Coastal Quant

@0xFaust12 @macrocephalopod @therobotjames @robertmartin88 @witchqueendot @Mtrl_Scientist Can always count on @macrocephalopod + @therobotjames for useful insights

@macrocephalopod @therobotjames @robertmartin88 @witchqueendot @Mtrl_Scientist @coastal_quant

@Merridew__ Historically DXYZ publishes holdings with the SEC 60 days after quarter end consistent with N-PORT rules. Expecting it to pop in late May, if VCX’s premium means anything…. Also don’t forget about the SpaceX shares it acquired in 2022!

@KrisAbdelmessih @Merridew__ The story I like to tell is: L/S funds run at some fixed dollar vol target -> market-wide vol rises so they need to sell positions to maintain dollar vol target -> they sell crowded names which leads to broad idio losses which is irrespective of their fund beta

I don't know how to think about this formally so I'll just throw it out there but I see highly volatile markets as a tax on everyone except mm'ers. It's like every strategy is whatever strategy it is minus some MM risk premium piece. In so far as the defensive leg of any strategy is held by a non market maker it also suffers from this. So even a strategy that runs low beta or net, if it's high turnover it still gets some amount of screwed in a world of 25 vix Does this sound sensible (although in this case being down 5% on a 0 beta is worse than even an aggressive version of mm tax) but generally curious what you think

@Merridew__ That Anthropic holding was publicized through a press release and hasn’t been officially filed yet. Would not be surprised to see a massive pop once people digest that Anthropic has become the biggest holding. Not to mention the second largest holding is SpaceX

@SowingAlphaSeed Cool visualization @0xFaust12

This factor rotation feature is really neat. factorstoday.com/viz/quadrant

@Merridew__ Straight up shorting VCX is basically an arb if you can hold for 6M until pre-IPO lockup expires The margin risk in the meantime though…

@Merridew__ I hate this fund

@Duderichy @thejaikannan startup invents taking risk!

anytime someone promises you to beat treasuries, you’re taking on additional risk if they’re not being honest with you about it, they’re scamming you

@__paleologo Not-so-subtle reference to our favorite SF fund @0xFaust12

If you hold a levered portfolio of ~7 stocks for a long time, chances are that you will look like a genius or like an idiot, whichever comes first. Kelly, Information Coefficient, etc. are not useless concepts. Many such cases these days.

@weaponizedFOMO @0xfdf Yep, this is a sim. Ex-ante IC is 0.01 per signal

@coastal_quant @0xfdf realized IC?? you mean realized correl?

Quick demonstration of weak signals that independent don't make money crossing the spread can combine to be profitable H/T @0xfdf Each signal is on a cross section of 100 stocks, IC 0.01, stock vol is 1%, bid/ask spread is 2%. Averaging features increases IC, overcoming spread

@macrojellyfish @macrocephalopod @Mang0_trad3r It’s cost not cosine

@macrocephalopod @Mang0_trad3r Why do you subtract cos(t)?

@Merridew__ @macrocephalopod Or instead of ER not being big enough to overcome the spread, say 1 day vol isn’t big enough to overcome the spread.

For the quant finance crowd: let's say you have a signal with high IC forecasting 1-day returns and high in-sample Sharpe ignoring costs. However bid/ask spreads are large so simulating a strategy including fees leads to very negative performance. How do you decide whether to: - Switch to maker instead of taker orders - Subset your signal to only tail positions - Slow down your signal / slow down your trading - Find longer term signals and use this faster one as an excution overlay - ... something else?

There’s some conflation between “statistical significance” and “implementability” here, but yes in typically systematic equity a correlation of 0.2% is quite strong. I do find the point about E[r] << StdDev(r) interesting. Let’s say you ensembled a bunch of non-implementable signals together to build some stronger forecast, but half of your cross-section still has an ER less than the bid/ask spread. Is it fair to just toss out those assets from your investment universe, since they can’t cross the spread even if you have 100% hit rate on forecasting their directional move?

@Merridew__ microcaps**

@Merridew__ (Not my signal) but try looking at a simple 1-day reversal on microchips. Total return is fine but idio return is better. You'll find you can predict next-day mid-to-mid moves pretty well with this one signal but hard to monetize because of spreads

Maybe I wasn't super clear but the signal here doesn't stop having positive IC OOS, the issue is the implementation of the signal - trading it in the face of wide spreads. Interesting point about IC decay too... in practice I've yet to see an equity signal that's had a statistically signficiant step-function type decay. Maybe if I squint closely at 13F data I could see it, alpha gone on release day and if you can preempt then you can make money.

@f4talStrategies @coastal_quant 💯 💯

@pierre_menard01 Crossing the spread much easier to model/backtest than adverse selection for sure. I could also pre-filter my investment universe to spreads that are "reasonable" or in the scale of my return (some spreads are so wide that 1 day worth of vol can't hope to cross it)

@coastal_quant depends on how youre modeling it. if you are assuming that you only get fills on wide markets by crossing bid-ask, then this makes sense. However, you may get more passive fills by posting near the market (although now adverse selection kicks in). would just have to test live imo

@AgustinLebron3 Super useful!

@coastal_quant 1. Switching to maker probably makes it worse. Harder to simulate AND likely makes even less money than taking. 2. See if filtering to high edge situations makes $. 3. Find other signals to combine it with.