Interest-Rate Modelling in Collateralized Markets: Multiple curves, credit-liquidity effects, CCPs
aa r X i v : . [ q -f i n . P R ] A p r Interest-Rate Modelling in Collateralized Markets:multiple curves, credit-liquidity effects, CCPs
Andrea Pallavicini ∗ Damiano Brigo † First Version: October 1, 2012. This version: April 5, 2013
Abstract
The market practice of extrapolating different term structures from different instru-ments lacks a rigorous justification in terms of cash flows structure and market observables.In this paper, we integrate our previous consistent theory for pricing under credit, collat-eral and funding risks into term structure modelling, integrating the origination of differentterm structures with such effects. Under a number of assumptions on collateralization,wrong-way risk, gap risk, credit valuation adjustments and funding effects, including thetreasury operational model, and via an immersion hypothesis, we are able to derive a syn-thetic master equation for the multiple term structure dynamics that integrates multiplecurves with credit/funding adjustments.
JEL classification code: G13.AMS classification codes: 60J75, 91B70Keywords:
Yield Curve Dynamics, Multiple Curve Framework, HJM Framework, Inter-est Rate Derivatives, Basis Swaps, Counterparty Credit Risk, Liquidity Risk, Funding Costs,Central Clearing Counterparties. ∗ Imperial College London and Banca IMI Milan, [email protected] † Imperial College London, [email protected] ontents The opinions here expressed are solely those of the authors and do not represent in any way those of theiremployers. . Pallavicini, D. Brigo, Multi-Curve models explained via Credit, Collateral and Funding Liquidity Starting from summer 2007, with the spreading of the credit crunch, market quotes of forwardrates and zero-coupon bonds began to violate standard no-arbitrage relationships. This waspartly due to the liquidity crisis affecting credit lines, and to the possibility of a systemicbreak-down triggered by increased counterparty credit risk. Indeed, credit risk is only onefacet of the problem, since the crisis started as a funding liquidity crisis, as shown for exampleby Eisenschmidt and Tapking [2009], and it continued as a credit crisis following a typicalspiral pattern as described in Brunnermeier and Pedersen [2009].This has been the most dramatic signal for inadequacy of standard financial modellingbased on idealized assumptions on risk-free rates and on unrestricted access to funding in-struments. Removing or relaxing such assumptions opens the door to financial models able toanalyze the inner mechanics of a deal: collateral rules, funding policies, close-out procedures,market fees, among others. Such features require to be included in our pricing framework.A few papers in the financial literature have tried either to re-define the theory fromscratch or to extend the standard framework to the new world. The works of Cr´epey [2011,2012a,b] open the path towards a pricing theory based on different bank accounts, accruingat different rates, and representing the different cash sources one may have at her disposal toimplement the hedging strategies, the funding policies and the margining rules. We cite alsothe works by Piterbarg [2010, 2012], which focus on perfectly collateralized deals and try toreformulate the basic Black Scholes theory, and the works of Burgard and Kjaer [2011a,b].On the other hand, the works of Pallavicini et al. [2011, 2012] stress the fact that all updatedfeatures can be included in terms of modified payoffs rather than through the need of a newand somehow ad-hoc pricing theory. In particular, Pallavicini et al. [2011, 2012] recognizethat a financial contract must include hedging, funding and margining fees as specific andprecisely defined additional cash flows.In this paper we try to distill the above contributions to highlight the common ingredientsthat a new pricing framework should include to be effective in a world where credit andliquidity issues cannot be disregarded. We formulate a term structure theory that is based onthe new pricing framework and that leads quite naturally to multiple curves. The cornerstoneof our analysis will be market procedures and market observables. We avoid employingpricing rules that are not supported by market strategies. For example, funding requires acounterparty willing to lend cash to the investor, hedging requires trading only the instrumentsoffered by the market, and so on. On the other hand, we wish to maintain the discussion assimple as possible and to use well developed pricing and hedging concepts as much as possible,so that we still use some theoretical quantities, such as a risk-neutral measure and a risk freerate. However, we use such unobservables only as instrumental variables and we check thatour final results are independent of them. Our final multiple-curves term structure theory,rigorously integrated in terms of credit, collateral and funding effects, will be based only onmarket observables and market procedures.In particular, we wish to analyze concrete pricing cases such as partial collateralized dealsand contracts cleared by Central Clearing Counterparties (CCP), which is a very sensitivetopic following pressure from Dodd Frank and EMIR/CRD4. See for instance Arnsdorf [2011],Pirrong [2011], Cont et al. [2011], and Heller and Vause [2012].We now move to describing more in detail the content of the paper.In Section 2 we present our pricing framework along with realistic approximations. . Pallavicini, D. Brigo, Multi-Curve models explained via Credit, Collateral and Funding Liquidity
LIBOR is now a rate assigned by the market and not derivedby risk-free forward rate agreements or risk-free one-period swaps .By including credit, collateral and funding effects in valuation we obtain the master equa-tion seen earlier in Pallavicini et al. [2011, 2012]. In the specific case where collateral is afraction of current all-inclusive mark-to-market, we obtain a simpler pricing equation basedon a hedge-funding-fees equivalent measure and on a generalized dividend process inclusiveof default risk, treasury funding and collateral costs. We do not consider collateral gap riskfor interest-rate products, as such risks are not essential for this asset class, as described inBrigo et al. [2011b] (contrary to credit derivatives, as seen in Brigo et al. [2011a]).In Section 3 we apply these approximations to the money market, in order to evaluatecollateralized interest-rate derivatives.We look at defining new building blocks that will replace the old unobservables-basedones (such as risk-free rate zero-coupon bonds, risk-free one-period swap rates, etc). Suchnew instruments are based on the collateral rate, which is an observable rate, since it iscontractually defined by the CSA as the rate to be used in the margining procedure. Thecollateralized zero-coupon bond can be proxied by one-period Overnight Indexed Swaps (orby quantities bootstrapped from multi-period OIS) when we accept to approximate a dailycompounded rate with continuously compounded rates. We then define fair OIS rates atinception in terms of collateralized zero coupon bonds.Thus, this setup allows us to define new LIBOR forward rates as equilibrium rates incollateralized one-period swaps.
The resulting rates depend on the collateralized coupon-bearing bonds. We may alsodefine a collateral based forward measure where LIBOR forward rates are expected valuesof future realized LIBOR rates. We then hint at the development of a market model theoryfor the new collateral-inclusive LIBOR forward rates and to a forward rates theory for theOIS-based instantaneous forward rates.This is where the multiple curve picture finally shows up: we have a curve with LIBORbased forward rates, that are collateral adjusted expectation of LIBOR market rates we take asprimitive rates from the market, and we have instantaneous forward rates that are OIS basedrates.
