A revised, and longer, list of indicators. In particular, we have expanded the sections on real estate and government finance, given these sectors’ importance to China’s macroeconomic outlook. Unfortunately, however, not all the changes are additions -- the authorities have ceased publication of some series that we found useful in the past.

Numerous additional charts and tables to summarize key data and display time series.

Further detail on the growing collection of proprietary indicators we have developed over the years. While our colleagues around the world have also developed proprietary indicators, and in many cases (e.g., the CAI and FCI) we apply those techniques to China, we have also developed many China-specific indicators.

Year-over-year versus “sequential” growth. We use the term “sequential” to describe period-on-period changes within a year (e.g., month-over-month or quarter-over-quarter, depending on the series). The Chinese government typically reports year-over-year series as a way to minimize seasonal influences. But year-over-year data can mask significant changes in sequential momentum, so we often calculate and refer to sequential figures. (Note that the term “base effect” refers to a particularly high or low sequential change from one year ago that affects the year-over-year calculation. For example, if GDP normally grows at a 4% annual rate but temporarily stalls at 0% quarter-over-quarter growth for one quarter, growth will be reported at 3% year-over-year in that quarter and the following three quarters, then one year later will jump back to 4% yoy due to the “base effect” as the weak quarter drops out of the calculation.)

Seasonal adjustment. This is critical when working with sequential data (more on this topic below).

Revisions. As in other countries, initial data reports may be revised as more comprehensive information becomes available or methodological revisions are made. The corollary is that the current data series do not necessarily reflect how the historical data looked at the time of release.

Survey versus “hard” data. Government agencies typically report samples or censuses of actual economic activity, which we sometimes refer to as “hard” data. In addition, a variety of government and private sector surveys (sometimes referred to as “soft” indicators) can give a useful qualitative sense of the direction of the economy. Often, these take the form of “diffusion indices” where respondents answer questions with either favorable, neutral or unfavorable responses; the percentage answering favorably plus half the percentage answering neutral are added to yield a score from 0 to 100. The widely quoted Purchasing Managers’ Indices (PMIs) take this form. While less precise than hard data and potentially subject to other biases (such as inflation), these reports can provide a timelier read on changes in direction and therefore are a useful reference for forecasting and policymaking.

First is the relevance score, which is based on the historical correlation with real GDP growth (quarter-over-quarter). This score can range from 0 (irrelevant) to 5 (most relevant), as illustrated in Exhibit 3. Given the central role of the GDP series in the MAP framework, the fact that China’s GDP statistics have tended to be relatively smooth historically (at least pre-pandemic) may reduce the number of significant indicators.

Second is the surprise score, measured as the difference between a particular release of that indicator and the Bloomberg “consensus” forecast for that indicator, measured in standard deviations. We assign a score from -5 to +5 depending on whether the actual figure is above or below expectations, and by how much. If the actual release is less than half a standard deviation from the consensus expectation, we will assign a score of 0. A difference of between 0.5 and 1 standard deviation will generate a surprise score of +1 or -1. Surprise scores rise with ½ standard deviation increments, with any surprise of greater than 2.5 standard deviations generating a score of 5.

Multiplying the relevance score by the surprise score gives a range of -25 to +25 for a given indicator; the aggregate of time series of MAP scores for those indicators included for China creates the China MAP score.

Bloomberg consensus forecasts are generally made using data from sell-side economists, most of whom work at global financial institutions. As a result, they may not precisely reflect the expectations of the broader investor community. In addition, these forecasts are often released in publications many days in advance of the official data, and subsequent changes in forecasters’ views may not always be updated in the published consensus.

There was a large shock during the sample period due to the GFC, and an even larger one during the Covid pandemic. Relative to the magnitude of the surprises during that period of time, any surprises today tend to appear small but are nevertheless significant for the market. Therefore, in setting our surprise score thresholds we have used the standard deviation of surprises based on data releases since 2010 (Exhibit 4).

“Trade Balance Bounces into the Year of the Rabbit”, US Daily, 6 April 2011

“A redesigned MAP of emerging Asia data”, Asia Economics Analyst, 10 May 2013

“A Better Global Economic MAP”, Global Economics Weekly 13/39, 5 December 2013

“Separating cyclical signal from seasonal noise”, Asia Economics Analyst, 27 June 2014

“Chinese New Year Seasonal Distortions Coming Home to Roost”, US Daily, 15 February 2016

“Revisiting post-reopening seasonal adjustment”, China Data Insight, 23 June 2023

China’s GDP data are mostly compiled in accordance with the System of National Accounts (SNA) 2008 standard by treating R&D expenditure as part of capital formation. The data are historically and internationally comparable.

Historically, the real GDP growth data were exceptionally smooth relative to other countries and to other indicators of activity, especially during economic downturns, contributing to skepticism among market participants over their accuracy. It was common for China to announce quarterly GDP growth with a variation of less than 0.5 pp before the Covid pandemic, especially during 2015-16, even as some high-frequency indicators occasionally experienced double-digit swings in growth. To get a better sense of cyclical momentum, we cross-check real GDP growth with other indicators including our proprietary Current Activity Indicator. During the Covid pandemic in 2020-22, the volatility in the real GDP growth data increased considerably due to periodic shifts in Covid-related restrictions.

GDP revisions are supposed to capture the newly available data in the whole economy, though in practice they are most relevant for tertiary industry (services). Measuring services-sector activity is inherently more difficult, and the Chinese statistical system – which grew out of the Soviet system that did not recognize services as value added – is particularly ill-equipped to do so. The shift from SNA 1993 to SNA 2008 in 2016 increased 2015 total GDP by 1.3% to US$11 trillion, and the real growth rate was also revised up slightly. The upward revision was due to the fact that China’s R&D expenditure growth has been consistently faster than that of overall GDP. In November 2019, the NBS revised up its 2018 GDP by 2.1% which made the goal of “doubling real GDP between 2010 and 2020” easier to reach.

GDP by industry

GDP by expenditure

GDP by income

Primary industry: Farming, forestry, animal husbandry and fishing (excluding related services).

Secondary industry: “Industry” (including mining, manufacturing, and utilities) and construction.

Tertiary industry: Any industry other than the primary and secondary industries. Activities in the services sector (such as wholesale and retail trade, finance, catering, and transportation) are captured in this category. As in many other countries, this sector has been growing as a share of overall economic activity in China.

All goods and services that consumers have purchased (with the exception of houses and housing construction materials) are included in household consumption (houses and housing construction-related items are mostly treated as gross fixed capital formation). Although retail sales data are often used as an indicator for household consumption, there are crucial differences between the two (see Section V. Consumption).

Government consumption measures the non-investment goods and services purchased by government.

The sum of household and government consumption equals final/total consumption.

Gross Fixed Capital Formation (GFCF) is a key component in the expenditure approach of national accounts reporting. It is different from fixed asset investment (FAI) because the latter includes spending on assets that do not directly contribute to GDP (e.g., purchases of land and used equipment), while GFCF does not.^[2] Other differences include a minimum project size cutoff for FAI, and property developer profits and IT investment being covered in GFCF but not FAI. Moreover, FAI data collection is susceptible to a lot of statistical noise (see Section III. Investment).

Changes in inventories refer to net changes during the observed period. China’s changes in inventories data are only available at annual frequency, and appear to be useful in assessing the direction, but not necessarily the degree, of inventory adjustments. Despite their importance, changes in inventories are generally the least reliable component of GDP by expenditure in China, as well as in many other countries, due to difficulties in data collection. Some countries estimate this as a residual item by balancing GDP by expenditure data and GDP by industry data. The NBS is supposed to make independent estimates for changes in inventories based on a wide range of data sources, though in practice the level of volatility and the gap between GDP by expenditure and GDP by industry suggest it may have an element of residual as well. To better track the contribution of inventory changes to GDP growth at a quarterly frequency, we have built a proprietary inventory tracker based on commodity inventory measures, PMI inventory sub-indices, and industrial final goods inventory and auto inventory data.

The sum of GFCF and the change in inventories is defined as Gross Capital Formation (GCF, or just “investment”, which remains a very large share of economic activity in China).

Net exports are the balance of trade in goods and services, which is equivalent to the balance of payments (BOP) definition for the trade balance. It differs from Customs trade data in three main aspects: (1) it includes trade in services, whereas the Customs data only cover trade in goods; (2) conceptually, it is based on the principle of exchanges between residents and non-residents, instead of goods moving across national frontiers; and (3) imports reported by Customs are based on CIF, while BOP/trade under GDP is based on FOB (see Section VI. External Sector).

GDP by expenditure data are only available annually in nominal levels. One needs to deflate official data to obtain real levels and growth rates. For example, the Consumer Price Index (CPI) can be used to deflate household and government consumption, the Producer Price Index (PPI) can be used to deflate GFCF, and goods trade price indices can be used to deflate goods trade. Since 2015, the NBS also publishes percentage point contributions of consumption, investment, and net exports to headline year-over-year real GDP growth which can be used to estimate the year-over-year growth in real consumption, real investment, and real net exports.

The income approach accounts for the income generated during the process of production activities for various industries. The value-added GDP data are composed of four parts: employees’ compensation, net taxes on production, depreciation of fixed assets and operating surplus.

The NBS does not release national aggregates for GDP by income, only for individual provinces. Users need to aggregate the provincial GDP-by-income data to get a national figure. However, releases of such provincial-level data are uneven and lagged. For example, for many provinces, 2017 data are the latest available as of this writing in 2024.

Historically the discrepancies between national and regional GDP data used to be significant, but narrowed in recent years. The sum of provincial GDP tends to be higher than the national GDP, and the aggregate of municipal GDP tends to exhibit the same pattern vs. the provincial GDP. A number of factors lead to these discrepancies. Among the most important are: (1) double-counting of the value added for enterprises that operate across different regions; (2) differences in data sources; and (3) human factors in compilation. The revised national GDP after the census is much closer to the aggregate of provincial GDP data before the revision. Although it has long been assumed that regional GDP data have overstated the growth and size of the economy, the revisions made by the NBS subsequently indicate early readings of national data may be too conservative as they tend to under report service-sector activities relative to the census, especially for emerging industries.

Another problem is the estimation of regional GDP per capita in China, especially at the city level, due to China’s large cross-region migrant population. As there are two regional population data series in China, those of actual residents and those with household registration (Hukou) in the reporting location, GDP per capita calculated using the two series can differ substantially, especially for cities with more migrant workers. In 2004, the NBS noted that for cities with large migrant populations, such as Shenzhen where the registered population is significantly smaller than the residential population, using the registered population may overstate GDP per capita. Therefore, it is important to use the actual resident population series to avoid serious over-estimation of GDP per capita, though even that series suffers from the problem of under-reporting. For example, someone who stays less than six months in a city is technically not a resident according to the statistical standards, but nevertheless contributes to local activity. Moreover, city-level GDP per capita data are not always reported on a timely basis.

Given population estimates reported are year-end data, to capture the "flow" feature of GDP for the whole year, the NBS will average the year-end residential population when calculating GDP per capita. For example, average 2020 population= (2020 year-end + 2019 year-end)/2.

Another concept closely related to GDP is gross national income (GNI), also called gross national product (GNP). The difference between the GDP and GNI/GNP is net factor income, which is the income of investment and labor by domestic residents earned abroad minus those of foreign residents earned in the country. Despite their apparent similarities, these two series measure two different aspects of the economy: GDP measures production and GNI measures income. Note China's primary income deficit has widened over the past few years, mainly driven by a wider investment income deficit, although net international investment positions have shown steady growth.

We have long viewed industrial production (IP) as among the more reliable monthly activity indicators China publishes because: (1) related to the structure of the Chinese economy, China’s statistical system has focused on tracking growth in industrial production since it was founded; and (2) historically there seemed to be less “smoothing” in this series than in some other politically more sensitive data series, such as GDP. However, IP became unusually smooth during the 2015-16 downturn. The reliability of the IP series appears to have improved in recent years with its volatility increasing dramatically during the Covid pandemic and with sequential moves largely consistent with high-frequency data such as coal consumption and steel production.

The IP data series generally tends to be more important than fixed asset investment and retail sales data in tracking GDP growth because it is in real terms and because, by being in value-added terms, it is more in line with the GDP concept. The only difference between IP and manufacturing output is that IP includes the mining and utilities industries.

The sectoral coverage of IP is selective in the following respects:

1. It covers only the industrial sector, which includes “mining and quarrying, manufacturing, and utilities” — otherwise known as “secondary industry” by GDP classification, excluding construction. There are 41 industrial divisions in total, in which manufacturing accounts for the vast majority of the components. Value added in a particular industrial division is the sum of value added from companies whose primary activities are in that division (in practice, this may include some ancillary activities which should technically be categorized in other areas). This issue is especially tricky when it comes to conglomerates, and as the level of sector detail increases.

