The GDP of Bridges to Nowhere
On January 18, 2018, China’s National Bureau of Statistics announced that the country’s GDP grew by 6.9 percent in 2017:
According to the preliminary estimation, the gross domestic product (GDP) of China was 82,712.2 billion yuan in 2017, an increase of 6.9 percent at constant price compared with last year. Specifically, the year-on-year growth of GDP for the first quarter was 6.9 percent, 6.9 percent for the second quarter, 6.8 percent for the third quarter, and 6.8 percent for the fourth quarter.
According to preliminary statistics, the aggregate financing to the real economy (AFRE) . . . was RMB 19.44 trillion in 2017 . . . Specifically, RMB loans to real economy registered an increase of RMB 13.84 trillion . . . foreign currency-denominated loans (RMB equivalent) . . . recorded an increase of RMB 1.8 billion . . . entrusted loans registered an increase of RMB 777 billion . . . trust loans registered an increase of RMB 2.26 trillion . . . undiscounted bankers’ acceptances recorded an increase of RMB 536.4 billion . . . net financing of corporate bonds stood at RMB 449.5 billion . . . equity financing on the domestic stock market by non-financial enterprises registered RMB 873.4 billion . . .
For those keeping track, the table below lists the relevant GDP and TSF data for the past four years:
|Table 1. China’s GDP and TSF Figures, 2014–2017|
|GDP (trillions)||63.65 RMB||67.67 RMB||74.41 RMB||87.71 RMB|
|Nominal GDP growth||–||6.3%||10.0%||11.2%|
|TSF (trillions)||16.46 RMB||15.41 RMB||17.80 RMB||19.44 RMB|
|Nominal TSF growth||–||-6.4%||15.5%||9.2%|
|Sources: National Bureau of Statistics of China; ChinaInternetWatch.com|
I was recently part of a discussion on a listserv that brings together Chinese and foreign experts to exchange views on China-related topics. What set off this discussion was a claim that the Chinese economy began to take deleveraging seriously in 2017. Everyone agreed that debt in China is still growing far too quickly relative to the country’s debt-servicing capacity, but the pace of credit growth seems to have declined in 2017, even as real GDP growth held steady and, more importantly, nominal GDP growth increased.
I was far more skeptical than some others about how to interpret this data. It is not just the quality of data collection that worries me, but, more importantly, the prevalence in China of systemic biases in the way the data is collected. Not all debt is included in TSF figures. The table above, for example, indicates a fall in TSF in 2015, but this did not occur because China’s outstanding credit declined.
It occurred rather because in 2015 there was a series of debt transactions (mainly provincial bond swaps aimed at reducing debt-servicing costs and extending maturities) that extinguished debt that had been included in the TSF category and replaced it with debt not included in TSF. The numbers are large. According to the China Daily, there were 3.2 trillion renminbi worth of bond swaps in 2015, plus an additional 600 billion renminbi of new bonds issued. If we adjust TSF by adding these back, rather than indicate a decline of 6.4 percent, we would have recorded an increase of 15.7 percent.
Credit Growth Can Embed Systemic Biases
Most analysts were already aware of the impact of provincial bond swaps on TSF and duly adjusted their data; but this suggests a wider problem, which is what I wrote about in my first contribution to the listserv discussion. I proposed two reasons why I am not convinced that observers have seen the beginning of any meaningful deleveraging.
The first has to do with systemic biases in the way credit is structured and counted. Chinese bankers—like those in the rest of the world, no doubt—have always gamed regulatory constraints when it comes to credit creation. Like bankers everywhere, they respond to institutional incentives by altering the ways in which they structure credit creation. These changes often have been driven as much by Chinese bankers’ need to please a varied group of regulators—whose own institutional biases are exacerbated by the competition, and even hostility, that exists among them—as by economic and financial factors.
That is why it is wise to be especially skeptical about recent evidence that Chinese banks have begun to take deleveraging seriously. Beijing has been worried about China’s growing debt burden since at least 2012. But in 2017, this issue has become almost an obsession in some quarters. Beijing has made a series of aggressive announcements in the past year testifying to this surge in concern, culminating in an October 2017 statement by PBoC Governor Zhou Xiaochuan, who warned that China could face its own “Minsky moment.”
Under these circumstances, then, it was always to be expected that banks would take greater pains than ever either to rein in credit growth or, if that was too difficult, to structure credit in ways that seemed to comply with regulatory concerns. One way of doing so is to push credit creation off balance sheets and into forms that are less likely to trigger regulatory reprisals. That Chinese bankers might be doing so is confirmed by both anecdotal and official evidence indicating faster-than-expected credit growth in categories that fall outside widely watched measures like TSF. The point is that the deceleration in credit growth implied by TSF data might indeed reflect the beginning of Chinese deleveraging, but it could also reflect the surge in regulatory concern. In the latter case, this would mean that China has experienced not the beginnings of deleveraging, but rather a continuation of the trans-leveraging observers have seen before. I will discuss some of the most obvious examples of how this may be occurring later in this essay.
