Unit 10, lesson 80

Funding capital and margin valuation adjustments

How a discounted ledger of supplied balances defines FVA, MVA and KVA, how DVA enters with first-to-default ordering, when the XVA terms add to a clean value and how partitioned balances prevent double counting.

  • Video 80
  • Counterparty exposure and valuation adjustments

A derivative has a clean value of 55 million, but holding and hedging it requires unsecured borrowing, posted initial margin and committed capital, and either party can default. Which of these payments should be subtracted from the 55 million, which should be added, and how can we be sure that no payment is charged twice? This lesson writes the costs as a discounted ledger of supplied balances, attaches the names FVA, MVA, KVA and DVA to separate payments of that ledger, and shows where the additive formula stops being valid.

Key terms

Funding valuation adjustment

Definition Fix a base curve with deterministic discount factors D(0,t)D(0,t), the curve on which the clean value VcleanV^{\text{clean}} is computed, and a horizon TT in years. Let Ft≥0F_t \ge 0 be a supplied unsecured funding balance in currency units, the cash that must be borrowed to hold and hedge the position, excluding initial margin, and let ftf_t be its incremental funding spread over the base rate, annualized with continuous accrual. Assume that FF and ff are deterministic, or at least fixed in advance, so that neither depends on the value being computed, and that ftFtf_t F_t is integrable on [0,T][0, T]. Over a short interval [t,t+dt][t, t + dt] the extra interest is ftFt dtf_t F_t\,dt, and discounting and adding these payments gives the funding valuation adjustment of this restricted linear ledger,

FVA⁡=∫0TD(0,t) ft Ft dt.\operatorname{FVA} = \int_0^T D(0,t)\,f_t\,F_t\,dt .

A rate per year times a balance in currency units times a time in years gives currency units, and discounting keeps them. For balances held over buckets of length Δi\Delta_i with the extra interest paid at the end dates tit_i, the integral becomes the sum of D(0,ti) fi Fi ΔiD(0,t_i)\,f_i\,F_i\,\Delta_i. In this ledger FVA is reported as a cost and subtracted. A negative balance, a cash surplus invested at a lending spread, gives a funding benefit, and when borrowing and lending spreads differ, the integrand becomes ftbFt+−ftlFt−f^{b}_t F_t^{+} - f^{l}_t F_t^{-} with x−=max⁡(−x,0)x^{-} = \max(-x, 0). The formula charges the spread up to TT in every state. Models in which a default ends the funding weight each date by the probability that no default has occurred yet. Any collateral remuneration that is already netted in the spread must not be charged again as a separate cost.

In plain words FVA is the present value of the extra interest, above the base rate, that is paid on the cash borrowed to carry the trade. Each date contributes its spread times its balance, discounted to today.

Example A constructed ledger has two one-year buckets with discount factors 0.980.98 and 0.950.95 at the payment dates. The non-margin funding balances are 33 and 22 million, financed at a spread of 0.010.01 per year, so

FVA⁡=0.98⋅0.01⋅3+0.95⋅0.01⋅2=0.0294+0.019=0.0484\begin{aligned} \operatorname{FVA} &= 0.98 \cdot 0.01 \cdot 3 \\[0.4em] &\quad + 0.95 \cdot 0.01 \cdot 2 \\[0.4em] &= 0.0294 + 0.019 \\[0.4em] &= 0.0484 \end{aligned}

million.

Margin valuation adjustment

Definition Let It≥0I_t \ge 0 be a supplied balance of initial margin that we post, in currency units, segregated so that it cannot fund anything else, and let mtm_t be the incremental spread over the base rate at which it is financed, net of any remuneration the margin earns. With the same discount factors and an integrable product mtItm_t I_t, the margin valuation adjustment is

MVA⁡=∫0TD(0,t) mt It dt.\operatorname{MVA} = \int_0^T D(0,t)\,m_t\,I_t\,dt .

FVA and MVA are separate costs only when the funding balance FtF_t excludes the margin, so that the two integrals finance disjoint amounts. The adjustment depends on the spread as well as on the balance. When the margin is financed at the base rate, mt=0m_t = 0, the MVA is zero although the margin is still posted. The margin balance itself comes from the margin model and the portfolio, and here it is supplied rather than modeled.