2. By business type, the value added of industries covers state-owned & holding enterprises, share-holding enterprises, private enterprises, and foreign, Hong Kong, Macau and Taiwan funded enterprises. The classification is based on the controlling stake. If the company is controlled by a minority shareholder, it will be classified as a company under that shareholder, as is often the case with state-controlled firms.

3. Minimum threshold: enterprises with annual sales of RMB20 million or above are included. Therefore, this series covers only a portion of the total value added by industry in the economy, though the portion is large (more than 80% in recent years). Given regular adjustments of the minimum threshold, the NBS releases comparable growth rates to eliminate statistical discrepancies in levels.

Export delivery value: Refers to the nominal value of industrial products for exports (including to Hong Kong, Macau and Taiwan). Prices are generally denominated in FX, and converted back to RMB based on the respective exchange rates. This category includes products from processing and assembling trades, etc. The export delivery data are different from customs trade mainly in two aspects. Firstly, Customs data includes all the merchandise goods that exit customs in a particular month while export delivery value only includes industrial goods produced for exports in a particular month. For example, Customs data includes exports from all kinds of enterprises and covers both industrial and non-industrial products (e.g., agricultural products, roughly 3% of total exports), while export delivery value only covers industrial products from industrial enterprises. Besides products sold to other countries, export delivery value also includes domestic sales of products which were initially planned for exports. Secondly, these two sets of data are different on treatment of processing trade. For processing trade with supplied materials, export delivery value only includes processing fees while Customs record the full value of the final product as export on a free-on-board (FOB) basis.

IP data are announced in terms of real single-month growth and real year-to-date growth. The reporting of the nominal level of value added by industry was discontinued in 2007 and official real level data have never been published. When nominal value-added series are needed one can “inflate” official real growth data by the PPI to generate a rough estimate, though this approach can yield results quite different from the official nominal series over the period when they were still published. This is due to technical issues such as whether the deflation is done first at the sector level and then aggregated, and/or whether value added is deflated with a single price index or output and intermediate input are deflated separately (“double deflation”, technically the preferable approach).

The most important sub-sectors of manufacturing are electronic equipment (including communications equipment, computers and other electronic equipment), transportation equipment (including auto manufacturing), smelting and pressing of ferrous/nonferrous metals, chemicals, electric machinery and equipment, and textiles.

The SIOI has a much shorter history than industrial production (IP), fixed asset investment (FAI) and retail sales. The NBS only releases the yoy growth data series for this index, while historical level data are not available (unlike IP, FAI and retail sales). As such, one needs to assume the monthly index levels for a specific base year, and back out level data for other years, to estimate the seasonally adjusted sequential growth.

Since the data series was introduced in late 2016, NBS has shown January-February SIOI growth combined, to smooth out distortions related to the shifting time of the Chinese New Year holiday.

The SIOI growth shares similar trends with retail sales growth in previous years. Their occasional divergences for some months may reflect things like growth differences between goods consumption and services consumption, growth differences between business services and consumer services, and major changes in price deflators.

The NBS also provides some sectoral breakdown for the SIOI growth, although historical level data are not available. These sub-sectors include wholesale & retail sales, transport, storage & post, hotel & restaurant, financial services, real estate services, IT and related services, leasing & commercial services, etc, with varying start month availability (between 2018 and 2020).

Electricity production: Refers to the power generated by industrial enterprises with annual revenue from principal business above RMB20 million.

Electricity consumption: Refers to the electricity consumption of the whole of society including the primary sector, industrial sector, tertiary sector and residents in urban and rural areas.

The signal to noise ratio of electricity production and consumption is relatively high — as the information collection is largely automated, there is relatively less room for local governments to distort the numbers. In fact, industrial electricity consumption per unit of GDP has been closely watched by local governments, with rising concerns about eliminating outdated capacity in heavy industries.

Periodic divergence between industrial electricity consumption and industrial production growth may in part reflect a shift towards less energy-intensive sectors, or a transition in the automobile sector towards electric vehicles. Other factors also affect electricity usage. For example, weather-related factors (e.g., hot summers which lead to more air conditioning demand) lead to strong seasonality in residential electricity consumption. Abnormal weather can therefore distort even seasonally adjusted data.

Total profit of industrial enterprises: Total profits are the sum of "operating profit" and net "non-operating profit", on a before-income-tax basis. The sample is based on industrial enterprises above the designated size threshold.^[3]

Operating revenue: Operating revenue refers to the total amount of income generated by business activities, including revenue from main business and other activities^[4].

The NBS also releases financial data stating the operational condition of industrial enterprises such as total assets, total liabilities, total owners’ equity, and main business income, etc.

We believe the data series are generally reliable, at least in terms of the overall growth rates, though probably less so at the industry level. SOEs are often perceived as over-reporting their profits and under-reporting losses because of performance assessment systems for their senior management, while private enterprises tend to under-report them to avoid taxes. But to the extent that there is no major change in the degree of over/under-statement (at least in the short run), the growth rates should be useful in terms of gauging profitability trends of large industrial firms. The trend in profit growth is also broadly consistent with that of enterprise income tax receipts reported by the Ministry of Finance, except for periods with major tax policy adjustments (e.g., tax waivers, cuts and postponement).

Profit/revenue data contain a lot of seasonal noise, e.g., profits and revenue levels usually show an uptick in December. Although seasonal adjustment should deal with this, given the pronounced seasonal effects an alternative approach to minimize this noise is to compare the data point with the same period in the previous year.

Profitability is useful in gauging the strength of the corporate sector. Profitable firms are able to, and often do, reinvest their retained profits. Other industrial financial data such as interest payments can also help assess the debt sustainability of the industrial sector.

The coverage is the same as that of the industrial production data, i.e., all industrial enterprises with annual sales of RMB20 million or more. Similar to industrial production, the minimum threshold for profit data sampling increased in 2011. (Previously the minimum threshold was annual sales of RMB5 million or above.) Official year-over-year growth rates are based on comparable samples, according to the NBS.

In 2012, the NBS released the new Industrial Classification for National Economic Activities which expanded the industrial coverage from 39 to 41 in total. Therefore, many sub-sectors’ data are not precisely comparable over longer periods of time even though the sectors may have very similar names; analysts should watch out for jumps in series around times of revised classifications.

Profits are on an accrual basis, and China onshore stock market listed companies follow the same standard, per enterprise accounting rules from the Ministry of Finance. However, NBS profits are pre-income tax, and for listed companies, investors usually pay attention to post-tax earnings.

The NBS typically releases the amount of losses (year-to-date) made by loss-making firms at the same time as it publishes the total profits of industrial enterprises. (As one would expect, the reported total profits number for all industrial enterprises already nets out the losses from loss-making companies.)

The main cost categories in the monthly profit report are cost of goods sold, finance costs (interest payments and fees paid to financial companies, which can often be significant especially when interest rates are high), and management/operation costs.

Of the two manufacturing surveys, China’s NBS manufacturing PMI -- also known as the official manufacturing PMI -- has historically performed slightly better, as its production and new orders indices appear to be the best coincident indicators of sequential industrial activity growth among all PMI related data.

This advantage could be because the NBS manufacturing PMI is more representative, due mainly to the fact that it is compiled by the official statistical authority:

1. The NBS manufacturing PMI is based on a sample size of 3000 firms while the Caixin manufacturing PMI is based on over 500 firms.

2. The NBS survey is likely to have a higher response rate. The response rate of the NBS survey is said to be as high as 99.6%, as the NBS has the legal right to demand that firms respond. While this can be a double-edged sword as some firms reporting unwillingly might be reporting with less care, the net effect of having a large effective sample size is probably a good thing. While the rate for the Caixin survey is unknown, it is unlikely to attain such a consistently high response rate.

China’s two manufacturing PMIs sometimes send different signals (Exhibit 11). Sample size difference and discrepancies in data-collecting periods may explain divergences. The Caixin survey is conducted in the middle of each month, while the NBS survey is conducted later, at around the 20-25^th of each month. When the economy is changing rapidly, this timing difference can be significant.

It is often said that the official PMI is a better reflection of large manufacturers and the Caixin PMI is a better reflection of smaller and often export-oriented producers. However, the index provider (Markit previously) disagreed with this characterization, and at least in terms of design -- both PMIs are designed to capture large manufacturers as well as SMEs. Empirically, the Caixin PMI outperformed the NBS PMI significantly in 2020H2 and 2024H1 when Chinese exports were very strong, suggesting the Caixin PMI may cover more export-oriented companies.

The NBS non-manufacturing PMI samples 4000 enterprises of different sizes in the construction and services sectors. The surveys ask 10 questions for 10 individual indices about production, new orders, input price, sales price, employment, business activity expectation, new export order, work backlog, inventory, and suppliers’ delivery time. The Caixin Services PMI sample covers more than 400 enterprises in the services sector.

Unlike the manufacturing PMIs which are conceptually identical, the NBS non-manufacturing PMI and the Caixin Services PMI are different—the official one is not just a services PMI but includes construction as well. It is therefore natural that these two indicators diverge more often than the manufacturing PMIs (Exhibit 12). But even if we look at the service sub-index of the official non-manufacturing PMI, it is still quite different from the Caixin Services PMI. The construction sub-index of the non-manufacturing PMI has a relatively weak relationship with the construction component of GDP.

The China Federation of Logistics and Purchasing (CFLP), which compiles the official PMIs, also publishes the emerging industries PMI (EPMI) each month. The EPMI is released around 10 days earlier than the NBS and Caixin manufacturing PMIs, and could serve as a leading indicator for these PMI prints. The data series was first introduced in 2014, mostly covers the country’s “strategic emerging industries” (mostly high-tech manufacturing), and is released on the 20th day of each month if it’s not a public holiday. We caution that there are some caveats when using EPMI in forecasting, due to its differences with NBS and Caixin manufacturing PMIs on samples, survey dates and seasonal adjustment methodologies.

PMIs can be useful leading indicators to gauge sequential growth momentum. They are also useful in gauging upstream inflation, via their input price indices, and in tracking inventory cycles, via their inventory sub-indices.

Both the NBS and the Caixin PMIs are compiled and summarized through the results of a monthly survey of enterprise purchasing managers. In reality, it is not necessarily the purchasing managers who are filling out the forms but other employees, especially those in the finance department. Statisticians send out questionnaires to firms every month to ascertain whether the situation in certain aspects of the business has improved, or if there is no change, or if it has deteriorated. Responses in each category are given the weights of 1.0, 0.5 and 0, respectively. An index between 0 and 100 is then compiled for sub-indices such as production, new orders, employment, prices (separately for input and output prices), inventory (separately for raw materials and finished goods inventories), and suppliers’ delivery times. The overall manufacturing PMI is a weighted average of the new orders (30%), production (25%), employment (20%), suppliers’ delivery time (inverted, 15%), and raw materials inventory (10%) components in accordance to the weights used internationally. Data are then reported as a reading between 0-100 for the overall indicator and different categories, by adding the share of respondents indicating “improved” plus half the share indicating “no change”.

The Caixin services PMI and NBS non-manufacturing PMI do not have aggregate readings and only report readings for different components. The “Business Activity Index” is often used as the headline reading. The Services Business Activity Index is to the services sector what the Manufacturing Output Index is to the manufacturing sector.

The PMI Composite index (also known as Composite Output index) is a weighted average of the Manufacturing PMI Output sub-index and the Services Business Activity Index (i.e., the services PMI).

The GS China Current Activity Indicator (China CAI) was created to provide an alternative measure with higher frequency and quality to identify shifts in the economic cycle. It attempts to encompass indicators from the main producing sectors of the economy – manufacturing, housing, and consumer – as well as the labor market. The China CAI is shown on a month-over-month annualized basis, after several rounds of adjustments by the GS global economics team to harmonize CAIs around the world and to deal with dramatic activity changes in response to the Covid pandemic (it is available on the GS research portal or on Bloomberg at ticker: GSCNCAI). The components of CAI include: industrial production, employment in composite PMI, Cheung Kong Graduate School of Business (CKGSB) Business Conditions Survey sales sub-index, the Caixin Services PMI, electricity consumption, automobile sales, the Caixin Manufacturing PMI output sub-index, real imports, real retail sales, floor space sold, floor space started, cement production, real exports, freight volume, passenger volumes, and floor space completed.

The CAI is calculated on a sequential basis. Statistically, our CAI is constructed as the first principal component of 16 standardized monthly economic indicators after seasonal adjustment, converted to GDP-equivalent terms through a regression of historical real GDP growth on this principal component.

We extend the CAI back to 2006, by backcasting a few series (mainly survey indicators) for which a complete history is not available. Other indicators on activity growth such as trade flows and sector-level data all showed a higher amplitude of fluctuation than the official GDP data, so we adjusted for volatility when constructing the CAI.

