The Monkey Cage

Now I might be missing something here, but if the forecasters were good at their jobs, shouldn’t the long term average of companies beating forecasts be the same as the long term average of companies doing worse than the forecasts?

It depends what they’re aiming for. If the distribution of innovations is asymmetric, then no, not necessarily. E.g. suppose that the deviation from the mean is +1 with probability 2/3 or -2 with probability 1/3 (but never any other value). Then a forecaster who minimizes mean squared error will find that reality beats their forecast 2/3 of the time.

In this case, however, I’d guess some sort of corrupt practice is at work (perhaps forecasters working for banks which also solicit business from the companies in question).

Here is what someone who is smarter than me says:

Measurement bias and Heisenberg type problem.

I run a public company. I know what the consensus is. Say 1.50 a share. If I’m close 1.47 to 1.53 depending on how I book and bill etc. I can manage earnings to beat 1.50. Say 1.51. So it can be manageable to beat earnings estimates by a little very often once I know the target. Notice how companies tend to beat by a little - miss short by alot. Don’t have data to support but I think I’m correct. Nothing that the analyst can do about it. Also, analysts tend to be guided to some extent by the Co. This is less innocent and was to some extent the focus of Reg FD.

I gess it is related to this topic: http://papers.nber.org/papers/w15189#fromrss

Here the abstract:
Market Selection:
The hypothesis that financial markets punish traders who make relatively inaccurate forecasts and eventually eliminate the effect of their beliefs on prices is of fundamental importance to the standard modeling paradigm in asset pricing. We establish necessary and sufficient conditions for agents making inferior forecasts to survive and to affect prices in the long run in a general setting with minimal restrictions on endowments, beliefs, or utility functions. We show that the market selection hypothesis is valid for economies with bounded endowments or bounded relative risk aversion, but it cannot be substantially generalized to a broader class of models. Instead, survival is determined by a comparison of the forecast errors to risk attitudes. The price impact of inaccurate forecasts is distinct from survival because price impact is determined by the volatility of traders’ consumption shares rather than by their level. Our results also apply to economies with state-dependent preferences, such as habit formation.

chaz got it. not only do companies have incentives (stock gains) to manage their performance to beat estimates (narrowly - there is no reward for hitting the estimate, but beating it by too much makes you look clueless, too), they also play a heavy role in the setting of those estimates in the first place. they provide “guidance,” in the form of a range, fully expecting to be able to come in at the top of the range. then they manage down, so as to just “beat” the “estimate.”

the secondary point chaz makes is probably right, too. once a company knows they’ve screwed up, they pull all their write-downs and re-org charges and what have you into that fiscal period, so as to make a clean start after their huge “miss.”

Thanks for the comments - this is really interesting. So basically the argument is that the fault for the upward bias lies with the companies themselves, and not the analysts.

If the companies are gaming the system in this way, then it suggests the numbers we should be watching is not the percentage of companies beating forecasts, but rather some sort of summary statistic of the extent to which companies are overshooting/undershooting estimates. If Chaz is right, that should average to zero in the long term, and thus being above zero at any given time would signify that companies truly are outperforming analysts expectations (or, to be honest, that there is noise in that particular observation).

Still, the big question seems to be if the market knows what Chaz knows - that companies doctor the numbers to come in above analysts’ estimates - then why does the market reward beating estimates?