Oh -- forgot to add that was over 4 months, but it's the basic idea. Use 4 months, 1 month, 2 days etc. and see how it changes over time. 24 for statistical significance -- cool, I presented a large number because usually on other posts I see the first thing someone gets slammed for is not having enough samples. Thanks for the comment about "the equivalent of 'portfolio allocation' ". That's what I was looking for -- a name of something I can look into to get more insight. I agree on backtesting a question -- any time I think I've got an idea, best to test it out before letting it grow and transmogrify. As for 30 - 40 strategies, I don't use all of them. But you know, optimizing weekly, and discontinuing the ones that break down, they start accumulating, and compulsive hoarding forbids common sense and hitting the delete key on useless old 2-week wonders that have long since been forgotten ... Unfortunately no way to provide proof I'm not "mizhael", other than to say no, I'm a different dork (in fact my first thread on ET). Yes, I do optimize and chew gum at the same time, from about 1:45 to 2:05 everyday, after a cigarette break. I can't believe it -- how did you know ??????
You have to understand that this is ET. In the nine years I have been here (I am an alias of an alias of an alias) I think I have ass-certained that no more than five people here are mathemagicians, and most of them are gone, except for Maestro and me. The vast majority therefore are innuminate. The innuminati, as it were. So when someone rags that such-and-such a number of samples is unadequate, they don't understand price or sadistics. In trading, NO number of samples is sadistically saddicient because the process of trading is not random. Price is quasi-pseudo-statistico-psycho-manipulated and obeys no known or unknown laws of nature. That is why if I find that when an incertain denomenon recurs within a sample size of twenty I feel no rambunction whatsoever testing down to a subsample size of four to six to make trading derisions. My mathematical fantasies beat the shit out of untuition every day. I am forever undebited to you for the uncept of transmogrification of edges. In the past, I have used the inadequate word evanescent, as I used to be a quantum physicist totally comfortable working in the invaginary plane. In the immortal lyrics of Stevie Nicks, who used to be hot and whom I evanescently wanted to fuck and is now so fat that I probably could: "Winners they will come and they will go." So you see, I have learnt the art of twisting the allegedly impartial backtest to tell me what is true that I want to be true. I know you are not Mizzy. Engrish clearly is your native langrich.
Thanks for the feedback, I find throw an idea out there, you get a couple of leads to follow. As for my strategies, ok, I'll take your suggestions, see if I can build on that. Thanks for your input and feedback, it has given me more than I started with. "No number of samples is statistically significant", "Price is ...manipulated and obeys no ... laws" -- ok, I will think about this, take time to cross-examine my thoughts, to sort out my assumptions and what is objective data. Maestro -- he got me hooked on ET !! I know this is ET and usually is more fun than informative, but he added some posts to one thread, and I read through the entire thread because it was so interesting. Humbly admit, I never figured out the conclusion he was pointing to, but following from the hints and concepts planted, and trying to figure them out, I learned a lot of interesting stuff as a byproduct. Pursuing the concepts in of themselves was their own reward, not necessarily reaching the end point. I probably won't understand all the math, and won't be able to prove anything to anybody. But studying it and trying to understand, that's something I enjoy. "transmogrification of edges"-- I will ponder if there is a deeper meaning here to lead me along. My comment did not concern "edges", like in connecting vertices, or Stevie Nicks's edges, rather the risk of an unverified idea that grows larger and larger ...
very simple, test the correlation between the two return profiles. If they are highly correlated (I understand it is undesirable but nonetheless) build a linear regression and chart the residuals and observe how they are distributed. Repeat the same after varying parameters for each strategy and observe how robust the distribution of the residuals is with those varying parameters. What you want to test for is relationships between returns on at least a daily basis, there is no point to check for relationships intraday between two strategies, at least in my opinion. Thus you do not need to worry that one strategy may not generate any trades on one day but the other does. Hope this gets you started...
"the risk of an unverified idea that grows larger and larger" I like that! I think I am trading one of those right now! As to Maestro, while I admire his facility, he is also facile, the only mathematical comedian I have ever encountered. He can make you believe anything. One is obliged to verify that it is true.
"build a linear regression and chart the residuals and observe how they are distributed" Are you suggesting anything more sophisticated than eyeballing the scattergram and calculating the variance?
he was asking for a starting point, which is what I suggested... For a start graphical observations are way sufficient. Robustness can be easily verified that way. Someone else suggested t-tests, which is another way. Residuals can be tested with Breusch-Pagan, if you are anal about the specific stat details, although the dynamics of markets generally take care of so much variability in statistical tests that, unless the model is not robust, such variability easily shows on simple Q-Q plots.
thanks for the suggestion. I'll have to research for more background info for my own understanding and come back to this, but I think I get the idea you're pointing towards. Further, it sounds like your vote is to weight the returns more heavily than the entry time, direction, or other factors. So from Arthur Deco's suggestion, get the correlation value from OpenOffice (not too difficult), then from asiaprop's suggestion, draw a scatter plot as well (also could be done in the same spreadsheet in less than a long time). Then vary parameters, see how the outcomes and correlation are affected. I'll start with this -- doable within one sitting, and will give me a number, and a visual to give an intuitive feel for the number. Initially I thought fixed stops and limits would have an effect on the OpenOffice correlation value, but now I suppose it would not -- it would reflect the holding time, a factor I have not yet considered. I'm not too worried about correlated or not; instead not knowing the degree to which they are correlated or not is the problem I'm trying to solve. I hope graphical observations are sufficient, I'd prefer a solution in fewer steps than a solution in more steps. But also feel free to drop the statistical test names -- that gives me something I can search for and learn about, the best part of reading through threads. Often, even if I cannot master it, it will introduce me to a new perspective on a problem.
Sorry, not sure how to verify that. Real life has presented so many unbelievable events and busted so many well-trusted truths that I have no confidence in what I believe or don't believe -- so if he made me believe something, it would impossible to tell if I actually believed or doubted it.