Greetings -- I disagree with Van Tharp on this point. I see no reason to limit the number of data points to 100.
Tharp does not distinguish between in-sample and out-of-sample. I suspect that all of his examples are either artificial or in-sample -- none are out-of-sample. If I have truly out-of-sample results, the more data points I have in the test, the more convincing the tests are. Assume I have run a walk forward test that has five years of out-of-sample results from a system that trades about once a week -- 250 data points. I want to use all of those results. I might want to slide a 50 or 100 point window and see what happens over time, but I do want to use all the data points. Let me say all of that in one sentence. If I have more than 100 truly out-of-sample data points, I want to use them all when testing the results. No matter how many data points I have, I want to use them all. Thanks, Howard On Mon, Oct 12, 2009 at 10:00 AM, bingk66 <[email protected]> wrote: > > > Hi Mike, > > If you are trying to cap the number of transactions 'N' to a max of 100, > shouldn't the multiplier be > > min(10, sqrt(N)) > > instead of max(10, sqrt(N)) > > So if N=225 as an example, sqrt(225) gives 15 and under those circumstances > could overstate the value of SQN and therefore you would use 10 instead of > 15. Using min gives 10 whereas max gives 15. > > Bing > > >
