longbridge-factor-research
A systematic framework for testing whether a quantitative factor adds predictive value for future returns — covering IC analysis, information ratio, decile portfolio construction, and factor decay.
Response language: match the user's input language — Simplified Chinese / Traditional Chinese / English.
When to use
- "帮我分析 PE 因子的 IC", "test IC for the PE factor on A-shares"
- "动量因子有效吗", "is momentum factor effective on HK stocks"
- "做个分层回测", "run a decile portfolio backtest"
- "这个因子多少期后失效", "how many periods until this factor decays"
- "IC 序列自相关怎么算", "calculate IC serial autocorrelation"
For multi-factor screening (not research), use longbridge-factor-screen. For ML-based strategies, use longbridge-ml-strategy.
Workflow
Step 1 — Define the factor
Clarify with the user:
- Factor name and calculation (e.g. trailing-12M PE, 1M price momentum, ROE YoY change).
- Universe: index constituent (e.g. CSI 300, HSI, S&P 500) or custom list.
- Test period (e.g. 2020-01-01 to 2024-12-31).
- Holding period (e.g. monthly rebalance).
Step 2 — Fetch universe constituents
longbridge constituent --help
longbridge constituent <INDEX> --format json
Extract the stocks array. Common indices: 000300.SH (CSI 300), HSI.HK, SPX.US.
Step 3 — Fetch factor values and returns
For each symbol in the universe:
longbridge calc-index <SYMBOL> --format json # valuation, growth metrics
longbridge kline <SYMBOL> --period day --count 252 --format json # price history for returns
Run --help on each command to verify available fields before parsing.
Step 4 — Compute IC at each rebalance date
IC_t = rank_correlation(factor_value_t, forward_return_t+h)
Where h = holding period. Use Spearman rank correlation (robust to outliers). Winsorize factor values at 1%/99% before ranking.
Step 5 — Summary statistics
| Metric |
Formula |
Good signal threshold |
| Mean IC |
Average of IC time series |
> 0.03 (positive) |
| IC Std Dev |
Standard deviation of IC |
Lower is better |
| IR (Information Ratio) |
Mean IC / Std Dev IC |
> 0.5 is promising |
| IC > 0 hit rate |
% of periods IC > 0 |
> 55% |
| ICIR (annualised) |
IR × √(periods per year) |
> 1.0 strong |
Step 6 — Decile portfolio backtest
- At each rebalance date, sort universe into 10 deciles by factor value.
- Track equal-weighted returns for each decile over the holding period.
- Key output: decile 1 vs decile 10 spread (long-short portfolio return).
- Compute cumulative return, Sharpe ratio, and max drawdown for the long-short portfolio.
Step 7 — IC decay analysis
Compute IC for multiple forward horizons (1M, 2M, 3M, 6M, 12M). Plot IC vs horizon. Fast decay = short-term factor; slow decay = longer-term signal.
Serial autocorrelation of IC series: autocorr(IC, lag=1). High autocorrelation → smoother signal, lower trading cost.
CLI
longbridge constituent --help
longbridge calc-index --help
longbridge kline --help
longbridge constituent <INDEX> --format json
longbridge calc-index <SYMBOL> --format json
longbridge kline <SYMBOL> --period day --count 252 --format json
Output
Present:
- Factor definition and universe summary.
- IC time series chart (describe in text if no chart tool).
- Summary statistics table (Mean IC, IC Std Dev, IR, hit rate).
- Decile return bar chart description (decile 1 to 10 cumulative return).
- IC decay table across horizons.
- Interpretation: is the factor effective? Recommended holding period?
Error handling
| Situation |
简体回复 |
繁體回覆 |
English reply |
command not found: longbridge |
请安装 longbridge-terminal 或检查 MCP 配置。 |
請安裝 longbridge-terminal 或檢查 MCP 配置。 |
Install longbridge-terminal or check MCP config. |
stderr: not logged in |
请运行 longbridge auth login。 |
請執行 longbridge auth login。 |
Run longbridge auth login. |
| Index not found |
请检查指数代码格式，如 000300.SH / HSI.HK / SPX.US。 |
請確認指數代碼，如 000300.SH / HSI.HK / SPX.US。 |
Check index ticker format, e.g. 000300.SH / HSI.HK / SPX.US. |
| Insufficient history |
该标的历史数据不足以进行回测，请缩短测试期或更换标的。 |
歷史數據不足，請縮短測試期或更換標的。 |
Insufficient price history; shorten the test period or change the symbol. |
Related skills
longbridge-factor-screen — screen stocks by factor values todaylongbridge-multifactor — combine multiple factors into a composite scorelongbridge-quant-stats — statistical tests (IC significance, t-test)longbridge-ml-strategy — machine-learning based strategy research
File layout
skills/longbridge-factor-research/
└── SKILL.md
1---2name: longbridge-factor-research3description: Factor research framework for evaluating single-factor effectiveness across A-shares, HK, and US stocks — information coefficient (IC), information ratio (IR), decile portfolio backtests, and IC decay (serial autocorrelation). Triggers: "因子研究", "IC分析", "信息比率", "分层回测", "因子有效性", "单因子测试", "因子衰减", "因子评估", "IC分析", "信息比率", "分層回測", "因子有效性", "單因子測試", "factor research", "information coefficient", "IC", "IR information ratio", "factor backtest", "decile portfolio", "factor decay", "factor effectiveness".4license: MIT5---6
7# longbridge-factor-research
8
9A systematic framework for testing whether a quantitative factor adds predictive value for future returns — covering IC analysis, information ratio, decile portfolio construction, and factor decay.