OIS rates are driven by collateral fees, whereas LIBOR forward rates are driven bothby collateral rates and by the primitive LIBOR market rates.We conclude this section by introducing a dynamical multiple-curve model for OIS andLIBOR rates. We reformulate the parsimonious HJM model by Moreni and Pallavicini [2010,2012] under our new pricing framework.In Section 4 we focus on uncollateralized, partially-collateralized and over-collateralizedcontracts.We remove perfect collateralization and adopt a partial one. This means that now weneed to evaluate the pricing adjustment by including also the corrections coming from thetreasury and hedging strategy funding costs.In particular, we have that
LIBOR forward rates associated to partially collateralized . Pallavicini, D. Brigo, Multi-Curve models explained via Credit, Collateral and Funding Liquidity one-period swap contracts acquire a covariance term that can be interpreted as a convexityadjustment. CCPs can be modelled in the above framework and, in particular, initial margins can beroughly modelled with a collateral fraction larger than unity. This leads to a generalizationof the previous immersion-based general formula.In Section 5 we discuss how to model credit spreads, funding rates and specific assumptionson the fraction of mark-to-market covered by collateral.Credit spreads should be calibrated via Credit Default Swaps or Defaultable Bonds. How-ever, this should be a global calibration because CDS are collateralized and are in principlepriced with the same general formula, inclusive of collateral and funding, that we use for allother deals. Bonds are not collateralized and are funded more heavily, so funding risk is alsothere. This global CDS or Bond calibration is not done usually even though interpretingCDS or Bonds as sources of pure credit risk calibration may lead to important errors, see forexample Fujii and Takahashi [2011b] or Brigo et al. [2011a].The term structure of funding rates depends on the funding policy and is model dependent.Stripping it directly from market liquid instruments is very difficult, especially because theinterbank market is no longer (fully) representative of such costs. Collateral portfolios playnow a key role too. Default intensities, collateral rates and liquidity bases will be key driversof the funding spreads. Finally, we consider using Value-at-Risk measures to determine theeffective fraction of mark-to-market we should hold as collateral, and introduce collateralhaircuts.Our final interest-rate curves are consistently explained by such effects and based onmarket observables.In Section 6 we conclude the discussion and list open points.
After the onset of the crisis in 2007, all market instruments are quoted by taking into account,more or less implicitly, credit and funding liquidity adjustments. Hence we have to carefullycheck standard theoretical assumptions which often ignore credit and liquidity issues. Wemust go back to market data by limiting ourselves to price only derivative contracts weare able to replicate by means of market instruments. The comparison to market data andprocesses is the only means we have to validate our theoretical assumptions, so as to dropthem if in contrast with observations.Our general results are not linked to a particular asset class, but in this paper we focusas main example on interest-rate derivatives, as a paradigmatic case.
Classical interest-rate models were formulated to satisfy no-arbitrage relationships by con-struction, which allow to hedge forward-rate agreements in terms of risk-free zero-couponbonds.Starting from summer 2007, with the spreading of the credit crunch, market quotes offorward rates and zero-coupon bonds began to violate usual no-arbitrage relationships. Themain driver of such behavior was the liquidity crisis reducing the credit lines along with thefear of an imminent systemic break-down. As a result the impact of counterparty risk onmarket prices could not be considered negligible any more. . Pallavicini, D. Brigo, Multi-Curve models explained via Credit, Collateral and Funding Liquidity F t ( T , T ) = 1 T − T (cid:18) P t ( T ) P t ( T ) − (cid:19) , where P t ( T ) is a zero coupon bond price at time t for maturity T , and F is the related LIBORforward rate. A direct consequence is the impossibility to describe all LIBOR rates in termsof a unique zero-coupon yield curve. Indeed, since 2009 and even earlier, we had evidencethat the money market for the Euro area was moving to a multi-curve setting. See Ametranoand Bianchetti [2009], Henrard [2009], Kijima et al. [2009], Mercurio [2009, 2010], Pallaviciniand Tarenghi [2010], and Fujii and Takahashi [2011a].A further example of evolving assumptions is given by the use of collateralized contracts.The growing attention on counterparty credit risk is transforming OTC derivatives moneymarkets. An increasing number of derivative contracts is cleared by CCPs, while most of theremaining contracts are traded under collateralization, regulated by Credit Support Annex(CSA). Both cleared and CSA deals require collateral posting, as default insurance, alongwith its remuneration. We cannot neglect such effects.As a consequence, once a derivative contract is closed, we must consider the cash flowsassociated to the collateral margining procedure. Thus, a zero-coupon contract behaves as adividend paying asset, and its discounted price is no longer a martingale, namely V ZC t B t = E t (cid:20) V ZC T B T (cid:21) , where V ZC t is the zero-coupon contract’s price and B is the risk-free rate bank account, thenumeraire of the risk neutral measure. In order to price a financial product (for example a derivative contract), we have to discountall the cash flows occurring after the trading position is entered. We can group them asfollows:1. product cash flows (e.g. coupons, dividends, premiums, etc.) inclusive of hedginginstruments cash flows;2. cash flows required by the collateral margining procedure;3. cash flows required by the funding and investing (borrowing and lending) procedures;4. cash flows occurring upon default (close-out procedure).We refer to the two names involved in the financial contract and subject to default riskas investor (also called name “I”, usually the bank) and counterparty (also called name “C”, . Pallavicini, D. Brigo, Multi-Curve models explained via Credit, Collateral and Funding Liquidity τ I ,and τ C respectivelythe default times of the investor