Quantitative Trading Strategies - How I went from 10k to 100k to 1M (part 3.1: event driven arbitrage)
Prediction markets have attracted considerable attention over the past year, driven by the proliferation of exchanges, liquidity venues and increasingly creative contracts. Pricing theory around many of these contracts remains nascent, and outsized opportunities still exist while more conventional market makers find their footing.
I have traded more than $100 million in volume across these platforms, generating more than $1 million in PnL. My operations have primarily revolved around three strategy families:
cross-exchange, combinatorial and statistical arbitrage;
event-driven arbitrage;
high-frequency market making.
The first family is somewhat capacity-constrained. Liquidity remains shallow across many nascent venues, and seemingly identical contracts are not necessarily fungible. Two exchanges may list economically similar questions while using different resolution sources, deadlines, wording or discretionary procedures. What looks like a locked arbitrage can therefore contain rather more basis risk. In fact, as I write this post, here is an obvious one between poly and hpl.
The second family is considerably more scalable, but requires serious thought around engineering processes. I ran reduced-form versions of these strategies in the past before eventually pausing them. Too many parts of the system were “non-automatable”: particularly around sourcing new contracts, registering new data sources et-cetera.
Recent improvements in agentic systems—particularly following GPT5.5 and Opus4.8 make the problem interesting again in my opinion.
The third strategy family, high-frequency market making, is the most scalable of the three. It also requires proper domain expertise, quantitative reasoning and a reasonably intimate understanding of exchange microstructure. This is what placed me near the top of the relevant leaderboards in terms of both volume and PnL.
We have, in fact, documented an entire lecture series detailing the design and implementation of my HFT ops, to be later posted here.
It will not be released until yours truly is no longer involved in said markets.
For now, we will focus on event-driven arbitrage. Our discussion here will take a more idealistic approach compared to what I ran, accounting for the scalability now possible under accelerated agentic capabilities. The objective is a pipeline capable of serving hundreds to thousands, of prediction-market contracts.