Is this a performance-sensitive exercise? Is the naive python implementation good enough? How frequently would you need this, and under what kind of load?
I take it the point is that Julia is generally faster without the optimization step, but the article seemed so heavily focused on this particular problem that I have to question whether it matters in this case.
Monte Carlo simulation in general is definitely performance sensitive. If you're averaging over realizations of a process, the accuracy of your estimate typically depends on the square root of the number samples you take. For a given time budget, every factor s improvement you make to the speed of the calculation earns you sqrt(s) more accuracy.
In a high frequency trading situation, if you can estimate a "correct" price faster than your competitor can, then you get to make the trade, and they don't. Doesn't get much more performance sensitive than that.
I take it the point is that Julia is generally faster without the optimization step, but the article seemed so heavily focused on this particular problem that I have to question whether it matters in this case.