I also quickly ran a regression, which indicated that an even more accurate forecast for quarterly GDP would be the following model:
Quarterly GDP = 0.25*NY Fed forecast + 0.66*Atlanta Fed forecast
Note though that this model was only based on 6 data points (6 quarters), because the NY Fed's model only became available in the second quarter of 2016.
The latest figures for the NY Fed and the Atlanta Fed forecasts were 1.6% and 2.5%, respectively, thus yielding to a forecast of just over 2% (0.25*1.6% + 0.66*2.5%) for the quarterly GDP figure for the third quarter.
Well, last Friday the BEA published its first estimate and it turns out that quarterly GDP came in at just below 3% (note though that sometimes these quarterly GDP figures are revised substantially in later months). So my "model" was off by a full percentage point.
Note that the Atlanta Fed model performed reasonably well while the New York Fed's estimate was pretty far off, which basically implies that one should give an even higher weight to the Atlanta model. Indeed, rerunning the regression yields a new estimate:
Quarterly GDP = 0.2*NY Fed forecast + 0.78*Atlanta Fed forecast
Let's see how this estimate hold up over time. I might be updating the model as I go along and maybe incorporate other estimates as well.