Our approach
How the numbers work
Same data, same code, same rules for everyone.
The honest card
The gap is what most backtests miss.
Survivorship
Testing only today's listed companies skips those that went bust, went private or left the index â often the worst losers â so almost any rule looks better. We test on the stocks available on each past date, delisted ones included where the data has them.
Overfitting
Try enough settings and one will look great by chance. We count your tries, test on years the rule wasn't tuned on, and discount the result. A rule that works for only one setting is describing the past.
Costs
Every trade pays commission, exchange fees, Hong Kong stamp duty and the bid-ask spread; frequent trading can lose most of its paper profit to them. We charge broker-like fees on every simulated trade and show what costs took.
Do the scores predict?
Each month since 2018, we ranked every stock by each score and compared its next 3-month return with the average stock.
Neither score beat chance at ranking future returns, so Overall weights them equally. The scores describe a stock; they are not a forecast.
| Score | Rank correlation | Strength (t) | Best tenth minus worst tenth | Months | Weight in Overall |
|---|---|---|---|---|---|
| Business | 0.008 | 0.6 | 0.4% | 100 | 50% |
| Trend | -0.012 | -0.3 | 0.1% | 87 | 50% |
Business
| Band | Higher after 3 months | Against the average stock | Fell 10% or more within 3 months | Stock-months |
|---|---|---|---|---|
| Business strong | 57% | 0.4% | 39% | 11619 |
| Business good | 56% | -0.2% | 38% | 11865 |
| Business mixed | 56% | -0.4% | 39% | 11873 |
| Business weak | 56% | 0.1% | 41% | 22886 |
Trend
| Band | Higher after 3 months | Against the average stock | Fell 10% or more within 3 months | Stock-months |
|---|---|---|---|---|
| Trend strong | 54% | -0.1% | 41% | 11288 |
| Trend good | 57% | -0.0% | 36% | 11529 |
| Trend mixed | 57% | 0.3% | 36% | 11534 |
| Trend weak | 56% | -0.1% | 41% | 22234 |
Strength below 2 could be chance. Returns include dividends; a stock that stopped trading keeps its last price. Overlapping monthly samples lower the strength accordingly.