A few weeks ago, I wrote about the interesting phenomenon of “behavioral leapfrog”the research puzzle | If you have not read that piece, it will make sense to do so before continuing. in the movement of earnings estimates by analysts.  Because of one comment that I received about it, the topic is worth revisiting.

Before dealing with that comment specifically, I’d like to draw your attention to a paper by Ambrus Kecskés, Roni Michaely, and Kent Womack, entitiled “What Drives the Value of Analysts’ Recommendations: Earnings Estimates or Discount Rate Estimates?”SSRN | Over the years (and with a variety of co-authors), Womack has produced a number of interesting papers regarding analyst recommendations. As background you ought to know that the academic literature over the years has consistently shown that there is little if any exploitable information in the level of the average recommendation on a given stock, but that the changes in the level of the average recommendation have some predictive power in explaining subsequent returns.

The authors explore the differences in upgrades or downgrades depending on whether they are accompanied by changes in earnings estimates, or whether they instead involve “discount rate changes,” which as defined in the paper include both what might classically be considered discount rate changes (with no change in underlying information, a change in recommendation would have to result from a change in discount rate) and changes in long-term growth rates, which, like discount rate changes, are “soft” and long-term.  In contrast, estimate changes are specific and more near-term, making them, in the hypothesis of the authors, less subject to cognitive biases.

(As for those growth rates, the authors echoed one of the assertions in “unpegged,” my attack on PEG ratios,the research puzzle | This has been my most widely-read posting to date. when they wrote that “analysts rarely change their discount rate estimates and growth rate estimates let alone justify them with detailed explanations or models.”)

The paper indicates that earnings-based recommendation changes are much more powerful when accompanied by estimate changes, reinforcing the fact that changes in earnings estimates (or cash flows) are what matter most.  (This is also supported by surveys of portfolio managers, who say that they pay much less attention to recommendation changes than they do to estimate changes.)

That brings us to the comment on my previous posting, which was from a person with experience as a sell side analyst.  He also shared his thoughts as “dsquared” on Felix Salmon’s review of “behavioral leapfrog.”Reuters | Salmon is one of the leading financial bloggers of the day. His main point was that rather than being behavioral in nature, the leapfrogging of estimates results from an application of game theory by the analysts, that they are responding rationally to the incentive structure under which they operate and that there is no incentive to be bold in making estimates because it doesn’t pay for them.

As someone who labored over an eight-part series on misaligned incentives in the investment business,the research puzzle | Here is an index of those postings. I should have thought to include that in my piece.  Watching and interviewing analysts over the years (those on the buy side as well as the sell side) leads me to think the behavioral aspects are broadly dominant, but dsquared makes an outstanding point and it is one of great importance to those involved in research management.  I believed that most firms incorporated sufficient incentives for analysts to be aggressive about getting earnings as close to the mark as they can — and to do so well before other analysts if possible.  dsquared indicated in his private comment to me that it wasn’t the case at his firm.  The bottom line in this regard is clear:  With all the empirical evidence pointing to earnings estimates as being critically important, research directors need to make sure that the incentives are in place to highly reward analysts that are good at estimating earnings.

The other point made by dsquared was that Salmon drew a questionable conclusion from my posting.  I agree.  While Salmon is free to state his views, to clarify my original conclusions it helps to summarize them from two different points of view.  If you are an observer of estimate changes, you need to understand that, be it for behavioral reasons or because of incentive structures, estimate changes tend to persist and the sooner that you can understand when a migration starts and why it might be occurring, the better chance you have to take advantage of it.  If you are a creator of estimates, it is important to break free from the behavioral fetters to make bold predictions.  To those I would add another, courtesy of dsquared:  If you are compensating people who create estimates, make the incentive structure do what you should want it to do.

There will be other posts down the road regarding estimates and recommendations.  Keep those electronic cards and letters coming in.