A Simple S&P System

About the Author Jeff Swanson

Jeff is the founder of System Trader Success - a website and mission to empowering the retail trader with the proper knowledge and tools to become a profitable trader the world of quantitative/automated trading.

follow me on:
  • Thomas says:

    May I use this example to ask a more general question? You optmize a higher frequency component (baseline system) together with a lower freqency component, the MA based market filter. Is there not -in general – a much longer periode of data for the market filter required to be significant?

    In the special case a simple MA may generate just enough trades in 14 years, but what about a little more complex filter like a MA crossover?

    Thomas

    • Hello Thomas. Not sure I understand your question. The baseline system look-back values were not optimized with the SMA filter. A short period SMA shows significance given the length of the historical data and the number of samples. Longer look-back periods showed a steady decline in performance. A more complex filter is worth testing, but I wanted to keep with the theme of the article which was “simple”.

      • Thomas says:

        Lets assume, a market filter switches the regime once a year. In order to be statitically significant it may require 30 or more years of data. My question is, if it is correct to optimize the baseline system together with the baseline system, if you use only 12 years of data, or if it is better to use standard parameter.

      • Thomas says:

        Corrected: Lets assume, a market filter switches the regime once a year. In order to be statitically significant it may require 30 or more years of data. My question is, if it is correct to optimize the baseline system together with the market filter, if you use only 12 years of data, or if it is better to use standard parameter for the filter.

        • In my opinion, it’s not the years of data that is important. It’s the number of trades and the market conditions covered. With over 100 trades (samples) spanning both prolonged bull/bear markets I would say our 12 years of data is statistically significant. You can optimize the regime filter along with the baseline. It’s more ideal to do it separately. Alternatively, using a walk forward optimizer to optimize the regime filter with the baseline would most likely give you the most realistic results.

          • Thomas says:

            Thank you for your proposal. I made a stand alone walk forward analysis of a simple MA crossover on the S&P500 and changed the size of the in-sample window to see how long does it take to train the market filter. In order to get a more ore less steady equity curve a 5 year window seems to be required. The resultig system makes around 2 trades per year.

            With this results in mind I would like to ask my question again. Assume you have a system with a statistically significant amount of trades. Don’t you loose significance if you add a low frequncy market filter and optimize the complete system?

          • In general yes. To another point in regards to your specific system, did you try your system on the S&P cash market? This goes further back than the futures market and may allow you to get more trades.

  • Marco says:

    Perfect system development as usual Jeff! What about using volume another filter different from price? Usually in my testings it helps a lot.

    http://nightlypatterns.wordpress.com

    • Marco, that certainly would be worth testing! As usual there are many things that could be done to potentially help improve the performance. Thanks for the idea.

  • >