Quant Fund

I'd been trading crypto by hand for a while and kept hitting the same wall: the market wasn't beating me, I was. I wanted to fix that with engineering: a system that spots opportunities without emotion and remembers everything. The deeper goal is to run a quant fund at micro scale, one person replicating what real funds do with quantitative methods. It leans heavily on Marcos López de Prado's Advances in Financial Machine Learning: the five layers, meta-labeling, and walk-forward tests to check whether a strategy actually has an edge.

Production runs on a small Linux server I administer myself, and the research work happens on my old Acer Nitro gaming laptop, wiped and turned into an eight-core compute node. Around the bot I built my own trading desk: a trade journal that auto-imports my Bybit trades (a daily status digest lands on Telegram), and a custom TradingView-style chart to see what the bot is doing instead of reading code.

Eight months in, the thing I actually learned is that finding signals is the easy half. The hard half is not fooling yourself. So every strategy is pre-registered against a pass/fail bar before I see the data that judges it, and every death goes in a register with its numbers and its cause. Twenty-two strategies are in that register. Zero have survived. That is an honest state, not a gap to fill.

Two hypotheses got measured and closed as negatives instead of quietly shelved, and the number that governs everything else came from 330 of my own real trades: roughly 24 basis points per trade before a result can even be told apart from noise.

Then I pointed the same suspicion at the system itself and audited it against how a small fund actually works, with every verdict run past two skeptics whose job was to refute it and to pick the more pessimistic reading. It found that my walk-forward validation never held anything out, its test window sitting inside its train window, while production parameters cited it as proof they were validated. Better to learn that from an audit than from a drawdown.