anatomy of a knockout
Contrary to what I’d written last time, I ran the test across some
1,190 equities yielding some 4,760 resultant strategies. Naturally,
when you look at a good number of cases like this, you’re going to
find some real doozies. Maybe there’s a good reason why some should
really yield above average returns with this strategy over some period of
time. I’ll leave that as the proverbial exercise for the reader…
If we look at some of the really good results like, say, this
humdinger for BID, we might think we’re onto something.
Maybe. Let’s go find a definite loser, like:
As different as night and day, right? Well, I’d certainly prefer to see the first in a monthly report, but if we look at them more carefully they really look a lot more similar. Here we take a look at the distribution of returns of trades executed by each of these respective strategies.
and this:
Now, this might not look very conclusive and perhaps it isn’t. But the two night and day strategies suddenly look a lot more alike.
They look like pretty darned random distributions to me.
And if I look at, say, 4,760 different such distributions I just might expect to see some that look pretty sweet (shifted slightly rightwards) and some that look decidedly sinister…
Because we want to believe, we might look at an absurdly unlikely series and say “Shazaam!” when we should really be saying something like “damned be Jacob Bernoulli and that law of large numbers he rode in on!”
Next time we’ll consider some strategies – and their distributions – which look rather different and which might lead us to consider different ways of reasoning about our historical results.
