djils

[djils @djils__]

@macrocephalopod @MostlyMonkey Cointegration?

[djils @djils__]

@macrocephalopod My bad, caught this late. This is a good one!

[djils @djils__]

@QuantEspresso @systematicls @quant_arb Surprising collab indeed!

[djils @djils__]

@macrocephalopod @oxbquant @quantymacro Totally get the overfitting point, but isn’t it still important to look at how each feature relates to the target first like checking correlation, lagged correlation, correlation in the tails etc?

[djils @djils__]

@dindjic @macrocephalopod Not sure that would help much

[djils @djils__]

@macrocephalopod @elletwocache Probably worth reminding the Databento folks that P72 and Cubist are essentially in the same bento box.

[djils @djils__]

@macrocephalopod A bit of luck might help too

[djils @djils__]

@macrocephalopod Not good, I guess… but a good-looking chart with a nice MATLAB vibe, though.

[djils @djils__]

@zetaquant @macrocephalopod @systematicls @ScottPh77711570 I wouldn’t dare let it sit on the same shelf as my favorite math books - say, Gilbert Strang’s Linear Algebra. That would just feel wrong

[djils @djils__]

@zetaquant @macrocephalopod @systematicls @ScottPh77711570 Do Do Prado’s books qualify as math books?

[djils @djils__]

@zetaquant @macrocephalopod @systematicls @ScottPh77711570 Also, one can eye-ball bugs in nearly every code sample lol

[djils @djils__]

@zetaquant @macrocephalopod @systematicls @ScottPh77711570 I skimmed through the first chapters of his book and found the reading artificially difficult, as if it was trying to formalize everything just for the sake of it.

[djils @djils__]

@Ligne14_RATP Vous rendez-vous compte de la misère que vous imposez aux usagers en pleine heure de pointe pour aller au travail ? Encore une fois, c’est tout à fait inacceptable.

[djils @djils__]

@macrocephalopod I remember that post. The funniest part was that that non-lagged equity curve was set as his X wallpaper :)

[djils @djils__]

@macrocephalopod There is still a working link: https://t.co/ApLZyvhJxU

[djils @djils__]

@macrocephalopod Too bad his interview on risk biscuits YouTube channel was deleted too. Pure gold.

[djils @djils__]

@macrocephalopod @EntropyChase Can you reasonably expect to build traditional CTA-like strategies with such a narrow universe?

[djils @djils__]

@EntropyChase Don’t forget market impact. Start with simple stuff. Assuming impact is proportional to square root of order size is a good starting point imo

[djils @djils__]

@EntropyChase That said, the signal could just as easily be binary (-1 or +1), as it doesn’t make much of a difference in practice.

[djils @djils__]

@EntropyChase Personally, I typically normalize the signal (z-score) by the instrument’s volatility and apply a scaling factor to achieve a daily target for the entire portfolio in USD terms.