Among my pet peeves as a market economist are all these China Indicators that keep popping up. While I can understand the frustration with the poor transparency in Chinese official statistics, I have always felt that trying to out-do China’s National Bureau of Statistics is something of a lost cause; Chinese data might be bad (for whatever reasons), but how can we – outsiders – expect to be able to create something ‘better’?
– Despite this Sisyphean task, all investment banks and research organizations as well as many think tanks, not to mention a not so small number of academics are trying to promote ‘their’ work as the ultimate China Indicator. But yes, before you remind me, we have also done a foray into this space, so I admittedly have little right to complain.
Rest assured, I have no issue with the creation of better-performing indicators. But what I fail to understand is, better than what exactly? What is your baseline? How do you validate any activity indicator, if you don’t trust the official activity data?
To my (limited) knowledge, only a few researchers have explicitly discussed the issue of validation of (e.g.) Chinese data, among them Pinkovsky and Sala-i-Martin (2014) with a path-breaking use of satellite data, and; more recently, a ‘forensic’ data approach in Chen et al (2019), who collect regional and value added tax-data on a very detailed level.
Another, and in my opinion, more pragmatic approach can be found in Fernald et al (2013), that uses external exports to China as a benchmark. Their hypothesis is that imports are a good benchmark for developments of demand as imports move very much in tandem with measured GDP-growth for all economies, in particular those with a developed statistical tradition. What we need to do, then, is to aggregate up the mirror image of Chinese imports; country-by-country exports to China.
Hence, what domestic data seems to correlate well with the calculated import proxy should be relatively free of bias or manipulation from domestic statistics producers in China. That said, and on a similar note, our preliminary results indicate that Chinese goods imports co-move quite well with our import proxy, suggesting that China’s foreign trade data (at least imports) might not be very manipulated after all.
 Fernald et al seem to aggregate external exports and that is easily done in the application as well. Here, nonetheless, I have used the principal components of all exports growth series.
Macrobond moment: By searching for the exact search string (following the organization in our database): “Foreign trade exports country China” in the search tab, I got all data and my chosen meta data to exclude, for example, countries with very short time series or ‘wrong’ frequencies.
Now, in order to rid the data of any trends, we have de-trended data used in all calculations with a HP-filter (standard settings). In addition, to avoid overfitting (multicollinearity), I used a principal components approach that might also be helpful in avoiding using data that has been manipulated by the statistical agencies.
In this next step, we need to have a look at the domestic indicators that seem to co-move the best with our import proxy and after using a number of combinations there are two set-ups that work ‘best’. In short, the series we end up with are: (1) Consumer Confidence (NBS); (2) Energy Production; (3) Goods Exports; (4) Fixed Assets Investments; (5) New construction of floorspace; (6) Industrial production; (7) Rail freight volumes, and; (8) Retail trade in consumer goods. (Indeed, it does have a Li Keqiang ring to it!)
When plotting the index on GDP, we see that they often move together very nicely. However, in the early parts of the sample, as well as the last few years, the covariation is weaker. It is over the latter part of the sample in particular that developments are really eye-catching. Although the GDP-growth is almost bang in line with its trend, our proxy (-ies) point to quite large cyclical variations (N.B., the diverging developments seem to precede the “trade war”). Interestingly, these variations also go hand in hand with more anecdotal evidence of how the Chinese economy is faring (such as PMI).
Macrobond moment: If you want to calculate the indicator to a monthly basis just change the frequency of the entire document to monthly, in the series list. Oh, if you do, don’t forget to add the regressor “Chinese New Year” to rid the indicator of spikes in Jan/Feb. And to update it later, there’s no need to feed data into excel or pass it through to R or E-views. Just one click and you have the China indicator updated.
We still have a lot of work to do to build this indicator into something we can publish on a more regular basis, but at least this initial analysis has provided some ideas on where we should go next. If it has provided any directions on the Chinese economy is still another issue, but we can’t help noticing that any of our specifications point to China being in some kind of cyclical downturn, maybe even a recession. Maybe it’s time again for Xi to flex some muscles? Or, even worse, maybe he already has?