I was referring to the process of mining the data. I had been using the nets to find dependencies. I just came up with a better process of doing it that saved lots of time. Didn't mean to imply that I found something better than nets.
I would add other non-correlated instruments if I could do 5-10 trades per-month in the instrument and it improved the sharpe ratio. To be honest I haven't spent much time looking for high frequency trading ideas outside of the SP market. I spent some time looking for longer term trading methods and found plenty that were profitable. From my perspective I'm nearing the end of my trading career. I don't see myself creating any new massive development projects in the rest of my life. I thought I'd share a little of what I learned along the way and maybe others will want to pick it up and run with it. If you have a single model maybe you'd like to share the markets you trade with it that aren't correlated and give us some idea of the modified sharpe ratio you've obtained. If you have the equity streams in weekly, quarterly, or whatever I'd be happy to run them through my program to spit out the optimal balance for you.
Thanks for answering my questions. As far as reading this stuff in a book, I don't read that many books....Tushar Chande wrote a book several years ago, I think called Beyond Technical Analysis, which alluded to developing a portfolio of systems to trade in different markets. Similarly, he used a rolling period of comparisons to evaluate system/market performance. I didn't read the book. My own efforts were based on variations of a single theme, geared toward equities and the performance of the market when tested against the theme and it's variations. I don't have the work on this computer but in the due course of the next week or so will look at it all again with the thought of contributing something. Thanks for the invite.
acrary, I came back to ET after a few months and noticed this thread. I remember several interesting posts of yours at Chuck LeBeau's forum a few years ago. So I have a general idea about your work. But I think it'd be better if you talked a bit about what type of behaviour you're looking for (breakout, reversal, buy-the-dip, extreme sentiment, exhaustion etc). And approx timeframe, i.e. day, swing, position trades.
If you could describe this process at some point, I am sure that many of us would be very interested and grateful for the contribution. I've also used neural nets to mine data in the past, and have found them to be insanely time consuming. Any way to achieve the good result with less time, would be incredibly helpful. Thanks, bulat
Systems or methods to harvest a certain set of conditions in market behavior every time they arise have always attracted research and adherents. Probably it is the most popular hope or approach by players. However I take a whole approach. I want to go first to the most profitable market. There I want to optimize (and then maximize) the means to remove all the profit offered by that market (stake multiplied by market moves) and as well I want to do that on a whole market day, day on day basis. I pursue that approach from the outset. Time is finite in any day and for my lifetime. So I must arrive each day with the biggest possible shovel. What should anyone do if they want to also follow that approach? Firstly consider implementing the paragraph above. Then what next? Very very short answer: recognize market behavior as a zero sum game in which you monetise the gyrations. I do not do weekly or monthly plays. It means I'm very risk averse. It would also be a diversion. I don't want to divert if day on day I'm taking away the optimum (or maximum). As always horses for courses and each to his own. And it is Acrary's thread to offer his benefits.
i'd describe the problem as, finding the efficient frontier for a portfolio of trading systems. here is a link about finding efficient frontiers (in the link it talks about petroleum company investments, not trading systems) http://www.sis.slb.com/content/software/valuerisk/expert_paper_monte_carlo.asp?printView=true& the link describes two alternatives to the approach acrary uses, which may or may not be more robust. it may be worth a look.
Just found this article: Q Our results clearly show that NNR models do indeed add value to the forecasting process. http://media.wiley.com/product_data/excerpt/55/04708488/0470848855.pdf UQ
It looks like Acrary stopped contributing to his thread. I was always looking for his posts with genuine interest