We tested 45 daily price/volume factors (return, volatility, liquidity, candlestick and technical families, drawn from the simons factor dictionary and the QuantsPlaybook broker-report replications) one by one and in combination: 1,444 configurations per market — single factors, every pair, the best triples and 144 walk-forward composites — each plain, with a 200-day market-trend filter (in cash while the equal-weight market is below its 200-day mean) and hedged.
Strategies were chosen on a training period only (A-shares 2022–2023, HK/US 2020–2022); everything after that is out of sample. The picks were run in two backtest engines, Backtrader and NautilusTrader: decide at the close, trade at the next open, sell first and never buy more than the cash available; an A-share limit-up open cannot be bought and a limit-down open cannot be sold. Both engines and an independent Python recount agree within one percentage point.
No future data: each factor on day t uses bars up to t only (checked factor by factor with a truncation test), and composite weights use only IC that had already finished. We also checked all 3.53 million live signal events: none enters on the signal bar itself, and an earlier signal still running is never counted in a later signal's win rate or MFE.
Period: 2024-10 → 2026-10-15
| Market | Book | Before costs | Max drawdown | After costs | Drawdown after costs |
|---|---|---|---|---|---|
| CN | MFE book: lowest win rate + MFE target | +27.2% | -28.7% | ≈+5.9% (est.) | — |
| CN | Market (equal weight) | +25.5% | -22.3% | — | — |
| CN | Composite (3 factors), top 20% | +29.9% | -18.1% | +21.2% | -18.2% |
| CN | Composite + trend filter | +46.1% | -18.1% | +37.3% | -18.2% |
| CN | Small caps (lowest turnover), top 20% | +50.9% | -23.5% | +46.6% | -23.7% |
| HK | MFE book: turnover + trail 15% | +49.4% | -28.3% | ≈+26.4% (est.) | — |
| HK | Market (equal weight) | +11.7% | -24.6% | — | — |
| HK | Small caps + 60-day residual momentum + trend filter | +52.5% | -17.5% | +44.5% | -17.6% |
| HK | Runner-up: small caps + 120-day momentum + trend filter | +69.1% | -23.3% | +60.9% | -23.4% |
| US | MFE book: worst MAE + MFE target | +195.8% | -36.9% | ≈+171.7% (est.) | — |
| US | Market (equal weight) | +53.9% | -21.3% | — | — |
| US | Kurtosis + small caps, top 20% | +82.7% | -24.2% | +77.2% | -24.4% |
| US | Composite, top 30% | +57.3% | -23.5% | +56.1% | -23.6% |
A-shares and HK: on exactly the same dates as the MFE / win-rate books, factor books made more with smaller drawdowns — both sides before costs, and both after. A-shares: composite with the trend filter +46.1% (drawdown −18.1%), small caps +50.9% (−23.5%), against the MFE book +27.2% (−28.7%). HK: small caps + 60-day residual momentum with the trend filter +52.5% (−17.5%), against +49.4% (−28.3%).
US: the MFE book's +195.8% is far ahead; the factor books (best +82.7%) only beat the market.
Costs: the MFE books trade a lot (149–211 trades of 25% of the book in two years). With the same cost model they lose about 21–23 points in A-shares and HK; the factor books lose 4–9.
Caveats: each MFE book's variant was picked after the fact from 24 on 2026-10-15, and later control studies found it indistinguishable from a random pick; it pays no costs itself (its after-cost figure is an estimate). The MFE figures here are the 2026-10-15 version: rebuilt on corrected inputs on 2026-10-21 they are CN −24.3%, HK −11.7%, US +277.3% (see the long-only backtest page). The A-share trend filter is a fixed rule, but the training period did not choose it (without it the composite roughly matches the market, with 4 points less drawdown). The HK pick leans to small caps, and HK has no delisted-stock backfill yet; adding 235 delisted A-shares changed A-share results by about 1 point. 1,444 configurations were tried, so even out-of-sample results deserve a discount. One period only.
Paper trading: three of these now run every day as paper books under the same rules as the backtest, costs included — CN "Factor composite + trend" and "Small caps (factor)", and HK "Small caps + residual momentum + trend", from 2026-10-16. Their holdings, every trade and their backtest record are on the portfolios page.
Correction (2026-10-18): the first version of this article priced HK stocks with a back-adjusted series that adds past cash dividends to the price; ratios of that series understate returns (for 43% of HK stocks, daily volatility by more than 20%). HK was re-run on a multiplicative, dividends-reinvested adjustment. The training period now picks small caps + 60-day residual momentum; the earlier pick, small caps + 120-day momentum, is listed as the runner-up, and the HK figures changed. A-share limit-up/limit-down checks now use the ex-rights reference price on ex-dates, which moved the A-share figures by up to 3.5 points. The conclusions are unchanged.
Data to 2026-10-15, run on 2026-10-17, corrected on 2026-10-18.