and counterparty. We fix the portfolio time horizon T > , G , Q ), with a filtration ( G t ) t ∈ [0 ,T ] such that τ C , τ I are ( G t ) t ∈ [0 ,T ] -stopping times. We denote by E t [ · ] the conditional expectation under Q given G t , and by E τ i [ · ] the conditional expectation under Q given the stopped filtration G τ i . Weexclude the possibility of simultaneous defaults, and define the first default event between thetwo parties as the stopping time τ := τ C ∧ τ I . We will also consider the market sub-filtration ( F t ) t ≥ that one obtains implicitly by assuminga separable structure for the complete market filtration ( G t ) t ≥ . G t is then generated by thepure default-free market filtration F t and by the filtration generated by all the relevant defaulttimes monitored up to t (see for example Bielecki and Rutkowski [2002]).The price ¯ V t of a derivative, inclusive of collateralized credit and debit risk (CVA andDVA), margining costs, and funding and investing costs can be derived by following Pallaviciniet al. [2011, 2012], and is given by the following master equation:¯ V t ( C, F ) = E t [ Π( t, T ∧ τ ) + γ ( t, T ∧ τ ; C ) + ϕ ( t, T ∧ τ ; F ) ] (1)+ E t (cid:2) { t<τ We consider r t and therelated numeraire B t and measure Q as unobservables, and will derive the theory based onthem, showing that the final equations will only depend on market observables. . Pallavicini, D. Brigo, Multi-Curve models explained via Credit, Collateral and Funding Liquidity C t and of the funding cash account F t , while the close-out amount ε t is defined by the CSA that has been agreed between the deal parties. Common marketprocedures, as we will see later on, may link the values of such processes to the price of theproduct itself, transforming the previous definition of Equation (1) into a recursive equation.This feature is hidden in simplified approaches based on adding a spread to the discountcurve to accommodate collateral and funding costs. A different approach is followed byCr´epey [2011], who extends the usual risk-neutral evaluation framework to include manycash accounts accruing at different rates. Despite the different initial approach, a structurethat is similar to our result above for the derivative price is obtained as a solution of abackward SDE. The results of Pallavicini et al. [2011, 2012] allow to define in an explicit way coupons andcosts in terms of the collateral process C t and close-out value ε τ . The authors proceed bydefining a sequence of margining, funding and hedging operations, and account for their costs.Such operations are taken on discrete time-grids.Since we are going to apply the Pallavicini et al. [2011, 2012] master equation to interest-rate derivatives, where collateralization usually happens on a daily basis, and where gap riskis not large, we prefer to present such results on a continuous time-grid. Furthermore, weassume that collateral re-hypothecation is allowed, as done in practice. See Brigo et al. [2011b]for a discussion on re-hypothecation. We writeΠ( t, u ) = Z ut dπ v D ( t, v ) , γ ( t, u ; C ) = Z ut dv ( r v − ˜ c v ) C v D ( t, v ) ,ϕ ( t, u ; F, H ) = Z ut dv ( r v − ˜ f v )( ¯ V v ( C, F ) − C v ) D ( t, v ) − Z ut dv ( r v − ˜ h v ) H v D ( t, v ) , where π t is the derivative coupon process, and the collateral, funding and market rates aredefined as˜ c t := c + t { C t > } + c − t { C t < } , ˜ f t := f + t { F t > } + f − t { F t < } , ˜ h t := h + t { H t > } + h − t { H t < } , with c ± defined in the CSA contract, f ± defined by the treasury, and h ± equal to the growthrate of market instruments traded to implement the hedging strategy, see Pallavicini et al.[2012] for full details.We can plug the above definitions into equation (1) and, by following Pallavicini et al.[2012], we are able to write the following continuous-time pricing equation.¯ V t ( C ; F ) := Z Tt E ˜ ht h (cid:0) { u<τ } dπ u + 1 { τ ∈ du } θ u ( C, ε ) (cid:1) D ( t, u ; ˜ f ) i (2)+ Z Tt du E ˜ ht h { u<τ } ( ˜ f u − ˜ c u ) C u D ( t, u ; ˜ f ) i . Pallavicini, D. Brigo, Multi-Curve models explained via Credit, Collateral and Funding Liquidity θ τ ( C, ε ) := ε τ − { τ C <τ I } L GD C ( ε τ − C τ − ) + − { τ I <τ C } L GD I ( ε τ − C τ − ) − , where we use the notation x − = min( x, { τ C <τ I } L GD C ( ε τ − C τ − ) + originatesthe CVA type term after collateralization, with 1 { τ I <τ C } L GD I ( ε τ − C τ − ) − originating the DVAtype term instead. The above expectations are taken under a pricing measure Q ˜ h under whichthe underlying risk factors grow at a rate ˜ h , and the discount factors are defined as given by D ( t, T ; x ) := exp (cid:26) − Z Tt du x u (cid:27) . Notice that the above pricing equation 2 is not suited for explicit numerical evaluations,since the right-hand side is still depending on the derivative price via the indicators within thecollateral, funding and market rates. We could resort to numerical solutions, as in Cr´epeyet al. [2012a], but, since our goal is pricing interest-rate derivatives, we prefer to furtherspecialize the pricing equation for such deals.In order to further specialize our master equation 2, we aim to write the price as anexpectation over the derivative discounted cash flows. This is achieved by adopting a fewfurther assumptions. In particular, we assume that gap risk is not present, and we considera particular form for collateral and close-out prices, namely C t . = α t ¯ V t ( C, F ) , ε τ . = ¯ V τ ( C, F )with 0 ≤ α t ≤ 1. This means that(i) Collateral is a fraction α t of the all-inclusive mark-to-market.