project
IMC Prosperity 4
result#194 algo, top 1.4% of 18,800 teams.
A multi round algorithmic trading competition where simple models, inventory discipline, and careful iteration mattered more than fancy machinery.
Problem
A multi round algorithmic trading competition across several asset classes, with changing products, partial feedback, and a leaderboard that rewards robust execution over one clean model.
Approach
I treated each round as a separate research problem: write a small simulator, inspect fills, start with simple fair value or mean reversion rules, then add inventory limits and product specific execution logic.
The strongest pieces were market making with inventory control and exploiting repeated structure across rounds. The weaker part was that not every product had enough feedback to justify complex modeling, so I kept those strategies deliberately simple.
What I tried that failed
Several ideas that looked attractive in local replay did not generalize once the next round changed the product set or the order flow.
The useful lesson was to prefer rules that degrade gracefully and to be suspicious of local profit that depended on one narrow fill pattern.
Finding
#194 on the algorithmic track and #256 overall out of more than 18,800 teams in IMC Prosperity 4, top 1.4%.
How to reproduce
The public leaderboard result is the external artifact. I have not linked private code because I want to review competition rules and remove brittle notebooks before publishing.