\omega,\omega'),d\xi)dt\) and \(\widetilde{\mathbb{Q}}|_{\mathcal{F}}=\mathbb{Q}\) holds true, we rely on [34, Theorem 3.6].
For all \(\omega\in\varOmega\), \(y\in\mathbb{R}_{+}\) and \(t\geq0\), let us first extend the definition of \(K_{t}(\omega,y, d\xi)\) as in (4.15), (4.16) to \(y\in\mathbb{R}_{-}\) by requiring that it is supported on \([-|y|, \infty)\) and by setting \(p_{t}(y,\omega )=p_{t}(-y, \omega)\) for \(y\in\mathbb{R}_{-}\). Due to Assumption 4.8, the measure \(K_{t}(\omega,Y^{\perp}_{t-}(\omega ,\omega'), d\xi)dt\) defined via the moment problem (4.15), (4.16) is a positive random measure on \(\mathbb{R}_{+} \times\mathbb{R}\). Since \(Y^{\perp}\) is càdlàg and \((\mathcal{G}_{t})\)-adapted and since \((p_{t}(\cdot,y))_{t\geq0}\) is \((\mathcal{F}_{t})\)-predictable and depends in a measurable way on \(y\), the process \(p_{t}(\omega, Y^{\perp}_{t-}(\omega ,\omega'))\) is \((\mathcal{G}_{t})\)-predictable and the same \((\mathcal{G} _{t})\)-predictability, and hence \((\overline{\mathcal {G}}_{t})\)-predictability, is inherited by \(K_{t}(\omega,Y^{\perp}_{t-}(\omega,\omega'), d\xi)\). Let us then define the \((\overline{\mathcal{G}}_{t})\)-predictable random measure \(\nu\) by
[34, Theorem 3.6] implies that there exists a unique probability kernel ℙ from \((\varOmega,\mathcal{F})\) to ℋ such that \(\nu\) is the \((\overline{\mathcal{G}}_{t})\)-compensator of the random measure \(\mu\) associated with the jumps of \(J\). On \((\widetilde {\varOmega}, \mathcal{G})\), we then define the probability measure \(\widetilde{\mathbb{Q}}\) by setting \(\widetilde{\mathbb {Q}}(d\widetilde{\omega})=\mathbb{Q}(d\omega)\mathbb{P}(\omega ,d\omega')\), whose restriction to ℱ is equal to ℚ. Moreover, since \(Y^{\perp}\) is \((\mathcal{G}_{t})\)-adapted and \(K_{t}(\omega ,Y^{\perp}_{t-}(\omega,\omega'), d\xi)\) is \((\mathcal {G}_{t})\)-predictable, the random measure \(\nu\) is also the \((\mathcal{G}_{t})\)-compensator of the jump measure of \(J\). Since for every \(\omega\in\varOmega\), \(y\in\mathbb{R}_{+}\) and \(t\geq 0\), the measure \(K_{t}(\omega,y,d\xi)\) is supported by \([-y,\infty)\), the process \(Y^{\perp}\) takes values in \(\mathbb{R}_{+}\).
It remains to show that \(Y^{\perp}\) is of finite activity (equivalently, that \(T_{\infty}=\infty\) \(\widetilde{\mathbb{Q}}\)-a.s.). Since \(g_{m+1}(\xi)\geq1\) for all \(\xi\in\mathbb{R}\) and due to condition (4.16), it holds that
due to the uniform boundedness of the processes \(\{(p^{m+1}_{t}(\cdot ,y))_{t\geq0};y\in\mathbb{R}_{+}\}\). This implies that \(\mu([0,T]\times\mathbb{R})<\infty\) \(\widetilde {\mathbb{Q}}\)-a.s. for each \(T \geq0\), and as a consequence, \(\widetilde{\mathbb{Q}}[T_{\infty}<\infty]=0\). □
Proof of Lemma 4.11
(i) For \(t\geq0\), let \(H\) be a bounded \(\mathcal{H}_{t}\)-measurable random variable, \(F\) a bounded \(\mathcal{F}_{t}\)-measurable random variable and \(A\in\mathcal{F}_{\infty}\). As can be deduced from the proof of Proposition 4.10, the \(\mathcal{F}_{\infty }\)-conditional law of \((\omega'(s))_{s\in[0,t]}\) under \(\widetilde {\mathbb{Q}}\) is \(\mathcal{F}_{t}\)-measurable (compare also with [20, part (iv) of Theorem 5.1]). In particular, this means that \(\mathbb{E}^{\widetilde{\mathbb {Q}}}[H|\mathcal{F}_{\infty}]=\mathbb{E} ^{\widetilde{\mathbb{Q}}}[H|\mathcal{F}_{t}]\). In turn, this implies that
By a monotone class argument, this means that \(\mathbb{E}^{\widetilde {\mathbb{Q}}}[G|\mathcal{F} _{\infty}]=\mathbb{E}^{\widetilde{\mathbb{Q}}}[G|\mathcal{F}_{t}]\) for every bounded \(\mathcal{G} _{t}\)-measurable random variable \(G\). It is well known (see e.g. [38, Proposition 5.9.1.1]) that the latter property is equivalent to the fact that all \((\widetilde {\mathbb{Q}},(\mathcal{F} _{t})_{t\geq0})\)-martingales are also \((\widetilde{\mathbb{Q}},(\mathcal{G}_{t})_{t\geq 0})\) martingales. Due to the fact that \(\widetilde{\mathbb{Q}}|_{\mathcal{F}_{\infty}}=\mathbb{Q}\), this implies that all \((\mathbb{Q},(\mathcal{F} _{t})_{t\geq0})\)-local martingales are also \((\widetilde{\mathbb {Q}},(\mathcal{G}_{t})_{t\geq 0})\)-local martingales. As a consequence, every \((\mathbb{Q},(\mathcal {F}_{t})_{t\geq 0})\)-semimartingale is also a \((\widetilde{\mathbb{Q}},(\mathcal {G}_{t})_{t\geq0})\) semimartingale. Moreover, since semimartingale characteristics can be characterized in terms of local martingales (see e.g. [36, Theorem II.2.21]), this implies that \((X,\hat{Y})\) is a semimartingale with respect to \((\widetilde{\mathbb{Q}},(\mathcal{G}_{t})_{t\geq0})\) with unchanged characteristics.
