Oct

23

Data, from James Sogi

October 23, 2006 |

Isn’t it best to follow Dr. McDonnell’s advice and use out-of-sample testing? In-sample may lead off the cliff when cycles changes. Trades that work across cycles are better than cyclic, are they not, despite their rarity? Though you capture the cycle, you don’t know they have changed until you’re drawn down. Shortened data also lowers the usable time frame for testing.

Victor Niederhoffer replies:

All I can say is:

You are right                                                          

And we are right                                                       

And all is right as right can be                          

And I expect you’ll all agree                                   

That Sogi was right to so decree

From — Our Great Mikado, Virtuous Man

I like to condition things on the last 100 or 200 days of market moves and run tests within that framework. Ideally, one should do everything on prospective basis, constantly updating the hypotheses based on data available to you at the time.

GM Nigel Davies responds:

To capture ever-changing cycles there’s a case for using very recent data wherever possible, and perhaps three years is enough. But then, how should one deal with the rally from July, which has made just about every aggressive short-term system look rather good? They also looked good in April.

Is there not a case for using test data up to the last time the market was at a similar rank (e.g. I have Friday’s close as #2 from the last 20/50/100 days) in order to get more realistic test results?


Comments

Name

Email

Website

Speak your mind

Archives

Resources & Links

Search