ISM Non-manuf - best survey bet for NFP

The Philadelphia Fed Business Outlook report, especially the Employees components, has garnered a lot of interest leading up to today's soon-to-be-released US Employment report. The reason being that it is supposedly a very good indicator of the change in payrolls. And, yes, judging just from a chart the two series (Change in Nonfarm Payrolls and Philly Fed Employment) do in fact seem to be highly correlated - at least when looking at only this millennium.

However, that does not imply that it is a "best indicator" or even the best among surveys when it comes to projecting Nonfarm Payrolls. Below we have modelled six employment indices of some of the most widely followed surveys versus (the change in) payrolls. We start off by allowing each model to use all the available data for that particular employment component meaning that the residuals (i.e. errors between the actual payroll number and the fitted one) are of varying length.

It is, obviously, quite hard to conclude anything from the above chart so we have calculated the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) of each model each well.

We see that the ISM Non-manufacturing Employment Index is in fact the best predictor of payrolls with an average error of 103,000 though its short existence may intefere with results. Of the older series (ISM Manuf., Philadelphia and Chicago) the best is in fact the Philadelphia Fed Employment Index with an average error of 127,900.

As noted above the short time of existence of the ISM Non-manufacturing report and also of the Empire Manufacturing report, which is the second best) may skew results. Hence we now model all series using only data from (July) 2001 and forward. This gives the following:

Or aggregated into our two measures, MAE and RMSE:

And there you go. Using only the last roughly ten-and-a-half year's of data the conclusion is the same. ISM Non-manufacturing still leads, Philadelphia Fed now comes second and ISM Manufacturing is a distant third. There is in other words some truth to the notion that the Philadelphia Fed report's Employment Index is a good indicator of payrolls - but it is not the best. We round off by offering the predictions for today's payrolls number based on the second set of models (data from 2001 and forward):

This short exercise only seeks to compare six widely known surveys. It is not an attempt to find the best overall model for predicting payrolls, which would require a much more vigorous approach, including out-of-sample tests and a more thoughtful selection of variables.