Utilities and information for the signals.numer.ai tournament.

eodhistoricaldata.com provides excellent ticker coverage (including delisted Signals tickers) for historical prices. While a free tier is available, I recommend the "EOD Historical Data - All World" package at $19.99/mo to make the most of dsignals.

dsignals solves the two main challenges with using eodhistoricaldata in Signals:

• build_eodhd_map.py generates the ticker mappings from bloomberg_ticker.
• download_quotes.py provides comprehensive global coverage by downloading from eodhistoricaldata on its supported stock exchanges, and from yahoo for Japan, Czech Republic and New Zealand.

To generate the up-to-date ticker mappings for the entire Signals universe (live and historical), run:

python build_eodhd_map.py

Step 1, this will download and merge tickers from three sources:

• live_universe (a small 40 KB file with the ~5,340 tickers for the current round)
• historical_targets (a large 150 MB file, and extract ~13,370 unique historical tickers)
• the bloomberg to yahoo map courtesy of Liam @ numerai

Step 2, tickers are mapped, and overrides in db/eodhd-overrides.csv are applied.

Step 3, db/eodhd-map.csv is generated in the following format:

There are two ways to specify your eodhistoricaldata API token for use by download_quotes.py:

• set "NUMERAI_EODHD_TOKEN" environment variable:

  linux: export NUMERAI_EODHD_TOKEN="your_eodhd_api_key"

  windows: set NUMERAI_EODHD_TOKEN="your_eodhd_api_key"

• or, edit download_quotes.py and replace "your_eodhd_api_key" with your API token.

To start the download, run:

python download_quotes.py

This will download price data from either eodhistoricaldata or yahoo, and save each ticker to a separate pickle file in the data/ticker_bin folder. As of February 2022, this yields price data for 11,200+ tickers.

You can modify the download behavior with two arguments:

Example code to receive a pandas DataFrame for a given bloomberg_ticker:

from download_quotes import read_quotes
quotes = read_quotes("MSFT US")

Alternatively, copy the make_filename_safe() function to your codefile to generate the name of the pickle file for a given ticker, and read the pickle file directly with pd.read_pickle(filename).