10
11> **Response language**: match the user's input language — Simplified Chinese / Traditional Chinese / English.
12
13## When to use
14
15- *"帮我分析 PE 因子的 IC"*, *"test IC for the PE factor on A-shares"*
16- *"动量因子有效吗"*, *"is momentum factor effective on HK stocks"*
17- *"做个分层回测"*, *"run a decile portfolio backtest"*
18- *"这个因子多少期后失效"*, *"how many periods until this factor decays"*
19- *"IC 序列自相关怎么算"*, *"calculate IC serial autocorrelation"*
20
21For multi-factor screening (not research), use `longbridge-factor-screen`. For ML-based strategies, use `longbridge-ml-strategy`.
22
23## Workflow
24
25### Step 1 — Define the factor
26
27Clarify with the user:
28- Factor name and calculation (e.g. trailing-12M PE, 1M price momentum, ROE YoY change).
29- Universe: index constituent (e.g. CSI 300, HSI, S&P 500) or custom list.
30- Test period (e.g. 2020-01-01 to 2024-12-31).
31- Holding period (e.g. monthly rebalance).
32
33### Step 2 — Fetch universe constituents
34
35```bash
36longbridge constituent --help
37longbridge constituent <INDEX> --format json
38```
39
40Extract the `stocks` array. Common indices: `000300.SH` (CSI 300), `HSI.HK`, `SPX.US`.
41
42### Step 3 — Fetch factor values and returns
43
44For each symbol in the universe:
45
46```bash
47longbridge calc-index <SYMBOL> --format json # valuation, growth metrics
48longbridge kline <SYMBOL> --period day --count 252 --format json # price history for returns
49```
50
51Run `--help` on each command to verify available fields before parsing.
52
53### Step 4 — Compute IC at each rebalance date
54
55`IC_t = rank_correlation(factor_value_t, forward_return_t+h)`
56
57Where `h` = holding period. Use Spearman rank correlation (robust to outliers). Winsorize factor values at 1%/99% before ranking.
58
59### Step 5 — Summary statistics
60
61| Metric | Formula | Good signal threshold |
62|---|---|---|
63| Mean IC | Average of IC time series | > 0.03 (positive) |
64| IC Std Dev | Standard deviation of IC | Lower is better |
65| IR (Information Ratio) | Mean IC / Std Dev IC | > 0.5 is promising |
66| IC > 0 hit rate | % of periods IC > 0 | > 55% |
67| ICIR (annualised) | IR × √(periods per year) | > 1.0 strong |
68
69### Step 6 — Decile portfolio backtest
70
711. At each rebalance date, sort universe into 10 deciles by factor value.
722. Track equal-weighted returns for each decile over the holding period.
733. Key output: decile 1 vs decile 10 spread (long-short portfolio return).
744. Compute cumulative return, Sharpe ratio, and max drawdown for the long-short portfolio.
75
76### Step 7 — IC decay analysis
77
78Compute IC for multiple forward horizons (1M, 2M, 3M, 6M, 12M). Plot IC vs horizon. Fast decay = short-term factor; slow decay = longer-term signal.
79
80Serial autocorrelation of IC series: `autocorr(IC, lag=1)`. High autocorrelation → smoother signal, lower trading cost.
81
82## CLI
83
84```bash
85longbridge constituent --help
86longbridge calc-index --help
87longbridge kline --help
88
89longbridge constituent <INDEX> --format json
90longbridge calc-index <SYMBOL> --format json
91longbridge kline <SYMBOL> --period day --count 252 --format json
92```
93
94## Output
95
96Present:
971. Factor definition and universe summary.
982. IC time series chart (describe in text if no chart tool).
993. Summary statistics table (Mean IC, IC Std Dev, IR, hit rate).
1004. Decile return bar chart description (decile 1 to 10 cumulative return).
1015. IC decay table across horizons.
1026. Interpretation: is the factor effective? Recommended holding period?
103
104## Error handling
105
106| Situation | 简体回复 | 繁體回覆 | English reply |
107|---|---|---|---|
108| `command not found: longbridge` | 请安装 longbridge-terminal 或检查 MCP 配置。 | 請安裝 longbridge-terminal 或檢查 MCP 配置。 | Install longbridge-terminal or check MCP config. |
109| stderr: `not logged in` | 请运行 `longbridge auth login`。 | 請執行 `longbridge auth login`。 | Run `longbridge auth login`. |
110| Index not found | 请检查指数代码格式，如 000300.SH / HSI.HK / SPX.US。 | 請確認指數代碼，如 000300.SH / HSI.HK / SPX.US。 | Check index ticker format, e.g. 000300.SH / HSI.HK / SPX.US. |
111| Insufficient history | 该标的历史数据不足以进行回测，请缩短测试期或更换标的。 | 歷史數據不足，請縮短測試期或更換標的。 | Insufficient price history; shorten the test period or change the symbol. |
112
113## Related skills
114
115- `longbridge-factor-screen` — screen stocks by factor values today
116- `longbridge-multifactor` — combine multiple factors into a composite score
117- `longbridge-quant-stats` — statistical tests (IC significance, t-test)
118- `longbridge-ml-strategy` — machine-learning based strategy research
119
120## File layout
121
122```
123skills/longbridge-factor-research/
124└── SKILL.md
125```
majiayu000