Project Overview

I plan to combine existing techniques to automated forecasting - especially for domains where it is important, but hard to get liquidity because it is not fun for humans. Coalition formation through Shapley values, systematic hyper parameter search, finetuning on resolved predictions using backtesting (questions after knowledge cut off), and incorporating the latest MLE leader approach in using UCT w. pruning as a subagent of the forecaster.

What does success look like?

The best signal is winning. Winning looks like (in descending order):

• Trading directly on Kalshi

• Place high in Metaculus Tournaments

• Backtesting

Why I am I the right person for the grant?

I have previous ML and HFT experience. Most importantly, I really care about how many containers the port of Ningbo handles in 2026, and other boring questions we need to answer to make prediction market useful for real world enterprises.