(ii) Close-out is the all-inclusive mark-to-market at the first default time.This approach is considered in Biffis et al. [2011] and in the longevity-risk chapter of Brigoet al. [2012]. An alternative approximation, not imposing a proportionality between theaccount pricing processes, and suited for contracts with gap risk, such as CDS, can be foundin Pallavicini [2013].We obtain, by switching to the market filtration F ,¯ V t ( C ; F ) = 1 { τ>t } Z Tt E ˜ ht h dπ u D ( t, u ; ˜ f + λ ) (cid:12)(cid:12) F i (3) − { τ>t } Z Tt du E ˜ ht h (˜ ζ u − λ u ) ¯ V u ( C, F ) D ( t, u ; ˜ f + λ ) (cid:12)(cid:12) F i with E t [ ·|F ] := E [ ·|F t ], and we define˜ ζ t := (1 − α t ) (cid:16) λ C Exotic interest-rate products maybe hedged with a strategy containing (one-period) OIS contracts. In particular, the corre-sponding borrowing/lending rate ˜ h t will be equal to the growth rate of the (one-period) OIScontract ˜ h OIS t ¯ V OIS t dt . = E ˜ ht (cid:2) d ¯ V OIS t (cid:12)(cid:12) F (cid:3) = e t ¯ V OIS t dt where we notice that we obtain a symmetric rate, namely the same for going long or short inOIS contracts when hedging. The extension to multi-period contracts is straightforward. Notice that we are only defining a price process for hypothetical collateralized zero-coupon bond. We arenot assuming that collateralized bonds are assets traded on the market. . Pallavicini, D. Brigo, Multi-Curve models explained via Credit, Collateral and Funding Liquidity LIBOR rates ( L t ( T )) are the indices used as reference rate for many collateralized interest-rate derivatives (IRS, basis swaps, . . . ). In particular we consider Interest Rate Swaps (IRS).IRS contracts swap a fix-payment leg with a floating leg paying simply compounded LIBORrates. IRS contracts are collateralized at overnight rate e t . Thus, a one-period IRS payoffwith maturity T and tenor x is given by xK − xL T − x ( T )where K is the fix rate payed by the IRS. Furthermore, we can introduce the (par) fix rates F t ( T, x ; e ) which render the one-period IRS contract fair, i.e. priced at zero. They areimplicitly defined as ¯ V IRS t ( K ) := E ˜ ht (cid:2) ( xK − xL T − x ( T )) D ( t, T ; e ) (cid:12)(cid:12) F (cid:3) with ¯ V IRS t ( F t ( T, x ; e )) = 0leading to F t ( T, x ; e ) := E ˜ ht (cid:2) L T − x ( T ) D ( t, T ; e ) (cid:12)(cid:12) F (cid:3) E ˜ ht (cid:2) D ( t, T ; e ) (cid:12)(cid:12) F (cid:3) = E ˜ ht (cid:2) L T − x ( T ) D ( t, T ; e ) (cid:12)(cid:12) F (cid:3) P t ( T ; e )The above definition may be simplified by a suitable choice of the measure under which wetake the expectation. In particular, we can consider the following Radon-Nikodym derivative d Q ˜ h,T ; e d Q ˜ h (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) t := E ˜ ht (cid:2) D (0 , T ; e ) (cid:12)(cid:12) F (cid:3) = D (0 , t ; e ) P t ( T ; e )which is a positive Q ˜ h -martingale. We name the corresponding equivalent measure the col-lateralized T -forward measure E ˜ h,T ; et [ · ].Thus, for any payoff φ T , perfectly collateralized at overnight rate e t , we can express pricesas expectations under the collateralized T -forward measure, and we get E ˜ ht (cid:2) φ T D ( t, T ; e ) (cid:12)(cid:12) F (cid:3) = P t ( T ; e ) E ˜ h,T ; et (cid:2) φ T (cid:12)(cid:12) F (cid:3) In particular, we can write LIBOR forward rates as F t ( T, x ; e ) := E ˜ ht (cid:2) L T − x ( T ) D ( t, T ; e ) (cid:12)(cid:12) F (cid:3) E ˜ ht (cid:2) D ( t, T ; e ) (cid:12)(cid:12) F (cid:3) = E ˜ h,T ; et (cid:2) L T − x ( T ) (cid:12)(cid:12) F (cid:3) (9)One-period forward rates F t ( T, x ; e ), along with multi-period ones (swap rates), are ac-tively traded on the market. Once collateralized zero-coupon bonds are derived, we canbootstrap forward rate curves from such quotes. See, for instance, Ametrano and Bianchetti[2009] or Pallavicini and Tarenghi [2010] for a discussion on bootstrapping algorithms. . Pallavicini, D. Brigo, Multi-Curve models explained via Credit, Collateral and Funding Liquidity Remark 3.2. (Hedging by means of IRS contracts). Exotic interest-rate products maybe hedged with a strategy containing (one-period) IRS contracts. In particular, the correspond-ing borrowing/lending rate ˜ h t will be equal to the growth rate of the IRS contract ˜ h IRS t ¯ V IRS t dt . = E ˜ ht (cid:2) d ¯ V IRS t (cid:12)(cid:12) F (cid:3) = e t ¯ V IRS t dt where we notice that we obtain a symmetric rate, namely the same for going long or short in(one-period) IRS contracts when hedging. The extension to multi-period contracts is straight-forward. Our aim is to setup a multiple-curve dynamical model starting from collateralized zero-couponbonds P t ( T ; e ), and LIBOR forward rates F t ( T, x ; e ). As we have seen we can bootstrap theinitial curves for such quantities from the market. Now, we wish to propose a dynamics thatpreserves the martingale properties satisfied by such quantities.Thus, without loss of generality, we can define collateralized zero-coupon bonds under the Q ˜ h measure as dP t ( T ; e ) P t ( T ; e ) = e t dt − α t ( T ; e ) · dW ˜ ht and LIBOR forward rates under Q ˜ h,T ; e measure as dF t ( T, x ; e ) = β t ( T, x ; e ) · dZ ˜ h,T ; et where W s and Z s are correlated standard Brownian motions with correlation matrix ρ , andthe volatility processes α and β may depend on bonds and LIBOR forward rates themselves.The following definition of f t ( T, e ) is not strictly necessary, and we could keep workingwith bonds P t ( T ; e ), using their dynamics. However, as it is usual in interest rate theoryto model rates rather than bonds, we may try to formulate quantities that are closer to thestandard HJM framework. In this sense we can define instantaneous forward rates f t ( T ; e ),by starting from (collateralized) zero-coupon bonds, as given by f t ( T ; e ) := − ∂∂T log P t ( T ; e )We can derive instantaneous forward-rate dynamics by Itˆo lemma, and we get under Q ˜ h,T ; e measure df t ( T ; e ) = σ t ( T ; e ) · dW ˜ h,T ; et , σ t ( T ; e ) := ∂∂T α t ( T ; e )Hence, we can summarize our modelling assumptions in the following way.1. We choose to hedge exotic options in term of linear collateralized products (OIS, IRS,. . . ) which we assume to be perfectly collateralized;2. Since such linear products can be expressed in term of simpler quantities, namely col-lateralized zero-coupon bonds P t ( T ; e ) and LIBOR forward rates F t ( T, x ; e ), we focuson their modelling.3. Initial term structures for collateralized products may be bootstrapped from marketdata. . Pallavicini, D. Brigo, Multi-Curve models explained via Credit, Collateral and Funding Liquidity df t ( T ; e ) = σ t ( T ; e ) · dW ˜ h,T ; et , dF t ( T, x ; e ) = β t ( T, x ; e ) · dZ ˜ h,T ; et (10)As we explained in the introduction, this is where the multiple curve picture finally showsup: we have a curve with LIBOR based forward rates F t ( T, x ; e ), that are collateral adjustedexpectation of LIBOR market rates L T − x ( T ) we take as primitive rates from the market, andwe have instantaneous forward rates f t ( T ; e ) that are OIS based rates. OIS rates f t ( T ; e )are driven by collateral fees, whereas LIBOR forward rates F t ( T, x ; e ) are driven both bycollateral rates and by the primitive LIBOR market rates.Then, we wish to stress an important property of the above dynamics, namely that it