A few weeks ago we posted this chart from of the Nomura economics team:
It shows the recent diverging trend between the Fed’s measure of industrial production and that of the Institute for Supply Management (ISM).
The explanation Nomura gave for the divergence was that of these two indicators, only the Fed measure had been properly adjusted (earlier in 2011) to account for seasonal biases in the data that have distorted a range of indicators since late 2008. These biases exist because the computational techniques used to seasonally adjust economic data inappropriately interpreted some of the downturn in the fourth quarter of 2008 as a new seasonal trend.
Nomura went on to predict that the December ISM reading, which was released yesterday and beat consensus at 53.9, would be exaggeratedly strong because of these biases. (The production level subindex, which is what Nomura shows in the graph above, was 59.9.)
Well, Goldman have a new note out today analysing the same thing, estimating that both the mild winter weather in the US and these seasonal biases gave the ISM reading an artificially strong boost, probably on the order of two or three points.
The most interesting part of the note is the explanation of the seasonal factors:
We suspect that the financial crisis distorted seasonal factors for many economic series, causing “seasonally adjusted” data to be relatively stronger in the winter months and weaker in the summer months. The reason is that the collapse in economic activity following the bankruptcy of Lehman Brothers in September 2008 was evident in sharp declines in many economic series that winter.
For example, the ISM manufacturing index fell nearly five points in September 2008, to 43.8, on its way to a low of 33.3 that December; it did not regain the 40 mark until the following May. Without special adjustments to correct for the crisis, standard seasonal adjustment algorithms would react to this plunge by subsequently “lowering the bar” for data in these months; the corollary to this would be a “tougher” seasonal adjustment in the summer months.
If the true seasonal pattern remained unchanged, the result would be seasonally adjusted series that tended to look weaker in the summer and stronger in the winter, with the seasonal factors gradually normalizing as they incorporate “normal” winter data in subsequent years. Exhibit 1 illustrates the implied seasonal factors for the ISM manufacturing index, which show a more pronounced swing since the crisis.
And here’s how they arrived at their alternative estimate, which adjusts for the mild winter and the seasonal biases:
We estimated the potential impact in two ways.
The first is to seasonally adjust the raw data through 2007 and then use the seasonal factors from 2007 for subsequent years. This approach has the merit of simplicity, but it will not compensate for longer-term trends in seasonality that may have continued since the crisis.
The second approach is to excise the financial crisis period (specifically, omitting one year of data beginning just before the Lehman bankruptcy) and estimate seasonal factors using this series. This method is more likely to capture longer-term trends but risks distortion because of the skipped year, in particular if there are unusual jumps at the point where the series are spliced (we choose breakpoints to minimize this).
Both approaches yield an alternative series that is as much as two to three points lower from November to March than the official seasonally adjusted index, and as much as two points higher in the summer and early fall. However, the month-to-month path differs somewhat; because neither method is obviously preferable to the other, we simply average the two to generate the gray line in Exhibit 2.
This crude analysis implies that without the impact of the crisis on seasonal factors, the seasonally adjusted December index might be one and a half points lower than the official figure. The impact seems to be concentrated in the new orders, production, and employment components; our alternative adjustments for the supplier deliveries and inventories components look very close to the official series.
Goldman goes on to the make the point that this could well apply to the manufacturing data from other countries, which also have been better than expected.
To repeat ourselves: to the extent that this is happening with other indicators — and we think it probably is — that doesn’t mean the recent improvement in the US recovery is a mirage. It’s real and looks set to continue barring a eurozone disintegration or some other cataclysm.
But its slope is flatter than the recent burst of surprisingly not-crap economic indicators would suggest.