China Galaxy Securities Factor Analysis Framework

What Problem It Solves

Factor research often stalls at the gap between computing a factor and judging whether it is usable. Factor values, preprocessing, IC, quantile returns, collinearity, orthogonalization, and multi-factor synthesis are often scattered across scripts with inconsistent assumptions. This skill organizes the China Galaxy Securities factor analysis workflow into a reusable framework that takes a factor matrix and produces diagnostic, interpretable results.

How It Works

It runs on AmazingData market data, with core steps such as:
- Factor preparation: after fetching data, reshape custom factors into a pd.DataFrame with index=date and columns=ticker.
- Preprocessing: apply winsorization, industry and size neutralization, standardization, and missing-value handling.
- Validity diagnostics: run IC analysis, regression, layering, NAV performance, crowding, and significance checks; for example, VIF > 10 signals severe collinearity, while a higher share of |T| > 2 usually indicates stronger factor significance.
- Multi-factor synthesis: perform collinearity checks, orthogonalization, weighting, and stock scoring. Common weighting methods include equal_weight, ic_mean, ic_ir, max_ic, and return_ir, and it can render a report.

Boundaries

This skill requires an AmazingData account and environment variables, and data calls must be made after login; single sign-on may limit concurrent sessions. It produces factor calculations and diagnostics, not investment advice. It is useful for reproducible factor testing, comparing multi-factor synthesis methods, and consolidating ad hoc scripts into a standard process.