Abstract. Cohen, Malloy and Nguyen (2020) find that firms which rewrite the language of their periodic filings subsequently underperform. On 10-Q and 10-K text it does not replicate: the measure is specification-sensitive, and the one significant version inverts. Survivorship, which deletes the informative short leg, is the likely cause. No signal is promoted.

1. An orthogonal anomaly

Most published equity signals are functions of price. This one reads the language of a filing: verbatim reproduction says nothing material has changed, while a rewrite says something is being managed. The mechanism is forensic (Mayew, Sethuraman and Venkatachalam, 2015).

2. Data and method

Change must be measured year over year, same period, or seasonal differences in disclosure count as change. The universe is current index constituents, so the panel is survivorship biased by construction.

Sample, method and headline statistics. Text change is one minus the cosine similarity of a filing to the same fiscal period a year earlier, measured on stripped text after discarding inline XBRL scaffolding and frequent stopwords. Returns are market adjusted and every measure is point-in-time. The promotion bar is the raw threshold after deflation for multiple testing.
MeasureValue
Filings tested (10-Q and 10-K, EDGAR, filed since 2015)14,155
Universe (large caps currently in the index)406
Forward return window, from the day after filing63 trading days
Naive quarter-over-quarter spread, low minus hight = 0.13
Year-over-year spread, stopwords removed−1.97% (t = −2.55, Newey–West)
Promotion barnear raw t = 5

3. Fragile, and inverted

Quarter over quarter, the spread of low minus high change is a precise zero. Year over year, the same spread turns significant and reverses.

Table 1. Forward return over 63 days, market adjusted, by quintile of disclosure text change from a year earlier; 9,644 filings with forward returns. The last row is not their difference: it is the Q1 average minus the Q5 average computed within each calendar month, then averaged over the 107 months. Differencing the quintile rows directly gives −1.33%, a different estimator.
Change quintileFwd. return
Q1 (least changed)−0.89%
Q2+0.05%
Q3−0.21%
Q4−0.26%
Q5 (most changed)+0.44%
Low minus high (monthly spread, tradable side)−1.97% (t=−2.55)

4. Survivorship deletes the short leg

Lazy Prices is largely a short-side result: the money is in avoiding the firms that rewrite disclosure because trouble is coming. Those firms underperform, leave the index, and often delist, so a panel of current constituents has already deleted them. Deleting the malign half is enough to flip the sign.

5. Verdict

Nothing is promoted. The strongest specification falls short of the promotion bar, points the wrong way, and names survivorship rather than information. The conclusion is not that Lazy Prices is false, since the original stands on a universe free of survivorship bias, which I do not have, but that it cannot be tested here. A clean test needs point-in-time constituents and delisting returns, my pipeline’s binding constraint.

A signal that inverts under survivorship is worth more than one that merely fails.

References

  1. Cohen, L., Malloy, C., and Nguyen, Q. (2020). Lazy Prices. Journal of Finance, 75(3), 1371–1415.
  2. Loughran, T., and McDonald, B. (2011). When Is a Liability Not a Liability? Textual Analysis, Dictionaries, and 10-Ks. Journal of Finance, 66(1), 35–65.
  3. Mayew, W. J., Sethuraman, M., and Venkatachalam, M. (2015). MD&A Disclosure and the Firm’s Ability to Continue as a Going Concern. The Accounting Review, 90(4), 1621–1651.