(ii) Since \(Y^{\perp}\) is a pure jump process, in order to prove its local independence with respect to \((X,\hat{Y})\), it suffices to show that \(Y^{\perp}\) and \((X,\hat{Y})\) never jump together. In view of (C.1), this reduces to showing that
Let \(\mathfrak{T}\) be the set of jump times of \(X\). Since \(X\) is càdlàg, the set \(\mathfrak{T}\) is countable (see e.g. [36, Proposition I.1.32]) and similarly as in [39, Theorem 4.7],
where the last equality follows from the fact that \(\mathbb {E}^{\widetilde{\mathbb{Q}}}[\textbf{1} _{\{\varDelta Y_{t}^{\perp}\neq0\}}|\mathcal{F}_{\infty}]=0\) for all \(t>0\) since due to Proposition 4.10, the jump measure of \(Y^{\perp}\) with respect to the larger filtration \((\overline{\mathcal {G}}_{t})_{t\geq0}\) (which satisfies \(\overline{\mathcal{G}}_{0}=\mathcal{F}_{\infty }\otimes\{\emptyset,\varOmega '\}\)) is absolutely continuous with respect to Lebesgue measure, so that \(Y^{\perp}\) does not have any fixed time of discontinuity (see e.g. [36, Lemma II.2.54]).
In order to prove the last assertion, note that condition (4.16) implies that condition \(I(0,1)\) from [40] is satisfied, since for all \(y\in\mathbb{R}_{+}\), \(T\geq0\) and \(i=1,\ldots,m\),
Moreover, condition (4.16) can be easily shown to imply that
for all \(T\geq0\). Hence, [40, Theorem 3.2] implies that
is a uniformly integrable \((\widetilde{\mathbb{Q}},(\overline {\mathcal{G}}_{t})_{t\geq 0})\)-martingale, for all \(i=1,\ldots,m\). In turn, because \(T\geq0\) is arbitrary, this proves the \((\widetilde{\mathbb{Q}},(\overline {\mathcal{G}}_{t})_{t\geq 0})\)-martingale property of \((\exp(u_{i}Y^{\perp}_{t}-\int _{0}^{t}\varPsi ^{Y^{\perp}}_{s}(u_{i})\,ds))_{t\geq0}\), for all \(i=1,\ldots,m\). Finally, since the latter process is \((\mathcal{G}_{t})\)-adapted, it is also a martingale in the smaller filtration \((\mathcal{G}_{t})_{t\geq0}\). □
Proof of Theorem 4.12
Due to Lemma 4.11, the local exponent of \(\hat{Y}\) with respect to the extended filtered probability space \((\varOmega,\mathcal {G},(\mathcal{G} _{t})_{t\geq0},\widetilde{\mathbb{Q}})\) is still given by \(\varPsi^{\hat {Y}}\) and it holds that \(Y^{\parallel}=(\hat{Y})^{\parallel}=\hat{Y}\). Since \(Y^{\perp}\) and \(Y-Y^{\perp}=\hat{Y}\) are locally independent (see Lemma 4.11), the consistency condition (3.11) directly follows from condition (4.15).
In order to prove the martingale property of the process in (3.10), note first that the \((\widetilde{\mathbb {Q}},(\mathcal{G}_{t})_{t\geq 0})\)-martingale property of \((\exp(u_{i}Y^{\perp}_{t}-\int _{0}^{t}\varPsi ^{Y^{\perp}}_{s}(u_{i})ds))_{t\geq0}\) (see Lemma 4.11), condition (iv) in Definition 4.1 and the local independence of \(Y^{\perp}\) and \((X,\hat{Y})\) on \((\varOmega,\mathcal{G},(\mathcal{G}_{t})_{t\geq0},\widetilde{\mathbb {Q}})\) imply that the process in (3.10) is a \((\widetilde{\mathbb{Q}},(\mathcal {G}_{t})_{t\geq0})\)-local martingale, for all \(i=1,\ldots,m\). Being a nonnegative local martingale, it is also a supermartingale by Fatou’s lemma. Hence, to establish the true martingale property, it suffices to observe that since \(\overline{\mathcal{G}}_{0}=\mathcal {F}_{\infty}\otimes\{ \emptyset,\varOmega'\}\), it holds for all \(T\geq0\) and \(i=1,\ldots,m\) that
where in the last equality, we have used \(\widetilde{\mathbb {Q}}|_{\mathcal{F}}=\mathbb{Q}\) and the \((\mathbb{Q},(\mathcal{F}_{t})_{t\geq0})\)-martingale property of (4.1). □
Rights and permissions
About this article
Cite this article
Cuchiero, C., Fontana, C. & Gnoatto, A. A general HJM framework for multiple yield curve modelling. Finance Stoch 20, 267–320 (2016). https://doi.org/10.1007/s00780-016-0291-5
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00780-016-0291-5
Keywords
- Multiple yield curves
- HJM model
- Semimartingale
- Forward rate agreement
- Libor rate
- Affine processes
- Multiplicative spreads