Happy New Year people!
As I was tidying up my tabs I revisited this piece on wavelet analysis for machine learning and I thought, I know some people who might be interested in that.
Wavelet analysis is like Fourier analysis except you also get information about where certain “frequencies” occur, and with the idea of position comes that of movement…
The wavelet equivalent of a spectrogram is a “scaleogram” (example below) and it occurred to me that one could perhaps predict the next scaleogram in a sequence, then do the inverse wavelet transform to get a price prediction, or one could do some sort of “image completion”/enhancement to get the next time period and hence value.
Either way, since the piece provides nice code examples I thought it could be interesting.
On a related note, I’ve just implemented a sliding window Fourier transform in Python (based on this article): it’s quite efficient but not optimal because the idea was to be able to process price updates as they are received without recalculating everything every time - if anyone’s interested in the code let me know. (It seems to produce the same results as numpy.fft.fft to ~10^-12, so it’s not too bad). I want to see whether something similar can be done for wavelets next, though TBH I’m not that hopeful.