Any EE engineers here tried spectrum based trading systems?

Discussion in 'Strategy Development' started by mizhael, Feb 23, 2010.

  1. Any one tried any FFT, frequency stuff?

    Any working examples? Just wanted to get an idea how it could work...

    Thanks...
     
  2. Yes. I did an FFT on closing price data and saw spectral peaks at 5 days and 20 days. LOL
     
  3. Sadly, there is nothing there. Except for people who sell books about it and software using it.
     
  4. All kidding aside, what's a moving average? It's a low-pass filter.
    What's an oscillator? It's a high-pass filter.

    I just posted this link recently. The guy has some very good papers talking about applying signal processing techniques to price data.
     
  5. vikana

    vikana Moderator

    I just want to point out the FFT and standard Fourier techniques don't have the resolution you need for trading.
     
  6. And they violate the causality requirement for time series.
     
  7. FFT/harmonics will rarely work. An example of where it will work is prices for heating oil... there should be strong seasonality every year.

    For equities/indexes look at forecasting time series, keywords: non-stationary, unstable, transition shifts, structural breaks, regime switching.

    Somewhat related thread with mention of cycle forecasting:
    http://www.elitetrader.com/vb/showthread.php?postid=2605630
     
  8. Given that you can implement any resolution (as long as it's a power of 2) and any wavelength of filter you desire with an FFT, this strikes me as an odd statement.

    Within a single transform, true, however not for the last data point in a generated series (as there is no future data).

    I'd be interested in reading your feedback on these comments.
     
  9. Murray Ruggiero

    Murray Ruggiero ET Sponsor

    FFT don't work on market data. I have worked with MEM, the berg algorithm and have my own version in CycleStudio for TradersStudio. Making MEM work with market data requires a lot of tricks and when developing my software it took 2-3 months to resolve all of these problems. John Ehlers is the one who first applied this method to market data and calls it MESA.

    I don't have MESA but my own version of MEM for market data.
     
  10. I can only quote Lao Tzu: "Those who know, do not speak. Those who speak, do not know."
     
    #10     Feb 24, 2010