Skip to content
All library documents

The Itô Quadratic Variation Term in Incentive-Compatible Consumption Dynamics

Article Quant Q&A · Author: Walrasian Auctioneer

Summary

The document derives a consumption process restriction from a supermartingale condition used in an optimal contracting problem. The key step applies Itô’s lemma to discounted consumption raised to a power, relating its differential to a martingale component and an increasing process. The original attempted derivation omitted the second-order term, which is generated by consumption’s quadratic variation.

After dividing through by the relevant power of consumption, the quadratic variation becomes the squared diffusion coefficient times time. This produces the additional half-times-(one plus risk-aversion parameter) volatility-squared drift term in the consumption growth equation. The result illustrates why nonlinear transformations of stochastic processes require accounting for diffusion variance, not just the first derivative. The explanation relies on the assumed semimartingale setup and does not develop the broader economic assumptions behind incentive compatibility or the supermartingale representation.

Key ideas

  • Applying Itô’s lemma to a nonlinear function of consumption requires a quadratic variation term.
  • The missing second-order term contributes a drift proportional to consumption’s squared diffusion coefficient.
  • Under the stated dynamics, relative quadratic variation equals the squared consumption volatility times time.
  • The increasing process and martingale representation link the consumption restriction to the contracting condition.

Tags

Full text
# Ito's Lemma for this problem


# Ito's Lemma for this problem












I'm attempting to prove a lemma from a paper, in the context of optimal contracts.

$r,\rho,\gamma,\alpha,\sigma$ are all known constants.

$dR_t = (\alpha + r)dt + \sigma dZ_t$ where $Z_t$ is a standard Brownian motion.

Lemma 1

Given an incentive compatible contract, the agent's consumption must satisfy $$\frac{dc_t}{c_t} = \left( \frac{r - \rho}{\gamma} + \frac{1+\gamma}{2} (\sigma^c_t)^2 \right) dt + \sigma^c_t \frac{1}{\sigma} \left( dR_t - (\alpha + r) dt \right) + dL_t$$ for some stochastic process $\sigma^c$ and a weakly increasing stochastic process $L$.

Proof

The authors provide the following steps:

- $e^{-(\rho - r)t}c_t^{\gamma}$ is a supermartingale, thus we can express it as $$ e^{-(\rho - r)t}c_t^{\gamma} = M_t - A_t$$ where $M_t$ is a martingale and $A_t$ is a weakly increasing process.

- Applying the martingale representation theorem to $M_t$, there exists a stochastic process $\sigma^M_t$ such that $$M_t = \int_0^{t} \sigma^M_t dZ_t$$ where $Z_t$ is a standard Brownian motion.

- They then apply Ito's Lemma to get the first equation by setting $\sigma^M_t = -\gamma \sigma^c_t e^{-(\rho - r)t}c_t^{\gamma}$.

I'm struggling at step 3, as I am not sure how the Ito differential looks like for $M_t$.

This is what I've done: $$- (\rho - r) e^{-(\rho - r) t}c_t^{-\gamma} dt - \gamma e^{-(\rho - r)t} c_t^{\gamma - 1} dc_t = dM_t - dA_t $$ Substituting in $dM_t$ and dividing by $K = e^{-(\rho - r) t} c_t^{-\gamma}$,

$$(r - \rho) dt - \gamma \frac{dc_t}{c_t} = K^{-1} \sigma^M_t dZ_t - K^{-1} dA_t$$ Define $\sigma^c_t = (-\gamma K)^{-1} \sigma^M_t$ and $dL_t = (\gamma K)^{-1} dA_t $, and thus $$ \frac{dc_t}{c_t} = \frac{r - \rho}{\gamma} dt + \sigma^c_t dZ_t + dL_t $$ Plug in $dZ_t = \frac{1}{\sigma} \left( dR_t - (r + \alpha) dt \right)$ (a previous result) and the result follows.

Where does the $\frac{1+\gamma}{2} (\sigma^c_t)^2$ term come from?

## Answer by Gordon (score 2, accepted)

https://quant.stackexchange.com/a/44537

You missed the quadratic term, specifically, rather than \begin{align*} - (\rho - r) e^{-(\rho - r) t}c_t^{-\gamma} dt - \gamma e^{-(\rho - r)t} c_t^{-\gamma - 1} dc_t &= dM_t - dA_t, \end{align*} we have \begin{align*} & - (\rho - r) e^{-(\rho - r) t}c_t^{-\gamma} dt - \gamma e^{-(\rho - r)t} c_t^{-\gamma - 1} dc_t + \frac{1}{2} \gamma (\gamma+1)e^{-(\rho - r)t} c_t^{-\gamma - 2} d\langle c, c\rangle_t \\ &=\ dM_t - dA_t. \end{align*} That is, \begin{align*} \frac{dc_t}{c_t} = \frac{r - \rho}{\gamma} dt +\frac{1}{2}(\gamma+1)c_t^{-2}d\langle c, c\rangle_t + \sigma^c_t dZ_t + dL_t. \end{align*} Moreover, \begin{align*} c_t^{-2}d\langle c, c\rangle_t = \big(\sigma_t^c\big)^2 dt. \end{align*} Therefore, \begin{align*} \frac{dc_t}{c_t} = \frac{r - \rho}{\gamma} dt +\frac{1}{2}(\gamma+1)\big(\sigma_t^c\big)^2 dt + \sigma^c_t dZ_t + dL_t. \end{align*}

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.