Unit 3, lesson 21

Futures daily settlement and convexity

How daily settlement and variation margin shape futures cash flows, why the futures price is a pricing-measure expectation, and how random rates create a forward-futures difference and a convexity adjustment.

  • Video 21
  • Forwards, futures and swaps

Before interest, a forward and a futures contract on the same index with the same expiry both pay the final spot price minus the price agreed at the start. The futures position, however, collects its gains and pays its losses as the price moves, so its cash arrives at different dates and earns or costs interest along the way. Should the fair futures quote then equal the fair forward price? This lesson follows the settlement cash flows, derives the futures price in an ideal settlement model, and shows that random interest rates separate the two quotes by a covariance term.

Key terms

Futures contract

Definition A futures contract is a standardized, exchange-traded agreement on an underlying asset with expiry TT, quoted through a futures price FtfutF^{\text{fut}}_t that changes over time. Opening or closing a position at the current futures price costs nothing apart from collateral. Instead of one payment at TT, each change in the futures price is paid in cash at the next settlement date, and the contract ends with a final settlement whose price, in the idealized contract of this lesson, equals the spot price STS_T. A contract covers mm units of the underlying, its multiplier, so one long contract gains mm currency units for each unit rise in the futures price. A clearing house takes the other side of every trade and collects margin, and in this lesson default is ignored.

In plain words A futures contract works like a forward that is settled a little at a time. Its futures price is a quote that moves every day, not a delivery price fixed for the life of the contract, and the position is paid or charged every change in that quote.

Example A constructed index futures contract with multiplier 5050 is bought at 40004000. If the next settlement price is 40124012, the long side receives 50⋅12=60050 \cdot 12 = 600 currency units, the short side pays them, and both positions continue at 40124012.

Daily settlement

Definition Let the settlement dates be 0=t0<t1<⋯<tn=T0 = t_0 < t_1 < \dots < t_n = T, often one per business day, and write FkfutF^{\text{fut}}_k for the settlement price at tkt_k, with Fnfut=STF^{\text{fut}}_n = S_T. Under daily settlement, one long contract with multiplier mm receives at each tkt_k the amount

ΔMk=m (Fkfut−Fk−1fut),\Delta M_k = m\,\bigl(F^{\text{fut}}_k - F^{\text{fut}}_{k-1}\bigr),

where a negative amount is a payment by the holder, and one short contract receives −ΔMk-\Delta M_k. After the payment the contract continues at the new price FkfutF^{\text{fut}}_k, so a position that has just been settled is worth zero. Without interest on the payments, their sum telescopes,

∑k=1nΔMk=m (ST−F0fut),\sum_{k=1}^{n} \Delta M_k = m\,\bigl(S_T - F^{\text{fut}}_0\bigr),

which is the payoff of a long forward on mm units with delivery price F0futF^{\text{fut}}_0. The total is the same, but it is paid in pieces on different dates.

In plain words Each settlement closes the old position at the new price and opens a fresh one at no cost. Adding up the daily changes without interest gives the same total as a forward entered at the first quote, but the cash shows up earlier, piece by piece.

Example Take the constructed settlement prices 100100, 110110 and 105105 at t0t_0, t1t_1 and t2=Tt_2 = T, with m=1m = 1. The long side receives 1010 at t1t_1 and pays 55 at t2t_2, a total of 5=105−1005 = 105 - 100. A forward entered at 100100 pays the same 55 as a single amount at TT.

Variation margin

Definition Variation margin is the cash paid at each settlement, the amounts ΔMk\Delta M_k of the previous card. It is separate from initial margin, the collateral posted when a position is opened, which is not a payment for the underlying and is returned when the position is closed, with interest on it as the exchange’s rules provide. Suppose each variation margin payment is invested or financed until TT in the money-market account BB, with B0=1B_0 = 1 and the same rate for borrowing and lending. The terminal wealth of one long contract is then

WT=∑k=1nΔMk BTBtk.W_T = \sum_{k=1}^{n} \Delta M_k\,\frac{B_T}{B_{t_k}} .

With zero interest it reduces to the telescoping sum m(ST−F0fut)m(S_T - F^{\text{fut}}_0). With positive rates an early gain is worth more at TT than the same gain received late, and an early loss costs more than a late one. Two price paths with the same start and the same end can therefore leave different wealth. This is a difference in timing and financing, not an arbitrage.

In plain words Gains that arrive early can earn interest, and losses that arrive early must be funded. Comparing a futures position with a forward means following that cash account, not only adding up the price changes.

