quantlab.factor.predefined.daily_anomaly_alpha#
Twenty daily equity anomalies from mechanisms the other factor bundles do not cover, in one KunQuant graph.
Each output is one published mechanism, computed from the daily SEP bars, DAILY’s market cap, the 8-K items of Sharadar EVENTS and the acquisitions, SPAC mergers and exchange changes of Sharadar ACTIONS, and signed so the source paper expects a higher value to predict a higher return. Windows, return frequencies and signs are the papers’; none was tuned.
r is the close-to-close return of adjClose, m the market’s: the
mean of r over every symbol of the panel weighted by the previous bar’s
market cap (for a 21-bar return, the market’s 21-bar return weighted by the
cap 21 bars before; the 21-bar returns overlap, one per bar). Turnover V
is adjClose * adjVolume / marketcap. An 8-K bar of news_drift_gap is
one on which, or on the bar after which, an item 1.01, 2.01, 2.02, 5.02,
7.01 or 8.01 filing is available. An acquisition is placed on the acquirer
from the acquired company’s ACTIONS row, which names the acquirer
(acquisition_counterparty) and what its holders received
(acquirer_payments).
Windowed statistics skip missing bars and need 80% of their window (the
downside beta 50 down bars); a statistic short of that is NaN. Event flags
(and the action flags) are 0 on a priced bar without one and NaN
before their source’s first one (fundamental_alpha.fill_flows). dividend_change and
dividend_initiation carry their state from the first bar of the read
window, so warmup_bars (WARMUP_BARS) must cover the longest window
and the 504-bar dividend gap.
The market and the ranks are taken over every symbol of the panel, so build it on the whole market and read it on a roster. The factor is batch only; the graph runs in double precision.
References#
Frazzini and Lamont (2007); Barber, De George, Lehavy and Trueman (2013); Johnson and So (2018); Hartzmark and Solomon (2013); Heston and Sadka (2008); Field and Hanka (2001); Harvey and Siddique (2000); Ang, Chen and Xing (2006); Da, Gurun and Warachka (2014); Akbas, Boehmer, Jiang and Koch (2022); Chan (2003); Lerman, Livnat and Mendenhall (2008); Ikenberry, Rankine and Stice (1996); Desai and Jain (1997); Asquith and Mullins (1983); Michaely, Thaler and Womack (1995); Spiess and Affleck-Graves (1999); Loughran and Vijh (1997); Dharan and Ikenberry (1995); Klausner, Ohlrogge and Ruan (2022); Gahng, Ritter and Zhang (2023); Ritter (1991); Loughran and Ritter (1995); Chordia, Subrahmanyam and Anshuman (2001).
Classes
Compute the twenty daily calendar, risk-shape, behavioural, corporate-action and liquidity anomalies. |
Functions
Place each acquisition on its acquirer: +1 if paid in cash, -1 if in stock or stock and cash. |