Because the CAI methodology is designed to be a high-frequency proxy for GDP, distortions in GDP data (e.g. data that are consistently “too smooth”) can in theory affect the CAI. In particular, the mean of growth as measured by the CAI over the sample is effectively equal to that of GDP growth over the sample, by construction.

The GS China inventory tracker is based on six underlying inventory indicators, including commodities (i.e., iron ore, aluminum), PMI sub-indices (i.e., raw materials, finished goods), industrial enterprises finished goods inventory, and auto inventory. After data cleaning, we derive the first principal component, which explains 25% of the total variation of the six series, and then map it into percentage of GDP terms as our tracker for inventory changes. Our inventory tracker can be used to gauge the contribution of inventory changes to China’s GDP growth.

Our inventory tracker also indicates slowdowns in inventory build during the Global Financial Crisis (GFC), around 2015-16 (despite implausibly stable reported real GDP growth) when the government implemented “supply-side reforms” in upstream sectors like steel and coal, in early 2019 at the height of US-China trade war, and in late 2022 when the Covid “exit wave” caused significant supply chain disruptions in China.

We acknowledge our principal component analysis (PCA) approach to tracking inventory is only a proxy and the mapping into the inventory component of real GDP is not perfect. However, we think it provides a useful way to track an important part of the economy that is opaque and can generate large swings in quarterly growth.

“China’s manufacturing PMIs: Which one should we look at and what are they telling us?” EM Macro Daily, 13 August 2013

“Taking stock of China activity: Updating our Current Activity Indicator”, 18 November 2015

“Tracking all over the world - Our New Global CAI”, Global Economics Analyst, 25 February 2017

“China Manufacturing PMI - Still an important signal”, China Data Insights, 27 September 2019

“China EPMI, a good leading indicator for manufacturing PMIs”, China Data Insights, 21 July 2022

“Tracking China’s Inventory Cycle”, China Data Insights, 7 June 2023

“Why have post-reopening industrial profits been so weak?”, China Data Insights, 15 June 2023

“Peeking into NBS' GDP revision practice”, China Data Insights, 31 August 2023

“Gauging China’s growth (again)”, Asia Economics Analyst, 7 September 2023

“A Closer Look at NBS 2022 GDP Revisions”, China Data Insights, 13 March 2024

FAI undertaken by nature of enterprise

1. State owned enterprises (SOEs): These are enterprises solely owned by the government. This category is also known as “State-Owned and State Holding Enterprises”, and includes all SOEs and enterprises over which the government has effective control, but excludes state sole proprietors that are part of “limited liability companies (LLCs)”, based on the NBS definition. New PPP (Public Private Partnership) projects complicate the picture, as projects with 50% state share are accounted for as SOE investments, which overstates the true share of the state. PPP projects experienced rapid expansion in 2017, stagnated in 2018-22, and declined thereafter.

2. Collectively owned enterprises (COEs): Another kind of publicly-owned enterprise which has become less common (as of 2023, FAI by COEs accounted for less than 1% of total FAI).

3. Foreign-funded enterprises.

4. Hong Kong, Macau and Taiwan-funded enterprises (note that foreign-funded enterprises and Hong Kong, Macau and Taiwan Funded Enterprises are two separate categories that do not overlap).

5. Share-holding enterprises: Enterprises that have a share-holding structure.

6. Private enterprises and individual businesses: Both are private businesses. Businesses run by one person and small private businesses that employ fewer than eight people are called “individual businesses”; larger businesses are called private enterprises. Private-sector FAI has remained weak in recent years despite fiscal stimulus implemented by the government. The private share of FAI increased from 31% in 2004 to 64% in 2015, but has declined since then.

7. Joint ownership enterprises: Typically enterprises set up by two or more independent firms. This was a popular form of enterprise in the earlier years of reform but is now much less common.

8. Limited liability companies (LLCs): Including state sole proprietors and other limited liability companies.

9. Others (e.g., miscellaneous projects which are not easy to be classified into the categories above).

FAI by work type

FAI by industry is available in terms of the primary, secondary and tertiary industries. Detailed breakdowns for specific industries are available from 2004. As is the case for IP by industry, FAI by a given industry is not investment in a particular type of project, but just FAI by companies in that industry. For analysis we typically aggregate the sectors into four major categories: manufacturing, infrastructure, property and others (“others” mainly includes services and agriculture-related sectors). In 2023, FAI in these four sectoral categories accounted for around 33%, 36%, 15% and 26% of total FAI, respectively, compared to 30%, 31%, 22% and 19% in 2007.

1. The share of manufacturing FAI has been close to infrastructure FAI over the past decade. Manufacturing investment is mostly carried out by private enterprises in China. Over the past few years, the solid growth of manufacturing investment was underpinned by POEs.

2. Our estimates for infrastructure FAI are based on our GS definition, which includes not only the three sectors under the NBS classification (i.e., transport, storage & postal service; water conservancy & environmental protection; and electricity, gas & water production and supply), but also four more industries that provide public goods mainly by the government sector (e.g., scientific research & polytechnic service; education; healthcare, social security & welfare; and culture, sport & entertainment). Most of the infrastructure investment in China was done by local governments (around 70% in 2023). Over the past two decades, the share of transportation and utilities in infrastructure investment has fallen from around 60% to around 40%. Recent policy communications suggest infrastructure spending will likely focus on both traditional and new infrastructure projects in coming years: the former may include water conservancy, emergency management and energy, while the latter could include 5G telecommunications, supercomputing, EV charging stations, ultra-high voltage power transmission, data centers, green capex and industrial parks.

3. Property FAI, often regarded as very important due to its interconnection with local government finances, now has a much smaller share than manufacturing and infrastructure FAI. Since the unprecedented property downturn that started in 2021, the share of property FAI has declined significantly. Furthermore, the value-added share of property FAI is likely to be significantly lower than in other components of FAI, since a significant share of property FAI is in land transfers (which are non-value-added and do not enter GDP accounting). Hence, the share of property investment in total gross fixed capital formation (GFCF) is notably lower than its share in total FAI (for more detail on property FAI see Section IV. Real Estate). On the back of the current property downturn, regulatory tightening in sectors such as education and internet that are dominated by private enterprises, and the ongoing fiscal stimulus, FAI growth has shown a notable divergence between infrastructure and property, and between SOEs and non-SOEs since 2022. Our research suggests that the weakness of private investment during 2022-23 was mostly due to the large contraction of property investment.

FAI by region: This set of data is available for the 31 provinces and many cities, on a monthly basis. The FAI share across provinces is largely in line with their GDP share. FAI breakdown by province can help estimate the difference in investment momentum across regions, especially when macro policies have an asymmetric impact on different regions.^[6]

Funds designated for FAI (this is sometimes referred to as FAI by source of funding, but we view this as a misleading term because these data refer to the amount of funds that became available for FAI, which is not the same as the amount of FAI completed):

1. State budget.

2. Domestic loans from banks.

3. Bonds: Bonds issued by corporates and financial institutions, including those by local government financing vehicles (LGFVs) and policy banks for FAI.

4. Foreign capital.

   • Among which: Foreign direct investment (FDI). In practice this can include significant capital from “round-tripping” (see more on FDI in Section VI. External Sector).

5. Self-raised funds.

   • Self-owned: This is intended to be mainly from retained earnings. In practice this likely includes a variety of funding channels including non-standard loans borrowed from financial institutions. This might be driven by regulatory requirements, such as those requiring a project to have a certain share of self-owned capital before it is qualified to borrow from banks.

   • Share issuance: Capital raised through public offerings.

6. Others (e.g., crowd-funding and donations).

There is considerable confusion and controversy over China’s FAI statistics, with different definitions and scope of coverage.

There are major differences between the definitions of FAI and of the investment concept (GFCF) in national accounts.

1. The monthly FAI data are not reported on a value-added basis. That is, they do not reflect the incremental new capacity added to the capital stock, but just total nominal investment spending reported by companies and governments. FAI includes spending on assets that do not directly contribute to GDP (e.g., land acquisition and old equipment purchase), while GFCF does not.

2. FAI includes a minimum project size cutoff (RMB5 million as of 2023) for non-property investment (and all property investment), while GFCF does not.

3. FAI does not include spending on intangible fixed assets such as computer software and IT investment, while GFCF does.

FAI data have a few measurement issues.

1. Both the monthly and annual investment series are in nominal terms. The NBS did release an official FAI deflator on a quarterly basis previously, but suspended the data series in 2020. Besides, capturing price changes in this sector is difficult – especially land prices.

2. The monthly yoy FAI growth rate, which is monitored closely by investors, has significant measurement issues. For example, the classification for monthly reported FAI data changed materially a number of times in the past. This change has made the yoy growth rates not completely comparable over time. In addition, the official year-over-year growth rate series are not compatible with the official level series, mainly due to sample changes over the years.

3. Over-reporting may be the major factor behind the high level of FAI growth, which has also been acknowledged by the government in the past. Problems include potential double-counting (e.g., of the same project by different regions) and misreporting (e.g., FAI data are based largely on questionnaire responses which do not always have to be backed by hard evidence). Given the statistical issues mentioned above, we suggest investors supplement FAI data with other indicators, such as capital goods imports, raw materials (e.g., steel and cement) production and apparent demand, and infrastructure-intensive bond issuance (e.g., local government special bond), to help form a better assessment of true investment growth momentum on a monthly basis.

The value of total planned investment in new projects and under construction reflects changes in the pipeline of investment projects.

Data are compiled together with FAI data and therefore many of the problems associated with FAI data also apply to these series.

“Projects started” data are often used as a leading indicator of investment activities. Policymakers and investors also monitor this series to assess likely future trends in fixed asset investment on a 1-2 year horizon. In reality, however, there is no stable lead-lag relationship between the two series.

FAI projects started and under construction are released in terms of value and number of projects.

Project started, planned (value): This refers to all projects planned to start during the reporting period (they may or may not be under construction at this stage but will be started before the reference period end). It excludes ongoing investment projects (including those suspended and restarted in this period). Previously, the NBS also released data on projects started and planned in unit terms, but this data series was suspended in early 2018.

Projects under construction (value): This contains ongoing investment projects and new starts. Previously, the NBS also released data on projects under construction in unit terms, but this data series was suspended in early 2018.

Total planned investment: Refers to FAI projects’ planned investment amount. Note these plans are not binding and therefore this is not a reliable indicator for future FAI. Availability of funding and subsequent changes in costs can have large impacts on actual investment.

The relationship between different series in this set of data is complicated because of regulatory approval issues. For example, sometimes local projects find it difficult or costly to obtain approval for a new project and therefore just report the new project as an extension of an existing project.

The investment value of projects approved by the NDRC may reflect the central government’s stronger determination to boost investment if it increases quickly in a short period of time, but its total amount in 2022 (RMB1.5 trillion; latest year-end data available) only accounted for 2.6% of total FAI.

Komatsu, a leading global manufacturer of construction machinery, provides data on the monthly average operating hours of excavators sold in China. It is updated on a monthly basis with a lag of around two weeks. This could help gauge the momentum of construction activity, but it is subject to distortions from changes in Komatsu’s market share in China (which has declined to 1% in 2023 from 15% in 2008) and not able to distinguish infrastructure construction from property construction. This data series is subject to significant seasonality. Some investors may also focus on sales of excavators and other construction machinery, but these durable capital goods have significant replacement cycles, which may distort their yoy sales growth in some years.

Steel and cement are crucial raw material inputs for construction activity, and there are several related indicators available from either official or private data sources, including steel and cement output, demand for construction-related steel (measured by production net of net exports and inventory changes), the cement inventory to storage capacity ratio, cement production capacity utilization ratio, and steel and cement prices.

“China: Is credit losing its cyclical growth impact?”, Emerging Markets Macro Daily, 20 May 2013

“China infrastructure investment: Is the high growth sustainable?”, Asia Economics Analyst, 13 April 2017

“Fast or slow, old or new: A macro view on China's infrastructure investment”, Asia in Focus, 15 April 2020

“China Green Capex: Renewable power investment and its impact on the economy”, Asia Economics Analyst, 16 January 2022

“China: Manufacturing sector to drive investment in 2022”, Asia in Focus, 22 February 2022

“China: Gauging the upside for infrastructure investment in 2022”, Asia in Focus, 28 March 2022

“China: Tracking the strength and pace of infrastructure stimulus”, Asia in Focus, 12 May 2022

“China: Assessing the implications of local governments’ 2023 targets”, Asia in Focus, 17 February 2023

“China: Beijing’s Balancing Act between Infrastructure Stimulus and LGFV Deleveraging”, Asia in Focus, 6 March 2024

“The shifting role of private investment in China”, China Data Insights, 25 April 2024

There are two sets of real estate investment data. The most widely tracked one is the total FAI conducted by real estate developers (also known as “property investment” or “property FAI”). The data are surveyed and released together with other real estate indicators (e.g., new home starts, completions, and sales), as well as other major economic activity data (e.g., industrial production, headline FAI and retail sales), by the NBS on a monthly basis. They cover land purchases, equipment purchases and construction activities for residential, commercial and office buildings. In 2023, around 90% of real estate investment was related to land purchases and construction. The breakdown details of property investment by ownership of enterprises (annual), work type (monthly), and sources of funding (monthly) are available. Like headline FAI data, the NBS only releases the year-to-date level and growth data for real estate investment at a monthly frequency, while single-month real estate investment growth rates require additional estimates given a specific base year.