In plain words Posted initial margin is cash or securities that we must find and cannot use elsewhere. MVA is the present value of the extra cost of financing it over the life of the trade.

Example In the same ledger the initial margin balances are 11 and 1.51.5 million, financed at a spread of 0.0120.012 per year, so

MVA⁡=0.98⋅0.012⋅1+0.95⋅0.012⋅1.5=0.01176+0.0171=0.02886\begin{aligned} \operatorname{MVA} &= 0.98 \cdot 0.012 \cdot 1 \\[0.4em] &\quad + 0.95 \cdot 0.012 \cdot 1.5 \\[0.4em] &= 0.01176 + 0.0171 \\[0.4em] &= 0.02886 \end{aligned}

million. If only the margin spread changes to zero, the MVA becomes zero while the margin balances stay the same.

Capital valuation adjustment

Definition Let Kt≥0K_t \ge 0 be the supplied capital that must be held against the position, in currency units, let hth_t be a hurdle rate, the annual return on that capital that a stated policy requires, and let ctc_t be the annual rate that the capital already earns. The capital valuation adjustment of this simplified policy is

KVA⁡policy=∫0TD(0,t) (ht−ct) Kt dt.\begin{aligned} &\operatorname{KVA}_{\text{policy}} \\[0.4em] &= \int_0^T D(0,t)\,(h_t - c_t)\,K_t\,dt . \end{aligned}

The capital requirement and the hurdle rate are inputs of an economic, regulatory or accounting policy. Risk-neutral pricing does not determine the hurdle rate, so the KVA is a charge under the stated policy and not a no-arbitrage price. Other definitions in use, for example ones that discount at the base rate plus the hurdle rate or let the capital depend on simulated future exposures, give different numbers.

In plain words Capital set aside for a trade is expected to earn a required return. KVA charges today for the gap between that required return and what the capital earns anyway, over the life of the trade.

Example In the ledger the capital balances are 0.80.8 and 0.60.6 million and the net policy rate h−ch - c is 0.080.08 per year, so

KVA⁡policy=0.98⋅0.08⋅0.8+0.95⋅0.08⋅0.6=0.06272+0.0456=0.10832\begin{aligned} \operatorname{KVA}_{\text{policy}} &= 0.98 \cdot 0.08 \cdot 0.8 \\[0.4em] &\quad + 0.95 \cdot 0.08 \cdot 0.6 \\[0.4em] &= 0.06272 + 0.0456 \\[0.4em] &= 0.10832 \end{aligned}

million. A net policy rate of 0.040.04 would halve it to 0.054160.05416 million, and no market price decides between the two rates.

Debt valuation adjustment

Definition Let τb\tau_b be our own default time and τc\tau_c the counterparty’s, with Q(τb=τc)=0\mathbb{Q}(\tau_b = \tau_c) = 0 under a pricing measure Q\mathbb{Q} with the money-market account BB as numeraire and B0=1B_0 = 1, and let RbR_b and RcR_c be the two recovery fractions. Use the risk-free closeout convention of Credit valuation adjustment and wrong way risk, with net claim Xt=Vtclose−CtX_t = V^{\text{close}}_t - C_t after usable collateral. If we default first, a negative net claim is a debt of ours of which only the fraction RbR_b is paid, so our gain is (1−Rb)Xτb−(1 - R_b) X^{-}_{\tau_b}. The debt valuation adjustment is

DVA⁡=EQ[1Ab (1−Rb) Xτb−Bτb],Ab={τb≤T, τb<τc}.\begin{aligned} &\operatorname{DVA} \\[0.4em] &= \mathbb{E}^{\mathbb{Q}}\Bigl[\frac{\mathbf{1}_{A_b}\,(1 - R_b)\,X^{-}_{\tau_b}}{B_{\tau_b}}\Bigr], \\[0.4em] &A_b = \{\tau_b \le T,\ \tau_b < \tau_c\}. \end{aligned}

The bilateral CVA uses the event Ac={τc≤T, τc<τb}A_c = \{\tau_c \le T,\ \tau_c < \tau_b\} and the loss (1−Rc)Xτc+(1 - R_c) X^{+}_{\tau_c} in the same way. Under the risk-free closeout convention, with the payment and collateral assumptions of the unilateral case, the value of the netting set including both defaults is the default-free value minus the bilateral CVA plus the DVA. The first-to-default indicators matter, because the first default ends the contract and no second default loss can follow it. A different closeout convention changes both terms.

