Introduction
Especially when it comes to the high-frequency world, trades quickly become crowded and die out. I’ll list out some arbitrages that used to work in their basic form and then died off:
Triangular arbitrage
Perpetual basis arbitrage
BTC lead-lag
Spot / spot arbitrage
This extends past pure arbitrages to more statistical trades or event-driven ones.
Of course, how you optimize a trade varies based on its attributes. A longer-lasting statistical alpha gives more room for execution enhancements than an event-driven alpha like Elon tweeting about DOGE, where the alpha realizes over a very discrete interval.
In this article, we dive deep into how to level up your alphas at very broad level. For specific examples where I’ve walked through this for individual trades, my small trader alpha series gives clear examples of many of these trades (Funding arbitrage, Spot/Spot, Triangular) which with the exception of funding arbitrage is dead in the traditional form, but as we show is still very profitable in modern day by levelling up the trade.
Index
Introduction
Index
Execution
Statistical
Exposures
Universe Expansion
Scope Expansion
Latency
Trading Costs
Leverage
Walkthroughs
Funding
Triangular
Spot
Final Comments
Execution
It’s often the case that we see teams evolve their HFT strategies over a common path, in these stages:
Taker / Taker
Maker / Taker
Maker / Maker
This is how arbitrageurs become market makers so frequently because they start off with a great trade that works fine. Then, as they expand it, there begins a transition to more making in the execution, until eventually the alpha dies and they are left with a full maker system that can no longer be called an arbitrage strategy and is now a full-fledged market-making strategy.
If you have an arbitrage that involves taking, you’ll start by making the first trade using a limit order and entering the maker/taker world. This is easier than all maker because you still ensure the trade remains fully hedged throughout.
With taker/taker, you control both legs, so you can ensure that you hedge after the first leg completes, and with maker/taker, whilst you don’t control the first leg, you can still make sure that the remaining legs will complete.
Maker/maker becomes a little bit more complicated, but by the time you transition to it, the general concept should be quite easy because, hopefully, you’ve solved the core optimization problems related to maker/taker already.
The main thing to consider is the signal decay. If your signal lasts a very long time, then you can afford to make into it. These are arbitrages that are uncompetitive since the exit liquidity on the exit legs will not be taken rapidly.
There is then also the question of what causes the discrepancy. If you get filled on a limit order that is 5% wide and then cross-exchange exit that trade, then it was probably the insanely wide limit order that created the profitability.
However, if you earn 10 bps on an arbitrage where the difference in midprices cross-exchange is 5%, then you’re still earning all the profits from the difference in midprices.
Thus, it’s important to establish how much of the actual edge you are gaining from this maker model. This isn’t just about the decision of how valuable making is, but more about alpha decay. Discrepancies in price tend to resolve a lot faster, but of course, if you are getting filled to make the edge, then it’s simply an analysis of your fill times.
Here’s what you’ll need to solve:
Optimal level of spread (trading off how much you get filled vs.. spread earnt)
How aggressive should I be based on signal decay (the balance between the signal going away over time and you getting better fills if you wait longer / wider)
Quoting constraints (spot requires inventory to quote, and margin considerations will all play a role)