Quote from sle:
70% edge on a sample of 120 months is just as "un"-glaring as 1% on a sample of 1000. This is becoming an increasingly academic discussion on low probability statistics. There are many thing people can do to evaluate statistical qualities of outer tails in distributions and there is a lot of non-financial work being done in that area.
In the end, most of your risk is not in evaluation of fair value (which is fairly straight forward), but rather in preventing a ruinous loss. That's where the skill of the portfolio manager comes in.
Won't the average gains and average losses be essential when evaluating the value? If a strategy proves to be correct 90% of the time over 1000 instances, but has an average win of 1 and loss of 5, I'm pretty sure we can put a lot of faith into this system.
Even though we don't know with a high confidence that it truly is correct 90% of the time, we can have a better understanding of the expected value because of the win/loss amount.