twisted_truth

Lightweight tools to collect and visualise M1 money supply against equity indices and constituents, and cryptocurrencies.

This small repo fetches M1 Money Supply from FRED, appends market data from Yahoo Finance, and provides simple utilities to visualise the series and compute long/short-term correlations.

Features

• Fetch M1 series from FRED and save a single CSV (data/data.csv).
• Append index and constituent close prices from Yahoo Finance.
• Visualise M1 vs an index, a constituent, and a cryptocurrency using matplotlib.
• Compute per-year and per-quarter Pearson correlations with optional time-lagging to output a Fisher z-transformed averaged correlation coefficent.

Key concept — Fisher z-transformed average correlation

• Problem: averaging Pearson r values directly is biased.

• Approach (used by src/calculating_correlation.py):

  1. Compute Pearson r for each group (e.g. each year or each quarter).
  2. Apply Fisher z-transform: z = arctanh(r) (maps r in (-1,1) to real line).
  3. Average the z values: mean_z = mean(z_i).
  4. Convert back to correlation: avg_r = tanh(mean_z).

  This stabilises variance and gives a sound aggregated correlation estimate.

Quick start

1. Create and activate a virtual environment:

python -m venv .venv
source .venv/bin/activate
pip install -U pip
pip install -e .

2. Provide secrets in a .env file at repository root:

cp .env.example .env

FRED_API_KEY=your_fred_api_key_here

3. Run the srcipts:

python run.py

Configuration

• All runtime parameters live under the configs/ directory (Hydra + OmegaConf).
• Common config keys live in configs/data_loader.yaml and configs/calculating_correlation.yaml.
• You can override any config value on the command line. Example:

python src/data_loader.py load_data.start_date=2020-01-01 load_data.end_date=2024-12-31

Notes & troubleshooting

• Network: yfinance and fredapi require outbound internet access.
• Forward-fill / placeholders: if the data loader cannot fetch new M1 values it may forward-fill the last value; that will produce zero-variance groups and cause NaNs in per-group Pearson correlations. If you see NaNs in correlation output inspect data/data.csv for long runs of identical M1_Money_Supply values and fix the upstream fetch (preferred) or drop/NaN those rows.

License

• License: MIT (see LICENSE)