A few days before important economic data comes out -- housing starts, unemployment, durable goods orders, etc. -- analysts guess where they're going to end up. Usually, they're not right on the money. And now, it turns out, their errors are almost predictable.
The research team at Goldman Sachs put out a paper Thursday morning analyzing the magnitude and direction of "surprise" in every release of data -- that is, the difference between the consensus expectation of economists polled by Bloomberg and where the indicator actually fell. They found that the forecasts tend to underestimate the outcome for several months in a row, and then overestimate it for several months in a row. In other words, if the forecasts were overly optimistic in one month, they're more likely to be overly optimistic the next month, as well.
That's surprising, because you might expect economists to adjust their forecasts based on being wrong the last time. Instead, they seem to think they were still close to right, until they finally change their minds and overcompensate in the other direction.
And guess what? The phenomenon seems to have gotten more pronounced in recent years. Here's the measure of the "autocorrelation," or the degree to which one consensus forecast will match up with the previous one:
That probably happened because the economic shocks of the last four years have just been so much greater than those of the relatively smooth years before them, and analysts have a harder time adjusting to structural changes.
Why does this matter? In part, because markets react to the difference between whether an indicator exceeds or disappoints analysts' expectations. So if you know that the forecast is more likely to be wrong in the same way that it was last month, you can make bets on the market reaction that are more likely to be right.
Mostly, though, it's just an interesting window into how people who are supposed to be able to tell where the economy is going have their own blind spots. And it will be fascinating to note, after realizing that bias, whether they adjust to get rid of it.