doesnot depend on unobservable rates such as the risk-free rate r t .Once this is done, the framework for multiple curves is ready and we may start populatingthe specific dynamics with modelling choices. We propose a possible choice in the followingsection. Remark 3.3. (Introducing a Collateral Measure). When we price exotic interest-ratederivatives and we hedge by means of OIS or IRS products (we can generalize the argumentto other basic money-market products such as FRA or others), we obtain that the borrow-ing/lending rate ˜ h t is equal to the overnight rate e t , so that we gets ˜ h t . = ˜ c t . = e t . Hence, under the previous assumptions we can write the pricing equation (6) as ¯ V t ( C ; F ) = 1 { τ>t } Z Tt E et (cid:2) dπ u D ( t, u ; e ) (cid:12)(cid:12) F (cid:3) and we can speak of pricing under a “collateralized measure”. The above equation is formallyequal to the pricing equation of the Black and Scholes theory, see Piterbarg [2012], but here (i)the collateral rate is specified by the CSA contract and it can be risky; (ii) the pricing measureis introduced by means of the Feynman-Kac theorem starting from a formulation of the theoryunder the risk-neutral measure; (iii) the risk-free rate is not a rate which can be observedon the market. To avoid confusion we prefer not to explicitly introduce such collateralizedmeasure. We can now specialize our modelling assumptions to define a model for interest-rate derivativeswhich is flexible enough to calibrate the quotes of the money market, but yet robust. Ouraim is to to define a HJM framework to describe with a single family of Markov processesall the yield curves we are interested in. We follow Moreni and Pallavicini [2010, 2012] byreformulating their theory under the Q ˜ h measure.In the literature many authors proposed generalizations of the HJM framework starting,see for instance Cheyette [2001], Andersen and Andreasen [2002], Carmona [2004], Andreasen[2006], or Chiarella et al. [2010]. In particular, in recent papers Mart`ınez [2009], Fujii et al.[2010], Cr´epey et al. [2012b] extended the HJM framework to deal with multiple-yield curves.See also Mercurio [2010], Mercurio and Xie [2012] for references on LIBOR Market Model.Let us summarize the basic requirements our model must fulfill (in this section we omitthe dependence on e or ˜ h to lighten the notation): . Pallavicini, D. Brigo, Multi-Curve models explained via Credit, Collateral and Funding Liquidity f t ( T );ii) existence of LIBOR rates, typical underlying of traded derivatives, with associatedforwards F t ( T, x );iii) no arbitrage dynamics of the f t ( T ) and the F t ( T, x ) (both being T -forward measuremartingales);iv) possibility of writing both the f t ( T ) and the F t ( T, x ) as function of a common familyof Markov processes, so that we are able to build parsimonious yet flexible models.We stress that our approach models only quantities which can bootstrapped in a modelindependent way from market quotes, and it includes natively the margining procedure withinpricing equations. We choose under Q ˜ h,T ; e measure, the following dynamics. df t ( T ) = σ t ( T ) · dW Tt (11) dF t ( T, x ) k ( T, x ) + F t ( T, x ) = Σ t ( T, x ) · dW Tt where we introduce the families of (stochastic) volatility processes σ t ( T ) and Σ t ( T, x ) , thevector of independent Q ˜ h,T ; e -Brownian motions W Tt , and the set of deterministic shifts k ( T, x ) , such that k ( T, x ) ≈ /x if x ≈ 0. We bootstrap f ( T ) and F ( T, x ) from market quotes.In order to satisfy requirement iv), getting a model with a reduced number of commondriving factors in the spirit of HJM approaches, it is sufficient to conveniently tie together thevolatility processes σ t ( T ) and Σ t ( T, x ) as in Moreni and Pallavicini [2010]. In doing so, weextend the single-curve HJM approach of Cheyette [1992], Babbs [1993], Carverhill [1994], andRitchken and Sankarasubramanian [1995]. First, we consider a common parametrization forrisk-free and LIBOR forward rate volatilities by introducing the family of stochastic processes σ t ( u ; T, x ) such that σ t ( T ) := σ t ( T ; T, , Σ t ( T, x ) := Z TT − x du σ t ( u ; T, x ) . (12)Second, we impose the separability constraint σ t ( u ; T, x ) := v t · ( q ( u ; T, x ) g ( t, u )) , g ( t, u ) := exp (cid:26) − Z ut ds a ( s ) (cid:27) , q ( u ; u, 0) := 1 , (13)where v t is a matrix adapted process, q ( u ; T, x ) and a ( t ) are deterministic array functions.The condition on q ( u ; T, x ) when T = u ensures that the standard HJM fulfills the usualRitchen-Sankarasubramanian’s separability condition. By plugging the expression for thevolatility into Equation (11), it is possible to work out the expression ending up with therepresentationln (cid:18) k ( T, x ) + F t ( T, x ) k ( T, x ) + F ( T, x ) (cid:19) = G ∗ ( t, T − x, T ; T, x ) · (cid:18) X t + Y t · (cid:18) G ( t, t, T ) − G ( t, T − x, T ; T, x ) (cid:19)(cid:19) , (14) . Pallavicini, D. Brigo, Multi-Curve models explained via Credit, Collateral and Funding Liquidity X t and the auxiliary matrix process Y t under Q ˜ h as given by X t := N X k =1 Z t ( h s g ( s, t )) · (cid:18) dW s + h s Z ts dv g ( s, v ) ds (cid:19) Y t := Z t ds ( h s g ( s, t )) · ( h s g ( s, t ))with X = 0 and Y = 0, as well as the vectorial deterministic functions G ( t, T , T ) := Z T T dv g ( t, v ) , G ( t, T , T ; T, x ) := Z T T dv q ( v ; T, x ) g ( t, v )It is worth noting that the integral representation of forward LIBOR volatilities givenby Equation (12), together with the common separability constraint given in Equation (13)are sufficient conditions to ensure the existence of a reconstruction formula for all OIS andLIBOR forward rates based on the very same family of Markov processes.In order to model implied volatility smiles, we can add a stochastic volatility process toour model. A tractable choice is to model the matrix process h t by means of a square-rootprocess as in Moreni and Pallavicini [2012]. In practice we set h t := √ v t R where R is a uppertriangular matrix, and the variance v t is a vector process whose dynamics under Q ˜ h measureis given by dv t = κ ( θ − v t ) dt + ν √ v t dZ t , v = ¯ v where κ , θ , ν ,¯ v are constant deterministic vectors, and Z t is a vector of independent Brownianmotions correlated to the W t processes as ρ ij dt := d h Z i , W j i t where ρ is a deterministic correlation matrix.We address the reader for calibration and pricing examples