Global macro data has surprised to the upside in recent weeks in general. For major economies the last 3 months has seen such an ‘impressive’ rise that it has reached a point at which (historically) market expectations have become relatively exuberant. As the chart below shows, not only is the macro surprise index near its normalized highs (suggesting there is not much room for further positive surprises here) but each time the pace of apparent improvement has been so fast, the US equity market has faded lower as “hard” data simply does not support the hope in the markets and “soft” data.
As BofAML notes, while a bit more than half of the recent data have been weaker than expected, the manufacturing and services PMIs have been very strong and the consensus believes what it wants to believe, adding that it is important to understand how crude these surveys are. A popular view is that these surveys are better than hard data. In our view, however, these data get way too much air time. They give a timely, rough read on the economy, but should get little weight once hard data are released.
Each time macro data has weakened and stocks have faded, central banks have rescued it – but with “taper” now required, we ask – is this as good as it gets?
Destroying the ‘myth’ of the exuberant PMI data…
While a bit more than half of the recent data have been weaker than expected, the manufacturing and nonmanufacturing purchasing manager’s indexes have been very strong, jumping 4.8 and 5.8 points, respectively, since June. By some accounts, these data are better indicators than the hard numbers that come out of the government. After all, they are released very early, they are raw unfiltered data (other than seasonal adjustment), they are never revised and they are simple to interpret. We disagree. In our view, they are useful as a rough and ready early read on the economy. However, once the corresponding official data are released, we put very little weight on these surveys.
It is important to understand how crude these surveys are. Each month, a few hundred purchasing managers are asked if a variety of activity variables are up, down, or the same relative to the prior month. Their responses are then converted into diffusion indexes: the sum of the number managers reporting activity is “increasing” and half of those reporting “the same.” Note that there is some guesswork involved: the survey is taken before the month is over and some of the questions cover areas of the firm that are difficult for a purchasing manager to get a timely read on. For example, a purchasing manager may not have a very precise idea of what is happening to hiring in a large, diverse firm. Moreover, since they don’t gather specific numbers for each series, they may have to make a rough guess, particularly if the trend is slightly up or down.
Fans of the two indexes point out that they are relatively stable, easy to interpret and never revised. However, in our view, the simplicity of the data is a drawback, not an advantage. It means no attempt is made to correct misreporting or to include late respondents. Moreover, the sample they use is not representative of the overall economy. They represent a broad cross-section of industries, but they oversample big firms and they make no attempt to adjust for the birth and death of firms. The US is a dynamic economy and these surveys will miss these compositional shifts. Indeed, a lot of the revisions to official data come from attempts to fix all these problems rather than ignore them.
A comparison with payroll employment underscores these drawbacks. The preliminary payroll report is based on data from 145,000 establishments with 557,000 individual worksites. Thus if the BLS wanted to, it could turn its raw data into simple up or down answers and then create hundreds of diffusion indexes just like the employment component of the ISM index. However, that would mean throwing out information on both the size of employment changes at each company and turning a big sample into a bunch of tiny samples.
One way to show the information advantage of the employment report is to show how it correlates with manufacturing output. Using data from 1990 to present, the employment component of the manufacturing ISM index has a correlation of 0.39 with monthly industrial production growth. How does the official data compare? First, using the Labor Department’s own diffusion index—based on 84 industries—the correlation improves to 0.46. Second, using the actual job growth data, the correlation improves to 0.60. And, finally, if we also take into account the length of the work week, the correlation for aggregate hours worked and industrial production is 0.69. Clearly, more information is better.
How do we interpret the latest ISM numbers? Table 2 above shows the results when we regress GDP growth on its own lags and then add the composite ISM. The results underscore the difficulty in forecasting GDP. Using the average ISMs for July and August, the model with just GDP lags predicts growth of 2% in 3Q, while the model with the ISM points to 4.0%. However, the error band for these forecasts is very high – the “standard error” for the first model is 2.4pp and the second model 2.1pp. In other words, using a two standard error confidence band, we can be “95% confident” that growth is somewhere between zero and 8%. On the other hand, it is encouraging that the ISM is statistically significant.
the US data mills churn out a lot of surveys. Since the last FOMC meeting, there have been four new ISM readings and a bunch of regional releases. A popular view is that these surveys are better than hard data. In our view, however, these data get way too much air time. They give a timely, rough read on the economy, but should get little weight once hard data are released.