Computer models outnumber experts in the accuracy of US GDP projections

Computer models outnumber experts in the accuracy of US GDP projections

Based on projections of US gross domestic product in the second quarter, made before the actual results were released on Thursday, predicting automated computer models is much more accurate than the calculations of economic experts.

According to a preliminary estimate by the U.S. Department of Commerce, the country's economy is declining for the second quarter in a row, with GDP falling by 0.9% annually in the second quarter. According to the Bloomberg survey, the average forecast was 0.4% higher, while 23 out of the 74 economists surveyed relied on a reduction.

The projection of the so-called "short-term forecast" models was closer to reality, for example, according to the projection GDP Now of the Federal Reserve Bank of Atlanta, GDP would have fallen by 1.2 per cent.

The projection of the computer model S&P Global Market Intelligence, developed by the Monthly Policy Analytics, was entirely consistent with the actual result of 0.9%, which is used by the U.S. government, banks, and the U.S. Federal Reserve itself.

Short-term projection models acquire more supporters as they improve their accuracy, and the estimates they provide become increasingly similar to actual results as data are collected. The GDP Now index, for example, projected a 1.8 per cent decline in GDP at the beginning of the week, but adjusted the projection to 1.2 per cent for the environment.

The S&P model provides projections that are within about 1.2 pp. from the actual GDP about three months before the U.S. Bureau of Economic Analysis published, with the gap narrowing to about half a percentage point closer to the date of publication, reported Ben Herzon, CEO of S&P.