Computers outperform humans for accuracy on US GDP forecasts

0

It might be better to look at calls from largely automated computer models rather than economists, judging by estimates of second-quarter gross domestic product ahead of the actual results released Thursday.

The US economy contracted for a second straight quarter, with GDP falling at a 0.9% annualised rate from the first three months of the year, the Commerce Department’s preliminary estimate showed. The median estimate in a Bloomberg survey was for expansion of 0.4%, and of the 74 projections by economists, 23 were for a decline.

Forecasts in so-called “nowcast” models, however, were closer to the outcome. The Federal Reserve Bank of Atlanta’s GDPNow index, for example, saw a 1.2% decline.

Another similar computer model scored a direct hit: The estimate from S&P Global Market Intelligence, initially conceived by Monetary Policy Analytics Inc. — co-founded by former Fed Governor Larry Meyer — predicted a 0.9% second-quarter contraction. Its clients include governments, banks and the Fed itself, which uses the data to glean insight on where the economy is going.

The latest GDP report was closely watched, given that two consecutive quarters of shrinkage is one rule of thumb that many use to gauge whether an economy is in a recession. The official determination of ends and beginnings of business cycles is made by a group of academics at the National Bureau of Economic Research.

Nowcast gauges have gained bigger followings as their accuracy improves, and the estimates they produce hew more closely to outcomes as they accumulate data. GDPNow, for instance, had forecast a 1.8% decline earlier in the week, but adjusted to reflect the 1.2% drop by Wednesday.

The S&P model produces projections that are within about 1.2 percentage points of actual GDP about three months before the Bureau of Economic Analysis release, with the gap narrowing to about half a percentage point closer to the release date, said Ben Herzon, an executive director at the firm.

Model workings

We use the “bean-counter method,” said Herzon. “What we do is look at the BEA source data and replicate their methodology.” The trick is estimating values for the latest month of data that haven’t yet been publicly released, he said.

The Atlanta Fed’s GDPNow model is largely publicly available and also mimics the methods used by the BEA to estimate real GDP growth. One key difference is that it isn’t subject to judgmental adjustments.

“The average absolute error of final GDPNow forecasts is 0.84 percentage point,” according to the Atlanta Fed, which emphasises that the result isn’t an official forecast by the bank, its president, the Fed system, or the Federal Open Market Committee.

The historically most accurate forecaster for GDP in the Bloomberg survey, Brett Ryan at Deutsche Bank AG, called for a 0.6% contraction. Deutsche Bank also has a nowcast GDP tracker that aggregates other models and weighs them “in a way that has been optimal historically,” Matt Luzzetti, a New York-based senior US economist at the firm, said in an interview at Bloomberg’s Washington office.

The Bloomberg Economics nowcast for US GDP pointed to a negative print of 1.5% and compared with Bloomberg economists’ official forecast of 0.7% expansion.

Other Fed banks also have various models such as the Chicago Fed national activity index, the Philadelphia Fed monthly coincident index and Aruoba-Diebold-Scotti business conditions index to help gauge economic activity.

For the third quarter, as things stand today, the JPMorgan Chase & Co. nowcast shows US GDP expanding 1%, while the S&P model sees a 0.9% increase. The Atlanta Fed will issue its first third-quarter estimate on Friday, and will update the model more than 30 times until they cumulate with the final third-quarter GDP release on December 22.

© 2022 Bloomberg

FOLLOW us ON GOOGLE NEWS

 

Read original article here

Denial of responsibility! My droll is an automatic aggregator of the all world’s media. In each content, the hyperlink to the primary source is specified. All trademarks belong to their rightful owners, all materials to their authors. If you are the owner of the content and do not want us to publish your materials, please contact us by email – [email protected]. The content will be deleted within 24 hours.

Leave a comment