Short-term momentum loses 90% in China. The same signal that works best at 6 months in US markets becomes deeply negative. We test every lookback from 3 to 24 months on 5,444 A-shares and show why the Chinese market inverts the rules. 208 months, 2009–2026.
The best result uses a 21-month lookback, skip 1 month, top 7 stocks equal weight, with a CSI 300 MA150 regime filter. Universe: all A-shares excluding ST-designated stocks (5,444 total). Period: January 2009 – May 2026.
Compare to the US, where 6-month momentum on large caps delivers +13.5% CAGR at Sharpe 0.68. China achieves similar returns at much higher risk, using a lookback window 3.5× longer.
21m-1 compounds to ¥809K from ¥100K. 12m-1 barely reaches ¥133K — below the CSI 300.
| Lookback | CAGR | Alpha | Sharpe | Sortino | Max DD |
|---|---|---|---|---|---|
| 3m-1 | −10.5% | −16.3% | −0.40 | −0.61 | −92% |
| 6m-1 | −10.2% | −16.0% | −0.40 | −0.52 | −91% |
| 9m-1 | −2.2% | −8.0% | −0.14 | −0.20 | −84% |
| 12m-1 | +4.4% | −1.3% | 0.08 | 0.13 | −60% |
| 15m-1 | +7.6% | +1.9% | 0.21 | 0.30 | −47% |
| 18m-1 | +6.1% | +0.4% | 0.16 | 0.20 | −64% |
| 20m-1 | +11.4% | +5.6% | 0.31 | 0.46 | −55% |
| 21m-1 ★ | +13.0% | +7.2% | 0.38 | 0.55 | −56% |
| 24m-1 | +10.8% | +5.1% | 0.30 | 0.42 | −63% |
| CSI 300 | +5.8% | — | — | — | −47% |
Short-term momentum is toxic in China. 3–9 month signals are strongly negative. Every lookback below 12 months destroys capital. This is the opposite of US markets, where 6-month momentum is the strongest variant.
Chinese A-shares are approximately 70% retail investor volume. When a stock spikes over 3–6 months, it draws in momentum-chasing retail buyers, creating overcrowded positions that reverse sharply. The stocks with the best recent performance are frequently the most dangerous to own next month.
By 20 months, the dynamics change. The retail noise clears, and what remains is genuine trend persistence — driven by fundamental improvement, earnings surprise, and the minority of institutional investors who take longer-horizon positions. This is the signal that 21m-1 captures.
In the US, institutional investors dominate volume and react quickly to earnings. Momentum is incorporated over 6–12 months. In China, the same fundamental signal takes three times as long to propagate through prices.
The CSI 300 experiences violent bear phases that destroy momentum portfolios: −65% in 2008, −43% in 2015, −35% in 2018, −25% in 2022. Without a regime filter, any momentum strategy simply rides these crashes to the bottom.
| MA Filter (on CSI 300) | CAGR | Sharpe | Max DD | % Invested |
|---|---|---|---|---|
| None — always invested | +2.4% | 0.01 | −89% | 100% |
| MA100 | +10.0% | 0.25 | −66% | 52% |
| MA150 ★ | +11.4% | 0.31 | −55% | 53% |
| MA175 | +10.7% | 0.29 | −55% | 51% |
| MA200 | +10.6% | 0.28 | −62% | 53% |
Without a regime filter, 20m-1 momentum loses nearly 90% peak-to-trough — almost double the CSI 300's worst drawdown of −47%. The concentrated 7-stock portfolio amplifies every crash. Going to cash when the index is below its 150-day MA turns a survival problem into a manageable strategy.
The strategy is invested roughly 53% of the time. When the CSI 300 is below its MA150, the portfolio sits in cash. This costs some upside during early recoveries but eliminates the catastrophic losses that make long-term compounding impossible.
The MA150 and MA175 filters perform similarly, both far better than MA200. The MA100 fires too early and cuts the portfolio out of valid uptrends. MA150 is the sweet spot.
US comparison: The MA200 overlay on the S&P 500 cuts US momentum drawdown from −57% to −28%, at a cost of ~1% CAGR. In China, the regime filter is not optimization — it is survival. Without it, the strategy is not viable.
| Dimension | US Markets | China A-Shares |
|---|---|---|
| Best lookback | 6 months | 21 months |
| 12m-1 (academic standard) | +13.0% CAGR, Sharpe 0.61 | +4.4% CAGR, Sharpe 0.08 |
| 6m-1 | +13.5% CAGR, Sharpe 0.68 | −10.2% CAGR, Sharpe −0.40 |
| Regime filter | Reduces DD by ~30pp | Required for viability |
| Market cap requirement | Large caps required (>$5B) | Large caps destroy alpha* |
| Dominant investor type | ~70% institutional | ~70% retail |
| Signal propagation speed | 6–12 months | 18–24 months |
* The large-cap / small-cap reversal in China is the subject of the next article in this series.
3 to 9 month lookbacks produce strongly negative returns. The academic 12-month standard barely breaks even at +4.4% CAGR — below the CSI 300 benchmark. This is not noise: it is systematic reversal driven by retail herding.
Returns improve monotonically from 12 to 21 months. The 21m-1 variant achieves +13.0% CAGR at Sharpe 0.38 — comparable to US large-cap momentum in absolute terms, though with higher risk. The signal weakens slightly at 24 months as it begins to capture mean-reverting long-term fundamentals.
Without the CSI 300 MA150 overlay, 20m-1 momentum loses 89% peak-to-trough. The filter costs nothing in CAGR (from +2.4% to +11.4%) and reduces drawdown from −89% to −55%. This is the largest single improvement in the entire study.
Every dimension inverts: shorter lookbacks that excel in the US are the worst in China. Large-cap filters that are essential in the US eliminate China's alpha entirely. The mechanism is different — this is not the same factor in a different country. China momentum reflects a different structural reality: retail-dominated, policy-sensitive, and institutionally under-covered.