cs702, fantastic comment. I am sorta poking around this area too. I'd be curious what benchmark you're using to evaluate performance amongst these repos? If you're up for it, shoot me an email -- my email is in my profile.
Related, are there better representations for music than standard notation (or MIDI)?
I'm wondering what the higher convolution levels could look like, if this was a CNN analyzing an image. Something between a the complete Ableton/Logic export and a MIDI file. Being able to capture the "feel" of a song (or a section within a song) strikes me as an important milestone towards designing really good generative music.
How do you think about backtesting? There are a few short-only shops that specialize in finding frauds. If you get their historical 13-Fs, how would you score against them in terms of precision/recall?
And I guess more broadly, how does alpha with your system compare to a portfolio that holds all short positions by big long/short funds (ex thematic shorts)? Meaning, those guys have full-time humans that focus on this... can you beat them? Very interesting if so.
Very true. I would gladly pay for a feed of financial news, but instead of "clicks" it optimizes on market movement. News articles from sources that, in the past, have predicted market activity.
I.e. "this blog mentioned NYSE:TEVA, and the next day the stock moved materially, therefore site_ranking++". (You'd probably have some TF/IDF saliency metric too, so that a site that mentions all stocks is penalized.)
You could expand the cutoff date by modeling lookalike audiences. A 2018 account that votes on similar things to a 2008 account might be admitted. This affords the moderator an easy “exploration” spigot they can tune up or down.