Solving stochastic control problems using Hitsuda representation

128 Views Asked by At

I would like to solve the following problem.
Consider a financial market with quadratic transaction costs, one risky asset with price dynamics:
$S_t = s_0 + \mu t + \sigma W_t$, for $t \geq 0 , \sigma > 0, \mu \in {R}$ and $W$ is Brownian Motion.
and (for simplicity) let the riskless asset bearing zero interest.

Now consider a risk neutral trader having access to the Filtration
$ G_t^{\Delta} := F_{t+\Delta}^S$, for ${\Delta} \in [0,\infty)$,
so he has more information than just having acces to the augmented filtration $(F_t^S)_{t \geq 0}$

I then tried to calculate the Valuefunction V for a timehorizon $T>0$, which led me to
$V_T^{\Phi_0, \phi} = \Phi_0 (S_T - S_0) + \int_0^T \phi_t(S_T-S_t)dt - \frac{\Lambda}{2} \int_0^T \phi_t^2 dt$ ,
where $\Phi_0$ is my position at the Start, $\phi$ is my traiding strategy and the last term is the quadratic transaction costs for a $\Lambda > 0$.

Since the trader is risk neutral we have a linear utility function i.e. u(x) = x
Also we need that $\phi$ is an admissible trading strategy, hence $G^\Delta$-optional with $\int_0^T \phi_t^2 < \infty$

I then tried to calculate: $max_{\phi} E[u(V_T^{\Phi_0, \phi})] = max_{\phi} E[V_T^{\Phi_0, \phi}]$

After using tower property and other properties this led to:
$max_{\phi}( \Phi_0(\mu T -s_0) + \int_0^T \mu (T-t) E[\phi_t] + \sigma E[\phi_t (W_{t+\Delta} - W_t)] - \frac{\Lambda}{2} E[\phi_t^2] dt)$

And here i am stuck tried to solve it using the Hitsuda represantion but i dont how to start.

Since i am very interested in the process of how to solve such a problem or related problems i would really like to just get a hint (not a whole solution) if someone has a idea where and how to start or maybe someone knows good lecture to unterstand the process of working with the hitsuda representation.

1

There are 1 best solutions below

0
On

Here is what the solution would look like in the standard non-anticipating case. Define the wealth at time $t$ as $Y_t$. This evolves as

$$dY_t=\int_o^t\phi_sdsdS_t-\frac{\Lambda}{2}\phi_t^2dt$$

Hence at terminal date $T$

$$Y_T=\int_0^{T}\int_o^t\phi_sdsdS_t-\int_0^{T}\frac{\Lambda}{2}\phi_t^2dt+Y_0$$

You can normalize $Y_0=0$. You have

$$E(Y_T)=\int_0^{T}\int_0^t\phi_sds\mu dt-\int_0^{T}\frac{\Lambda}{2}\phi_t^2dt=\int_0^{T}(\int_0^t\phi_sds\mu-\frac{\Lambda}{2}\phi_t^2) dt$$

The problem is

$$\max_{\phi_t}E(Y_T)$$

The first order condition is

$$\mu-\Lambda \phi_t=0$$

and hence

$$\phi_t=\frac{\mu}{\Lambda}$$