Skip to content
All library documents

Deriving the Futures Convexity Adjustment from a Pricing PDE

Article Quant Q&A · Author: siou0107

Summary

The document explains how to obtain a partial differential equation for a discount factor used to express a futures price. It starts with a pricing PDE for a claim depending on the spot price and interest rate, then substitutes the similarity form in which the futures price is spot divided by the factor. Differentiating this form with respect to time, spot, and rate and inserting those derivatives into the original PDE produces the equation for the factor.

The worked response identifies two apparent notation errors in the displayed equation: the quadratic rate term uses the factor itself in the denominator, and the mixed term uses the rate volatility. These terms arise from the second rate derivative and the spot-rate cross derivative. The source gives an algebraic derivation, not a broader treatment of convexity adjustments. It does not discuss boundary conditions, calibration, numerical solutions, or empirical validation, so the result should be read as a derivation within the stated model assumptions.

Key ideas

  • The futures price is represented as spot divided by a function of the interest rate and time.
  • Substituting this similarity form into the pricing PDE yields a PDE for the discount factor.
  • The quadratic derivative term comes from differentiating the reciprocal factor twice with respect to the rate.
  • The mixed spot-rate derivative creates the term involving rate volatility and correlation.
  • The displayed equation's symbols for the denominator and volatility are identified as errors in the response.

Tags

Full text
# Forward price vs. futures price - Wilmott


# Forward price vs. futures price - Wilmott












I am reading Paul Wilmott's book PWOQF2, and there is something I don't get in his derivation of the convexity adjustment between forward and futures prices (chap. 30).

He models $S$ and $r$ following SDEs $$dS_t = \mu S_t dt +\sigma S_t dX_1$$

$$dr_t = u(r,t)dt + w(r,t) dX_2$$

$$d\langle X_1, X_2 \rangle_t = \rho dt$$

under the physical measure, the risk-neutral measure dynamics being the same up to the market price of risk term $\lambda$.

He shows the well-known result for forward price, i.e.

$$\text{Forward price} = \frac{S}{Z}$$ where $Z$ is the relevant zero coupon bond price. Until then everything's fine.

He then writes the futures price as $F(S, r, t) = \frac{S}{p(r,t)}$, where $p$ is some kind of discount factor. Following his usual routine, we get the pricing PDE for a derivative depending on $S$ and $r$: $$\frac{\partial F}{\partial t} + \frac{1}{2}\sigma^2S^2\frac{\partial^2 F}{\partial S^2}+\rho\sigma Sw\frac{\partial^2 F}{\partial S \partial r} + \frac{1}{2}w^2\frac{\partial^2 F}{\partial r^2} + rS\frac{\partial F}{\partial S} + \left( u - \lambda w \right)\frac{\partial F}{\partial r} = 0$$

And then he derives a PDE for $p$: $$\frac{\partial p}{\partial t} + \frac{1}{2}w^2\frac{\partial^2 p}{\partial r^2} + \left( u - \lambda w \right)\frac{\partial p}{\partial r} - rp \underline{-w^2\frac{\left(\frac{\partial p}{\partial r}\right)^2}{q} + \rho\sigma\beta\frac{\partial p}{\partial r}} = 0$$ commenting "Just plug the similarity form into the equation to see this".

My questions are :

- What similarity form? And into which equation? (Not clear at all to me…)

- Do you guys have any ideas where the $q$ and $\beta$ in the underlined terms (the famous convexity adjustment) come from? They never appear in the equations given at the beginning of the section (yes, the long summary was about that)

Thanks a lot for your help!

## Answer by siou0107 (score 2)

https://quant.stackexchange.com/a/49904

Modulo the two $\beta$ and $q$ errors, the proof is not that complicated actually. The similarity solution is simply $F(S, r, t) = \frac{S}{p(r, t)}$ and it has to be input into the previous pricing PDE. We thus replace : $$\frac{\partial F}{\partial t} = -\frac{S}{p^2}\frac{\partial p}{\partial t}$$ $$\frac{\partial F}{\partial S} = \frac{1}{p}$$ $$\frac{\partial^2 F}{\partial S^2} = 0$$ $$\frac{\partial F}{\partial r} = -\frac{S}{p^2}\frac{\partial p}{\partial r}$$ $$\frac{\partial^2 F}{\partial r^2} = -\frac{S}{p^4}\left[p^2 \frac{\partial^2 p}{\partial r^2}-2p\left( \frac{\partial p}{\partial r} \right)^2 \right] \equiv -\frac{S}{p^3}\left[p \frac{\partial^2 p}{\partial r^2}-2\left( \frac{\partial p}{\partial r} \right)^2 \right]$$ $$\frac{\partial^2 F}{\partial S \partial r} = -\frac{1}{p^2}\frac{\partial p}{\partial r}$$ $$\Rightarrow -\frac{S}{p^2}\frac{\partial p}{\partial t} - \frac{\rho\sigma Sw}{p^2}\frac{\partial p}{\partial r} - \frac{1}{2} w^2 \frac{S}{p^3}\left[p \frac{\partial^2 F}{\partial r^2}-2\left( \frac{\partial p}{\partial r} \right)^2 \right] + r\frac{S}{p} - (u - \lambda w)\frac{S}{p^2}\frac{\partial p}{\partial r} = 0$$ You multiply by $-\frac{p^2}{S}$ and we're done. $$\frac{\partial p}{\partial t} + \frac{1}{2} w^2 \frac{\partial^2 p}{\partial r^2} + (u - \lambda w) \frac{\partial p}{\partial r} - rp \underline{- \frac{w^2}{\color{red}{p}} \left( \frac{\partial p}{\partial r} \right)^2 + \rho\sigma \color{red}{w} \frac{\partial p}{\partial r}} = 0$$

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.