GDP Growth No Proxy for Growth in Debt-Servicing Capacity
My second reason for skepticism is one I have written about elsewhere, in both a November 2017 Financial Times article and a September 2017 blog entry. When one compares credit growth to growth in debt-servicing capacity, not only is it uncertain how quickly credit is growing in China but, more importantly, it is even less certain how quickly the country’s debt-servicing capacity is growing.
The standard proxy for growth in debt-servicing capacity is GDP growth, but this is only valid in economies in which GDP growth data is a systems output that measures the underlying performance of the economy. But this is not true of China, as I explained in the Financial Times article:
Typically, analysts assume that changes in reported GDP reflect movements in living standards and productive capacity. In China, however, this is not the case. Local governments are expected to boost spending by whatever amount is needed to meet the country’s targets, whether or not it is productive. [In China] GDP growth is not the same as economic growth.
In my email, I went on to discuss why this matters so much and why it is incorrect to think of China’s GDP growth as growth in China’s underlying economy (or in its debt-servicing capacity, or its productive capacity, or however else one prefers to think of GDP). GDP growth measures the increase in economic activity, whether that activity is building bridges to Brooklyn or bridges to nowhere. It does not directly measure the increase in economic value-creation, as observers usually assume it does. It is only because in most economies there are constraints on the ability to fund nonproductive activity that observers generally ignore the difference between economic activity and value creation.
Because of these constraints, most economic activities create value and those that do not are reversed in fairly short order. If these constraints are not in place, however, analysts can no longer ignore this difference because the economy can then engage in nonproductive activity that for many years can force up the debt burden and add to GDP. While the purpose of this activity may be to generate employment, if the activity is not productive, the GDP it creates—along with the employment it generates—will be reversed in the future once debt capacity has been reached.
Not everyone on the listserv fully understood this argument about Chinese deleveraging. Because I received a lot of responses, both negative and positive, I decided to follow up with a second email, in which I use an approach very different from any I have used before to try to show that this is just a logical argument.
I thought it might be useful if I were to reproduce on my blog the two responses that I wrote, although only after eliminating any reference to participants in the discussion (this is a private listserv), editing, and extending for clarity. Here are the two emails:
The First Email: On Credit Growth
Actually, I think the claim that credit growth has slowed is more complicated than what you suggest. The growth in TSF, plus most standard adjustments, seems indeed to have decelerated in 2017 relative to 2016, and this is especially impressive when one compares the growth in credit with rising nominal GDP growth (which is the only right way to do it). But as I see it, there are at least two problems, besides obvious timing mismatches, with treating this as a meaningful deceleration in credit growth.
First, Chinese banks and financial institutions have always been very creative in terms of extending credit in ways that get around regulatory constraints. There is no reason to assume that the enormous amount of attention paid to credit growth recently has had no impact on this kind of behavior. For example, Andrew Collier, of Orient Capital, has been particularly thorough in attempting to understand these kinds of transactions. In a January 2018 report, he discusses the recent surge in lending that has occurred under China’s public-private partnership (PPP) program. He sets the number of completed PPP projects at 3.2 trillion renminbi, with a total of over 14 trillion renminbi in projects planned or under evaluation.
Because most of these transactions are structured in ways that help banks keep them off their balance sheets, the consensus among analysts with whom I have discussed PPP is that a lot of these transactions are not showing up in TSF or in any of the other standard measures of credit creation. What is more, these PPP transactions are often structured in ways that may create additional contractual obligations to fill the cash-flow gap for user-pay ventures, and these obligations do not show up at all in the debt statistics. Already, as Reuters has reported, “China’s finance ministry . . . has grown increasingly concerned about rising hidden debt risks from potential abuses of the program.”
The recent explosion in debt securitizations, as another example, also seems to reflect mainly the need to structure loans in ways that do not appear in the main measures of credit creation. The numbers here are large too. A January 2008 Orient Capital research note reports the following:
Developers have learned how to reduce their official borrowing by offloading balance sheet debt to consumers through securitization, as a way to evade borrowing restrictions. For example, in December 2017, China Merchants Group, a state-owned firm, said it would work with China Construction Bank to issue asset-backed notes worth Rmb20 billion to be sold in the interbank market, in what would be China’s largest rental housing securitization. These notes are not included in TSF.
