Usually exact test are rare. Even though in general you may form the likelihood ratio test statistic, the exact distribution is intractable and the critical region relies on simulation/numerical method. More common approach is forming the asymptotic tests.
But these are not the most important issues. The most important one is that: what is your data structure? model? assumptions?
Can you define your parameters more clearly? any likelihood function you can write down?
I don't have specific data or assumptions. This is the problem submitted by one of professors as a matter of defending the thesis. But now I see that "+" is not shown properly and easy question has changed to huge one ...