Thread Reader


Feb 28
11 tweets

1/ A picture is worth a thousand...ideas for a new HFT strategy in crypto? Ever wonder how to go from raw market data to an automated strategy in crypto HFT? Here's an actual case study our team did.

2/ This graph is the BBO (i.e. price of highest bid and lowest offer) of BTCUSDT on #Binance, zoomed in. The y-axis is in USDT, with an arbitrary offset. The x-axis is time, around a 20ms range. Standard stuff. But HFT is all about thinking deeply about simple things.
3/ Mysteries: y-axis: that spread is tiny. One cent on BTCUSDT is... less than 0.01bps? What is this liquidity?? x-axis: When the offer ticks up the bid follows in... 3ms? If you've ever measured the round-trip order latency on AWS, you'll know it's never this fast.
4/ I'd encourage you to pause here if you haven't noticed or thought about these effects before. Learn to fish!
5/ So what gives? Let's analyze the two effects separately. y-axis: The spread is only so tight because of maker rebates. For the highest volume makers on binance spot, that's 2bps. So the spread is meaningless. All we know is that the fair price is within 2bps of the BBO
6/ It's important to validate hypotheses. Let's see...Binance introduced zero fee trading on BTCUSDT at some point. For market makers, they had to remove the rebates, otherwise the exchange would lose money on each trade. And what happened? A normal spread, supporting our story
7/ Now for the x-axis. Here's the graph again. The bid can't be moving up in response to the offer moving, or else the latency showing in the public data would be at least the round trip binance order latency (trust me it's much higher than this). So what's the alternative?
8/ Well the order that updates the bid must be in flight before the offer moves. Are other strategies preemptively sending orders to predict events? Being at the front of a new level is incredibly valuable, because there is less adverse selection to the fills.
9/ You can combine these two effects and formulate a simple strategy to try. There are literally hundreds of details to get right, but at least there's a concrete hypothesis and direction. Go rent that AWS machine, collect the data, and start iterating.
10/ In conclusion, even the simplest graph that you make on your first day as an HFT intern has profound mysteries. This is how quant strategies are born. You look at data, notice weirdness, and try to understand the people behind the movements.
11/ May the alpha be with you. There's nothing quite like the satisfaction of investigating a mystery and turning it into a strategy. It's hard work and a competitive PvP game, but it's great fun and profits too.


@hyperliquidx building a pretty good dex. Previously: Founder of top 10 volume crypto HFT firm, quant @wehrtyou. I tweet about trading, eng, and defi.
Follow on Twitter
Missing some tweets in this thread? Or failed to load images or videos? You can try to .