So after my Bloomberg View piece came out citing a self generated statistic that Chinese household debt to household income was above 100%, I had a number of eagle eyed reader send me a piece from the South China Morning Post from the same day. In the SCMP piece, they present a graph that shows Chinese household debt to household disposable income at just above 50%. Readers were wondering how could I explain the enormous discrepancy between my self generated number and the number that was cited in the SCMP.
This worry about household debt levels in China and the most common mistake is that people use per capita GDP rather than household income. For numerous reasons, there are enormous differences between per capita GDP and actual household income numbers. Even this recent SCMP piece about the rapidly rising household GDP number mistakenly uses household debt to GDP rather than household income.
Before I explain the discrepancy, let me stress, I personally am quite accepting of differences in how to interpret the data and whether additional data changes our view. However, especially when focusing on China, presenting the most accurate data and knowing what it does and does not say, is something I take very seriously. So I was also personally intrigued by the discrepancy.
I cannot say with 100% accuracy how the SCMP figure was generated but I can come quite close. The first data source cited is the Bank for International Settlements which generates a dataset with a figure for market value of household debt as a percentage of GDP. Though it does not specifically say, I would assume that GDP here is nominal.
There are a couple of points worth mentioning about this statistic. First, the BIS figure on household debt as a percentage of GDP does not perfectly match the figure in the SCMP but it matches within at most 10%. The BIS lists Chinese household debt as a percentage of GDP at 44.4%. The SCMP figure appears to be just a little bit above 50% and does not have a data label so I cannot say for certain. However, later in the article the writer claims that Chinese “household debt-to-GDP ratio is only 40 per cent” even though the BIS places it at 44.4%. Later the writes claims that Chinese “household debt-to-disposable income is 56 per cent” though again it is not entirely clear how this figure is arrived at.
What makes the authors figures even more suspect is the transformation into “household debt to disposable income by country” that he cites. If we follow the sources used by the author, we are able to locate within the UN National Accounts data a gross household disposable income number which would appear to represent the number used by the author.
This is where the author appears to get the cited statistic and take amazing statistical liberties. The UN data indicates that in 2013 (the last available year in the UN data set) China had 35.7 trillion RMB of gross disposable household income (more about this specific number later). At the end of 2013, Chinese households had 19.7 trillion RMB of household debt. If we divide 19.7 trillion by 35.7 trillion we get a number of 55.1% which is very very close to the statistic used of 56%.
However, this number is grossly and intentionally misleading. The author never prominently notes that the data used on China, his primary subject, is from 2013. He only notes in the last note of the figure that “the rest are as of 2013”. The author is writing about second half 2017 discussing current economic situation and never prominently mentions that the data he is basing his argument on is nearly 4 years old? The authors intention was clearly to mislead readers rather than educate them as to what best available data tell us right now.
In fact, we have best available data right for the year ending 2016. If we take the PBOC data on Loans to Households we get a total of 33.4 trillion RMB in debt outstanding at the end of 2016 which is for all intents and purposes statistically identical to the BIS figure of 32.95 trillion. Now what we need to do is find recent data on the amount of disposable household income in China. According to the National Bureau of Statistics China, per capita disposable income in China in 2016 was 23,821 RMB. With an official 2016 population of 1.38 trillion, this gives us a total disposable income of 32.9 trillion RMB. Next we take the total PBOC household debt number of 33.4 trillion and divide by the NBS number of total household income to arrive at a household debt to disposable income number of 101%. If we extrapolate out through the first half based upon the rate of growth in disposable income through H1 and use the June 2017 household debt, this number comes in around 104-105%.
What is interesting is that even if we take the official Chinese data used to calculate household debt to household income ratio back in 2013, we get 79.7% not the 55.1%/56% number used by the author. So where did the SCMP and the author go wrong?
In addition to the misleading date, the author confuses a measure of GDP for household income. The author uses a measure of household income with GDP measures that is based upon the estimated value of household consumption within GDP. The reason this matters is that the NBS compiles other data on household income that shows relatively different numbers. So far, I have been unable to locate the exact “gross disposable income” number in Chinese data that seems to be used within UN data. This is used primarily in a form of GDP accounting that is not widely recognized from the expenditure approach. I have however, been able to match the consumption number the UN uses to the NBS consumption expenditure within GDP data. This
The NBS however, compiles survey data where they actually go out and conduct surveys on rural and household incomes rather than compiling it at a GDP level. The UN data on gross disposable income collected via GDP overstates household income by roughly 43% according to the NBS survey data. What is important is that this measure of income actually compiles data on income from all sources such as wages and salaries, transfers, and income from business and property. Similarly the same data also compiles detailed data on the expenditure side with significant detail by category. This does not match identically but close enough the highly regarded China Household Finance Survey conducted by the Southwester University of Finance and Economics that we can take this survey data as much closer to reality than the 1993 methodology using headline GDP data from 2013.
The fundamental problem is that the author uses headline GDP data for household income rather than that survey data on what households actually make. It should be noted though that the use of 2013 data is misleading. In both fundamental data errors, there is significant laziness when significantly better quality and newer data sources exist. The household debt levels for Chinese households is above 100% of household income.