Example On the constructed path 100100, 110110, 105105, suppose the money-market account grows by the factor 1.021.02 between t1t_1 and TT. The early gain of 1010 becomes 10.210.2 at TT, and after the final payment of 55 the wealth is 10⋅1.02−5=5.210 \cdot 1.02 - 5 = 5.2. The extra 0.20.2 is interest on cash received early, not a further price change. On the path 100100, 9090, 105105 with the same financing, the early loss of 1010 is borrowed and grows to 10.210.2, so the wealth is 15−10.2=4.815 - 10.2 = 4.8. Both paths end at 105105, and the raw total is 55 on each.

Futures price

Definition Consider settlement dates t0<⋯<tn=Tt_0 < \dots < t_n = T and a money-market account BB with B0=1B_0 = 1, where the interest for each interval (tk−1,tk](t_{k-1}, t_k] is known at tk−1t_{k-1}, so that BtkB_{t_k} is Ftk−1\mathcal{F}_{t_{k-1}}-measurable. Assume borrowing and lending at that rate, no default, no cost of posting margin, and an equivalent pricing measure Q\mathbb{Q} for the numeraire BB, so that a payment XX at tkt_k has the value

Btk−1 EQ[XBtk | Ftk−1]B_{t_{k-1}}\,\mathbb{E}^{\mathbb{Q}}\left[\frac{X}{B_{t_k}} \,\middle\vert\, \mathcal{F}_{t_{k-1}}\right]

at tk−1t_{k-1}. Assume also that EQ[∣ST∣]<∞\mathbb{E}^{\mathbb{Q}}[\lvert S_T \rvert] < \infty, that the settlement prices are integrable and that the final settlement price is STS_T. If opening a futures position at any settlement date costs nothing, then

Fkfut=EQ[ST∣Ftk],F^{\text{fut}}_k = \mathbb{E}^{\mathbb{Q}}\bigl[S_T \mid \mathcal{F}_{t_k}\bigr],

and in particular F0fut=EQ[ST]F^{\text{fut}}_0 = \mathbb{E}^{\mathbb{Q}}[S_T]. The reason is that a position opened at tk−1t_{k-1} and closed at tkt_k pays only ΔMk\Delta M_k at tkt_k and costs nothing, so EQ[ΔMk/Btk∣Ftk−1]=0\mathbb{E}^{\mathbb{Q}}[\Delta M_k / B_{t_k} \mid \mathcal{F}_{t_{k-1}}] = 0. Since BtkB_{t_k} is known at tk−1t_{k-1}, it comes out of the conditional expectation, which leaves EQ[Fkfut−Fk−1fut∣Ftk−1]=0\mathbb{E}^{\mathbb{Q}}[F^{\text{fut}}_k - F^{\text{fut}}_{k-1} \mid \mathcal{F}_{t_{k-1}}] = 0. The futures price is therefore a Q\mathbb{Q}-martingale that ends at STS_T. For continuous settlement with a short-rate process, the same formula is a standard result under compatible financing and the regularity that its martingale argument needs, and it is stated here without proof. The expectation is taken under the pricing measure, so it is not a forecast of STS_T under P\mathbb{P}.

In plain words Each gain is paid at once and the interest for the next interval is already known, so the futures price has to be a fair game under the pricing measure from one settlement to the next. A fair game that must end at STS_T starts at the Q\mathbb{Q} expectation of STS_T.

Example In a constructed two-interval model the first interval has zero interest, so Bt1=1B_{t_1} = 1. At t1t_1 one of two states is revealed, each with Q\mathbb{Q}-probability 12\tfrac12, and with it the final spot price, ST=120S_T = 120 in the up state and ST=80S_T = 80 in the down state. Since STS_T is then known, the futures price at t1t_1 equals STS_T. The initial quote is F0fut=12⋅120+12⋅80=100F^{\text{fut}}_0 = \tfrac12 \cdot 120 + \tfrac12 \cdot 80 = 100, and with it the first settlement gain, 2020 or −20-20, has Q\mathbb{Q}-mean zero. The weights 12\tfrac12 are pricing weights, not forecasts of the two states.

Forward-futures difference

Definition In the model of the previous card, let DT=1/BT>0D_T = 1/B_T > 0 be the random discount factor of the money-market account to TT, and assume EQ[DT]<∞\mathbb{E}^{\mathbb{Q}}[D_T] < \infty, EQ[∣ST∣]<∞\mathbb{E}^{\mathbb{Q}}[\lvert S_T \rvert] < \infty and EQ[DT∣ST∣]<∞\mathbb{E}^{\mathbb{Q}}[D_T \lvert S_T \rvert] < \infty. A forward with delivery price KK paid at TT is worth EQ[DT(ST−K)]\mathbb{E}^{\mathbb{Q}}[D_T(S_T - K)] at time 00, and since KK is fixed this is zero exactly for

F0,Tfwd=EQ[DTST]EQ[DT].F^{\text{fwd}}_{0,T} = \frac{\mathbb{E}^{\mathbb{Q}}[D_T S_T]}{\mathbb{E}^{\mathbb{Q}}[D_T]} .

