QuantRocket 1.6.0 Release Notes

Announcements
#1

QuantRocket 1.6.0 is now available. All users are encouraged to update to the latest version. See how to update.

Highlights

  • Support for machine learning strategies using walk-forward optimization. See the usage guide.
  • Forward-looking earnings announcements dates from Wall Street Horizon, via IB. See the usage guide.

API changes

:warning: This release introduces a breaking API change for Moonshot strategies that rely on the MASTER_FIELDS parameter to access securities master data in the body of their strategy. However, reverting to the old behavior is simple. See below for more details.

  • The master_fields parameter for get_historical_prices is deprecated. Instead, if you need securities master fields aligned to prices, please make a separate call to the newly added function get_securities_reindexed_like (see usage guide). This API change is motivated by performance considerations: querying numeric fields (like prices) and string fields (like symbols, exchanges, etc.) in the same query forces Pandas to load all the fields, including numeric ones, as the object dtype rather than float or integer, resulting in slower loads and higher memory consumption. This problem is avoided by loading numeric and non-numeric fields in two separate queries.
  • Related to the above change, the Moonshot.MASTER_FIELDS parameter now defaults to None whereas previously it defaulted to ["Currency", "Multiplier", "PriceMagnifier", "PrimaryExchange", "SecType", "Symbol", "Timezone"]. If you were accessing any of these master fields in the body of your Moonshot strategy, you will receive a KeyError. The preferred solution is to update your code to use get_securities_reindexed_like (see usage guide). Alternatively, to preserve the old behavior, set MASTER_FIELDS manually:
class MyStrategy(Moonshot):

    MASTER_FIELDS = ["Currency", "Multiplier", "PriceMagnifier", "PrimaryExchange", "SecType", "Symbol", "Timezone"]

Documentation

Complete 1.6.0 release notes

quantrocket/fundamental:1.6.0

  • Forward-looking earnings announcements dates from Wall Street Horizon, via IB. See the usage guide.
  • a new force parameter for collect_reuters_estimates and collect_reuters_financials forces fundamentals to be queried for all specified securities even if they were collected recently. (The default behavior is to skip securities that were updated in the last 12 hours.) Also available for the new function collect_wsh_earnings_dates.

quantrocket/history:1.6.0

  • support for sharding intraday databases by year. See the usage guide.
  • fix improper database index to yield faster query performance for Sharadar databases

quantrocket/jupyter:1.6.0

  • update to quantrocket-client 1.6.4
    • new function: get_securities_reindexed_like. See usage guide.
    • a new ffill parameter for get_reuters_estimates_reindexed_like allows disabling forward-filling if you only want the estimate/actual populated on the date it was first available. (Forward-filling is enabled by default.) See the API reference.
    • a new shift parameter for get_reuters_estimates_reindexed_like allows disabling the shifting of estimates and actuals by one period (enabled by default). Shifting is intended to prevent lookahead bias, but some announcements occur before the market open so shifting is overly cautious in those cases. See the API reference.

quantrocket/master:1.6.0

  • add environment variable IB_WEB which can be used to override the IB website base URL used for collecting listings. Can be used to point directly to an IB subsidiary website. See community post for background. To override the URL, create or edit your docker-compose.override.yml file and add the following override for the master service:
version: '2.4'
services:
  master:
    environment:
    - IB_WEB=https://www.interactivebrokers.co.uk

quantrocket/moonshot:1.6.0

  • Support for machine learning strategies using walk-forward optimization
  • update to moonshot 1.6.2
    • support machine learning strategies with new MoonshotML class
Listings collection fails silently