Comment by SukadarBukadar
13 hours ago
How is it useful when the underlying distribution is unknown, specially if it likely isn't stationary? For example, when doing such retirement portfolio projections or testing stock market trading rules, how do we know that the simulations we took into account are representative of what could have happened? Maybe some paths were unreachable and some underrepresented...
You go to a higher level of abstraction and define a distribution over the underlying distribution.
Though such predictions are infamously bad at extrapolating without some strong theories behind the kind of distributions you could expect.
Probabilistic Monte Carlo simulations are a way to check your guesses, if you claim you can't even guess at how things work you should consult a professional astrologer.
No, I claim that I can't check it without making another, more difficult guess. If my guess is a trading rule, to check it I need to (independently?) model the whole distribution of the multivariate stochastic process that produces it, which is more complex. My question was: how can I claim that my Monte Carlo check is reliable if I need to _guess_ how to check instead of _knowing_ how to check?
It’s a more useful tool than no tool. We are talking about predicting the future after all…