quantlab.dataset.massive.trade_bars#

The Trade bar panel built from Massive’s daily trade files, on the Massive ticker axis.

MassiveTradeBarDataset reads the raw tier the Massive client writes (quantlab.dataset.massive.raw) one trading day at a time, resamples the trades with quantlab.dataset.massive.resample.TradeBarResampler, and returns the dense (timestamp, symbol) panel the store holds (ADRs 0030, 0031).

Sessions are the XNYS calendar’s, half days included; the window is session_start to session_end, the regular session by default. A raw day that is not a session is refused.
The symbol axis is the Massive ticker, exactly as the trade files spell it (BRK.A, AApB), every ticker kept, exchange test symbols included (ADR 0031). Conversion reads no Sharadar data. A column is a ticker, not a security: one ticker can belong to different securities on different days, so reading the store as securities goes through the Ticker map (quantlab.dataset.massive.ticker_map). tickers restricts a conversion to a roster of Massive tickers.
A cell without an eligible trade has NaN prices and zero volumes and n_trades; a ticker on the axis that did not trade on a day is zero volume on every bar of it.
Per-day statistics (trades in, dropped by each rule, outside the session, excluded from volume, outside the roster) are written to the JSON sidecar <store>.massive_stats.json (stats_path), merged day by day.
Each one-minute day is checked against Massive’s own minute aggregates of that day when the raw tier holds them (quantlab.dataset.massive.vendor_check), ticker by ticker: agreement counts, the worst difference of each variable and the bars present on one side only go to the day’s vendor_check statistics. Differences are reported, never raised.
The store grows a day at a time. A config whose range holds the next day, passed to update(granularity="day"), appends that day: the day’s new tickers are appended at the end of the store’s axis in place (ADR 0032), with zero volume and NaN prices over the days before, so a backfill and a daily update are the same operation. Raw trade files are kept (ADR 0031).
The settings are the store’s identity. The bar interval, the session window, the roster and the counting rules (_build_settings) are recorded in the sidecar and in every read’s data fingerprint; a conversion into a store recorded with other settings is refused before anything is written. Each bar interval has its own store (trade_bar_store_name).
Coarser bars come from Resample. Each variable declares its aggregation (DEFAULT_RESAMPLE_HOW) and bars are cut per session, right closed and labelled at their end, so resample("5m") of the one-minute store equals converting the trades at 5m. "1d" gives one bar per session, labelled with its date.

Convert one day per window (granularity="day"): a day of the whole market is tens of millions of trades.

A day can be converted elsewhere. convert_day reads and resamples one day without touching the store, and returns a picklable ConvertedTradeDay; add_converted_day hands it to the instance that appends, which then does not read the day again. A backfill converts many days in worker processes this way and appends them in date order.

Examples#

Needs a trade file and a condition table under raw_data_dir_path:

config = MassiveTradeBarsDatasetConfig(
    zarr_file_path="/data/zarrs/massive_trade_bars_1m.zarr",
    raw_data_dir_path="/data/downloads/massive",
    start_date="2024-11-29",
    end_date="2024-11-29",
)
MassiveTradeBarDataset(config).from_raw_data_chunked(granularity="day")

Classes

`ConvertedTradeDay`	One converted trading day: its bars on the Massive ticker axis and its statistics.
`MassiveTradeBarDataset`	Dense Trade bar panel on the Massive ticker axis; see the module docstring.

Functions

`no_trade_fills`	Return the values of a Trade bar column without a trade: 0 volumes and counts.
`trade_bar_store_name`	Return the name of the Trade bar store of one bar interval.