Backtesting individual stocks

Discussion in 'Strategy Development' started by CA04, May 29, 2011.

  1. CA04

    CA04

    Looking at individual stocks.... I don't see the point of back-testing 1 or 5 or even 10 years back. All I want is a general idea of what is happening NOW.

    I'm looking at how a stock has acted over the previous 10 sessions and getting descent results going into the 11th day...

    Picking 10 days was arbitrary... all I knew was that I wanted to test a very short-term window...

    Does anybody look back very short term to get a general impression of how an individual stock is currently acting?
     
  2. Well "Now" is highly dependent on what other stocks are doing at the same time...

    IBM EMC JNPR etc..

    will behave in a group manner.. so you'd want to look for highly correlated activity and follow that momentum.
     
  3. dyson

    dyson

    Back-testing is a tool used by the systemic/mechanical trader to find a pattern that gives them a probabilistic edge. Trading is all about probability. The more data points used to validate the pattern, the easier it is to determine the pattern's reliability.

    Running a back-test on a small data set (i.e. 10 days) will increase the noise. You may find a pattern that worked great for that small data window, but has failed reliability in the past.

    See signal to noise ratio.

    You may also be confusing back-testing with curve-fitting based on the most recent data set.

    10 days or 50 days or even 100 days is still a small data set. For example with only 100 days of data, you cannot even determine the 200 day moving average for a chart.

    I don't think you will find a system trader who uses such a small data set. Depending on the interval of the data (tick, 1 min, 5 min, 60 min), the minimum would be from 2 years to 10 years.

    With 5 years of data, you can run the back-test on 3 years and a forward-test on the remaining 2 years.
     
  4. BwPirt

    BwPirt

    How else would you know your strategy is worth anything if you can't backtest?
     
  5. BonScott

    BonScott

    I think you should test in a bit more detail.
    I think testing over several years will give you the confidence that your ideas work.

    I have been testing my ideas against 4 years of historical data on approx 2500 stocks in the NASDAQ just to get an idea of where I am going.

    I am using Excel, because that is what I know how to use.
    I wish I could program, then it would be a lot quicker.

    When I have finalised my ideas, then I will backtest them against another 6 years of data, including other exchanges.
    Probably 10 years of data on 7000 stocks if I can.

    10 days analysis on one stock may work, but my "Mr Safe" personality does not allow me to commit thousands of dollars unless I'm sure in my head that I will get back thousands more.

    Good luck with it. I truely hope you suceed, because I have got so much great information from the people on this site.
     
  6. Backtesting is only give you a significant confidence about the behavior of your trading in the PAST. However, this would have no any relevance with the present or the future.