The other data on real estate investment is the FAI by industry – real estate (as a composite industry in the tertiary sector). The definition of this dataset is different because it supposedly includes real estate related investment by all types of enterprises, not just property developers. However, the general trends of these two indicators are similar.

Both series of real estate investment have similar drawbacks as other FAI data series. For example, the official year-over-year growth rate series is not entirely compatible with the official level series, mainly due to sample changes over the years (for details on other statistical issues with FAI data, see Section III. Investment). Furthermore, these two indicators do not capture rural properties without property rights. Rural properties without property rights are only for self-use, and transactions with buyers outside the community are deemed illegal. As such, there is a lack of data for rural properties in China.

The NBS made occasional revisions to the definition of private investment, which may also affect the data quality. For example, in March 2024 the NBS revised the definition by excluding land redevelopment related spending from real estate investment. Although the NBS flagged that the reported yoy real estate investment growth estimates are on a comparable basis, it did not release the revised historical level data series under the new definition.

Land transaction area (NBS): This is defined as land area (squared meters) purchased by real estate developers during the reporting time period. As all land in China is owned by the state, buyers can only purchase the right to use land instead of ownership. These rights range from 20-70 years, hinging on how the land will be used (e.g., for the construction of industrial, commercial, or residential buildings). In the latest version of Civil Code of the People's Republic of China, effective in January 2021, the land-use right will be automatically renewed when the right leases are up.^[7]

Land transaction value (NBS): This refers to the final amount (value) of the land-use right transacted in both the primary and secondary markets by real estate developers, based on the flows of their actual payment. Land purchase value is on a comparable basis with land purchase area in terms of statistical coverage, so we can derive the average price of land purchases from these two indicators. However, this is not precisely the same as a land price index because the quality of land transacted may not be comparable. Land purchase area and value data used to be available at the city level for 40 major cities, released by the NBS, but these city-level data series were suspended in January 2019. The NBS also suspended the release of land transaction area and value data series in January 2023.

Developer land purchase value (NBS): These refer to the total amount of land transactions, in value terms, based on the contracts signed by property developers in the primary and secondary markets. "Developer land purchase value” is much larger than “land transaction value” above because it includes various taxes, fees and spending associated with land compensation, land preparation, and land management. This series is recorded at the time of disbursement by property developers and may lag the transaction time somewhat.

Government revenue from land sales (Ministry of Finance): This is reported by the Ministry of Finance (MOF) on a monthly basis – along with other government income and expenditure data – and is a major component of government-managed fund revenue. The MOF series is slightly different in scope from the land-use right transfer value reported by the NBS and third-party data vendors. For instance, the former series also includes income from renting land by land administrations. Land sales remain a very important source of income/expenditure for local governments. Refer to Section X. Government Finance for more details.

Land transaction area and value by city (CREIS/Soufun/Wind): CREIS/Soufun^[8] collect and summarize the data (data access requires a subscription) on government revenue from land transactions in more than 300 cities in China (based on 302 cities and is often referred to as “CREIS 300-city revenue from land sales”, compiled and tracked by our GS China property research team). The definition of the data is broader than the NBS land purchase value/volume data mentioned above, given that these data cover land purchases by different types of enterprises, not just developers. Wind also compiles a sample of 100 cities to track land transaction value and area, on both a weekly and monthly basis. The Wind data have breakdowns by city tier (i.e., Tier-1, Tier-2 and Tier-3). The benefit of CREIS/Soufun/Wind data is that they are more timely (e.g., daily for the original CREIS/Soufun data and weekly for the Wind data) and tend to be a leading indicator (e.g., CREIS series on land transactions are recorded when transactions occur, NBS land purchases are recorded when investment is completed, and MOF government revenue from land sales are recorded when the payment is settled). The drawback is that CREIS/Soufun/Wind data aggregate a large number of cities but do not include all transactions nationwide.

The discrepancies between these land transaction measures mainly come from different timings to register land sales (e.g., registered when contracts are signed or when funds are paid), different groups of land buyers captured by the sample (property developers only or all types of buyers), and some additions and/or deductions before the purchase fees are transferred to the MOF's account for land sales revenue.

Among all the land sales indicators mentioned above, we tend to rely more on the land transaction volume and value series from CREIS. The data coverage of land purchase value and volume reported by property developers is much narrower, and the NBS land purchase data lag the CREIS data series slightly as discussed above.

The introduction of a centralized land auction mechanism for major cities in 2021 increased the transparency of land supply, reduced panic land purchases among property developers, and significantly lowered land auction premia. However, as land supply needs to be concentrated in several batches in a year under the new regime, occasional shocks (e.g., Covid lockdowns during 2020-22, natural disasters) would affect the timing of land actions and thus cause near-term distortions to yoy growth in land sales revenue.

The three indicators above are all reported by property developers, and released monthly together with the real estate FAI indicator. There are other indicators related to broader property activity: “housing under construction” and “housing completion”, for example, cover not just commercial housing but also other types of housing constructed by all types of enterprises.

The government also publishes data on social housing. There are annual data on the total number of flats planned to be started, as well as the total number of flats completed. In recent years, these data points were mostly unveiled by the Ministry of Housing and Urban-Rural Development on ad hoc occasions. Although social housing related data are not quite as important as commercial housing related data, and the former's data quality could be lower than the latter, construction of social housing still drives upstream/downstream industries (such as construction materials) and thus is useful to track. Social housing units could also be converted from commercial housing — during the property down-cycle in 2014-2015 and also in recent years, this was one inventory destocking measure adopted by many local governments. More recently, the PBOC announced in May 2024 a relending program encouraging local SOEs to purchase completed but unsold properties and to turn them into social housing to destock existing property inventories.

Property transactions data are generally reliable, especially when compared with new starts and completions data. But people sometimes under-report the transaction value to reduce the transaction tax, though this mostly occurs in the secondary market.

The year-over-year growth rates and the levels reported in the NBS property data used to be consistent with each other until 2023. In May 2023, the NBS revised down the comparison base when reporting yoy property activity growth data, especially for new home sales, to adjust for false sales data.^[10] The practice continued through the remainder of 2023 and resulted in a meaningful divergence between the officially-reported level data and growth rate data. However, the NBS has not released the revised level data for previous years. We adjust the property-related activity data series from 2023 in our analysis to take into account the revisions. We estimate new home sales volume in 2022 has been revised down by around 10%, especially for April-May 2022 (amid the Covid-related Shanghai lockdown).

These price indices generally follow a well-designed methodology, and most indices control to varying degrees for the impact of quality, though their original data sources differ somewhat. For some of the indices such as the NBS 70-city property price index, given that there is a higher degree of estimation involved in compiling the index and pressures on local governments to control property prices (especially new home prices), there are market concerns that the indices are over-smoothed, showing smaller fluctuations in prices than actual property price changes. For example, the rise in property prices may be hidden when developers are forced to sell properties at prices below the market equilibrium because of local government pressures, and the excess demand leads to alternative expenses such as requirements for property purchasers to deposit funds well in advance to be able to purchase the property. During downturns, the decline in property prices may be disguised by free parking spots and extensive internal renovations offered by developers in lieu of price cuts. These distortions are typically more relevant for new properties.

Price indices on second-hand properties are therefore often a useful reference when they diverge from primary market property prices. On secondary-market home prices, besides the NBS 70-city secondary home price index, some third-party data vendors such as Centaline, Zhuge, and Beike (mostly real estate agents, information platforms and consulting firms) also have their own measures. Home price data from private sources usually have a much shorter time series than the NBS series, but they tend to be released earlier than the official indices. In previous years, secondary home prices have also been affected by many non-market-based policy interventions such as government guided prices and reference prices (related to tax payments and mortgage borrowing for secondary home transactions). We also note some third-party data vendors such as Beike suspended their release of secondary home prices in late 2023 when secondary home prices showed notable declines.

The NBS changed the methodology for the Property Price Index in 2011. Property sales prices were split into primary and secondary housing price indices. The index initially covered 35 major cities and was expanded to 70 cities in 2005. Rural areas are not covered. Data are collected via a mixture of reporting forms from real estate companies and site visits by NBS staff. In January 2018, the NBS suspended the release of the “new residential property (including social housing) price index” series. Since then, it has only released the “new residential commodity property (excluding social housing) price index” series. The patterns of these two data series are similar given the small share of social housing in total housing.

The divergence of property price trends among different city tiers in China reflects differences in housing market fundamentals. Top-tier cities face more supply restrictions and resilient demand, and thus upward pressures on housing prices are the strongest. Lower-tier cities tend to have diverse circumstances but on average face less restrictive supply and weaker demand, and thus housing price growth tends to be slower and/or prices tend to fall more in property market down-cycles.

The table below shows the NBS classification for the 70 large and medium-sized cities by tier. Although there is widespread agreement on the definition of Tier-1 cities (Beijing, Shanghai, Guangzhou, and Shenzhen), categorization beyond that point is not always consistent among different sources, particularly for lower-tier cities.

“How China's property policy tightening lowered sales and prices”, Asia Economics Analyst, 26 September 2017

“Tracking residential housing’s impulse to Chinese growth”, Asia Economics Analyst, 27 March 2018

“China property policy: capturing the big picture from localized measures”, Asia Economics Analyst, 29 November 2020

“Q&A on Evergrande’s macro implications”, Asia in Focus, 24 September 2021

“Credit supply holds the key to China housing outlook in 2022”, Asia Economics Analyst, 11 October 2021

“How big is China's property sector?”, China Data Insights, 11 October 2021

“Lessons from Japan: Credit Tightening in the Property Market”, Asia in Focus, 29 November 2021

“Demystifying the discrepancy in different land sales measures”, China Data Insights, 14 April 2022

“Understanding the recent rise and fall of high-frequency property sales data”, China Data Insights, 14 July 2022

“China: 'L-shaped' Property Sector Recovery Ahead without a Quick Fix”, Asia Economics Analyst, 11 June 2023

“Understanding differences in China’s home price measures”, China Data Insights, 9 July 2023

“Q&A on China's property downturn and its implications”, Asia Economics Analyst, 23 August 2023

“China: Tracking the impact of ongoing housing easing”, Asia in Focus, 8 October 2023

“Comparing China and US Housing Downturns: Different Fiscal Backdrop, Same Need to Prevent Spillovers”, Asia Economics Analyst, 12 February 2024

“A Closer Look at NBS 2022 GDP Revisions”, China Data Insights, 13 March 2024

“China: Housing easing underway, but no signs yet of game-changing measures”, 14 April 2024

Total Retail Sales of Consumer Goods measure goods and restaurant services sold at the retail level (as opposed to wholesale), including both online and offline sales. These goods and restaurant services may be purchased by households, firms or the government.

A separate indicator named “Online Retail Sales of Consumer Goods” refers to goods transacted over online platforms.

Before the Covid pandemic, the main issue with retail sales data was that growth appeared overly smooth. We suspect the problem is caused by the fact that sales by companies below the minimum threshold are compiled using sample surveys, which involve a significant degree of discretion, whereas sales by companies above the minimum threshold are compiled using census surveys. Reported retail sales growth was also higher relative to the growth rate of the economy for many years, even after adjusting for inflation. During 2005-2015 for example, real retail sales (nominal retail sales growth adjusted for price factors) grew 15% per year, significantly above the 10% per year real GDP growth. In addition, sample changes over the years caused discrepancies between the reported retail sales level and reported retail sales yoy growth rate. Volatility in the series increased dramatically with the onset of the Covid pandemic along with more volatility in the underlying economy induced by Covid-related lockdowns, and the signal to noise ratio has therefore improved in recent years.

Although monthly retail sales data are often used as the main indicator for private consumption, users should be aware of a few data issues:

1. The most serious problem is that services consumption (apart from restaurant services) is not included in the retail sales data. Experiences of other countries show that the share of services in total consumption typically rises as the economy develops. However, retail sales data are unable to reflect this increasingly important component of consumption.^[13]

2. Retail sales also include non-household (i.e., government and corporate) purchases, but it is currently impossible to obtain a breakdown between household and non-household purchases. While household consumption probably constitutes the largest part of the reported retail sales, some of the retail sales expenditure is classified as government consumption vs. private consumption under the GDP by expenditure framework.

