What is the optimal stopping point for an experiment when expecting unknown event

96 Views Asked by At

Assume we notice that stock prices are rising and we can deduce we are in a bubble. Assume we start at $w(0)=0$ worth at time $t=0$ and the value grows linearly with time $(w(t)=t)$. We know that there will be a point where the prices will plummet and we lose everything if we haven't sold by that time $(w(t_f)=-1/2 t_f^2)$.

If the time till the burst of bubble is exponentially distributed with parameter $\lambda$, so $t_f \sim exp[\lambda]$. Assume also that the parameter can be approximated with some accuracy, so we know what $\lambda$ is.

So the worth function would be something like: $w(t) = \begin{cases} t, & t<t_f \\ -1/2 t^2, & t=t_f \\ 0, & t>t_f \end{cases}$

What would be the optimal time $t_s$ to sell all the money if all we know is that the fall will happen following exponential distribution with parameter $\lambda$? Optimal would mean, I think, that $E(w(t))$ is maximized. I think we would have to compute integral $$\int_0^{t_s}\int_0^{\infty} \lambda e^{- \lambda t_f} w(t) \ dt_f \ dt$$ and find $t_s$ that maximises that.

1

There are 1 best solutions below

0
On

I don't understand your value for $t = t_f$, but no matter, it won't contribute to the integral. Your double integral is wrong, though. The expected value should be $$ E(w(t_s)) = t_s \int_{t_s}^\infty \lambda e^{-\lambda t}\ dt = t_s e^{-\lambda t_s}$$ and the maximum of this is at $t_s = 1/\lambda$.