To take yet a third example, in 2017 at least 1 trillion renminbi of debt was converted into equity by the banks that had extended the loans. While this reduces the reported amount of outstanding debt, if the concern is the ability of borrowers to generate the returns needed to service the debt that funded these projects, converting them into equity does not reduce the riskiness of the banking system, nor does it reduce net indebtedness for the country overall. Most, if not all, of these transactions, create unreported contingent liabilities even while reducing the growth rate of debt—in this case by more than five percentage points.
These are pretty big numbers, and there are likely to be many similar examples, many of which will not be recognized for months or even years. The problem is that the banks are in a difficult position. The recent surge in regulatory demands to rein in various types of credit creation conflicts with the pressure on banks from local authorities to fund even more nonproductive economic activity. There are only four ways for the financial system to respond to these conflicting pressures, of which the recent data seems to support only the last two:
- Banks can rein in credit growth to please the regulators and anger local authorities by allowing GDP growth to decline sharply.
- Banks can maintain high GDP growth rates to please local authorities and anger regulators by allowing credit to surge.
- Banks can maintain high GDP growth rates to please local authorities but, while allowing credit to surge, they structure new credit in ways that disguise credit growth.
- Finally, banks can radically transform the ways they issue credit so that it is possible to keep GDP growth rates high even as the real debt burden declines.
Clearly, the last of these four is the optimal response, but Beijing has been trying unsuccessfully to do this for nearly ten years, and neither the arithmetic nor the historical precedents give observers much hope that this can be done to any serious extent without a radical transformation of the country’s development model. That leaves analysts with two possibilities: either credit growth has been disguised, or there has been the necessary transformation in the capital allocation process just in time for the 19th Party Congress. A year ago, a very cynical professor at Peking University promised me that now that Beijing had become so concerned about debt, there would inevitably be an improvement in the measures of credit creation, whether or not there were any real changes in the pace of credit growth: this seems always to be a safe prediction.
The First Email: On GDP Growth
There is a second important problem, as I see it, with treating the recent data as indicating a meaningful deceleration in credit growth. The proper way to measure the improvement in China's debt burden is to measure the relationship between credit creation and debt-servicing capacity. This requires that observers have not only an appropriate measure of new credit in each period but also an appropriate measure of the growth in debt-servicing capacity.
In China, this is an even bigger problem than quantifying the growth in credit. For most economies, analysts typically use GDP as a proxy for debt-servicing capacity, but while this is more or less appropriate in most economies (in spite of a barrage of criticisms recently in the Financial Times and elsewhere about the usefulness of GDP), measures of GDP growth in most economies are at least systems outputs, which means that they can serve as proxies, however inexact, for activity in their respective economic systems.
In China, however, GDP growth is a systems input. This means analysts cannot treat it in the same way. An input measure cannot tell observers how a system is performing—only an output measure can.
If the authorities are willing to engage in loss-making activities to achieve the GDP growth target, there are two relevant characteristics of an economy like China’s that change the nature of the GDP measure: first, economic activity is much less affected by hard-budget constraints than it is in most other economies; and second, bad debt is much less likely to be written down. Government officials everywhere, and not just in China, would probably be happy to engage in loss-making activities to achieve higher current GDP growth rates and lower current unemployment rates, even though these benefits are only temporary and must be reversed in the future. But they can only do so within the limits of the budget and debt-capacity constraints under which they operate.
Relaxing these limits is what allows them to achieve higher GDP growth rates than what underlying economic growth would dictate. GDP is, according to one widely accepted definition, “the sum of gross value added by all resident producers in the economy,” but it is not always obvious how to determine this market value. Consider two factories that cost the same to build and operate. If the first factory produces useful goods, and the second produces unwanted ones that pile up as inventory, only the first boost the underlying economy. While inventory piles up, however, both factories will increase GDP in exactly the same way.
In most economies, the operator of the second factory is subject to hard budget constraints and must eventually close the factory. Once this happens, the factory stops producing GDP. In addition, if the facilities are not reassigned to other productive uses, they must be written down, along with the worthless inventory. This process reverses the GDP formerly created by the second factory because it reduces the profits of the entity that owns it (or that lent money to fund it as the loan is written down). Business profits are included in the value-added component of GDP when GDP is calculated, so because of the subsequent losses, the second factory does not add to GDP except over very short periods, after which it is reversed.
This is why countries like China, whose economies are not subject to these two constraints, are able to achieve GDP growth targets that for many years exceed the underlying growth of the economy. The simplest way to think about it, I think, is that if one wants a number that means what GDP growth means in most other economies, China’s reported GDP growth is only a starting point. It must be adjusted by some other relevant systems output number. My best guess is that one should start with reported GDP growth and subtract from it your best estimate of the amount of debt in each period that should be written down to zero.