In plain words DVA values the part of our own obligations that we would fail to pay if we defaulted first. It is a gain to us, and under the same conventions it equals the CVA that the counterparty would compute on us.

Example In a constructed one-date model, the states in which we default first have probability 0.050.05 under Q\mathbb{Q}, and in each of them we owe a net 2.52.5 million after collateral. Our recovery is 0.40.4 and the discount factor to settlement is 0.970.97, so

DVA⁡=0.97⋅0.6⋅2.5⋅0.05=0.07275\begin{aligned} \operatorname{DVA} &= 0.97 \cdot 0.6 \cdot 2.5 \cdot 0.05 \\[0.4em] &= 0.07275 \end{aligned}

million. In the ledger of the XVA card a DVA of 0.070.07 million and a CVA of 0.180.18 million are supplied as credit inputs, not derived from the funding spreads.

XVA

Definition XVA is the collective name for the valuation adjustments applied to a clean value, among them CVA, DVA, FVA, MVA and KVA. Suppose each adjustment is the discounted value of its own payment stream, the streams are disjoint, all of them are computed for the same party with the same base curve, measure and closeout conventions, and the balances that generate the funding, margin and capital payments are supplied rather than solved for. Linearity of discounted sums then gives the restricted ledger

Vadj=Vclean−CVA⁡+DVA⁡−FVA⁡−MVA⁡−KVA⁡,\begin{aligned} V^{\text{adj}} &= V^{\text{clean}} - \operatorname{CVA} \\[0.4em] &\quad + \operatorname{DVA} - \operatorname{FVA} \\[0.4em] &\quad - \operatorname{MVA} - \operatorname{KVA}, \end{aligned}

with CVA, FVA, MVA and KVA reported as costs and DVA as a benefit. Other texts report some terms as signed numbers, so the sign of a term has to be read from its ledger. Outside these hypotheses the additive formula can fail, and a split into named terms is not unique. Suppose the unsecured balance depends on the value being computed, for example Ft=Vt−Ct−HtF_t = V_t - C_t - H_t, with reusable collateral CtC_t and the part HtH_t that is financed through the hedge instruments themselves, as in a repo, and suppose that borrowing and lending spreads differ. The funding cost rate

ftb (Vt−Ct−Ht)+−ftl (Vt−Ct−Ht)−\begin{aligned} &f^{b}_t\,(V_t - C_t - H_t)^{+} \\[0.4em] &\quad - f^{l}_t\,(V_t - C_t - H_t)^{-} \end{aligned}

then depends on VtV_t itself. The value therefore solves a nonlinear equation, which in a stochastic model is a backward stochastic differential equation. The values of a buyer and a seller can differ, and that solution has no universal decomposition into named additive adjustments.

In plain words The named adjustments are labels for separate payments in one ledger. When those payments are fixed in advance and do not overlap, their present values add to the clean value with the signs of costs and benefits. When the financing depends on the value, the ledger has to be solved as one equation.

Example The constructed ledger of the previous cards has clean value 55 million, CVA 0.180.18, DVA 0.070.07, FVA 0.04840.0484, MVA 0.028860.02886 and KVA 0.108320.10832, all in million currency units. The adjusted value is

Vadj=5−0.18+0.07−0.0484−0.02886−0.10832=4.70442\begin{aligned} V^{\text{adj}} &= 5 - 0.18 + 0.07 \\[0.4em] &\quad - 0.0484 - 0.02886 \\[0.4em] &\quad - 0.10832 \\[0.4em] &= 4.70442 \end{aligned}

million. With the margin spread set to zero and everything else held fixed, the value rises by the MVA to 4.733284.73328 million. That controlled change is valid because the other balances were supplied. If new financing terms also changed the hedge or the future collateral calls, the balances would have to be solved again.