Writing both quotes over the positive denominator EQ[DT]\mathbb{E}^{\mathbb{Q}}[D_T] gives the forward-futures difference

F0,Tfwd−F0fut=Cov⁡Q(DT,ST)EQ[DT].\begin{aligned} &F^{\text{fwd}}_{0,T} - F^{\text{fut}}_0 \\[0.4em] &= \frac{\operatorname{Cov}^{\mathbb{Q}}(D_T, S_T)}{\mathbb{E}^{\mathbb{Q}}[D_T]} . \end{aligned}

If the zero-coupon bond maturing at TT is priced by Q\mathbb{Q}, then EQ[DT]=P(0,T)\mathbb{E}^{\mathbb{Q}}[D_T] = P(0,T), and for a traded asset without income, priced by Q\mathbb{Q}, also EQ[DTST]=S0\mathbb{E}^{\mathbb{Q}}[D_T S_T] = S_0, so the forward price is S0/P(0,T)S_0/P(0,T) as in the previous lesson. The difference is zero exactly when DTD_T and STS_T are uncorrelated under Q\mathbb{Q}, in particular when interest rates are deterministic, since DTD_T is then a constant. It is negative exactly when the covariance is negative, for example in a model where the states with high final prices are the states with high rates over [0,T][0, T], which make DTD_T small.

In plain words The forward price weights each state’s final price by what a payment at TT is worth in that state, while the futures quote weights the states by Q\mathbb{Q} alone. When high final prices come with high rates, the forward gives those states less weight and comes out below the futures quote.

Example In the two-state model of the previous card, let the money-market account grow by the factor 1.11.1 after t1t_1 in the up state and by 11 in the down state, so DT=10/11D_T = 10/11 or DT=1D_T = 1. Then

EQ[DT]=2122,EQ[DTST]=104011,\begin{gathered} \mathbb{E}^{\mathbb{Q}}[D_T] = \tfrac{21}{22}, \\[0.4em] \mathbb{E}^{\mathbb{Q}}[D_T S_T] = \tfrac{1040}{11}, \end{gathered}

about 0.9545450.954545 and 94.54545594.545455, and the forward price is 208021≈99.047619\tfrac{2080}{21} \approx 99.047619, against the futures quote 100100. The covariance is 104011−2122⋅100=−1011\tfrac{1040}{11} - \tfrac{21}{22} \cdot 100 = -\tfrac{10}{11}, and dividing by 2122\tfrac{21}{22} gives the difference −2021≈−0.952381-\tfrac{20}{21} \approx -0.952381. If the underlying asset pays no income, its value today is 104011\tfrac{1040}{11}, and the same forward price is that value divided by the bond price 2122\tfrac{21}{22}. If both growth factors were 11, the covariance would vanish and both quotes would be 100100.

Convexity adjustment

Definition A convexity adjustment is the amount that converts a futures-based quote into the forward quote for the same underlying payoff and dates. With the sign convention forward minus futures, for an asset-delivery contract in the model above it is the covariance ratio of the previous card. Its size and sign depend on the joint law of rates and the underlying under Q\mathbb{Q}, on the settlement schedule and on how margin is financed. The term is used mostly for interest rate futures. Their quoted price is an affine function of a reference rate, for example 100100 minus a rate in percent, and their daily margin follows a different timing from the single settlement of a forward rate agreement. The adjustment between such a futures rate and the corresponding forward rate has to be derived for the actual payoff in a specified rate model, and the covariance identity of this lesson is not a general formula for it.

In plain words A convexity adjustment measures, in units of the quote, how far daily settlement with random financing moves a futures quote away from the forward quote. Without the contract’s settlement rules and the model behind the pricing measure, a quoted adjustment is an incomplete statement.

Example Swap the growth factors of the two-state model, so the account grows by 11 after t1t_1 in the up state and by 1.11.1 in the down state. Then DT=1D_T = 1 when ST=120S_T = 120 and DT=10/11D_T = 10/11 when ST=80S_T = 80. The futures quote stays 100100 and EQ[DT]=2122\mathbb{E}^{\mathbb{Q}}[D_T] = \tfrac{21}{22} is unchanged, but EQ[DTST]=106011\mathbb{E}^{\mathbb{Q}}[D_T S_T] = \tfrac{1060}{11}, so the forward price is 212021≈100.952381\tfrac{2120}{21} \approx 100.952381 and the adjustment is 2021≈0.952381\tfrac{20}{21} \approx 0.952381. The same two rate levels attached to the other states reverse the sign.

