← All projects

Finance & Research2026

Equity Lab

Tracks large-holder shifts in Taiwan stocks using TDCC shareholder-distribution data, generating rule-based signals and event backtests—intentionally skipping ML in the core so every signal stays traceable and reproducible.

ResearchPrivate

Why

Why I built it

Before asking whether large-holder shifts predict price behavior, bracket revisions, stock universes, and corporate actions must be clean—otherwise great backtests are usually data bugs.

What I built

What I built

  • Normalizes historical TDCC bracket changes, reconstructs large-holder positions and shareholder counts, and pins down the common-stock universe from official filings.
  • Resumable full-market backfill, ownership indicators, and rule-based event backtests retaining raw source records and dates.
  • Strict split across dev, calibration, and holdout sets guards against overfitting while validating the rule logic on a 12-stock holdout set.

Role & AI

How I split the work with AI

I define the ownership questions, signal definitions, and backtest assumptions, build the pipeline with coding agents, and audit results for sample or data-prep distortion.

Trade-offs

What testing made me change

  • A personal research codebase focused on clean data definitions and reproducible signals.

Outcome

Outcome

Ingestion, normalization, backfill, and event backtesting run end to end, turning Taiwan stock ownership hypotheses into traceable reports.

PythonFastAPIPolarsDuckDBParquetpytest