quantlab.factor.predefined.alternative_alpha

Literature-backed short-selling, news, 8-K event and 10-K text factors in one KunQuant graph.

Every output comes from data beyond prices and the financial statements, signed so that a higher value predicts a higher return in its source paper:

output	formula	reference
short_interest_ratio	-short_interest / shares	Asquith, Pathak and Ritter (2005)
days_to_cover	-short_interest / mean volume, 21 bars	Hong et al. (2016)
short_interest_change	-(SIR - SIR 21 bars before)	Desai et al. (2002)
short_constraint	-SIR / institutional ownership	Asquith, Pathak and Ritter (2005)
short_volume_ratio	-sum short volume / sum volume, 21 bars	Diether, Lee and Werner (2009)
news_attention	-log(1 + articles), 63 bars	Fang and Peress (2009)
news_sentiment	(positive - negative) / rated, 63 bars	Heston and Sinha (2017)
earnings_announcement_return	mean market-adjusted 3-bar return of the 8-K item 2.02 filings in the last 63 bars	Brandt et al. (2008)
activist_13d	Schedule 13D filings, 252 bars	Brav et al. (2008)
distress_events	- delisting notices, impairments, non-reliance, bankruptcy, accelerated obligations, 252 bars	Hennes, Leone and Miller (2008)
equity_issuance_8k	- unregistered equity sales, 252 bars	Hertzel et al. (2002)
lazy_prices_business	10-K Item 1 cosine similarity with the year before	Cohen, Malloy and Nguyen (2020)
lazy_prices_risk	10-K Item 1A cosine similarity	Cohen, Malloy and Nguyen (2020)
lazy_prices_business_jaccard	10-K Item 1 Jaccard similarity	Cohen, Malloy and Nguyen (2020)
lazy_prices_risk_jaccard	10-K Item 1A Jaccard similarity	Cohen, Malloy and Nguyen (2020)
risk_disclosure_growth	- log growth of the Item 1A word count	Cohen, Malloy and Nguyen (2020)

shares is DAILY’s market cap over SEP’s unadjusted close and SIR the short-interest ratio. Institutional ownership is SF3A’s value_held over DAILY’s market cap on the 13F quarter’s end (fundamental_alpha.value_on_dates). The earnings announcement return of an 8-K item 2.02 (event_22) filing available on bar d is the 3-bar return to the close of d + 1, less its equal-weighted mean over every symbol with one; the output averages it over the filings of the last 63 bars and is NaN without one.

The 10-K outputs are the text-change panel’s values (quantlab.dataset.massive.ten_k), shown from the filing until the next one: a firm that rewrites its annual report (“changers”) underperforms.

The panel merges Sharadar SEP (adjClose, close, volume), DAILY (marketcap), SF3A holdings, Sharadar EVENTS, the Massive short interest, short volume and news panels and the 10-K text-change panel, each point in time. Event, short volume and news counts are 0 on a priced bar without one and NaN without a price or before their source’s first non-zero bar (fundamental_alpha.fill_flows); the news sentiment counts start on their own, later, first bar. The short interest itself is shown as released (NaN where the visible release has no row).

The factor is batch only. Inputs are exported as float32.

References

Asquith, Pathak and Ritter (2005); Hong, Li, Ni, Scheinkman and Yan (2016); Desai, Ramesh, Thiagarajan and Balachandran (2002); Diether, Lee and Werner (2009); Fang and Peress (2009); Heston and Sinha (2017); Brandt, Kishore, Santa-Clara and Venkatachalam (2008); Brav, Jiang, Partnoy and Thomas (2008); Hennes, Leone and Miller (2008); Hertzel, Lemmon, Linck and Rees (2002); Cohen, Malloy and Nguyen (2020).

Classes

AlternativeAlpha	Compute the sixteen short-selling, news, 8-K event and 10-K text characteristics in KunQuant.
AlternativeAlphaParameters	Windows of AlternativeAlpha, read from FactorConfig.kwargs.