FFT (Fast Fourier Transform) constructs a “best cyclic approximation to the data” that can be constructed with N cycles. And you get to pick the N value. Better yet is that the output is a smoothed representation of the raw data without any lags. Wow, no lags! However, FFT assumes that the cyclic behavior is repetitive from the beginning of time to the end of time. That’s great for fitting data, but not generally reliable for forecasting markets. Also, every time you add or drop a datapoint, a subsequent cyclic approximation will have different values over the entire period.

Suggestion: FFT is fine for seasonally adjusting past macroeconomic data, but your expected value of using it for trading will be negative.





Speak your mind

1 Comment so far

  1. Kermit on September 28, 2011 12:02 pm

    For an interesting read on how cycles in chaotic systems work, read James Gleick’s excellent book “CHAOS: The Making of a New science.” What is particularly interesting (for me, at least) is that he describes the cyclic behavior of a very small container of liquid helium (as I remember). The chapter (again, as I remember) was called “The Experimenter.”

    IMHO, that the cyclic behavior in a container of liquid (when heated) looks the same as cyclic behavior in many markets is by itself fascinating. Gleick’s description of the cycles, however, is very instructive. It is also easy to see why they can be so difficult to use in models.


Resources & Links