That said, the monthly retail sales data have the advantage of being timely. Policymakers also pay close attention to these figures, roughly on par with IP, FAI and trade data.

Sporadic waves of Covid in 2020-2022 and stay-at-home policies disrupted offline activities in China, drawing increased engagement to online platforms. As consumption behavior shifted from offline to online purchases in China, the share of online goods sales in total retail goods sales rose to 31% in 2023 from 23% in 2019.

All large firms above the designated size report sales data monthly. Data from small firms below the designated size are compiled through sampling. In practice, it is not clear how often, how much, and how well the sampling of small firms is actually carried out. Data from major online platforms are compiled for online goods sales, but the list of surveyed platforms may change over time.

Retailers with annual revenue from primary business of RMB5 million and above, hotels and restaurants with annual revenue from primary business of RMB2 million and above, and wholesalers with annual revenue from primary business of RMB20 million and above are jointly defined as enterprises above designated size.

Retail sales data are available in nominal terms only. The NBS discontinued the release of the Retail Price Index at end-2022. To deflate the series, the Consumer Price Index for goods could be used, though the breakdown and weights of CPI goods baskets and retail sales categories differ somewhat.

There are three kinds of breakdown available for retail sales data:

1. By location, i.e., consumer goods sold in urban and rural areas. However, urban retail sales may be goods purchased by rural households who live in urban areas and do not necessarily represent urban household consumption, and vice-versa.

2. By commodities. There are two layers of breakdown data available - the first layer is total retail sales by goods vs. catering, and online retail sales by goods and services. The second layer of breakdown is retail sales by specific commodities (e.g., food, clothes, home appliances, autos, etc.). The second set of data are only available for sales made by firms at or above the designated size and may not always give an accurate picture of overall consumption growth. Goods sales by above-designated-size retailers accounted for 40% of total retail goods sales in 2023.

Urban and rural income is classified in terms of total and disposable income.

Total income is composed of pre-tax wages, business profits, return on assets (e.g., interest, dividends, rents) and other “transfer income” (gifts, insurance claims, retirement pensions, transfer payments from members in other households etc.). Prior to 2012, samples were collected from 66,000 households which were presumably rotated every three years. Starting from Q4 2012, the NBS issued a new survey format to reflect the reform of urban-rural integration, which adjusted the sample to 2 million households nationwide for general investigation and selected 160,000 households for direct survey.^[14] One-third of the surveyed households are rotated each year.

Disposable income is total income for final consumption expenditure and households’ savings excluding income tax and social security contributions. It includes both cash income and in-kind income.

Two factors seriously affect the reliability of household surveys:

1. Households participating in the survey are required to provide very detailed notes on their expenditure, which is time-consuming. Households are rewarded financially for taking part in the survey, but the financial payments are small. For example, financial payment was RMB70 per month for households in selected cities in Zhejiang Province in 2020. Therefore, the incentive for households to respond accurately may not be high. This problem is especially serious for urban high-income households.

2. Lack of confidentiality may lead people to under-report their income/expenditure or simply refuse to participate in surveys. Again, this is likely to be especially serious for high-income groups and households with gray or illegal income. As a result, there is likely a downward bias in reported income levels. However, the direction of any bias on growth rates is less clear as under the aggressive anti-corruption campaign of recent years, under-reported gray and illegal income growth likely fell dramatically. Since these sources of income were never reported in the first place, the impact may not show up in official statistics.

Household consumption expenditure survey measures households’ expenditure on consumption (including both money and non-money expenditure) in eight broad categories:

1. Food (including tobacco, liquor, and catering)

2. Clothing

3. Housing^[15]

4. Household appliances, articles, and services

5. Healthcare, medicines, and medical equipment and services

6. Transportation and communication

7. Recreation, education, and cultural goods and services

8. Others

Key changes since the implementation of the Urban-Rural Unified Household Survey: 1) income and spending sub-categories have been standardized; 2) urban population covers migrant workers living in urban areas, but migrant workers are excluded from the rural population; 3) college students supported by the surveyed households but living in separate residence areas are counted as members of the households.

The compilation methodology for expenditure data is the same as for disposable Income data (described above), and therefore entails the same problems.

Household survey data provide useful information that is not available from other data sources. The household survey consumption per capita data is the only data set that captures all consumption categories, including both goods and services, and therefore should provide the most comprehensive gauge of the state of household consumption at a quarterly frequency. Household consumption in GDP data incorporates data from retail sales as well as household surveys. As the Chinese economy rebalances towards more consumption, this data set has become more important.

In addition, one can calculate the household saving rate (the difference between disposable income and consumption, divided by disposable income) based on the income and consumption data. This is the most timely read of households’ savings/consumption behaviors. In general, the household saving rate has been trending up over the past decade on the back of continued urbanization (urban households have higher income and higher savings rates than rural households). The household saving rate rose during 2020-2022 amid the Covid pandemic, and declined in 2023 along with China’s reopening as the Covid-related restrictions were lifted.

This series reports the year-over-year growth rate of the total retail sales of the top 50/100 retailers in China. The top retailers include both offline retailers such as chain stores and department stores and online retailers such as Tmall and JD.com. These goods may be purchased by households, firms, or the government. Retail sales data by product are available, though there are missing values in this data. Similar to the official retail sales data, this series is reported in nominal terms.

Total retail sales of the top 50/100 retailers data are subject to sample changes, as the list of retailers included is updated regularly based on the latest ranking. In addition, releases of this indicator are occasionally delayed for a few weeks. CNCIC suspended the release of the full top 100 retailers data in April 2024, although the top 50 are still available.

Total retail sales of the top 50/100 retailers are reported separately by non-official sources (CNCIC) and therefore could provide cross-checks for the official retail sales/household consumption data^[16]. For example, year-over-year growth in total retail sales of the top 50/100 retailers dropped from around 20% in 2011 to 0% in 2014-15 amid a housing downturn and overall economic slowdown. Growth in the official retail sales series over this period was more stable by contrast, decelerating gradually from 17% yoy in 2011 to 11% yoy in 2015.

CAAM reports car sales in units on a monthly basis, capturing different types of cars, such as passenger and commercial cars, through wholesale channels. The CAAM is regulated by SASAC and authorized by the government to collect auto production and sales data from auto manufacturers. CPCA is a data exchange platform among automakers and not accredited by the government. It reports auto sales in units on a weekly basis, capturing passenger car sales through both retail and wholesale channels. For the value of sales of automobiles, the NBS monthly retail sales data report sales by above-designated-sized enterprises.

Autos-related data tend to be highly cyclical, though sales are sensitive to policy measures such as auto purchase tax changes and changes in purchase restrictions in large cities. Policymakers may also apply differentiated policies (e.g., on new energy vehicles vs. traditional vehicles) to provide targeted support for certain types of vehicles. New energy vehicle sales have expanded at a very fast pace in recent years on the back of policy support. For example, license plates for new energy vehicles are not restricted in top-tier cities, unlike traditional vehicles. New energy vehicle sales rose from 0.04% of total auto sales in 2011 to 32% of total auto sales in 2023, based on CAAM data.

Auto sales data are generally reliable over time. CAAM and CPCA data are directly reported by automobile companies, though subject to potential temporary distortions due to a lack of cross-check mechanism. Companies that already reached their annual sales targets, for example, may delay booking some sales to make it easier to reach the target in the following year.

Auto sales data have gained importance in recent years. From the demand side, automobile consumption accounts for around 10% of total household consumption and plays an important role in overall household consumption growth. On the production side, automobiles accounts for around 6% in overall industrial value-added, and is a key sub-industry in the industrial sector. According to the National Development and Reform Commission (NDRC), as of 2020, auto and related sectors accounted for roughly 10% of GDP.^[17] As policymakers push for energy transition and emissions reduction, new energy vehicle production and sales data could shed light on the progress of China’s economic transformation.

The NBS publishes the consumer confidence index (CCI) on a monthly basis. It is a survey-based diffusion index that ranges from 0 to 200, with 0 implying extreme pessimism and 200 implying extreme optimism. The index is imputed based on a monthly telephone survey of more than 6,000 urban and rural consumers in 15 provinces. Survey questions include respondents’ assessment of current employment situation, income level, employment prospects, income expectations and willingness to spend.

The NBS CCI is based on a telephone survey. It is less precise than hard data (e.g., retail sales) and potentially subject to biases such as sampling and response rates, but it can be a useful gauge of consumer sentiment, especially at times when there is a major shift in consumer confidence.

Consumer confidence is an important concept as in theory it drives household savings, spending, and investment decisions. Our research shows that consumer confidence does matter to household consumption after controlling for other variables like household disposable income. Since the Shanghai Covid lockdown in Q1 2022, the NBS consumer confidence index has remained depressed despite China’s lifting of the zero-Covid policy at the end of 2022. Muted domestic demand, a weak labor market, and the continued fall in house prices hurt consumer confidence and hindered willingness to spend.

“China consumption worries: Goods spending decelerates as credit impulse fades”, Asia Economics Analyst, 8 September 2018

“How has online shopping affected CPI inflation in China?”, Asia Economics Analyst, 3 October 2019

“A bit more confident about the Chinese consumer confidence data”, China Data Insights, 2 December 2019

“Household cash flow: Still the key driver for consumption in China”, Asia Economics Analyst, 24 May 2020

“China: Households’ excess savings around the pandemic”, Asia in Focus, 15 March 2021

“A Pulse Check on Chinese Household Consumption Growth: Still Sluggish”, Asia Economics Analyst, 4 July 2022

“China post-reopening consumption recovery: Large potential, lingering scars”, Asia Economics Analyst, 18 January 2023

“FAQs on “excess savings” in China”, China Data Insights, 7 February 2023

“Lower housing cost burdens can lower the savings rate in China, but confidence remains key”, Asia Economics Analyst, 21 January 2024

Merchandise trade by China Customs measures the value of goods transactions (both in RMB and USD terms) across national borders. Exports are valued on an FOB (free on board) basis, which includes costs to deliver goods onto the vessels but not further costs, such as insurance or freight. Imports are calculated on a CIF (cost, insurance and freight) basis, which includes insurance and freight charges.

Trade data in the Balance of Payments standard, as published by SAFE, place greater emphasis on transactions between residents and non-residents, rather than physical movements across borders as shown in the China Customs data. (Though making this distinction can be difficult in practice.) The credit and debit sides of trade under SAFE data are both accounted on an FOB basis.

Customs trade data are among the most reliable macro-economic data in China. Their high volatility is a clear indication of absence of the smoothing prevalent in other economic data. However, in 2012-13 and 2015-16, there were distortions to trade reporting due to importer/exporter incentives to move capital either onshore or offshore. For example, when there were strong capital inflows into China, exporters tended to over-report exports in order to facilitate higher payments from offshore. When there are outflow pressures, importers tend to over-report imports to disguise the movement of capital offshore. Exporters sometimes also create “false exports” to gain government subsidies such as export rebates. Therefore, at the end of this section, we have estimated an alternative series that relies on information from China’s trading partners.

Besides over-/under-reporting, there are significant discrepancies between bilateral trade data compiled by China and the corresponding data from its counterparts due to re-exports via Hong Kong. There are various ways to adjust for these data problems. For example, the statistical authorities in Hong Kong provide detailed data on the region’s re-export trade by country and surveys of re-export margins. The difficulty with this analysis is that much of the false reporting is done via companies set up in Hong Kong, often solely for this purpose. These companies tend to use high-value-added goods that are easy to report large values for, but are often hard to judge in terms of underlying value.

Commodity breakdowns are available by the Standard International Trade Classification (SITC), the Harmonized System (HS), the Broad Economic Categories (BEC) (e.g., consumption goods, capital goods and intermediate goods, etc.), and customs regime (e.g., ordinary trade, processing trade, etc.).

China Customs releases monthly trade indices on export value, volume, and unit value with a slightly longer lag (usually 20-25 days). Prior to 2014, the indices were denominated in USD. From 2014 onwards, the indices have been denominated in RMB.

Some of China’s imports are for eventual export (e.g., raw materials or intermediate inputs into manufacturing goods exports). Total imports therefore can be divided into imports for processing trade and those for domestic use. We derive our estimates of total imports for processing trade by aggregating the following import categories:

1. Imports for processing and assembling

2. Equipment imported for processing and assembling

3. Customs warehousing trade

4. Entrepot trade by bonded area

5. Imports for outward processing

The share of “imports for processing” in total imports has declined in recent years. “Imports for processing” declined to around 34% of total imports in 2023, from around 51% in 2005 when the series started.

Note that there are a few differences between the Customs trade data and the net exports data in GDP. Net exports in GDP capture the net trade balance in goods as well as in services, whereas Customs trade data only cover trade in goods. Furthermore, in GDP data, both exports and imports are valued on an FOB basis, whereas Customs imports data are valued on a CIF basis. GDP standard is consistent with BOP standard. The NBS does not separately release data on exports of goods and services and imports of goods and services.

