A small, anonymous Singapore-based quant collective. We publish research, not promises.
AI Trading Lab started in late 2024 as a private research group between three friends who had each independently gotten burned by "signal groups," Telegram gurus, and strategy marketplaces that looked bulletproof in backtest and fell apart within six weeks of going live.
AI 自動交易研究所 2024 年尾由三個朋友創辦 — 大家都曾經被「訊號群組」、Telegram 大師、策略商店坑過。回測完美，實盤六週就爆。
We got together and asked a simple question: if every published backtest looks amazing but almost none of them survive live trading, what are they all doing wrong?
於是我哋問自己一個好簡單嘅問題：如果每一份公開回測都咁靚仔，但絕大部分都喺實盤陣亡，佢哋到底錯咗啲乜？
The answer, after months of reading papers and rebuilding other people's strategies, was almost always the same thing: over-fitting disguised as "optimisation," validation leaks, and backtests that quietly exclude the worst month. The fix wasn't some secret indicator. It was discipline — walk-forward testing, honest cost modelling, and publishing the failures alongside the wins.
答案好一致 — 過度擬合偽裝成「最佳化」、驗證洩漏、回測悄悄剔除最差月份。解決方法唔係新指標，係紀律：滾動驗證、誠實成本建模、失敗同成功一齊公開。
Not because we're hiding — because it forces the work to stand alone.
唔係要收埋身份，係要等研究本身講嘢。
Without a personal brand to protect, there's no temptation to cherry-pick, rescue a dead strategy, or pretend a failed experiment didn't happen.
You shouldn't follow a strategy because of who built it. You should follow it because the statistics, costs and validation hold up under scrutiny.
No "funnels," no paid mentorship, no Lambo screenshots, no upsells. The day this lab tries to sell you a course is the day we've failed.
Every chart, every metric, every walk-forward fold on this site is generated from a Python notebook we can re-run on demand. We don't publish what we can't recompute.
每一張圖、每一個指標、每一次滾動驗證，全部由 Python notebook 生成，隨時重跑。唔可以重現嘅嘢，我哋唔會出街。
Spreads, commissions, swaps, and estimated slippage are all built in from the first line of code. A "gross of cost" return number is not a return number; it's a sales pitch.
點差、佣金、隔夜息、滑點 — 從第一行代碼就加埋。冇計成本嘅回報不是回報，係銷售話術。
We show every month, including the ugly ones. Drawdowns are the truth about a strategy; equity curves are a marketing slide.
我哋會展示每一個月，包括難睇嘅。回撤才是策略的真相；資金曲線只是行銷素材。
We will never promise a specific return, a specific win rate, or a specific date. We will show you the work and let you decide what it's worth.
我哋永遠唔會承諾某個回報、某個勝率、某個日期。我哋展示工作，由你判斷價值。
Four members, four specialisations, zero real names.
四位成員，四個專業，零真名。
Statistics and strategy design. Ex-sell-side quant. Builds the walk-forward framework and decides which strategies enter paper testing.
Position sizing, correlation budgeting, and execution-cost modelling. The reason every strategy on this site has a drawdown estimate you can actually trust.
Tick data pipeline, MT5 integration, Python research environment. If a chart renders on this site, Data 03 built the pipeline behind it.
Turns the team's notebooks into something you can actually read on a phone at 11pm. Bilingual EN/中文. Ruthless about jargon.
Standard tools used properly, not exotic tools used as a flex.
用普通工具做正確嘅事，唔係用奇怪工具嚟炫技。