Check yourself

  1. You hold one long futures contract with multiplier 1010, opened at the settlement price 200200. The next three settlement prices are 196196, 203203 and 205205. Find each variation margin payment and their total without interest, and compare the total with a forward entered at 200200.
Show the answer

The payments to the long side are

10⋅(196−200)=−40,10⋅(203−196)=70,10⋅(205−203)=20,\begin{gathered} 10 \cdot (196 - 200) = -40, \\[0.4em] 10 \cdot (203 - 196) = 70, \\[0.4em] 10 \cdot (205 - 203) = 20, \end{gathered}

so the holder pays 4040 at the first settlement and receives 7070 and 2020 at the next two. Without interest the total is 50=10⋅(205−200)50 = 10 \cdot (205 - 200), the payoff of a long forward on 1010 units with delivery price 200200. The short side has the opposite payments. With interest, each payment would also earn or cost interest until expiry, so the terminal wealth would differ from 5050.

  1. A colleague argues that forward and futures prices on the same asset and expiry must always be equal, because both contracts pay STS_T minus the initial quote in total. When is the conclusion right in the model of this lesson, and which quote is larger when DTD_T and STS_T are negatively correlated under Q\mathbb{Q}, as when high final prices come with high interest rates?
Show the answer

The totals agree only before interest. The futures payments arrive on different dates and are invested or financed at random rates, and the colleague’s argument ignores that timing. In the model of this lesson the difference is

F0,Tfwd−F0fut=Cov⁡Q(DT,ST)EQ[DT],\begin{aligned} &F^{\text{fwd}}_{0,T} - F^{\text{fut}}_0 \\[0.4em] &= \frac{\operatorname{Cov}^{\mathbb{Q}}(D_T, S_T)}{\mathbb{E}^{\mathbb{Q}}[D_T]} , \end{aligned}

so the two quotes are equal exactly when this covariance is zero. That holds when interest rates are deterministic, so that DTD_T is a constant, and more generally whenever DTD_T and STS_T are uncorrelated under Q\mathbb{Q}. With negative correlation the covariance is negative, and the forward price is below the futures quote. The dependence that matters is the one under the pricing measure Q\mathbb{Q}, not under P\mathbb{P}.

  1. In a constructed two-interval model the first interval has zero interest and reveals one of two states, with Q(u)=0.6\mathbb{Q}(u) = 0.6 and Q(d)=0.4\mathbb{Q}(d) = 0.4. The final spot price is ST=110S_T = 110 in state uu and ST=85S_T = 85 in state dd, and after t1t_1 the money-market account grows by the factor 11 in state uu and 1.11.1 in state dd. The underlying asset pays no income. Find the futures quote, the forward price and their difference, and check the difference with the covariance formula. Then check that the forward price equals the value today of the asset divided by the bond price P(0,T)P(0,T).
Show the answer

Since STS_T is known at t1t_1, the futures price at t1t_1 is STS_T, and the initial quote is F0fut=0.6⋅110+0.4⋅85=100F^{\text{fut}}_0 = 0.6 \cdot 110 + 0.4 \cdot 85 = 100. As a check, the first settlement gain is 1010 or −15-15, and 0.6⋅10−0.4⋅15=00.6 \cdot 10 - 0.4 \cdot 15 = 0. The discount factor is DT=1D_T = 1 in state uu and DT=10/11D_T = 10/11 in state dd, so

EQ[DT]=5355,EQ[DTST]=106611,\begin{gathered} \mathbb{E}^{\mathbb{Q}}[D_T] = \tfrac{53}{55}, \\[0.4em] \mathbb{E}^{\mathbb{Q}}[D_T S_T] = \tfrac{1066}{11}, \end{gathered}

about 0.9636360.963636 and 96.90909196.909091. The forward price is their ratio, 533053≈100.566038\tfrac{5330}{53} \approx 100.566038, and the difference from the futures quote is 3053≈0.566038\tfrac{30}{53} \approx 0.566038. The covariance is 106611−5355⋅100=611\tfrac{1066}{11} - \tfrac{53}{55} \cdot 100 = \tfrac{6}{11}, and dividing it by 5355\tfrac{53}{55} gives 3053\tfrac{30}{53} again. The difference is positive because the high final price comes with the larger discount factor. At t1t_1 the asset is worth STDTS_T D_T, which is 110110 in state uu and 85011≈77.272727\tfrac{850}{11} \approx 77.272727 in state dd, so its value today is EQ[DTST]=106611\mathbb{E}^{\mathbb{Q}}[D_T S_T] = \tfrac{1066}{11}, and the bond price is P(0,T)=EQ[DT]=5355P(0,T) = \mathbb{E}^{\mathbb{Q}}[D_T] = \tfrac{53}{55}. Their ratio is 533053\tfrac{5330}{53}, the forward price.

How this lesson connects