In that light, the recent divergence of the Atlanta Fed Nowcast and the New York Fed Nowcast are extremely interesting, with the former displaying an improved outlook and projecting a quarterly growth rate of now 4.8%, up from about 4.3% at the beginning of the August. The New York Nowcast, on the other hand, has shown a declining growth rate of now just about 2%, down from exactly 3% at the beginning of the quarter.
New York Fed GDP Nowcast
The correlation between the forecast error of the Atlanta Fed Nowcast and the NY Nowcast is 33%, meaning that most of the time both models are producing forecast errors in the same direction. The recent divergence of the two models over the last two months is thus especially interesting. Unfortunately, I haven't had the time to dig deeper into the two models and can thus not comment to what extent they differ and what the source of the different prediction for this quarterly GDP figure is.
1) The dynamic factor models as a tool for GDP predictions have not been around for very long. They are purely data driven and thus completely differ from the DSGE models Central Banks have used over the last few decades. Many of those DSGE models have failed quite spectacularly during the financial crisis (Not in terms of failing to predict the crisis. Crises are by definition more or less unpredictable. Don't believe anybody who tells you otherwise. But the models failed in predicting some of the important dynamics in the aftermath of the Great Recession, such as the failure of interest rates and inflation to rise, etc.). I have not seen any study so far that compares the accuracy of those DSGE models with the data driven dynamic factor models, but my gut feeling is that the data driven approach can be superior. The DSGE methodology rests on some highly dubious assumptions on how the macroeconomy works. There is no doubt that the core neo-keynesian model is quite flawed, a topic I have written about before. See here, for example.
2) Quarterly GDP figures are highly volatile. One might think that an average forecast error of 0.6% is quite high, but given the state of current macroeconomic models this is at the moment the best macroeconomists can do. Other variables are much more sticky, especially inflation and also interest rates to some extent, at least in recent years, thus making a prediction somewhat easier.
3) Financial markets efficiently aggregate information, at least most of the time. Financial market predictions about the future path of interest rates, the Fed funds future market, has repeatedly turned out to be more accurate than even the Fed's internal forecasts on where interest rates will be in one year's time.