Because the amount of bad debt in each period is almost certainly a growing number, it must follow logically that the GDP growth number observers really want, rather than the one they have—that is, GDP growth as a systems output that can serve as a proxy for debt-servicing capacity—is a declining number, and perhaps even a quickly declining number.
I personally suspect that China’s actual GDP growth is less than 3 percent, although of course, I cannot prove it. What I can prove, and have proved if you agree with my assumptions that Chinese economic activity is much less limited by hard-budget constraints than other economies and that debt is much less likely to be written down correctly, is that it is definitely not 6.9 percent.
The Second Email: Bookshop Targets
[In one of the responses to my first email, someone agreed that it was “short-sighted” to place too much faith in the validity of GDP.]
I would say it is even more than short-sighted to assume that because they share the same name, GDP as a systems input can be compared to GDP as a systems output. It is literally wrong. Because this seems confusing in the macroeconomic context, I thought maybe an example outside macroeconomics might be helpful.
Suppose one wanted to measure urban literacy as a function of a city’s size, or of the average income of its residents. One way to do so is to look at dozens, or perhaps hundreds, of cities in terms of population, or average income, and compare them with the number of bookshops in each city.
In that case, the number of bookshops, which is a systems output—generated organically by the size of the city, the rate of literacy, the income of residents, and other relevant factors— would serve as a proxy for literacy, while population or income would serve as proxies for whatever variable one wants to measure. Like with GDP, there are a whole set of problems with using the number of bookshops as a proxy for literacy, but it is a reasonable proxy, and perhaps in our case, it is the best measure that can be found. What is more, because one can reasonably assume that the problems with using bookshops as proxies for urban literacy are not systematically biased, these problems can be addressed and partially resolved by using a large sample base.
But now assume that the government passes a literacy law that requires that every city must have exactly one bookshop for every 10,000 residents—no more, no less—and there is a government agency whose purpose is to make sure that for every city there is the number of bookshops required by law. To that end, all bookshops become government-owned and operated, without hard budget constraints.
Clearly, the number of bookshops in each city has now become an input to the system of literacy and is no longer an output generated by the organic growth of the city and the underlying need for bookshops. This immediately renders the number of bookshops useless as a proxy for literacy, unless one can find another output measure that can be used to adjust the number of bookshops, perhaps the number of titles on offer or revenue per shop. Without some kind of adjustment, however, anyone who used the number of bookshops as a proxy for literacy would find his work widely ridiculed.
The Second Email: GDP Targets
This is effectively what happens in any system when an input variable is treated as if it were an output. It is not a perfect analogy but—except, of course, for the part in which analyses that use the number of bookshops as a proxy for literacy are widely ridiculed—it is nonetheless similar to what happens when the health of the Chinese economy is measured by the reported GDP data, or when second-order measures, such as the dependence of Chinese growth on debt, is estimated by looking at credit growth in relation to GDP growth.
By the way, while this has always been a conceptual problem with the use of GDP in China, it has not always been a practical problem. When China was severely underinvested, which was the case in the 1980s and 1990s, GDP growth could be accepted as a reasonable measure of the real growth in the underlying economy. This is because even though budgets than were no more constrained than they are now, one could assume that the distortions they introduced were quite minimal. When China was underinvested, investments were nearly always productive, and so the ability to ignore budget constraints and hide the costs of nonproductive investment in the form of rising debt had little effect on the GDP data—or, to put it differently, rising debt did not reflect a rising debt burden.
But that is no longer the case. Because debt is now rising faster than debt-servicing capacity, one can think of the gap between the two as the capitalizing of what should be an expense—nonproductive spending—the result of which is higher reported growth and higher reported wealth: GDP as input in each period has become the sum of GDP as output and the number of expenses that have been incorrectly capitalized.
In sum, three assumptions are all that is needed to effect this transformation of GDP. First, assume that a significant share of economic growth is generated by entities that, were they subject to hard-budget constraints, would have been closed down because the economic benefits they create are less than the cost of the investment (assume all externalities are part of the revenues generated by the system). Second, assume that the bad debt generated by the system (by which I mean the excess portion of any debt used to fund projects that add less value to the economy than the cost of the project) is not written down within the reporting period in which it was extended. And third, assuming that China continues to have as much debt capacity as needed in the current period to fund the amount of activity required to meet the GDP growth target.