Exports/Imports of services refer to income/payment from/to foreigners on intangible products such as transport, tourism, entertainment, telecommunication and financial services.

After expanding through 2018, China’s service trade deficit narrowed sharply in 2020 after the onset of the Covid pandemic when strict travel restrictions were imposed. In 2019, travel services imports totaled $250bn. During the Covid pandemic, they averaged only $120bn per year, most of which were tuition and medical services payments made by Chinese residents overseas. China’s service trade deficit widened in 2023 after the end of its zero-Covid policy.

Service trade data are probably subject to a significantly higher level of misreporting, because it is usually harder to verify the underlying fair value of services provided than goods traded. Although less subject to reporting distortions that are aimed at benefiting from government subsidies (as local governments are much more focused on goods trade) on balance services trade data appear to mask substantial net capital outflows. Spending per traveler roughly doubled in 2013-16, for example, suggesting the possibility that a significant part of “travel expenditure” is really hidden capital outflow. According to some academic research, correcting for this would suggest a current account surplus 1-1.5% of GDP larger (and commensurately higher capital outflows).^[18] When faced with capital outflow pressures, the authority tends to crack down on outflows through services trade as well. For example, in 2016/2017, bank card overseas withdrawal and large-amount purchases (which would show up in services imports) were under tighter restrictions.

SAFE began to compile BOP data in 1985 in accordance with the 4^th edition of the IMF BOP Manual, and subsequently adopted the 5^th edition in 1996 and the 6^th edition (BPM6) in 2015.^[19] In this edition, the IMF changed “Income” to “Primary Income”, and “Current Transfer” to “Secondary Income”.

SAFE must rely on other government agencies for data on various sub-components, which are often not compiled in accordance with IMF standards and are difficult to adjust (for example, the FDI data). We discuss the difference in FDI data from SAFE and Ministry of Commerce later in this section.

Misreporting of data to evade capital controls are common and likely worsened after 2015 given significant capital outflow pressures. The misreporting of goods trade data is well known, but the problem is likely to be more serious in terms of services trade because it is more difficult for the government to prove any wrongdoing.

Balance of payments data (China’s are summarized in Exhibit 44) consist of two main components.

1. Current Account (CA): This records the flow of international trade (both goods and services), primary income (i.e., income that accrues to foreign-owned inputs to production and other assets, e.g., profits generated by foreign-owned enterprises and investment income paid to foreign-owned property and financial securities), and secondary income (i.e., income transferred to or received from foreigners without “quid pro quo”, e.g., remittances from workers working overseas).

2. Capital and Financial Accounts: This essentially records the flow of investment. Main components are: direct investment, portfolio investment, other investment, reserve assets, net errors and omissions.

The merchandise trade by BOP accounting is conceptually different from the monthly trade data from the Customs Administration as it is supposed to capture transactions between residents and non-residents instead of across national borders. Besides, its valuation standards are different: BOP measures both exports and imports on an FOB basis, whereas Customs values exports on an FOB basis and imports on a CIF basis. As FOB excludes costs of insurance and freight, imports are smaller in BOP accounting than in Customs accounting (by around 5%). SAFE also makes other minor technical adjustments to the customs trade data in accordance with IMF standards, such as subtracting goods exported but then returned. SAFE counts only the cargo with ownership change. Goods trade with no ownership change is accounted for as service trade in BOP.

The services component in the current account consists of items such as transportation, travel, and insurance. Data are compiled by SAFE directly, as well as by various other government agencies, such as the Ministry of Transportation and the Ministry of Culture and Tourism. Most of the service trade in China is through the “travel” channel, and “travel” under services trade includes spending while traveling, and education tuition abroad.

The income account records compensation for employees working abroad and returns on investments. The numbers of Chinese residents working abroad and foreign residents working in China are both relatively small, and most primary income is investment income. In addition to primary income, secondary income through current transfer is also part of the current account.

The capital account includes capital transfers and reduction or cancellation of debts. Financial accounts include international reserve assets and non-reserve financial accounts. Non-reserve financial accounts in turn include direct investment, portfolio investment, financial derivatives and other investment.

Reserve assets are external financial assets held by the monetary authorities, including gold, foreign exchange, special drawing rights (SDRs) with the IMF, and the use of the Fund’s credits. Reserve assets are transaction-based and therefore not affected by valuation effects from asset price and exchange rate fluctuations.

Net Errors and Omissions (NEO) (also called “statistical discrepancy”) balances the credit and debit items in the BOP. It captures the errors and inconsistencies in data recording and processing in all sections of the BOP.

Foreign Direct Investment (FDI) measures investments made by foreign residents who seek to have significant long-term interest in and have direct influence over a domestic enterprise. FDI also reflects the confidence of international investors in the Chinese economy. In addition, FDI data are useful in estimating total capital flows.

There are multiple sources reporting China’s FDI data. Investors usually pay most attention to the FDI numbers from the BOP releases, given that it is compatible with GDP accounting, can be compared with other countries’ FDI data, and is published quarterly. The MOFCOM provides another official dataset on FDI with breakdowns by sector and by country/jurisdiction of origination. But it follows a different accounting principle, with reinvested earnings excluded. The IMF Coordinated Direct Investment Survey (CDIS) data cover over 100 countries and regions and offer mirror data from other economies, which can be used to formulate an “outside-in” measure to cross-check China’s official data. From a bottom-up perspective, Bloomberg reports investment value on its Mergers & Acquisitions (M&A) page, which can be aggregated into a timely series of China’s inward FDI. In addition, OECD and United Nations Conference on Trade and Development (UNCTAD) provide cross-country panel data on FDI as well.

Compilation methodology: Most FDI data are compiled based on one of two principles: asset/liability principle or directional principle, except for Bloomberg data. FDIs under the directional principle account for net investment on the basis of ultimate parent companies, while those under the asset/liability principle will simply add up net assets/liabilities for outward/inward FDI. For example, if a Chinese parent company receives investment from its own foreign subsidiary, this flow will be captured under the asset/liability principle but will not be captured under the directional principle. The data prepared by SAFE and OECD follows the asset/liability principle. In contrast, MOFCOM and UNCTAD compile FDI data based on the directional principle. IMF CDIS shares the same inputs as SAFE, but the data is presented under the directional principle. Lastly, Bloomberg’s M&A investment amount is based on project-level accounting, effectively adding up announced deals.

Data coverage: We summarize data coverage by availability of flow/stock data, detailed breakdowns and mirror data. SAFE provides FDI flows under BOP and FDI stock under International Investment Position (IIP) without any breakdowns. MOFCOM provides breakdowns of FDI flows by sector and by country/jurisdiction of origination in their annual statistical bulletin. The monthly FDI flows do not include direct investment in financial services (e.g., banks), although the size would be relatively small (around 4-6% of inward FDI in 2018-2022). IMF CDIS offers mirror data of FDI stock, which can be compared with the MOFCOM's FDI data by-country breakdowns.^[20] Bloomberg data are FDI flows with industry classifications. Yet, its numbers reflect planned cross-border M&A investment flows only, missing other FDI flows such as greenfield projects. Both UNCTAD and OECD provide FDI flow and stock data without further breakdowns. Among these data sources, only SAFE, IMF CDIS and OECD incorporate reinvested earnings in their FDI measures, as required by the Balance of Payments and International Investment Position Manual, 6th edition (BPM6) standard. They also effectively assume all non-repatriated earnings are reinvested.

Data availability: Reporting frequency ranges from monthly to annual. MOFCOM provides monthly data from 1984 with a one-month lag. Bloomberg also publishes monthly data from 1998 with daily updates of the latest released FDI projects. SAFE’s FDI data come on a quarterly basis from 1998 with a one-quarter lag. The IMF CDIS, UNCTAD and OECD offer annual data. UNCTAD has the longest data history, tracking the FDI series from 1990 onwards, followed by OECD, which provides FDI numbers from 2005 onwards. Both have a publication lag of around one year. IMF CDIS is the least timely data source, with the earliest statistics dated back to 2009 and the most recent data for 2022 (a 2-year publication lag).

As with GDP data, there are considerable issues in FDI data compilation at the regional level. The authorities have strengthened the requirements for the verification of data reporting. For example, data on cash FDI are double-checked against capital flow records at the SAFE, and data on goods FDI are double-checked against the records of China Customs. Data that cannot be verified are, in principle, discarded. Moreover, FDI’s importance as a part of the assessment metric for local government officials has declined, and hence the incentives to over-report have dropped relative to earlier years.

Assessment of the true strength of FDI is also complicated by the issue of round-tripping, which refers to funds originated in China but reinvested back as FDI. Money laundering and preferential policies for foreign enterprises are the most important reasons for round-tripping.

Data prior to 2015 only included FX-denominated external debts. The SAFE started to publish data with both CNY and FX denominated external debt since 2015. At the end of 2023, the outstanding external debt of China (excluding that of Hong Kong and Macao, but including debt both in FX and RMB) was US$2.45 trillion.

In regard to currency type, the total external debt is composed of foreign debt in RMB and foreign debt in other currencies. By the end of 2023, 47% of external debt was denominated in CNY. Of FX-denominated external debt, 84% was denominated in USD, 7% in EUR, 4% in HKD, 3% in JPY, respectively.

By original maturity, short-term refers to external debt with a term of one year or less. Medium- and long-term refer to external debt with a contract term of more than one year. As of 2023, short-term external debt took around 56% of China’s total external debt, and long-term external debt 44%.

The external debt is also classified by institutional sector, with general government accounting for 18%, central bank 4%, deposit-taking corporations except central bank 41%, other sectors 25% (including other financial corporations and non-financial corporations), and intercompany lending under direct investment 12%, as of the end of 2023.

As we think official data do not capture some of the FX debt raised by offshore Chinese companies, we compile our own estimate of Chinese corporates and households’ FX debt by aggregating onshore FX loans, claims on the Chinese non-bank sector by BIS reporting banks (offshore, non-Chinese), FX bonds, and trade liabilities. Our approach indicates that Chinese corporates’ and households’ total FX-denominated debt stood at around US$2.1 trillion as of 2023 year-end.

PBOC FX reserves data can be opaque and influenced by factors other than capital flow fundamentals. Because reserves data are based on market prices and denominated in USD, exchange rate and asset price fluctuations affect reserve values. When calculating changes in reserves, we usually adjust for the estimated effect of exchange rate fluctuations, but the impacts of asset price changes are difficult to estimate because of the lack of detailed information on portfolio holdings.

China started to disclose its historical dollar share of FX reserves from 2019 in SAFE’s annual reports with a five-year lag. The share remained in the range of 57% - 59% in 2014-2018, down from 79% in 1995. However, SAFE does not provide further details on shares of other currencies, or USD shares for the most recent few years. USD holdings remain prominent in the official FX reserves, despite continued selling of US Treasury securities over the past few years. We estimate USD assets account for roughly 60% of China’s official FX reserves via both China’s data and US Treasury International Capital (TIC) data. The TIC data can also be used to gauge the composition of the USD portfolio in China’s FX reserves.

In order to estimate valuation effects on reserves, we assume the currency composition of China’s FX reserves is similar to that of the global average and follows the Currency Composition of Official Foreign Exchange Reserves (COFER) published by the IMF.

Given possible PBOC balance sheet management and shifts in banks’ net open position, we prefer the SAFE data on banks’ FX settlements on behalf of their onshore clients as a gauge of the FX-RMB conversion trend among onshore non-banks.

PBOC’s reserve data and IIP data on reserve assets are based on market price, and thus will be impacted by valuation effects. Reserve changes in BOP data on the other hand are free from valuation effects as they are supposed to measure flows.

China’s FX reserves have been surprisingly stable at just over USD 3 trillion since 2017 despite the large inflow pressures in 2020-2021 and outflow pressures in recent years. Our analysis suggests that the FX holdings of China’s commercial banks may serve as a buffer for capital inflows/outflows, with official FX reserves changing relatively little as a result. China’s commercial banks accumulated a large amount of FX from 2H 2020 to 2021 on the back of China's elevated goods trade surplus. The authorities can guide banks to sell FX first before selling the government’s own FX reserves to defend the currency, if necessary.

CNY is the code determined by the International Organization for Standardization (ISO code) for the renminbi/yuan. In practice, it refers to the Chinese currency traded onshore in mainland China.

Despite some gradual steps toward liberalization, the CNY market remains heavily managed by Chinese officials, who maintain strict capital controls and set a daily “CNY fix” against a basket of world currencies and a pre-determined trading band around that fix, which the “CNY spot” must settle within. The trading band was initially established at 0.3% but was expanded to 0.5% in May 2007, to 1.0% in April 2012, and to 2.0% in March 2014.

