Experts and complex, chaotic, constantly evolving, loosely coupled systems.

As I mentioned earlier,

The most difficult systems to model are those which are complex, chaotic, constantly evolving, loosely coupled systems. Such as health, diseases, economics, climate, voting patterns, etc.

And now the Labor Department releases its monthly figures providing a case study of the point.

From Unemployment in U.S. Unexpectedly Falls in May: Live Updates

The unemployment rate fell to 13.3 percent in May, the Labor Department said Friday, an unexpected improvement in the nation’s job market.

Economists had expected the rate to increase to as much as 20 percent, after it hit 14.7 percent in April, which was the highest since the government began keeping official statistics after World War II.

The economy added 2.5 million jobs after a record loss of 20.7 million in April.

“These improvements in the labor market reflected a limited resumption of economic activity that had been curtailed in March and April,” the Labor Department said in its release.

The report noted that “employment rose sharply in leisure and hospitality, construction, education and health services, and retail trade,” even as jobs in the government continued their decline.

Forecasting unemployment is hugely consequential. There is a lot of money riding on it. One month is usually a manageable time-frame.

They forecast a rise in unemployment to 20% and we actually had a decline in unemployment to 13.3 percent.

They got the sign wrong (declining rather than rising) and they missed the level by 34%.

When everyone is watching the economy closely. When there is lots of ready data. When there is a lot of money to be made by forecasting correctly.

This is not an indictment of models and forecasting. They are necessary, in part because the discussion of how to model deepens our understanding of what it is that we are modeling.

And pivot points are among the most difficult things to predict, both when and by how much.

The criticism is against the naive estimation that any model is producing either a precise or accurate forecast. The focus has to be on the process of the modeling rather than the output. And it rarely is.

Complex, chaotic, constantly evolving, loosely coupled systems are inherently difficult to model and we should never forget that and never attach too great a certainty to the outcome of the model.