Changing Measure to Make an Exponential Brownian Process a Martingale
Summary
The document works through finding a probability measure under which a process formed from Brownian motion and time is a martingale. The accepted solution applies the product rule to separate the process into two factors, then rewrites its differential so the drift is carried by a shifted Brownian increment. A constant-drift Girsanov change of measure makes that shifted process Brownian under the new measure, leaving the target process as a stochastic integral.
A second solution reaches the same differential by splitting the process into two terms and showing that their extra drift terms cancel on subtraction. The discussion is a worked stochastic-calculus example, not a trading strategy or empirical result. Its conclusion relies on the stated finite time horizon and the validity of the measure change; the document does not discuss broader integrability conditions or generalize the construction to other processes.
Key ideas
- Girsanov’s theorem can change the drift of Brownian motion by changing the probability measure.
- The product rule exposes a shifted Brownian increment in the process differential.
- Under the new measure, the process is represented as a stochastic integral with respect to Brownian motion.
- A separate decomposition confirms the drift cancellation by subtracting two component differentials.
Tags
Full text
# Change of measure for a stochastic process to be a martingale
# Change of measure for a stochastic process to be a martingale
$\text { Give a measure change so that } X_{t}=e^{B_{t}}\left(B_{t}-t / 2\right) \text { is a martingale, } 0 \leq t \leq T$
My attempt
Using Ito's lemma on $X_{t}$ we get:
$-\frac{e^{B t}}{2} d t+\left(X_{t}+e^{B_{t}}\right) d B_{t}+\left(\left(2 B_{t}-t+4\right)e^{B_{t}} d t\right. \\ =(\left.2 B_{t}-t+4-\frac{1}{2}\right) e^{B_{t}} d t+\left(X_{t}+e^{B_{t}}\right) d B_{t}$
Then I used Girsanov's theorem:
$d \hat{B}_{t}=d B_{t}+\int_{0}^{+} H_{S} ds$
$H_{t}=\frac{\left(2 B_{t}-t+4-\frac{1}{2}\right) e^{B_{t}}}{X_{t}+e^{B_{t}}}$
I am some what scared to go any further because it seems like I am heading in the wrong direction.
## Answer by Gordon (score 4, accepted)
https://quant.stackexchange.com/a/64026
Let $Y_t= e^{B_t}$ and $Z_t = B_{t}-t / 2$. Then, \begin{align*} dX_t &= Z_t dY_t + Y_t dZ_t + d\langle Y, Z\rangle_t\\ &=(B_{t}-t / 2)e^{B_t}\big( dB_t + 1/2\,dt \big) + e^{B_t}\big(dB_t -1/2\, dt\big) + e^{B_t} dt\\ &=e^{B_t}(B_t-t / 2+1)dB_t + e^{B_t}(B_t/2-t / 4 -1/2+1)dt\\ &=e^{B_t}(B_t-t / 2+1)d\big(B_t+1/2t\big). \end{align*} We define the probability measure $Q$ on $\mathscr{F}_T$ by \begin{align*} \frac{dQ}{dP}=e^{-\frac{1}{8}t-\frac{1}{2}B_t}. \end{align*} Then $\{W_t, \, t\ge 0\}$, where $W_t = B_t+1/2\,t$, is a standard Brownian motion under $Q$. Moreover, \begin{align*} dX_t = e^{W_t-\frac{1}{2}t}(W_t-t+1)dW_t. \end{align*} That is, $\{X_t, \, 0 \le t \le T\}$ is a martingale.
## Answer by ir7 (score 2)
https://quant.stackexchange.com/a/64030
(Now that I saw Gordon's solution, I can finish my attempt; I had noticed $dB_t +1/2dt$ immediately from product rule for $V_t$, zero quadratic covariation between $t/2$ and $e^{B_t}$, but hours later :) I was still perplexed by $U_t$.)
$$ X_t = U_t - V_t $$ $$ V_t = e^{B_t}t/2$$ $$ U_t = e^{B_t}B_t $$
$$ dV_t = \boxed{1/2e^{B_t} dt} + 1/2V_t(dB_t + 1/2dt) $$
$$ dU_t = (U_t + e^{B_t}) dB_t + 1/2(U_t + 2e^{B_t})dt $$
$$ = (U_t + e^{B_t})(dB_t +1/2dt) + \boxed{1/2e^{B_t} dt} $$
So, by subtraction (and lucky cancellation of the boxed terms):
$$ dX_t = (U_t + e^{B_t} - 1/2V_t)(dB_t + 1/2dt) $$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.