The PBOC adjusted the CNY fixing regime in August 2015 so that CNY fixing reflected the previous day’s closing price of USDCNY and overnight USD moves. In May 2017, the PBOC further introduced a countercyclical factor in the CNY fixing to lean against herding behaviors in the market. The PBOC has utilized countercyclical factors since then to guide market expectations on the exchange rate direction. The PBOC doesn’t release any data on the countercyclical factors. We measure the countercyclical factor by calculating the difference between the official daily CNY fixing and our estimation of the fixing based on official documents of CNY fixing mechanism. That is, the CNY fixing should be determined by the previous close of USDCNY spot and the overnight move of USD against a basket of currencies. Our estimates suggest that the countercyclical factors have risen since 2023 and reached record levels in mid-2024, suggesting the authorities’ preference to slow CNY depreciation against USD.

CNH refers to RMB that is traded outside of the Chinese mainland. Establishment of the offshore CNH market by Chinese policy makers reflected their desire to pursue greater use of RMB for international trade and financial transactions (i.e., the “internationalization” of the RMB) post the 2008 financial crisis. The CNH market was established in July 2010 when the PBOC and HKMA jointly announced that RMB would be deliverable in Hong Kong.

Although the CNH market remains concentrated in Hong Kong, RMB has since become deliverable in Singapore, Taiwan, Paris, Luxembourg, London, etc. Any offshore corporate entity or individual investor can participate in CNH by establishing non-resident accounts in a country where CNH is delivered, typically through local banks that have a relationship with banks in those countries/regions.

Unlike the CNY market, the CNH market is not directly managed by the Chinese authorities and instead is determined by the supply of and demand for CNH. It is therefore a “floating” currency much like the US dollar (with the important caveat that the PBOC and SAFE regulate RMB flows between onshore and offshore accounts, and official entities may participate in the market to influence the exchange rate). Having said that, CNH and CNY are essentially the same currency, and the active international trade between mainland China and the rest of the world, as well as financial investment channels such as Stock Connect, Bond Connect, and Wealth Management Connect Program, help keep the two exchange rates closely aligned.

The Chinese government has for some time been revamping its foreign exchange mechanism in an effort to make the RMB more market-oriented and relatively stable against a basket of currencies. Between mid-2023 and mid-2024, policymakers prioritized slowing CNY depreciation against USD amid elevated capital outflow pressures, leading to CNY appreciation against the CFETS basket. Chinese exports still remain competitive in the global markets, thanks to low domestic inflation in China and falling export prices.

On August 11, 2015, the PBOC announced a major reform to the formation of the RMB’s central parity rate against the US dollar, by referring to the closing price on the inter-bank foreign exchange market of the previous day. The PBOC considered this a “one-time correction” to remedy previously accumulated differences between the central parity rate and the spot market rate. However, an abrupt weakening in the RMB occurred during the days following the announcement of this reform, triggering considerable market volatility and additional capital outflows. Starting from December 2015, the PBOC released a new index called the CFETS RMB index based upon international trade weights after adjusting for re-export factors. The basket weight is updated annually to reflect the latest trade flows of China vs. the rest of the world. Compared with CFETS, the BIS and IMF also compile RMB trade-weighted indices. Exhibit 48 shows the currency share in each basket and Exhibit 49 shows the movement of indices.

In 2012-13, the prevalence of export/import over-invoicing to bring in/move out funds to invest in the “carry trade”/transfer assets offshore distorted officially reported trade data. In an attempt to identify the underlying trends in export and import flows, we compiled an “outside-in” trade measure based on trading partners’ reported data on trade with China.

We made two improvements to our previous measures over time. First, we allow for lags to reflect shipping time in matching China’s data with trading partners’ data. Secondly, we collect data from more countries and now include 21 of China’s major trading partners that together make up 79% (67%) of the value of total Chinese exports (imports) in 2022.

Because exports are usually reported on an FOB basis and imports on a CIF basis (see the discussion above on goods trade indicators and re-exports), there will be a gap between China’s reported exports and trading partners’ reported imports from China. In addition, re-exports and delayed data availability may also create discrepancies in level terms between the China’s official trade data and our “outside-in” trade measure. However, these shouldn’t affect year-over-year changes on a persistent basis.

We collect trade data from 21 major Chinese trading partners (including the Euro area as a single “partner”). Where necessary, we convert import data into US dollars. Note that all figures discussed here are nominal dollars (not adjusted for inflation).

We use 2010-2019 monthly data to estimate the lead-lag relationship between China’s official import/export data vs. trading partner reported export/import data due to different shipping time needed for different trading partners.

Adding up the import/export data for the trading partners (after adjusting for the lead-lag relationship described above) gives us “adjusted exports/imports to/from China’s major trading partners”. Comparisons of the outside-in estimates to China’s official data are shown in Exhibit 50 and Exhibit 51.

SAFE dataset on onshore FX settlement: FX settlement and sales on behalf of clients by banks refer to the transaction of FX settlement, sales and other business conducted by banks for their clients. Deals on their own behalf and interbank market transactions are not taken into account.

Cross-border RMB flows: In late 2015 through 2016, while there was a large amount of net cross-border RMB flow from onshore to offshore, there was no corresponding observed increase in foreigners’ holdings of RMB assets. It is possible that some Chinese financial institutions buy RMB in the offshore market and either sell it back in the onshore FX market or invest it in onshore RMB assets. In addition, foreign equity and bond investments in China through the “Stock Connect” and “Bond Connect” channels are conducted in the offshore RMB market (i.e., in CNH instead of CNY). These RMB/FX transactions conducted do not necessarily show up on the PBOC’s balance sheet. Therefore, we believe that tracking the data on cross-border RMB flow (foreign-related receipts and payments reported by SAFE) is also important to developing a comprehensive view of the underlying flow picture.

“How fast are Chinese exports really growing?”, Emerging Markets Macro Daily, 17 March 2014

“How does a weaker RMB impact China credit”, Asia Credit Line, 14 August 2015

“Sources and sizes of China’s capital outflows”, Asia Economics Analyst, 26 January 2016

“China capital flows update—how cross-border RMB flow might mask outflow pressures”, Asia Economics Analyst, 4 July 2016

“Index inflows and rate differentials support our constructive view on CNY”, Asia in Focus, 20 July 2020

“Gauging Downside Risks to Chinese Exports”, Asia Economics Analyst, 6 April 2023

“Analyzing recent puzzles on China trade data”, China Data Insights, 24 May 2023

“Q&A on the recent RMB depreciation and policy reaction”, Asia in Focus, 23 August 2023

“Deciphering China’s Inward Foreign Direct Investment Data”, China Data Insights, 24 October 2023

“The Enigma of China’s FX Reserves – Size, Adequacy, and Composition”, Asia Economics Analyst, 12 January 2024

The People’s Bank of China (PBOC) reports three series on China’s money supply:

1. M0: currency in circulation (bills and coins)

2. M1: M0 + demand deposits (excluding household demand deposit)

3. M2: M1 + quasi-money (time, savings, and other deposits, excluding fiscal deposits)

China’s money supply data are subject to significant distortions, in large part arising from the attempts of financial institutions to evade regulatory controls. For example, commercial banks used to depress month-end money and credit data and minimize regulatory costs such as required reserves; the PBOC responded by changing the required data from month-end to month-average in September 2014. This alleviated the old problem but the time series is no longer directly comparable. Since 2016, the PBOC has started to assess banks’ performance under the new macroprudential assessment framework (MPA), including indicators for capital and leverage, asset and debt, liquidity, pricing, asset quality, risk of cross-border financing and the implementation of credit policy. The quarterly MPA tends to distort quarter-end data.

Changes in broad money supply provide useful leading information for short-term economic activity because banks remain the dominant financial intermediaries in China. M2 remains one of the intermediate policy targets for the central bank, but its importance has been declining due to the ongoing reform of the monetary policy framework.

Breakdowns are available from M0 to M2. M2 is the broadest measure currently available but is becoming inadequate as direct financing (such as government/corporate bond issuance) has increased in importance over the past decade.

The PBOC has expanded the definition of money supply five times due to China’s evolving financial markets. Specifically, the PBOC incorporated clients’ margin deposits into M2 on the development of the stock market in 2001. Domestic RMB deposits in foreign financial institutions have been included into money supply since 2002. From October 2011, M2 has included both non-bank financial institutions deposits and housing provident fund deposits. In January 2018, the PBOC employed money market funds held by non-bank sectors to measure money market related deposits (previously measured by deposits of money market funds) and incorporated it into M2. Since December 2022, e-CNY in circulation has been included in M0. In June 2024, PBOC governor Pan Gongsheng acknowledged that the M1 definition was outdated, and the central bank may expand the scope of M1 to include household demand deposits.

Before 2018, the government used to set a specific target for M2 growth at the annual “Two Sessions” in early March, which would usually be consistent with its desired real GDP growth and inflation targets. In recent years, the government did not set a specific target for M2 growth and only stated an objective “to keep money and credit growth broadly consistent with nominal GDP growth”. At the 2023 Central Economic Work Conference, the phrase was changed to “keeping total social financing and M2 growth broadly in line with economic growth and inflation target” amid deflationary pressures and weak nominal GDP growth. Most recently, the PBOC has started to downplay the importance of quantity-based indicators, emphasize the composition and structure of money and credit, and push forward the transition towards a more price-based monetary policy framework.

Loan data published by the PBOC include loans made by depository institutions and non-depository financial institutions (e.g., trust and insurance corporations). Recipients of loans are non-financial institutions and individuals, including non-residents. Loans to non-bank financial institutions have been added since the start of 2015. Loans denominated in CNY and foreign currencies are compiled separately and aggregated to provide total loans data.

Similarly, deposit data include deposits in CNY and other currencies. There is further information on deposits broken down by different sectors of the economy (e.g., households, government, and non-financial corporates) and by type of deposits (demand deposits, time deposits and other deposits).

Balance sheet of Monetary Authority: The main liabilities are currency issued and commercial bank reserves, and main assets come from foreign assets (funds outstanding for foreign exchange) and claims on other depository corporations (liquidity injection via monetary policy tools, such as OMO, MLF, PSL, relending, etc.).

Balance sheet of Other Depository Corporations: Other depository corporations include policy banks, commercial banks, credit unions and finance companies. Main liabilities come from deposits of households and non-financial institutions (around 60%), interbank borrowings (around 10%), and bond issuance (around 10%).

Source and uses of funds of financial institutions: This describes major sources and uses of commercial banks and other financial institutions. Sources can be deposits and financial bonds. Uses include loans and portfolio investments.

The new loans and changes in loan stock announced by the PBOC are not always consistent with each other. The main reason for this discrepancy is non-performing loan (NPL) write-offs, which have been subtracted from new loans data, while changes in loan stock are calculated after adjusting for these write-offs. It is difficult to reconcile fully the two series using announced data.

Like M2 data, loan growth data are useful for gauging the current monetary policy stance, as well as possible macro policy changes, when combined with other government-driven activity indicators such as infrastructure FAI. Given that China’s capital markets are still under-developed, bank loans are the major source of external funding for the non-financial corporate sector. However, their importance has been falling as more alternative financing methods have become available and as the PBOC gradually shifts its focus from quantity-based metrics to a price-based framework.

Loans are broken down by borrower type – households, nonfinancial enterprises and government agencies/organizations, and nonbank financial institutions. To some degree, loans are broken down by tenor as well: short-term vs. medium to long-term loans for households, and short-term loans, bill financing, vs. medium to long-term loans for nonfinancial enterprises and government agencies/organizations. In addition, loans to some specific sectors (e.g., property, green and inclusive financing) are available, and the PBOC also publishes loans by sector on an annual basis with a lag of about one year and a half. But there is insufficient detail from the PBOC’s data to allow one to identify finer categories such as LGFV borrowings. The National Financial Regulatory Administration (NFRA) releases quarterly data of NPL ratios for banks.^[22]

The PBOC has expanded its definition of non-depository financial institutions for data releases of loans and deposits over the past decades, which also contributed to the inconsistency between the level and growth rate of loans and deposits. The latest two revisions happened in 2015 and early 2023. In 2015, the PBOC incorporated loan companies into non-depository financial institutions. In early 2023, the definition was further expanded to include consumer finance companies, wealth management companies, as well as financial asset investment companies.

The PBOC publishes the initial release of new loans and deposits flows first, with detailed breakdown data (such as balance sheets of the PBOC and other depository corporations) published a few days later.

Foreign Exchange Loans include foreign-currency-dominated loans extended to domestic and foreign residents/institutions by domestic and foreign financial institutions based in mainland China.

TSF statistics are intended to be a comprehensive measure, but historically there have been multiple rounds of revision on TSF data. In 2018, asset-backed securities and loan write-offs were included in TSF, and in 2018-2019, net government bond issuance and net corporate bond issuance statistical standards were also adjusted. In April 2024, the PBOC adjusted 2024 year-to-date corporate bond net financing according to the latest industry classification results. These revisions also created occasional distortions to TSF growth measures.

