This report studies a 20-stock China A-share lithium battery industry universe and builds a quantitative framework that combines multi-factor stock selection with CTA-style trend exposure control. The research finds that simple momentum chasing was not the best approach in this sample. Instead, low volatility, profitability quality, earnings improvement, and valuation repair signals were more effective. Adding a CTA trend module materially improved drawdown control and strategy stability.
本文以 A 股锂电产业链 20 家头部上市公司为研究对象,构建覆盖行情、估值、成长、质量、波动率与相对强弱等维度的多因子数据库,并开展因子有效性检验、多因子组合回测、CTA 趋势择时策略以及多策略融合研究。
研究发现,A 股锂电行业并不适合简单追涨式动量策略。短中期动量因子在样本期内呈现负向预测特征,说明行业内部存在一定反转效应。相较之下,低波动、毛利率、净利润增速和 EP 因子表现更优,说明在高波动成长行业中,市场更偏好风险可控、盈利质量较高、盈利改善明显且估值具备修复空间的标的。
在策略层面,ICIR 加权多因子策略能够跑赢锂电等权基准,但仍存在较高行业 Beta 暴露和较大回撤。进一步引入 CTA 趋势策略后,双均线趋势模型能够在行业下行阶段降低仓位、在行业上行阶段参与趋势行情,从而改善最大回撤和夏普比率。
| Metric | Value |
|---|---|
| Universe | 20 A-share lithium battery industry leaders |
| Final sample period | 2023-04-04 to 2025-12-31 |
| Final daily observations | 10,782 |
| Rebalance interval | 20 trading days |
| Selected stocks per rebalance | Top 5 |
| Multi-factor transaction cost | 0.15% one-way |
The equal-weight lithium benchmark is constructed from the same 20-stock universe. It represents a simple buy-and-hold industry basket without stock selection or timing.
Rank IC uses future 20-trading-day returns as the validation label. ICIR is calculated as mean IC divided by IC standard deviation.
| Factor | IC Mean | ICIR | Direction | Interpretation |
|---|---|---|---|---|
| 20D low volatility | 0.1517 | 0.4607 | Positive | Lower-volatility names performed better |
| 60D low volatility | 0.1449 | 0.4586 | Positive | Medium-term low volatility was also effective |
| Gross margin | 0.0802 | 0.3447 | Positive | Profitability quality mattered |
| Net profit growth | 0.0947 | 0.3381 | Positive | Earnings improvement mattered |
| ROE | 0.0656 | 0.2199 | Positive | Profitability remained useful |
| 20D momentum | -0.0603 | -0.2019 | Negative | Short-term reversal effect |
| 60D momentum | -0.0492 | -0.1707 | Negative | Medium-term reversal effect |
| EP | 0.0525 | 0.1693 | Positive | Valuation repair signal |
Final ICIR-weighted score:
Score =
0.2733 * LowVol20
+ 0.2045 * GrossMargin
+ 0.2006 * NetProfitGrowth
- 0.1198 * MOM20
- 0.1013 * MOM60
+ 0.1005 * EP
The CTA switch model holds the Top 5 multi-factor portfolio when the CTA signal is positive and moves to cash when the CTA signal is negative.
| Strategy | Total Return | Annual Return | Volatility | Sharpe | Max Drawdown |
|---|---|---|---|---|---|
| Multi-factor + CTA switch | 81.02% | 25.13% | 26.44% | 0.95 | -24.75% |
| Pure ICIR multi-factor | 56.65% | 18.48% | 35.63% | 0.52 | -50.35% |
| Equal-weight lithium benchmark | -3.39% | -1.30% | 38.14% | -0.03 | -62.09% |
The fusion strategy improved both return and drawdown relative to the pure multi-factor strategy. This supports the idea that multi-factor selection answers "what to buy", while CTA timing answers "when to take industry beta exposure".
The best moving-average parameter set in the sweep was MA10 / MA30 under a long-only mode.
| CTA Strategy | Total Return | Annual Return | Volatility | Sharpe | Max Drawdown |
|---|---|---|---|---|---|
| MA10 / MA30 CTA | 84.43% | 27.35% | 29.73% | 0.92 | -21.47% |
| Benchmark over CTA comparison window | 7.61% | 2.94% | N/A | N/A | -56.32% |
| Strategy | Total Return | Annual Return | Volatility | Sharpe | Max Drawdown |
|---|---|---|---|---|---|
| 60% multi-factor + 40% CTA | 72.55% | 22.89% | 30.73% | 0.74 | -38.16% |
| Pure CTA trend strategy | 89.46% | 27.31% | 29.08% | 0.94 | -20.88% |
| Pure ICIR multi-factor | 56.65% | 18.48% | 35.63% | 0.52 | -50.35% |
| Equal-weight lithium benchmark | -3.39% | -1.30% | 38.14% | -0.03 | -62.09% |
The results suggest three practical lessons:
- Lithium battery stocks showed strong industry-cycle behavior during the sample period.
- Low-volatility and quality-growth factors were more useful than naive momentum signals.
- CTA trend control helped reduce downside exposure when the whole industry moved into a downtrend.
- The universe is a manually selected 20-stock lithium battery sample, so survivorship and selection bias may exist.
- Tushare data can be revised or updated, so rerunning the scripts may produce slightly different results.
- The backtests do not fully model limit-up/limit-down constraints, suspension risk, market impact, or execution latency.
- The results are for research and education only and are not investment advice.