Mixed signals from the US

Most economic time series, in all countries, are based on sample surveys.  A group of respondents is asked  questions about the state of their business.  Because it would be prohibitively expensive to ask everybody, only a small percentage of the total is asked what is happening or what happened over the time period.  The statisticians then estimate what their small sample of actors in a sector means for the whole sector.  This introduces random errors into the data.  In addition, just like waves and ripples on the sea, there are random fluctuations in the actual economy, not just in our measurement of it. 

There are two ways to get round these issues.  The first is to 'extreme-adjust' an individual time series.  This technique measures the random error of each individual observation of the time series by comparing it with a smoothed trend (usually some version of a Henderson curve) and if an individual error term of an observation is more than 2.5 standard deviations away from the average of over, say, the last five years, that observation of the time series is replaced with the value of the moving average.  This technique removes "spikes" in the data caused by bad weather, strikes, or simple mismeasurement.

The other way is to use different surveys.  For example, retail sales data come from surveys of shops (online and brick and mortar).  Industrial production comes from surveys of factories.  If you take an average of these two time series drawn from completely different samples, you reduce the random error, and get a clearer picture of what is happening in the economy.

There are two national opinion surveys in the United States, as well as several regional surveys.  The two national opinion surveys are produced by different bodies, using different samples.  The oldest is the ISM survey, produced by the Institute of Supply Management (previously the 'National Association of Purchasing Managers', or NAPM.)  Initially the survey just covered manufacturing, but about 20 years ago, the service sector was added.  The other is produced by IHS Markit, and is called the PMI, or Purchasing Managers Index. 

Right now these surveys are showing different things.  Here is the ISM manufacturing survey, with the official data and my calculation of the extreme-adjusted data:

As you can see, this indicator is still falling.  The ISM survey of services is however rising.  And this is very unusual.  Normally, because manufacturing is much more cyclical than services, it is manufacturing (and construction) that tend to drive the business cycle, with the result that upturns and downturns in services lag behind similar shifts in manufacturing.  You can see that in the chart below:

The other survey I look at, the PMI, is much stronger.  Is shows a renewed upturn.

And the average of the manufacturing and services PMIs is also rising.

Should we go with the ISM manufacturing survey?  It is the longest, with data going back, I believe, to 1937.    For the data I have available, the ISM manufacturing has called all major economic cycles and many minor ones too.  On the other hand, the PMI data are additional information, helping reduce random fluctuations when we add them to the ISM data.  Below I show the manufacturing surveys for both series, extreme adjusted, and their average, and also the whole economy series.  The balance of probabilities is that the US recovery is advancing.  I'll do some deeper analysis over the next few days.