China's IP for May

Year-over-year Chinese industrial production growth rates fell from 5.4% in April to 5% in May, after an anomalous spike in March to 8.5% from the band of 5.3 to 6.1% it's been in since June last year.  The 8.5% spike makes it difficult to estimate a trend, because it is such a big jump compared to the range of previous and subsequent months.  Is it real, or just some random statistical aberration?

What I do to remove apparently anomalous spikes from the data is to extreme adjust the time series.  This is a technique invented by the US Bureau of the Census, which tests the irregular component (error term) of a time series, and scales back individual data points to the underlying trend-cycle where the error term for that particular data point is more than 2.5 standard deviations from the average error term over the whole extent of that time series.  Whew!

I also fit a moving average to the underlying data, which is an attempt to represent the underlying "trend-cycle" of the time series.  The chart below shows the official data, which in common with the way data used to be announced in communist countries, is released as a year-on-year percentage change, not as an index level.  Note how necessary extreme-adjustment is to remove "spikes" in the data.   See how, if we ignore the March "spike" in 2019, the trend as interpreted by eye seems to still be down.  The statistics reckon that it might be turning, as shown by the smoothed extreme adjusted series, i.e., the blue line.  What the extreme-adjustment process has done is to reduce not just the March upward "spike" but also the May downward "spike". 

However, extreme adjusting a rate of change is problematic, especially when there are huge secular shifts in growth rates, as there have been in China.  Relative to its surrounding data points, the 2012 spike is no larger than the 2019 one.  Yet my extreme-adjustment program would treat the 2012 spike as more anomalous than the 2019 one because it is larger in absolute terms.  If we extreme-adjust an index, on the other hand, it's the percentage (logarithmic) deviation which is adjusted, which would treat both anomalies more closely.

So to get a better estimate of underlying Chinese IP, I generate an index of the level of Chinese industrial production, using the year-on-year percentage change.  I then seasonally adjust this resulting series.

This version of the year-on-year percentage change shows the beginnings of a modest uptrend which began in February, after the downtrend through 2018.  My own Chinese leading index points towards an upturn, but I have no way of knowing just how disruptive the Trump trade war is to the Chinese economy.  I think the damage is significant, with the real risk that this new Chinese economic upturn just fizzles, despite the Chinese government's attempts to re-stimulate growth.  Which would not be good news for the world economy, given slowdowns across the OECD area.