Double counting

Definition Double counting occurs when two named adjustments charge, or credit, the same payment in the same states. The ledger adds correctly only when every balance is partitioned into parts that each carry one financing rate and one name. For example, if a reported funding balance Ftincl=Ft+ItF^{\text{incl}}_t = F_t + I_t includes the initial margin, then for any spread ff,

∫0TD(0,t) ft Ftincl dt=∫0TD(0,t) ft Ft dt+∫0TD(0,t) ft It dt,\begin{aligned} &\int_0^T D(0,t)\,f_t\,F^{\text{incl}}_t\,dt \\[0.4em] &= \int_0^T D(0,t)\,f_t\,F_t\,dt \\[0.4em] &\quad + \int_0^T D(0,t)\,f_t\,I_t\,dt , \end{aligned}

so adding a full MVA to an FVA computed on FinclF^{\text{incl}} finances the margin twice. Other overlaps are harder to see. A lender’s spread can contain compensation for loss if we default, and then a funding benefit and the DVA can describe the same gain in our own default states. Whether they overlap depends on which liabilities and closeout cash flows the valuation includes, not on the labels. Adding a standalone unilateral CVA and a standalone DVA without first-to-default indicators charges a second default after the first default has ended the contract. Collateral remuneration that is already netted in a spread must not be charged again.

In plain words A second name for a payment does not create a second payment. Before adding adjustments, write down the balance, the rate and the dates that each one uses, and check that no part of a balance appears twice.

Example Suppose the funding balances of the constructed ledger had been reported as 44 and 3.53.5 million, including the margin of 11 and 1.51.5 million, and charged at the funding spread 0.010.01. That FVA is 0.98⋅0.01⋅4+0.95⋅0.01⋅3.5=0.072450.98 \cdot 0.01 \cdot 4 + 0.95 \cdot 0.01 \cdot 3.5 = 0.07245 million, and adding the MVA of 0.028860.02886 million gives 0.101310.10131 million for funding and margin together. The partitioned ledger charges 0.0484+0.02886=0.077260.0484 + 0.02886 = 0.07726 million. The difference of 0.024050.02405 million is exactly the margin charged a second time at the funding spread, 0.98⋅0.01⋅1+0.95⋅0.01⋅1.50.98 \cdot 0.01 \cdot 1 + 0.95 \cdot 0.01 \cdot 1.5.

Check yourself

  1. A constructed ledger has clean value 88 million and two one-year buckets with discount factors 0.970.97 and 0.930.93. The non-margin funding balances are 44 and 2.52.5 million at a spread of 0.0120.012, the initial margin balances are 22 and 22 million at a spread of 0.010.01, the required capital is 1.21.2 and 11 million with hurdle rate 0.100.10 and earned rate 0.020.02, and the credit inputs are CVA 0.250.25 million and DVA 0.090.09 million. Find the FVA, the MVA, the KVA and the adjusted value. What value would a colleague get by subtracting the DVA?
Show the answer

The three cost terms are

FVA⁡=0.97⋅0.012⋅4+0.93⋅0.012⋅2.5=0.04656+0.0279=0.07446,\begin{aligned} \operatorname{FVA} &= 0.97 \cdot 0.012 \cdot 4 \\[0.4em] &\quad + 0.93 \cdot 0.012 \cdot 2.5 \\[0.4em] &= 0.04656 + 0.0279 \\[0.4em] &= 0.07446 , \end{aligned}

then MVA⁡=0.97⋅0.01⋅2+0.93⋅0.01⋅2=0.0194+0.0186=0.038\operatorname{MVA} = 0.97 \cdot 0.01 \cdot 2 + 0.93 \cdot 0.01 \cdot 2 = 0.0194 + 0.0186 = 0.038, and with the net policy rate 0.10−0.02=0.080.10 - 0.02 = 0.08,

KVA⁡policy=0.97⋅0.08⋅1.2+0.93⋅0.08⋅1=0.09312+0.0744=0.16752,\begin{aligned} \operatorname{KVA}_{\text{policy}} &= 0.97 \cdot 0.08 \cdot 1.2 \\[0.4em] &\quad + 0.93 \cdot 0.08 \cdot 1 \\[0.4em] &= 0.09312 + 0.0744 \\[0.4em] &= 0.16752 , \end{aligned}

all in millions. The adjusted value is

Vadj=8−0.25+0.09−0.07446−0.038−0.16752=7.56002\begin{aligned} V^{\text{adj}} &= 8 - 0.25 + 0.09 \\[0.4em] &\quad - 0.07446 - 0.038 \\[0.4em] &\quad - 0.16752 \\[0.4em] &= 7.56002 \end{aligned}

million. Subtracting the DVA gives 7.380027.38002 million, which is too low by twice the DVA, 0.180.18 million. The DVA is a benefit to us in the states where we default first, so in this sign convention it is added.