But once those three assumptions reach their limits, as they eventually must, the excess of reported GDP growth over the GDP growth that would have been reported had these conditions not existed will be amortized in the form of a negative excess. Reported GDP in each period will still be the sum of GDP as output and the number of expenses that have been capitalized, but now that this capitalized amount is being amortized, it will become a negative number.
When that happens, GDP growth will drop sharply and will even begin to understate the real growth in debt-servicing capacity. Until then, the conclusion that GDP in China does not mean what it is used to mean elsewhere is just a matter of logic. It is not a hypothesis or a theory.
 According to a March 2017 South China Morning Post article, there were a total of 8 trillion renminbi in provincial debt-for-bond swaps and an unspecified amount of additional bond issues under the program, so these numbers must be added back to TSF not just for 2015 but also for 2016 and perhaps 2017. Because the newly issued bonds tended to have longer maturities than the underlying loans, it is reasonable to assume none of them were paid down in 2016 and 2017.
 The TSF measure itself reflects this behavior. The PBoC created the TSF measure in 2011, largely because Chinese bankers had been de-emphasizing renminbi-denominated bank loans, the main measure regulators had previously used to monitor credit expansion, in favor of other forms of credit that were not so carefully monitored. From 2002 until 2009, renminbi-denominated bank loans fell from over 90 percent of all the categories later included in TSF to just 55 percent (I am quoting from memory, so please double-check the data). Credit in TSF categories other than renminbi loans had been growing three or four times faster, in other words, than were renminbi loans. There were, no doubt, many reasons for their relatively rapid growth, but the fact that they were not being monitored to the same extent was among the most important, at least according to informal discussions with Chinese bankers.
 It may seem willfully perverse to most analysts to suggest that a debt-equity swap does not reduce debt, but that is because most analysts do not think systemically and fail to consider the overall impact of these transactions on debt-servicing costs and on contingent liabilities of the government. Debt-equity swaps remove contractual obligations from the borrower’s balance sheet and replace them with equity in the borrower. If done correctly, with an eye not to achieving political or regulatory objectives but rather to eliminating financial distress costs, these can improve the enterprise value of the borrower; to the extent that the lender participates in the upside (and if the performances of the various equity positions emerging from these swaps are uncorrelated), the lender’s net asset position can also improve. But the banks themselves must fund their debt and equity positions with debt, mostly in the form of deposits, interbank liabilities, and bonds. Assuming that the total amount of bad debt in the banking system exceeds total bank capital—something which is almost certainly true—the conversion of debt which cannot be serviced into an equity position that is unlikely to generate much more (and in an economic downturn, which is when we are most concerned about the debt burden, we can assume that the decline in value of these equity positions will be highly correlated) leaves the net indebtedness of the banking system unchanged, and so the contingent liabilities of the government are unchanged even as reported debt in the system declines.
 There are two very different ways of thinking about credit growth in China, and these differing perspectives determine the way an analyst is likely to approach and interpret the data. Mainstream economics has real difficulty incorporating credit growth into its models, and many analysts think that the explosive growth in credit is not fundamental to China’s rapid GDP growth. In that case, it is possible that, with the right set of productivity-enhancing reforms, credit growth begins to decline sharply while nominal GDP growth remains constant or even rises.
The balance-sheet model that I use ties an acceleration in credit with any attempt to keep GDP growth above what I have elsewhere called, following the lead of the Deep Throat blog, “productive growth.” According to this model, a small amount of deceleration in credit growth can occur as additional credit efficiency is squeezed out of the system, but without a sharp decline in GDP growth, substantial and sustainable credit growth deceleration cannot occur except after a major transformation of China’s growth model, one condition of which is net wealth transfers from local governments to median households of at least one to two percentage points of GDP annually. As GDP growth remains high, and as there is no evidence of substantial wealth transfers, my default reaction to data that suggests a rapid deceleration of credit growth is skepticism.
About Michael Pettis
Michael Pettis Archive
|09/06/2018||Beijing’s Three Options: Unemployment, Debt, or Wealth Transfers||story|
|08/09/2018||The U.S. Trade Deficit Isn’t Caused by Low American Savings||story|
|07/10/2018||Tariffs and Trade Intervention||story|
|04/04/2018||High Wages Versus High Savings in a Globalized World||story|
|01/18/2018||What Would Happen If China Stopped Buying U.S. Treasury Bonds?||story|
|04/20/2017||China Update - Why Is the Consensus Now So Optimistic?||story|
|04/03/2017||Mexico’s Positive Impact on the US Trade Balance||story|
|02/02/2017||Is Peter Navarro Wrong on Trade?||story|
|01/09/2017||Three Possible Scenarios for China||story|
|01/03/2017||A US Retreat on Global Trade Will Not Lead to a Shift in Power||story|