Sell-side analysts' earnings estimates for individual companies, typically reported by I/B/E/S, are a key ingredient to many quantitative models. However, revisions to analyst estimates tend to lag changes in market expectations. Basing a model's recommendation on these estimates, particularly for valuation, can often lead to certain stocks seeming very cheap only because the stock price has declined dramatically and analyst estimates have not yet had a chance to catch up. Likewise, for market timing, the entire market can seem cheap immediately after a crash ahead of a predicted recession.

How should one deal with the lag when constructing valuation indicators? Should you attempt to correct for the lag by attempting to predict the coming changes in earnings estimates? If so, how should you do this prediction? Is there any available research on analyst earnings revisions?


I wouldn't put too much faith in IBES forecasts. You may remember this situation:


(In case the above link doesn't work, Google "Rewriting History Alexander Ljungqvist").

You'll find lots of excuses for worthless forecasts:


Below is a graph that I saved from some study. As I recall, their point was that analyst forecasts are typically nothing more than slightly modified naive forecasts (i.e. next year's earnings will be about the same as this year's earnings....and this was for analyst forecasts of S&P500 earnings). I'll post a reference if I can find it again.

enter image description here

The bottom line is, you're better off if you ignore analysts. From your own models you'll find that forecasting earnings for individual companies is a crap shoot. And, by far, the easiest earnings sequence to forecast is for the S&P500. However, that's by no means easy. Structural shifts happen without warning and model inputs that were useful for decades can easily drift into bizarre territory.

As far as your questions go, if you have evidence that some analyst's earnings forecasts have influence on a stock/index, your model would probably be more about the analyst than actual earnings.

Edit 1 (10/11/2011) ===================================================

Additional information on IBES earnings data:


Analyst earnings expectations:





  • $\begingroup$ Bill, I was not asking for a tirade against IBES. Unless you can point to something better, I have to do the best with what we have. My models, btw, do show significant added value coming from using those analyst estimates. The "slightly modified" naive forecast may not be very different in a statistical sense, but it has huge pricing and relative value implications. $\endgroup$ Oct 11 '11 at 19:35
  • 1
    $\begingroup$ @Tal: That wasn't a tirade against IBES, I simply linked a couple of papers. As stated above, the analysts may be the issue, not earnings. $\endgroup$
    – bill_080
    Oct 11 '11 at 19:52
  • $\begingroup$ BTW, the article you reference refers specifically to analyst recommendations, not analyst earnings estimates. Thomson/IBES claims the earnings estimates data don't have many of the same issues. $\endgroup$ Oct 11 '11 at 20:00
  • 1
    $\begingroup$ OK, but at the end of the day, I am going to use these estimates and I'd like to know how they respond to changing conditions. Do you know of any research on that specific point? $\endgroup$ Oct 11 '11 at 20:01
  • 1
    $\begingroup$ @Tal: All I can say is, try it for a while. You won't have to wait long. Keep your own version of the "Detail History" (quarterly should be good enough) and watch the Exclfil* file closely. When it changes, e-mail IBES and ask why, and ask if it affects the composite earnings forecast. $\endgroup$
    – bill_080
    Oct 11 '11 at 20:24

I would normalize valuation variables over the business cycle. These normalized variables exhibit mean-reversion. For example, use price-to-peak earnings rather than P/E. Here is a good illustration of the idea.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.