X-12 Arima Is Back: A Look At ADP's "Seasonal Adjustment" Protocol

Tyler Durden's picture

You know it: our old friend the BLS's very own Arima X-12 makes a very unexpected appearance. Why a private entity, the ADP, which has no links to the Bureau of Labor Statistics is using the same adjustment process used by the government agency, to divine its final, seasonally adjusted number, especially when it refuses to disclose its unadjusted data, is anyone's guess. Or is the ADP number now nothing but a reinforcing surrogate to double the credibility of the BLS data, whose credibility in recent months has also hit unseen lows? It certainly would explain the recent revision in ADP methodology, and the fact that administration sycophant, Moody's Mark Zandi is now the "brains" behind this meaningless number (not to mention the resulting humiliation for all those who had though that ADP data, like that from the NAR, is even remotely credible).

From the ADP's "Methodology-Full Detail"

Seasonal adjustment


Employment growth in each of the 90 size class/industry cells is seasonally adjusted using the Census Bureau’s X-12 ARIMA method, with the default ARIMA outlier criterion modified to resemble the corresponding CES criterion. Deseasonalized trends for the industrial cells are recalculated with each new month of data.


Each observation in an industrial cell is then compared with the trend value, and outlier observations are removed. Matched employment growth for the 90 size classes is calculated for the second time using the cleaned data. Each cell employment growth is seasonally adjusted again using the same X-12 ARIMA method.


An additional adjustment is made for months in which there are five weeks between survey reference weeks. This is done by regressing the growth rate in each cell on a dummy variable, which if significant, is used to eliminate the long-month effect.

And where it gets funnier:

Regression and Results


Moody’s Analytics uses a restricted structural VAR model to estimate the monthly change in private payrolls by regressing the most recent payroll growth reported by the BLS for each super sector on a constant term and: (1) ADP matched-pair growth rates by industry; (2) lagged values of BLS estimates of growth of employment by industry with industry specific restrictions; and (3) the Philadelphia Federal Reserve’s Aruoba-Diebold-Scotti Business Conditions Index (ADS). The equations for all industries are estimated simultaneously. Thus, the model is made up of 10 endogenous variables—the 10 BLS super sector industries, and 11 exogenous variables—the 10 ADP super sector industries and the ADS index.


The ADS index was chosen because it incorporates the timeliest data and is updated on a daily basis as new data are released. The estimation uses a weekly average of the daily index for the week that includes the 12th of the month. The index is based on the following seven indicators: weekly initial jobless claims, monthly payroll employment, industrial production, personal income less transfer payments, manufacturing sales, trade sales, and quarterly real GDP. The average value of the ADS index is zero.


Progressively bigger positive values indicate progressively better-than-average conditions, whereas progressively more negative values indicate progressively worse-than-average conditions. The index is seasonally adjusted,

... Aaaaaand that's where it ends. Cut off mid sentence. Dont believe us? Check page 3 of 10. And this is the source of the "data" used by the likes of Goldman Sachs to revise their NFP forecasts for tomorrow (higher from 175K to 200K as the case may be).

So much for the ADP's revised "credibility" which is nothing but a dummy attempt by the BLS to double up the "validity" of its own data by getting a comparable data point the day prior, by a company that effectively merely piggybacks on BLS "source data." How to check? Tomorrow's NFP print will be within 15-20K of today's ADP seasonally adjusted 215K print.

At least the NFP provides what its unadjusted source data is: ADP refuses to even do that.