This bundle contains three complete Python scripts that authenticate to FinMind, fetch Taiwan market data, and visualize with pandas and seaborn. Each script is heavily commented (300+ lines) for learning and adaptation.

⚠️ Disclaimer: Educational use only. Not investment advice.

Contents

  • finmind_prices_ma_corr.py
    Fetches daily prices (TaiwanStockPrice) for multiple tickers, computes returns and moving averages, and produces:
  • Close + MA lines (focus ticker)
  • Daily return distributions (FacetGrid)
  • Pairplot of returns across tickers

  • Correlation heatmap across tickers

  • finmind_monthly_revenue_analysis.py
    Pulls monthly revenue (TaiwanStockMonthRevenue) for multiple companies, computes MoM/YoY growth, and charts:

  • Monthly revenue trend lines
  • YoY bar plots (recent 24 months)
  • YoY heatmap by month/year (single ticker)

  • YTD cumulative revenue comparison by year

  • finmind_institutional_flow_vs_price.py
    Fetches institutional investors buy/sell (TaiwanStockInstitutionalInvestorsBuySell) and daily prices, engineers net buy features (5d/20d sums), and explores relationships with forward returns via:

  • Price vs. NetBuy (z-scored overlay)
  • Regressions (NetBuy_5d ↔ fwd returns)
  • KDE joint density
  • Rolling correlation (60D window)

Prerequisites

  • Python 3.9+ recommended

  • Packages: bash pip install finmind pandas seaborn matplotlib numpy python-dateutil

  • Credentials (choose one):

  • Token (Recommended): set environment variable
    FINMIND_TOKEN="your_api_token"
  • User/Password: set environment variables
    FINMIND_USER="your_email" and FINMIND_PASS="your_password"

All scripts try token first, then user/password, else anonymous (strictly rate-limited).

Quick Start

  1. Unzip this package.
  2. Set your environment variables (macOS/Linux example): bash export FINMIND_TOKEN="YOUR_TOKEN_HERE" # or: export FINMIND_USER="YOUR_EMAIL" export FINMIND_PASS="YOUR_PASSWORD"
  3. Run any script: bash python finmind_prices_ma_corr.py python finmind_monthly_revenue_analysis.py python finmind_institutional_flow_vs_price.py

Each script will pull data, show several charts, and print summary tables to your terminal.

What Each Script Demonstrates

1) Prices, Moving Averages, Correlations

  • Dataset: TaiwanStockPrice
  • Features:
  • Returns (pct_change)
  • Moving Averages MA(5/20/60)
  • Visualizations:
  • Focus ticker Close + MAs
  • Return distributions (FacetGrid)
  • Pairplot on a subset of tickers
  • Correlation heatmap across tickers
  • Customization: edit default_tickers, start_date, or ma_windows in script.

2) Monthly Revenue: MoM, YoY, YTD

  • Dataset: TaiwanStockMonthRevenue
  • Features:
  • MoM = pct_change(1)
  • YoY = pct_change(12)
  • YTD cumulative by year
  • Visualizations:
  • Revenue line chart
  • YoY bar plot (last 24 months)
  • YoY heatmap by month/year (single ticker)
  • YTD lines by year
  • Notes: First 12 months have NaN YoY by design.

3) Institutional Flow vs. Price / Returns

  • Datasets:
  • TaiwanStockPrice
  • TaiwanStockInstitutionalInvestorsBuySell
  • Features:
  • Aggregate net buy across available institution columns
  • Rolling sums over 5D and 20D
  • Forward returns over 5D and 10D from price series
  • Visualizations:
  • Price vs. NetBuy (z-score overlay)
  • Regression plots (relationship exploration)
  • KDE joint density
  • Rolling correlation as a monitoring tool

Common Problems & Fixes

  • ModuleNotFoundError: FinMind
    → pip install finmind

  • Login fails
    → Double-check FINMIND_TOKEN (or FINMIND_USER/FINMIND_PASS). Try token first.
    → Ensure quotes are correct and variables are exported in the same shell session.

  • Rate limiting (HTTP 429)
    → Reduce number of tickers or date range.
    → Increase polite_sleep_sec.
    → Use token auth for better quotas.

  • Empty / unexpected columns
    → Print df.columns to inspect schema; datasets can evolve.
    → Update column references accordingly.

  • Charts are slow or memory-heavy
    → Sample your data for pairplots (.sample(n=3000) in the script).
    → Limit date range or number of tickers.

Security Tips

  • Keep your token and passwords out of version control.
  • Prefer environment variables over hard-coding credentials.
  • If sharing notebooks, remove any saved tokens.

Extending These Scripts

  • Write output CSV/Parquet files and reuse them to avoid re-downloading.
  • Add CLI arguments via argparse (tickers, date ranges, windows).
  • Replace seaborn with matplotlib if you need strict style control.
  • Build dashboards (e.g., Streamlit, Dash) on top of these functions.
  • Add statistical tests, rolling betas, or event studies.

Notes on Interpretation

These are exploratory analyses. Relationships (e.g., net buy vs. forward returns) can be noisy and unstable. Always validate, backtest, and consider transaction costs, survivorship bias, and regime changes.

License

MIT (for these example scripts). The data usage is subject to FinMind terms of service.