2 ms·
I thought the kelly criterion assumes a known probability distribution outcome? How do you size the bets?
by cinntaile 5y ago
I thought the kelly criterion assumes a known probability distribution outcome? How do you size the bets?
- kqr 5y agoSure, it does. But it does so in the same sense that MPT requires known mean and variance. In either case, we go with estimations. In many practical cases, the joint distribution of outcomes is easier to estimate safely than the variance. The joint distribution of outcomes can be approximated by something very close to the actual observed distribution of outcomes. However, the variance has, in a sense, to be derived from a fitted model. This is one step further removed from reality which risks introducing more incorrect assumptions. Other reasons are that the variance doesn't even exist for some assets (too heavy tailed distributions) and that errors in estimating the joint distribution can be smaller thanks to at least some level of central limit theorem.