yfinance-cache 0.4.8

Smart caching wrapper for 'yfinance' module

yfinance-cache

Persistent caching wrapper for yfinance module. Intelligent caching, not dumb caching of web requests - only update cache where missing/outdated and new data expected.

Only price data caching fully implemented. Everything else is cached once but never updated (unless you delete their files) - I ran out of time to implement e.g. financials cache update.

Persistent cache stored in your user cache folder:

• Windows = C:/Users/<USER>/AppData/Local/py-yfinance-cache
• Linux = /home/<USER>/.cache/py-yfinance-cache
• MacOS = /Users/<USER>/Library/Caches/py-yfinance-cache

Price cache

Idea behind this cache is to minimise fetch frequency and quantity. Yahoo API officially only cares about frequency, but I'm guessing they also care about server load from scrapers.

How is this caching different to caching URL fetches? Simple - they don't adjust cached data for new stock splits or dividends.

What makes the cache smart? Adds 'fetched date' to each price data, then combines with an exchange schedule to know when new price data expected.

Note:

• '1d' price data always fetched from start date to today (i.e. ignores end), as need to know all dividends and stock splits since start.
• price repair enabled, to prevent bad Yahoo data corrupting cache. See yfinance Wiki for detail

Financials cache

I planned to implement this after prices cache, but ran out of time. Strategy to minimise fetch frequency is to fetch at/after the next earnings date, inferred from Ticker.calendar and/or Ticker.earnings_dates.

Interface

Interaction almost identical to yfinance. Differences highlighted underneath code:

import yfinance_cache as yfc

msft = yfc.Ticker("MSFT")

# get stock info
msft.info

# get historical market data
hist = msft.history(period="1d")

# bulk download
yfc.download("MSFT AMZN", period="1d")
...
# etc. See yfinance documentation for full API

Refreshing cache

df = msft.history(interval="1d", max_age="1h", trigger_at_market_close=False, ...)

max_age controls when to update cache. If market is still open and max_age time has passed since last fetch, then today's cached price data will be refreshed. If trigger_at_market_close=True then refresh also triggered if market has closed since last fetch. Must be Timedelta or equivalent str, defaults to half of interval.

The returned table has 2 new columns:

• FetchDate = when data was fetched
• Final? = true if don't expect future fetches to change

Adjusting price

Price can be adjusted for stock splits, dividends, or both.

msft.history(..., adjust_splits=True, adjust_divs=True)

Verifying cache

Cached prices can be compared against latest Yahoo Finance data, and correct differences:

# Verify prices of one ticker symbol
msft.verify_cached_prices(
	rtol=0.0001,  # relative tolerance for differences
	vol_rtol=0.005,  # relative tolerance specifically for Volume
	correct=False,  # delete incorrect cached data?
	discard_old=False,  # if cached data too old to check (e.g. 30m), assume incorrect and delete?
	quiet=True,  # enable to print nothing, disable to print summary detail of why cached data wrong
	debug=False,  # enable even more detail for debugging 
	debug_interval=None)  # only verify this interval (note: 1d always verified)

# Verify prices of entire cache, ticker symbols processed alphabetically. Recommend using `requests_cache` session.
yfc.verify_cached_tickers_prices(
	session=None,  # recommend you provide a requests_cache here
	rtol=0.0001,
	vol_rtol=0.005,
	correct=False,
	halt_on_fail=True,  # stop verifying on first fail
	resume_from_tkr=None,  # in case you aborted verification, can jump ahead to this ticker symbol. Append '+1' to start AFTER the ticker
	debug_tkr=None,  # only verify this ticker symbol
	debug_interval=None)

With latest version the only genuine differences you should see are tiny Volume differences (~0.5%). Seems Yahoo is still adjusting Volume over 24 hours after that day ended, e.g. updating Monday Volume on Wednesday.

If you see big differences in the OHLC price of recent intervals (last few days), probably Yahoo is wrong! Since fetching that price data on day / day after, Yahoo has messed up their data - at least this is my experience. Cross-check against TradingView or stock exchange website.

Performance

For each ticker, YFC basically performs 2 tasks:

1 - check if fetch needed

2 - fetch data and integrate into cache

Throughput on 1 thread decent CPU: task 1 @ ~60/sec, task 2 @ ~5/sec.

Installation

Available on PIP: pip install yfinance_cache

Limitations

• only price data is checked if refresh needed
• intraday pre/post price data not available