Everything here is tagged with its real status. Nothing on this page is finished yet.
Out of a university competition on CSI 300 index enhancement: 1st of 153 teams on the public leaderboard, 27th of 85 on the private one. The paper asks whether price–volume signals add anything to a LightGBM cross-sectional baseline on a six-year walk-forward (2017–2022), and uses all 85 teams' scores to measure how much public-leaderboard advantage is overfitting.
experiment design and evaluation protocol; an independent script that re-verifies factor RankIC, coverage and the leaderboard regressions; writing. Model code was built with AI tools and checked by me.
Read the project pageFIFA's Annex C assigns the eight best third-placed teams to group-winner slots through a fixed table over all 495 qualifying-group scenarios. I treat the table as a bi-objective combinatorial mechanism: exposure fairness across source groups versus stability when one qualifier changes. Exact exposure optimum 1/308; a Delsarte linear-programming lower bound on stability loss via the Johnson scheme.
model, proofs and computation. Before submitting, I'm having the two core theorems checked by people outside the project and re-deriving them myself without the draft.
Read the project pageA study project running alongside my Berkeley courses: rolling-window minimum-variance backtests on US sector ETFs (sample vs. Ledoit–Wolf covariance) on one side, the Merton problem and drift uncertainty on the other. The point is to see the same estimation problem from the data side and the continuous-time side.
Read the project page