to Moreni and Pallavicini [2010]for the cap/floor market, and to Moreni and Pallavicini [2012] for the swaption market. In the previous section we have seen that liquid instruments quoted on the money marketcan be considered as fully collateralized. The HJM framework we have developed so far isbased on such assumption, and it allows us to calculate prices of fully-collateralized derivativecontracts. On the other hand, in practice we have often the need to evaluate uncollateralizedor partially-collateralized contracts, since many counterparties, in particular corporates orsmaller banks, may have difficulties to subscribe to standard CSA agreements, which requirea reserve of eligible assets to be used in the margining procedure. We need also to evaluateover-collateralized deals when dealing with CCPs because of initial margin, as we shall see inthe following sections. . Pallavicini, D. Brigo, Multi-Curve models explained via Credit, Collateral and Funding Liquidity When we trade an exotic interest-rate derivative contract we can hedge interest-rate risks bygoing long or short in money-market liquid instruments. On the money market the liquidcontracts are usually collateralized on a daily basis at overnight rate as we have seen inSection 3. Thus, whatever the collateralization procedure of the exotic contract is, we assumeto implement the hedging strategy by means of overnight collateralized contracts.Hence, prices can be calculated by means of Equation (5), namely we have¯ V t ( C ; F ) = 1 { τ>t } Z Tt E ˜ ht h dπ u D ( t, u ; ˜ f + ˜ ζ ) (cid:12)(cid:12) F i where the risk factors grow at rate ˜ h , which is equal to the overnight rate e t when the riskfactor is one of the money-market collateralized contracts (for instance the one-period OIS orIRS). Notice that, in general, the collateral accrual rate ˜ c t , entering the pricing equation dueto the definition of ˜ ζ t given by Equation (4), may be different from the overnight rate e t . We can apply the pricing formula (5) to evaluate the money market products we introduce inthe previous section, namely interest-rate swaps. Here, for example, we consider the par rate¯ F t ( T, x ; e ) for a (partially-collateralized) single-period IRS. Such rate implicitly defined as1 { τ>t } E ˜ ht h (cid:0) x ¯ F t ( T, x ; e ) − xL T − x ( T ) (cid:1) D ( t, T ; ˜ f + ˜ ζ ) (cid:12)(cid:12) F i = 0The above definition can be simplified by moving to Q ˜ h,T ; e measure applying Equation (5).1 { τ>t } E ˜ h,T ; et h (cid:0) x ¯ F t ( T, x ; e ) − xL T − x ( T ) (cid:1) D ( t, T ; ˜ f + ˜ ζ − e ) (cid:12)(cid:12) F i = 0leading for τ > t to ¯ F t ( T, x ; e ) := E ˜ h,T ; et (cid:2) L T − x ( T ) D ( t, T ; ˜ q ) (cid:12)(cid:12) F (cid:3) E ˜ h,T ; et (cid:2) D ( t, T ; ˜ q ) (cid:12)(cid:12) F (cid:3) (15)where we define the effective dividend rate˜ q t := ˜ f t + ˜ ζ t − e t (16)which includes the effects of credit risk, funding, and the mismatch in collateralization betweenthe exotic deal and the instruments used in its hedging strategy.We can now express the par rates in term of collateralized LIBOR rates, we have for τ > t ¯ F t ( T, x ; e ) = E ˜ h,T ; et (cid:2) L T − x ( T ) D ( t, T ; ˜ q ) (cid:12)(cid:12) F (cid:3) E ˜ h,T ; et (cid:2) D ( t, T ; ˜ q ) (cid:12)(cid:12) F (cid:3) = E ˜ h,T ; et (cid:2) L T − x ( T ) (cid:12)(cid:12) F (cid:3) E ˜ h,T ; et (cid:2) D ( t, T ; ˜ q ) (cid:12)(cid:12) F (cid:3) + Cov ˜ h,T ; et [ L T − x ( T ) D ( t, T ; ˜ q ) ] E ˜ h,T ; et (cid:2) D ( t, T ; ˜ q ) (cid:12)(cid:12) F (cid:3) = F t ( T, x ; e ) E ˜ h,T ; et (cid:2) D ( t, T ; ˜ q ) (cid:12)(cid:12) F (cid:3) + Cov ˜ h,T ; et [ F T − x ( T, x ; e ) D ( t, T ; ˜ q ) ] E ˜ h,T ; et (cid:2) D ( t, T ; ˜ q ) (cid:12)(cid:12) F (cid:3) . Pallavicini, D. Brigo, Multi-Curve models explained via Credit, Collateral and Funding Liquidity F t ( T, x ; e ) = F t ( T, x ; e )(1 + γ t ( T, x ; e )) (17)where γ t ( T, x ; e ) is the partial collateralization convexity adjustment given by γ t ( T, x ; e ) := Cov ˜ h,T ; et [ F T − x ( T, x ; e ) , D ( t, T ; ˜ q ) ] F t ( T, x ; e ) E ˜ h,T ; et (cid:2) D ( t, T ; ˜ q ) (cid:12)(cid:12) F (cid:3) . (18)Hence, the price of a partially-collateralized one-period IRS contract paying a fix rate K can be calculated as¯ V IRS t := 1 { τ>t } E ˜ ht h ( xK − xL T − x ( T )) D ( t, T ; ˜ f + ˜ ζ ) (cid:12)(cid:12) F i (19)= 1 { τ>t } x (cid:0) K − ¯ F t ( T, x ; e ) (cid:1) ¯ P t ( T ; e )where we have defined for convenience for τ > t the (partially) collateralized adjusted zero-coupon bond as ¯ P t ( T ; e ) := P t ( T ; e ) E ˜ h,T ; et (cid:2) D ( t, T ; ˜ q ) (cid:12)(cid:12) F (cid:3) (20)These results can be straightforwardly extended to multi-period contracts.Notice that partially collateralized zero-coupon bonds and forward rates depend on theprice process of the contract paying them. Thus, they have different values for differentcontracts. We can interpret them respectively as a per-contract Z -spread-adjusted bonds andconvexity-adjusted forward rates. Thus, when collateralization is not perfect, we obtain thateach contract has its own curve. Funding costs may arise also in collateralized contract, if they are closed with a Central Clear-ing Counterparty (CCP). Indeed, CCP requires that counterparties post an initial margin toclose the deal. Along with the initial margin CCPs require also regular posting of collaterals(variation margin) to match the mark-to-market variation of the deal. The analysis of theimpact of variation margin procedures can be found in Cont et al. [2011], where convexityadjustments and NPV effects are discussed for different clearing houses. These two effects,in the case of one-period contracts, correspond respectively to the adjustments of LIBORforward rates, given in Equation (17), and to the adjustment in zero-coupon bonds, given inEquation (20). In this section, we focus on initial margins and their contribution to fundingcosts.Initial margins are collected to hedge potential future counterparty exposures resulting inreplacement costs. In particular, a CCP is vulnerable to losses on defaulting counterparty ex-posures between the time of the last variation margin payment of the defaulting counterpartyand close-out valuation. This is known as margin period of risk, see Heller and Vause [2012].Furthermore, initial margin may cover gap risks. Initial margins are calculated at contractinception, and, then, possibly updated during the deal life-time. The