HdW Stock Data

First draft of hdw-stock-data 0.1.0: retrieve stock observations directly from Yahoo's chart HTTP endpoint, normalize OHLCV, preserve timestamp instants, and optionally write an atomic CSV snapshot. It uses requests, not yfinance. This project is independent of hdw-crypto-data.

Install

Python 3.11 or newer is required. From the repository root:

py -3.13 -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install -e ".[dev]"

Use

from hdw_stock_data import YahooStockLoader

result = YahooStockLoader().load(
    "^GSPC", interval="1h", days=730,
    preferred_tz="Europe/Amsterdam", save_to="snapshots",
)
sp500 = result.dataframe["close"].rename("sp500")
print(result.descriptor)
print(result.filepath, result.warnings)

Only 1h is validated for this release. Both client and loader default to 730 days and reject longer ranges before retrieval. This is a tested capability limit, not a guarantee that every symbol has 730 days of observations. At the 730-day limit, the client moves the requested start forward by five minutes to provide a margin inside Yahoo's rolling cutoff. Shorter requests retain their full duration.

Data contract

The frame has a sorted, unique, timezone-aware DatetimeIndex named dt and numeric open, high, low, close, volume columns. Missing OHLC rows are removed; missing volume is retained. The last valid duplicate is kept. Invalid OHLC relationships, negative volume, nonfinite values, and invalid timestamps are rejected. No shifts, resampling, or gap filling are applied.

number_of_trades contains pd.NA with nullable Float64 dtype for chart compatibility. Trade counts are unavailable, not zero. This additional column is optional to the market-data validator. result.data aliases dataframe.

The descriptor records actual normalized bounds and count, currency and exchange when available, provider yahoo, and output timezone. adjusted=False means the package performs no price adjustment; source adjustments are not undone.

Verification

py -3.13 -m unittest discover -s tests -v
py -3.13 -m build
py -3.13 -m twine check dist/*

Tests use mocked HTTP responses and local files. Live requests are opt-in:

py -3.13 -m examples.fetch_stock AAPL --days 5

Run the initial stock and index smoke test (AAPL, ASML.AS, ^GSPC, ^VIX):

py -3.13 -m examples.test_assets
py -3.13 -m examples.test_assets --days 5 --save-to snapshots

This retrieves hourly bars, prints actual bounds and the latest five rows for each symbol, and exits with status 1 if any retrieval fails. The default output timezone is Europe/Amsterdam; override it with --timezone UTC. You can also provide a different list of symbols as positional arguments.

See architecture, API, and testing plan, and showcase setup. No live-provider coverage is claimed by the offline test suite.