  1. In the ledger of question 1, another desk reports the funding balances as 66 and 4.54.5 million, which include the initial margin, charges them at the funding spread 0.0120.012, and then also adds the MVA of 0.0380.038 million. By how much does it overcharge funding and margin together? Would dropping the MVA and keeping its FVA fix the ledger?
Show the answer

The desk’s FVA is 0.97⋅0.012⋅6+0.93⋅0.012⋅4.5=0.06984+0.05022=0.120060.97 \cdot 0.012 \cdot 6 + 0.93 \cdot 0.012 \cdot 4.5 = 0.06984 + 0.05022 = 0.12006 million, so with the MVA it charges 0.158060.15806 million. The partitioned ledger of question 1 charges 0.07446+0.038=0.112460.07446 + 0.038 = 0.11246 million. The overcharge is

0.15806−0.11246=0.0456=0.012 (0.97⋅2+0.93⋅2)\begin{aligned} &0.15806 - 0.11246 = 0.0456 \\[0.4em] &= 0.012\,(0.97 \cdot 2 + 0.93 \cdot 2) \end{aligned}

million, the margin financed a second time inside the FVA, and it would lower the adjusted value to 7.514427.51442 million. Dropping the MVA leaves 0.120060.12006 million, still 0.00760.0076 million too high, because the margin is then financed at the funding spread 0.0120.012 instead of its own spread 0.010.01, and 0.002 (0.97⋅2+0.93⋅2)=0.00760.002\,(0.97 \cdot 2 + 0.93 \cdot 2) = 0.0076. The fix is to split the balance into its margin and non-margin parts and charge each part at its own spread.

  1. Two trades share one funding pool over a one-year bucket with discount factor 0.980.98. Trade A needs 33 million of unsecured borrowing, and trade B leaves a cash surplus of 22 million. Borrowing costs a spread of 0.0150.015 over the base rate, and a surplus earns a spread of 0.0050.005. Find the funding cost of each trade on its own, their sum and the funding cost of the pool. Why is the FVA of a pool not in general the sum of standalone FVAs, and when is it?
Show the answer

Write the funding cost rate of a balance xx as g(x)=fbx+−flx−g(x) = f^{b} x^{+} - f^{l} x^{-} with fb=0.015f^{b} = 0.015 and fl=0.005f^{l} = 0.005, so that the bucket costs 0.98 g(x)0.98\,g(x). On its own, trade A costs 0.98⋅0.015⋅3=0.04410.98 \cdot 0.015 \cdot 3 = 0.0441 million, and trade B earns 0.98⋅0.005⋅2=0.00980.98 \cdot 0.005 \cdot 2 = 0.0098 million, a cost of −0.0098-0.0098 million, so the standalone costs add to 0.03430.0343 million. The pool has the net balance 3−2=13 - 2 = 1 million and costs 0.98⋅0.015⋅1=0.01470.98 \cdot 0.015 \cdot 1 = 0.0147 million. The standalone sum is larger by

0.0343−0.0147=0.0196=0.98⋅(0.015−0.005)⋅2\begin{aligned} &0.0343 - 0.0147 = 0.0196 \\[0.4em] &= 0.98 \cdot (0.015 - 0.005) \cdot 2 \end{aligned}

million, because in the pool 22 million of A’s borrowing is replaced by B’s surplus and saves the difference between the two spreads. When fb≥flf^{b} \ge f^{l}, the function gg is convex and positively homogeneous, so g(a+b)≤g(a)+g(b)g(a + b) \le g(a) + g(b), with equality when the two balances have the same sign or the two spreads are equal. With one symmetric spread the cost is linear, and the standalone FVAs add, for example 0.98⋅0.01⋅(3−2)=0.00980.98 \cdot 0.01 \cdot (3 - 2) = 0.0098 million either way. With asymmetric spreads, how the pool’s FVA is allocated to its trades is a choice, not a consequence of the ledger, and when the balances depend on the value itself the nonlinear equation of the XVA card replaces the sum.

How this lesson connects