amount of initial margincan be estimated by a CCP according to Value-at-Risk or expected shortfall risk measure.We can deal with initial margin by assuming that an over-collateralization is requiredby the contract, namely we can generalize the argument leading to Equation(3) to allow for . Pallavicini, D. Brigo, Multi-Curve models explained via Credit, Collateral and Funding Liquidity α t > 1. By a direct calculation starting from Equation (2) we obtain¯ V t ( C ; F ) = 1 { τ>t } Z Tt E ˜ ht h dπ u D ( t, u ; ˜ f + ˜ ξ ) (cid:12)(cid:12) F i (21)where we define˜ ξ t := (1 − α t ) + (cid:16) λ C Even more problematic is the funding rate modelling.Funding rates are determined by the Treasury department according to the Bank fundingpolicies. Thus, a term structure of funding rates is known, but it is far from being unam-biguously derived from market quotes. For instance, long maturities in the term structure offunding rates are calculated by rolling over short-term funding positions, and not by enteringinto a long-term funding position. It is very difficult to forecast the future strategies followedby the Treasury. Thus, the term structure of funding rates is model-depend. The optionmarket (e.g. contingent funding derivatives) is missing.A tempting possibility is using the LIBOR rates as a proxy of funding rates. This choiceis widely spread, but it is very problematic, since it implies that the funding policies ofthe Treasury department is based on inter-bank deposits (not to speak of possible frauds inLIBOR published rates). After the crisis only a small part of funding comes from this source:the main source of funding for an investment Bank is the collateral portfolio which is mainlydriven by the credit spreads of the underlying names.Once a sensible model is defined for the funding policy we should consider that the Trea-sury department may implement a maturity transformation policy along with a fund transferprice (FTP) process. Such procedures may alter in a significant way the funding rate.Here, we wish to select a sensible choice for the dynamics of funding and investing ratesto perform numerical simulations. A possibility is to use for the funding rate f + t an affinecombination of the investor’s default intensity λ It and the average default intensity λ Pt of apool of financial names, representing the names underlying the collateral portfolio. Such anaverage can be obtained from CDS quotes or from index proxies, such as i-Traxx financialsub-index. While for the investing rate f − t we can use a different affine combination which . Pallavicini, D. Brigo, Multi-Curve models explained via Credit, Collateral and Funding Liquidity λ Pt . Thus, we can define the funding and investing rates as given by f − t := e t + w − ( t ) + w P ( t ) λ Pt (23)and f + t := e t + w + ( t ) + w P ( t ) λ Pt + w I ( t ) λ It (24)where e t is the overnight rate, and the w ’s are deterministic functions of times, which canbe calibrated to Treasury data and bond/CDS basis. Moreover, the w ’s define in an implicitway the correlation between the funding/investing rates, the overnight rate and the defaultintensities. The last unknown process to be modelled is the collateral fraction. We consider three realisticcases: (i) no collateralization, (ii) perfect collateralization, and (iii) CCP collateralization withinitial margin calculated with VaR risk measure. The first two cases are simply obtained bysetting respectively α t . = 0 and α t . = 1. The third case is more elaborated, since we shouldassume C t := α t ¯ V t ( C ; F ) . = ¯ V t ( C ; F )+ (cid:16) − Q − ¯ V t ( C ; F ) ( q ) (cid:17) + { ¯ V t ( C ; F ) > ( − Q − ¯ Vt ( C ; F ) ( q ) ) + } + (cid:16) Q ¯ V t ( C ; F ) ( q ) (cid:17) − { ¯ V t ( C ; F ) < ( Q ¯ Vt ( C ; F ) ( q ) ) − } leading to α t . = 1 + ς + t ( q ) + ς − t ( q ) (25)where we introduce the collateral haircuts ς ± t ( q ) ∈ [0 , 1) as given by ς ± t ( q ) := ∓ ( Q ∓ ¯ V t ( C ; F ) ( q )) − ¯ V t ( C ; F ) 1 {∓ ¯ V t ( C ; F ) < ( Q ∓ ¯ Vt ( C ; F ) ( q )) − } (26)while the quantile function Q for a random variable X at a particular level q is defined as Q X ( q ) := inf { x : q < P { X < x }} and P is the physical probability.Yet, this approach can be difficult to follow within a risk-neutral pricing framework. Thus,we suggest an alternative approach which focuses on prices instead of considering probabilities.Such approach can be used to gauge the impact of funding the initial margin, but to exactlymatch the algorithm used by the CCP we should resort to Equation (25).The meaning of the haircuts ς ± t ( q ), introduced starting from quantiles, is to cover thesurviving party from settlement liquidity risk, namely from the risk that the replacementdeal will be close some days after the default event, so that its mark-to-market may bedifferent from the loss evaluated at the default event. . Pallavicini, D. Brigo, Multi-Curve models explained via Credit, Collateral and Funding Liquidity C t . = 1 { ¯ V t ( C ; F ) > } E ˜ ht h max { ¯ V t ( C ; F ) , ¯ V t + δ ( C ; F ) D ( t, t + δ ; ˜ f ) } i + 1 { ¯ V t ( C ; F ) < } E ˜ ht h min { ¯ V t ( C ; F ) , ¯ V t + δ ( C ; F ) D ( t, t + δ ; ˜ f ) } i = ¯ V t ( C ; F ) E ˜ ht " ¯ V t + δ ( C ; F ) D ( t, t + δ ; ˜ f )¯ V t ( C ; F ) − ! + where δ is the margin period of risk, see Brigo et al. [2011b], and can range from few days upto 20 days. In term of collateral fraction we obtain α t . = 1 + ς t (27)where we introduce the collateral haircut ς t > ς t := E ˜ ht " ¯ V t + δ ( C ; F ) D ( t, t + δ ; ˜ f )¯ V t ( C ; F ) − ! + Then, in order to avoid non-realistic values for the collateral haircut, and to mimic thequantile-dependent version, we choose the following definition which ensures ς ∈ [0 , ς t := 1 + E ˜ ht " ¯ V t + δ ( C ; F ) D ( t, t + δ ; ˜ f )¯ V t ( C ; F ) − ! + − ! − (28) We model multiple LIBOR and OIS based interest rate curves consistently, based only onmarket observables and by consistently including credit, collateral and funding effects. Furtherwork includes a more realistic model for initial margins in CCP’s, extending our analysis to gaprisk, collateralization in different currencies, and more realistic collateral processes inclusiveof minimum transfer amounts, thresholds and haircuts. Also, cases where collateral is notcash can be considered. Acknowledgments We are grateful to Marco Bianchetti, Antonio Castagna, Marc Henrard, Nicola Moreni,Daniele Perini and Giulio Sartorelli for helpful discussions. . Pallavicini, D. Brigo, Multi-Curve models explained via Credit, Collateral and Funding Liquidity References F. M. Ametrano and M. Bianchetti. Bootstrapping the illiquidity: Multiple yield curvesconstruction for market coherent forward rates estimation. In F. Mercurio, editor, ModelingInterest Rates: Latest Advances for Derivatives Pricing . Risk Books, 2009.L. Andersen and J. Andreasen. Volatile volatilities. Risk Magazine , 12, 2002.J. Andreasen. Stochastic volatility for real. ssrn.com , 2006.M. Arnsdorf. Central counterparty risk. arXiv , 2011.S. Babbs. Generalised vasicek models of the term structure. Applied Stochastic Models andData Analysis , 1:46–92, 1993.T. Bielecki and M. Rutkowski. Credit Risk: Modeling, Valuation and Hedging. SpringerFinance, Berlin, 2002.E. Biffis, D. P. Blake, L. Pitotti, and A. Sun. The cost of counterparty risk and collateralizationin longevity swaps. 