Designing Studies of Historical Earnings Surprises
Summary
The document outlines questions and practical choices involved in studying how earnings surprises relate to stock behavior. A study can measure how often firms beat or miss expectations, the size of the gap, price moves around announcements, later performance, and differences across industries. One suggested workflow is to gather reported earnings and analyst estimates, define surprise thresholds using the distribution of forecast errors, then compare stock prices around each event.
The discussion cautions that the result depends on what counts as earnings and as consensus. EPS may be distorted by unusual items, while revenue or company guidance may better reflect the news investors respond to. Mean, median, estimate range, influential analysts, and informal expectations can produce different surprise measures. The replies offer data-source suggestions but no empirical findings or validated strategy. They also note that announcement moves are difficult to attribute solely to reported results because guidance and other information arrive at the same time.
Key ideas
- A historical surprise study needs explicit definitions for the event, the forecast benchmark, and the price response window.
- Reported EPS can be affected by unusual items, so revenue and company guidance may also be relevant.
- Consensus can be measured with the mean, median, estimate range, or other benchmarks, each capturing different expectations.
- One proposed workflow compares forecast errors with their historical distribution and then studies stock moves around events.
- Price changes after announcements can reflect guidance and other news as well as the earnings result.
Tags
Full text
# Measuring historical earnings surprises, their frequency and severity # Measuring historical earnings surprises, their frequency and severity This is my first post to Quantitative Finance, so I hope my question is formatted the right way. I am starting to research the effects of earnings surprises on certain equity indices. Is there a source, such as an academic paper or database, for: Companies that have had a high incidence of positive and/or negative earnings surprises over the last 5 to 10 years? Statistics on the frequency and severity of earnings surprises on the day of and a few days after the event? Statistics about the medium-term performance (the following quarter, for example) of stocks with large negative or positive surprises? Industries with more earning surprises than others? Thank you very much in advance for your help. ## Answer by berkorbay (score 2) https://quant.stackexchange.com/a/14639 Recently I found a book on earnings trading but did not have time to read thoroughly. Trading on Corporate Earnings News - John Shon I also had spent some time to see earnings surprise effects and it is a quite interesting but not easy to use topic. There is certainly a jump if the estimates and announced earnings have a large mismatch but the magnitude and direction are hard to quantify. I also recommmend Fama's efficient markets survey. It is generally about market but it also makes a point about the market's quick reaction to news such as earnings. p.s. I used Google's returns to make a point. It is totally crazy. ## Answer by SCallan (score 1) https://quant.stackexchange.com/a/10169 There are many issues with doing this type of study. Will you be looking only at EPS surprises? Many times the bottom line EPS is not relevant to the earnings of the company since it contains extraordinary items. You may want to consider revenue surprises. Investors are typically looking at a lot more than just the EPS when earnings are released. They're looking at the company's guidance for future earnings. So it's difficult to say what percentage of a stock's price movement is due to the earnings result and what percentage is reacting to the company's guidance for its future earnings potential. StarMine (Thompson Reuters) has been following analyst estimates. With any data set of analyst estimates, you need to consider whether you'll use the mean estimate, the median estimate and whether you'll consider the range of estimates. Sometimes a star analyst's estimate carries more clout than the average of all analyst estimates. Also, sometimes investors believe the consensus estimate and sometimes there is more of a "whisper" number that is not officially publicized but generally expected. For example, people may expect a company will beat analyst estimates, yet the analysts haven't come out and increased their estimates. You could also consider using company guidance vs. reported earnings. ## Answer by PabTorre (score 0) https://quant.stackexchange.com/a/9925 building these statistics is just a matter of getting the right data... the actual earnings are easy to find on EDGAR... using their very friendly and fast FTP site. The estimates could take a little more work. yahoo earnings has a few entries, but it is not very complete, and you have to crawl day by day instead of doing it by symbol... which can be a pain... msn money seems to have pretty complete info and the navigation is done by symbol so it sounds quicker, specially if you only need a few stocks and don't care about little details like the time when the reported... http://investing.money.msn.com/investments/earnings-estimates?symbol=aapl http://biz.yahoo.com/research/earncal/today.html once you have the data, you would need to define what a suprise is (maybe estimate a distribution of the error on the estimate and get those that are more than 2 std from the mean?) and voila! you have your event. At that point you just need to get daily stock prices and see the moves in the days after each event. last step is, once you have your results, share them back with the community by uploading them to quandl :)
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.