Skip to content
All library documents

Why Earnings-Date Forecasts Can Reveal Information to the Market

Article Quant Q&A · Author: Viking
Read the original

Summary

The document considers whether a long history of earnings announcement timestamps can be used to predict future dates. Its answer argues that timing is not merely a calendar forecasting problem: firms may advance or delay announcements, and the timing itself can carry information about the forthcoming earnings news. It cites research finding that firms that announce earlier tended to report better earnings outcomes than firms that announce later, including stronger return on assets and analyst-based surprises.

This creates a tension for a forecasting strategy. If a model predicts the eventual date perfectly using only publicly available calendar information, the forecast may offer little new information. If the date differs from the expected schedule, that deviation may be informative precisely because it was not predictable. The response does not propose a specific machine-learning method or present a test on the questioner’s timestamp database. Its insight is about the informational limits of date prediction and the potential signal in schedule changes; the cited finding does not establish a universal trading rule.

Key ideas

  • Earnings announcement timing may convey information about the earnings news that follows.
  • Research cited in the document links earlier announcements with stronger earnings outcomes than later announcements.
  • A date forecast based only on existing calendar information may have little incremental value if it is fully predictable.
  • A deviation from an expected announcement schedule may be informative because the timing was not anticipated.
  • The document does not provide a forecasting algorithm or validate one on timestamp data.

Tags

EquitiesEvent-drivenStatisticsSentiment
Full text
# How do I predict future earnings dates if I have a database of all prior earnings dates?


# How do I predict future earnings dates if I have a database of all prior earnings dates?












So I have a database of all earnings announcements for all US stocks down to the millisecond for the past 10 years, and I want to make reasonable predictions on when exactly next earnings will be released. Any ideas on how I could accomplish this? Perhaps something with machine learning? Any help is appreciated!

## Answer by lehalle (score 4, accepted)

https://quant.stackexchange.com/a/79873

You question is unfortunately not as simple as it seems. You probably know that you can buy such calendars, that are progressively replacing "expected earning dates" by "announced earning dates".

The main difficulty stems from the informational content of the date not being the one expected. Indeed, if you can predict the exacte date, and if it deviates from the "expected date" (the one that you could get from historical calendars, using the "best ML tool"), they you will get information about the content of the earning announcement!

From this paper: Johnson, Travis L., and Eric C. So. "Time will tell: Information in the timing of scheduled earnings news." Journal of Financial and Quantitative Analysis 53, no. 6 (2018): 2431-2464.

> Our first tests show that high-R SCORE firms (i.e., “advancers”) subsequently report better earnings news than low-R SCORE firms (i.e., “delayers”) at their earnings announcements. Specifically, advancers report statistically and economically greater return on assets (ROA), same-quarter growth in ROA, and analyst-based earnings surprises compared with delayers. Together, these results highlight the predictive power of scheduling disclosures for firms’ earnings news and thus provide strong evidence that earnings scheduling is itself an information event that is commonly observable weeks ahead of firms’ actual announcement dates.

It implies there is no correct answer to your question

- one the one hand if you can predict the next earning date (with no other informations that calendars) then it has no value,

- on the other hand given that you cannot predict this date, then it has some value...

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.