2011.D. Brigo and F. Mercurio. Interest Rate Models: Theory and Practice with Smile, Inflationand Credit . Springer Verlag, 2006.D. Brigo and A. Pallavicini. Counterparty risk under correlation between default and interestrates. In J. Miller, D. Edelman, and J. Appleby, editors, Numerical Methods for Finance .Chapman Hall, 2007.D. Brigo, A. Capponi, and A. Pallavicini. Arbitrage-free bilateral counterparty risk valua-tion under collateralization and re-hypothecation with application to CDS. MathematicalFinance , 2011a. Accepted for publication.D. Brigo, A. Capponi, A. Pallavicini, and V. Papatheodorou. Collateral margining inarbitrage-free counterparty valuation adjustment including re-hypotecation and netting. Working Paper , 2011b. URL ssrn.com .D. Brigo, M. Morini, and A. Pallavicini. Counterparty Credit Risk, Collateral and Fundingwith pricing cases for all asset classes . Wiley, Forthcoming, 2012.M. Brunnermeier and L. Pedersen. Market liquidity and funding liquidity. The Review ofFinancial Studies , 22 (6), 2009.C. Burgard and M. Kjaer. Partial differential equation representations of derivatives withcounterparty risk and funding costs. The Journal of Credit Risk , 7 (3):1–19, 2011a. URL http://ssrn.com/abstract=1605307 .C. Burgard and M. Kjaer. In the balance. Risk Magazine , October, 2011b.R. Carmona. Hjm: a unified approach to dynamic models for fixed income, credit and equitymarkets. In R.A. Carmona, E. C¸ inlar, I. Ekeland, E. Jouini, J.A. Scheinkman, and N. Touzi,editors, Paris-Princeton Lectures on Mathematical Finance . Springer Verlag, 2004.A. Carverhill. When is the short rate markovian. Mathematical Finance , 2:135–154, 1994. . Pallavicini, D. Brigo, Multi-Curve models explained via Credit, Collateral and Funding Liquidity Journal of Fixed Income , 1,1992.O. Cheyette. Markov representation of the heath-jarrow-morton model. ssrn.com , 2001.C. Chiarella, S.C. Maina, and C. Nikitipoulos Sklibosios. Markovian defaultable hjm termstructure models with unspanned stochastic volatility. ssrn.com , 2010.R. Cont, R.P. Mondescu, and Y. Yu. Central clearing of interest rate swaps: A comparisonof offerings. ssrn.com , 2011.S. Cr´epey. A BSDE approach to counterparty risk under funding constraints. Working Paper ,2011. URL grozny.maths.univ-evry.fr/pages_perso/crepey .S. Cr´epey. Bilateral counterparty risk under funding constraints Part I: Pricing. Forthcomingin Mathematical Finance , 2012a.S. Cr´epey. Bilateral counterparty risk under funding constraints Part II: CVA. Forthcomingin Mathematical Finance , 2012b.S. Cr´epey, R. Gerboud, Z. Grbac, and N. Ngor. Counterparty risk andfunding: The four wings of the TVA. Working Paper , 2012a. URL grozny.maths.univ-evry.fr/pages_perso/crepey .S. Cr´epey, Z. Grbac, and H.N. Nguyen. A multiple-curve hjm model of interbank risk. Mathematics and Financial Economics , 6:155–190, 2012b.J. Eisenschmidt and J. Tapking. Liquidity risk premia in unsecured interbank money markets. Working Paper Series European Central Bank , 2009.D. Filipovic and A.B. Trolle. The term structure of interbank risk. ssrn.com , 2012.M. Fujii and A. Takahashi. Clean valuation framework for the usd silo. Working Paper ,2011a. URL ssrn.com .M. Fujii and A. Takahashi. Collateralized cds and default dependence. Working Paper , 2011b.URL ssrn.com .M. Fujii, Y. Shimada, and A. Takahashi. Collateral posting and choice of collateral currency. Working Paper , 2010. URL ssrn.com .D. Heller and N. Vause. From turmoil to crisis: Dislocations in the fx swap market. BISWorking Paper , 373, 2012.M. Henrard. The irony in the derivatives discounting part ii: The crisis. ssrn , 2009.M. Kijima, K. Tanaka, and T. Wong. A multi-quality model of interest rates. QuantitativeFinance , 9:133–145, 2009.T. Mart`ınez. Drift conditions on a hjm model with stochastic basis spreads. 2009. URL .F. Mercurio. Interest rates and the credit crunch: New formulas and market models. Bloomberg Portfolio Research Paper , 2009. . Pallavicini, D. Brigo, Multi-Curve models explained via Credit, Collateral and Funding Liquidity Risk Magazine , 12, 2010.F. Mercurio and Z. Xie. The basis goes stochastic. Risk Magazine , 12, 2012.N. Moreni and A. Pallavicini. Parsimonious hjm modelling for multiple yield-curve dynamics. Submitted to Quantitative Finance , 2010.N. Moreni and A. Pallavicini. Parsimonious multi-curve hjm modelling with stochastic volatil-ity. In M Bianchetti and M. Morini, editors, Interest Rate Modelling After The FinancialCrisis . Risk Books, 2012.A. Pallavicini. A comprehensive framework for bilateral collateralized cva and funding costs.In Finance and Stochastics Seminar . Imperial College, 2013. Presented at the conferencein London, January 16.A. Pallavicini and M. Tarenghi. Interest-rate modelling with multiple yield curves. 2010.A. Pallavicini, D. Perini, and D. Brigo. Funding Valuation Adjustment: FVA consistent withCVA, DVA, WWR, Collateral, Netting and re-hyphotecation. arxiv.org, ssrn.com , 2011.A. Pallavicini, D. Perini, and D. Brigo. Funding, Collateral and Hedging: uncovering themechanics and the subtleties of funding valuation adjustments. arxiv.org, ssrn.com , 2012.C. Pirrong. The economics of central clearing: Theory and practice. ISDA Discussion PapersSeries , 2011.V. Piterbarg. Funding beyond discounting: collateral agreements and derivatives pricing. Risk Magazine , 2:97–102, 2010.V. Piterbarg. Cooking with collateral. Risk Magazine , 8, 2012.P. Ritchken and L. Sankarasubramanian. Volatility structures of forward rates and the dy-namics of the term structure.