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

Step 6 — Decile portfolio backtest

1. At each rebalance date, sort universe into 10 deciles by factor value.
2. Track equal-weighted returns for each decile over the holding period.
3. Key output: decile 1 vs decile 10 spread (long-short portfolio return).
4. 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:

1. Factor definition and universe summary.
2. IC time series chart (describe in text if no chart tool).
3. Summary statistics table (Mean IC, IC Std Dev, IR, hit rate).
4. Decile return bar chart description (decile 1 to 10 cumulative return).
5. IC decay table across horizons.
6. Interpretation: is the factor effective? Recommended holding period?

Error handling

Related skills

• longbridge-factor-screen — screen stocks by factor values today
• longbridge-multifactor — combine multiple factors into a composite score
• longbridge-quant-stats — statistical tests (IC significance, t-test)
• longbridge-ml-strategy — machine-learning based strategy research

File layout

skills/longbridge-factor-research/
└── SKILL.md