quantlab.dataset.massive.news#

Massive’s news as a daily count and sentiment panel on the permaticker axis (#357).

Massive’s /v2/reference/news articles, as MassiveClient.pull keeps them in the REST raw tier (quantlab.dataset.massive.raw.rest_page_files), become four variables on SEP’s trading calendar, one column per permaticker:

  • news_count: the articles naming the security that count on the day;

  • news_positive, news_negative, news_neutral: those of them whose per-ticker insights sentiment for the security is that label. An article without an insight for the ticker is counted in news_count only.

A cell without an article is 0. An article counts on the trading day of its published_utc in New York time when published before that session’s close, and on the next trading day when published at or after the close or on a day without trading, so news after the close never reaches the day’s bar. The close is the session’s actual one on the XNYS calendar (XnysSessionCalendar): 16:00, and 13:00 on an early close.

An article’s tickers are its tickers and the tickers of its insights; each is placed on the permaticker the Ticker map gives it on the article’s trading day, through quantlab.dataset.massive.placement.TickerPlacement, and the article counts once for each permaticker placed (two tickers placed on one security count once). A ticker the map cannot place is left out and counted in <store>.unplaced.json. Only the universe’s permatickers (by default SEP’s domestic common stock) are columns.

The store is built through the last day SEP, the news pull and the Ticker map all reach, so update() appends a day only once each of them holds it. The pull’s watermark is a day in UTC; every article counting on a trading day was published by its close in New York that day, before the UTC day ends, so a watermark through the day holds them all.

Examples#

Build the store and read a security’s daily counts:

config = MassiveNewsConfig(
    zarr_file_path="/data/quantlab/market/massive/massive_news_1d/massive_news_1d.zarr",
    raw_data_dir_path="/data/quantlab/downloads/sharadar",
    massive_dir="/data/quantlab/downloads/massive",
    ticker_map_path="/data/quantlab/market/massive/massive_ticker_map/massive_ticker_map",
    start_date="2016-10-10",
)
panel = MassiveNewsDataset(config).update().panel("2024-01-02", "2024-12-31")
panel["news_count"].sel(symbol=199059)

Classes

MassiveNewsDataset

Daily panel of Massive's news counts and sentiment labels, by permaticker.
On this page
Show Source