strategies research

Discussion in 'Strategy Development' started by kevinzhu, Sep 30, 2010.

  1. kevinzhu



    I work on strategies professionally and I want to ask if there is anyone who has done some serious research trying to find certain patterns from either option or stock historical behaviors.
    I mean using formal backtesting and portfolio simulation technologies, not low cost package software toys.

    I have done a lot quantitative research on stock index, and am interested in sharing experience, data, code, and potentially other things if I can find anyone who share similar interests.

  2. rosy2


    what do you have to share?
  3. These will find automatically strategies based on your risk/reward parameters.

    TSL is quite expensive but has many capabilities. I have not used it as I cannot afford it and I am not familiar with any curve-fititng issues that may be present. Some people claim that genetic algorithms is just another way of curve-fitting, more sophisticated.

    PAL is limited to price patterns or what I call micro-patterns. Useful in many ways but requires some work to avoid getting fooled by randomness. Easy to use to study patterns in stocks but extra caution should be exercised as all that glitters is not gold.

    The problem with patterns is when they fool you as statistically significant but they are actually random that have survived by luck and are subject to quick mean reversion, usually in the next two or three instances. If you can get a grasp of this problem and find even a sub-optimal solution you can make money.
  4. kevinzhu


    Thanks for your post.

  5. kevinzhu


    Thanks for your reply. I have option data from 2002. I have done some coding work around those data. I also tried reproducing some results from academic publications, to just name a few things

  6. Basically, As bill says. People will test a thousand patterns or indicators. Then they will rank the results and get excited by say, the best 50. But isn't is expected statistically that 50 out of the 1,000 would be in the top 5%?

    And then you take these 50 and forward test them a few months. And out of these, 2-3 will seem like the top 5% of the 50. And the newbie trader gets excited. But as Ecclesiastes says: Is it useless, useless. A chasing of the wind..."

    In fact, John Wilkes Booth's last word I believe were basically"Useless, useless" as he looked at his hands, paralyzed from a gunshot.

    There are going to be few traders who will backtest (or even walkforward test) their way to lifetime profits and success. You either invest the time, energy and money to learn the markets, or you become one of the 99%.

    Bone, you there to for the 5th time, to offer $10K to show up at the airport and compare penis lengths so we can look at your photoshopped 18 year statements devoid of any metrics, which you refuse to publish here? *smile*
  7. This is what it boils down to. No miracles to be expected. My estimate of the expected time for learning what is going on with an average 10 hour/day devotion to study and analysis is 7 years with a standard deviation depending on depending on IQ. It took me about 6 years to understand what is going on.
  8. rosy2


    great. do you have anything useful? i have tick data for the past several years on a number of instruments and have coded thousands of strategies that dont work.
  9. I have met several individuals who wasted time and money to test strategies on Brownian motion (tick data) and never succeeded, as expected. I also did the same for some time but I quit fast. If you move your ambitions to 30 minute timeframes and above there are still many edges to be discovered.
  10. Coded thousands of strategies?

    Maybe you meant "machine-generated thousands of strategies"?
    #10     Oct 1, 2010