The PBOC publishes the stock and flow data of TSF separately each month. However, the new TSF flows and changes of TSF stock can diverge from time to time. The divergence mainly comes from FX loans and corporate bond financing, for which the PBOC uses different statistical methodology for flow and stock data.

M2 and TSF growth can diverge from time to time, as M2/TSF refer to liabilities/assets of the financial sector’s balance sheet, respectively. By definition, TSF includes overall financing by the real economy (non-financial sectors), while M2 includes borrowings between financial institutions as well. FX purchases by the PBOC and fiscal deposit changes can affect M2 growth but not TSF growth, while net equity financing and shadow banking credit extension (undiscounted bankers’ acceptance bills, trust loans, entrusted loans) can add to TSF but not M2.

Loan prime rate (LPR): This is the interest rate banks offer to their prime clients. LPR serves as the pricing reference for bank lending, especially after the LPR reform in mid-2019, including lending to corporates and households. Currently, the LPR consists of rates with two maturities: 1-year LPR and 5-year LPR. Mortgage rates are benchmarked against the 5-year LPR. By definition, LPR is based on the quotes made by quoting banks by adding a few basis points to the interest rate of MLF. The LPR is calculated by the National Interbank Funding Center (NIFC), serving as the pricing reference for bank lending. At present, the LPR quoting banks are comprised of 20 banks. The quoting banks submit their quotes to the NIFC on the 20th day of every month (postponed in case of holidays), in increments of 0.05 percentage points. The NIFC will calculate the arithmetic average of rates after excluding the highest and lowest submissions, and find its nearest integral multiple of 0.05% to be the LPR, which will then be published at 9 a.m. on the same day. In theory, the LPR is the average interest rate submitted by large banks; in reality, the PBOC can guide banks on their quotes and therefore impact the LPR.

The rediscount rate is the rate at which central banks discount commercial banks' unexpired paper. Since this rate affects borrowing costs for commercial banks, it can be used as a policy tool. However, this rate is infrequently adjusted in China.

The re-lending rate is the rate the PBOC uses to lend to financial institutions. Loans from the PBOC are a regular source of funding for Chinese commercial banks to support areas of policy priorities (e.g., SMEs, agriculture and rural development).

The PBOC pays interest on required reserves and excess reserves that financial institutions hold at the PBOC. The interest rate for excess reserves creates the floor for China's short-term interest rate.

Deposit rate: In late 2015, the PBOC removed the official ceiling for the deposit rates, but an implicit ceiling set by “the self-regulatory pricing mechanism for market interest rates” (市场利率定价自律机制) remains in place. The implicit ceiling for the deposit rates was anchored to the benchmark deposit rates, multiplied by a designated factor. But the benchmark deposit rates have remained unchanged since 2015. In June 2021, the PBOC improved the formation of the implicit ceiling for deposit rates by shifting towards benchmark deposit rates plus a few basis points. In April 2022, under the PBOC’s instruction, the “deposit rate self-regulatory mechanism” was established. This mechanism guides banks to set deposit rates based on the 10-year central government bond yields and the 1-year LPR. This implies that when corporates and households’ funding cost is guided lower by the PBOC through lowering MLF rate and LPR, banks can also reduce their deposit rates to protect their net interest margins to some degree.

China's interbank market was established in 1996, and market participants include all financial institutions. The interbank rate is a good indicator of liquidity in the market. Interbank Offered Rates and Interbank Repo Rates are the two main rates quoted in the market.

Interbank markets have been fully liberalized, with the interest rate paid on excess reserves the lower bound and the 7-day SLF rate the upper bound for interbank rates, creating an effective, albeit wide, interest rate corridor. Repo rates are typically lower than the offered rate, as they have higher liquidity and are collateralized. The 7-day repo rate for all financial institutions (R007) is the best proxy for the overall interbank market rate.

Large and medium banks are net suppliers of liquidity in the interbank market, while small banks and non-bank financial institutions are net borrowers (Exhibit 62). Around 90% of interbank funding trades are based on repo rates. Within repo trades, around 60% of them are based on repo rates for all financial institutions (R rates), while the rest are based on repo rates for depository institutions (DR rates). Interbank term funding is confined to the front end – more than 80% of funding trades are overnight borrowings.

These accounts record financial flows amongst five key economic sectors: non-financial enterprises, the financial corporate sector, the general government sector, households, and the rest of the world (ROW). The main financial instruments covered include deposits, loans, securities, insurance technical reserves, foreign direct investment, etc. Their main value is to provide a bird’s-eye view of the interdependence between the real economy and the financial economy from a macroeconomic perspective, describing inflows, outflows, and stock of funds in the economic system.

For China, the main features shown by the flow of funds data (Exhibit 64) are that: (1) Households have been a constant supplier of funds, initially almost all to corporates, then for a time to foreigners and lately more to the Chinese government; (2) The financial deficit (i.e., when investment>savings) in the non-financial corporate sector in China gradually narrowed from -12.3% in 1992 to -2.5% in 2020 before widening to -8.2% in 2022; (3) The financial deficit of the government sector in China was relatively stable at about -1% from 1992 to 2004. However, from 2005 to 2015, the government sector changed from running a financial deficit to a surplus (except 2007), meaning that the Chinese government’s asset increases exceeded its liability increases. The government’s financial deficit reemerged after 2015 and has widened significantly since 2018, peaking at -8.2% in 2020 and remaining elevated at -6.9% in 2022; (4) The rest of the world sector has absorbed funds except for 1993, that is, there has been a net outflow of domestic funds from China, with the financial balance widening from -1.3% of GDP in 1992 to a peak of -9.8% in 2008 before narrowing to roughly -2.3% in 2022.

China’s Flows of Funds Accounts (FFA) is similar to the Japanese model, which is divided into physical transactions and financial transactions. The NBS is responsible for compiling data on physical transactions, while financial transactions are prepared by the PBOC. Due to limited original data on various sectors and industries (e.g., inventories), the classification of sectors and transaction items in the financial transaction section is usually more detailed than in the physical transaction section of FFA.

Compared with FFA in the US and Japan, FFA in China is still at the developing stage in terms of sector classification and transaction items, does not yet have quarterly or stock data, and has no reconciliation account. Moreover, the time lag of FFA releases in China is roughly two years for physical transaction tables, quite long compared with typical release lags of less than one year in other countries. Finally, China’s FFA lacks more granular decomposition of sectors.

“China’s monetary policy transmission efficiency weakened after Covid”, Asia Economics Analyst, 18 February 2022

“Navigating China rates: A shifting market to open further”, Asia Economics Analyst, 5 September 2022

“Demystifying the widening M2-TSF growth differential”, China Data Insights, 22 March 2023

“China H2 monetary policy outlook: Expecting further easing to facilitate demand stimulus”, Asia in Focus, 2 August 2023

“Understanding China’s Flow-of-Funds Accounts”, China Data Insights, 12 October 2023

“Why China’s monetary easing has become less effective post-Covid”, Asia Economics Analyst, 18 December 2023

“Demystifying China’s falling M1-M2 growth differential: Drivers and implications”, Asia in Focus, 20 January 2024

“Navigating banking system liquidity in China”, Asia Economics Analyst, 3 March 2024

“Why is the PBOC reluctant to cut policy rates?”, Asia in Focus, 2 June 2024

The Consumer Price Index (CPI) measures the price of a basket of goods and services that a typical household purchases. The basket is updated every five years. The last major update was in 2021, when the NBS updated the goods and services covered to reflect several new consumption patterns (e.g., food takeaways, new energy vehicles, ridesharing services), and adjusted the weights of each category according to an updated mix of household purchases. The latest index weights therefore may reflect some pandemic-related changes in spending patterns. Unlike most DM economies, the NBS does not publish the weighting scheme of the CPI basket. Weights mentioned in this section are our estimates (see Compilation and reporting below for details of our methodology and our most recent estimate of the basket’s composition). Category weights vary over time as prices change.

Despite some data quality issues, we believe the CPI inflation rate is largely reliable, at least in terms of the direction of change. Some common criticisms of China’s CPI—not all necessarily warranted—include:

1. Some prices, such as gasoline prices, are regulated despite being benchmarked to the broad trend of global oil prices.

2. Its components and weights are revised infrequently, so the basket composition might diverge somewhat from actual consumption behavior.

3. Compared with other countries, the NBS does not publish data on weights of different items in the CPI baskets. The NBS also does not publish “core goods” CPI inflation.

4. The NBS revised down the weight of pork prices significantly in January 2021 to reflect the fading effect of the African Swine Fever (ASF) outbreak in 2019-2020, but the magnitude of the downgrade appears too large, mitigating the impact of hog cycles on China’s reported CPI inflation.

It is standard international practice to exclude property prices from the CPI basket. The CPI basket does have a housing component, which captures price changes in rent, implied rent for self-owned houses, utilities, and real estate management fees. Inflation for private housing (implied rent for self-owned houses) has correlated closely with rent inflation (market rent) since 2011. However, home purchases are usually considered an investment following the international standard, and thus are not counted in the CPI basket.

In terms of other components, such as medical services and education, it is possible that the CPI data may fail to capture all the price changes. For example, CPI captures standardized education costs, such as school tuition fees; however, most of the “grey” charges levied by schools (tuition for supplementary classes, other administrative charges, etc.) are not included. This is likely to bias reported inflation downward at least slightly, and possibly significantly.

CPI inflation is also distorted by regulatory measures in various administered categories, including gasoline, electricity, education services, medical goods/services and telecom services. Even if global oil prices were to jump, regulatory measures would initially cap the extent of price increases faced by consumers. That said, the pass-through of international oil prices to domestic refined product prices is fairly timely and meaningful, occurring every 10 working days if changes are large enough to warrant an adjustment in the period. The NDRC requires refineries to start to reduce profit margin towards zero once global crude oil prices reach $80/bbl and caps petroleum product prices via fiscal subsidies once crude oil prices reach and/or exceed $130/bbl. There is also a price floor for refined products when crude oil prices are below $40/bbl.

The CPI basket has eight broad categories. Under each category, there are a few more detailed subcomponents (e.g., the “Transportation and communication” category is further broken down into “transportation equipment”, “vehicle fuel”, “vehicle use and maintenance”, “communication equipment”, “communication service” and “postal service”). The components of the CPI and their weights are determined by the regular household expenditure surveys and other non-regular surveys. The latest round of quinquennial updating was carried out at the beginning of 2021.

As in many other developing countries, food accounts for a relatively large share of consumer expenditure in China. However, the food share in the CPI basket has fallen as living standards in the country have improved. Currently, food (not including dining out) accounts for about 19% of the total CPI basket in China.

Pork accounts for around 2-3% of China’s CPI basket on average historically, higher than other major economies. As pork prices declined notably in 2021, the weight of pork prices fell to and has remained below 2% since mid-2022. Hog cycles can be important in driving CPI inflation. In late 2019 and early 2020, for example, an outbreak of African Swine Fever (ASF) caused pork prices to more than double, sending CPI inflation above 4%.

Data are collected at different retail outlets by statisticians 2-3 times a month. Goods that have large weights and are subject to frequent price changes, such as vegetables, are sampled every five days. On the other hand, goods/services whose prices are regulated by the government or are relatively stable, are only sampled once a month.

The NBS does not officially publish weights of the items in the CPI basket. It does mention weight changes in a few major categories following each quinquennial basket update. The NBS also comments on the contribution to year-over-year headline CPI for a few categories (mostly in food) in its monthly release, which allows us to track weights of select items more frequently. For subcomponents of each category, we follow NBS documentation and academic papers to construct a detailed weight scheme in a two-step procedure. Firstly, we use regression-based estimates and NBS releases to gauge the weights of major categories. Second, we estimate the relative importance of subcomponents in each category. China’s Yearbook of Household Survey provides shares of subcomponents in household consumption as a benchmark of CPI basket weights. That said, the weights of each category will vary over time as prices change. The coefficients only represent averaged weights over certain periods, and the actual weights may swing significantly for specific categories, such as pork prices. Exhibit 66 below summarizes our estimates of subcomponent weights in the 2016 and 2021 baskets with major categories highlighted in blue.

Core CPI is often a useful indicator of the underlying inflationary pressures because it excludes high volatility of food and energy prices. Even though China’s energy prices are regulated, they still show higher volatility than many other components. There is no separate energy component in the CPI, but energy inflation and its weight can be inferred from other data. If we estimate energy’s weight by regressing the headline CPI on food CPI, core CPI, and China domestic gasoline prices, the implied weight for energy is around 2%; if we estimate energy’s weight by summing up the estimated weights for “utilities” and “transportation fuel and parts” (which would suggest at least some over-estimation as these categories contain elements other than energy), the implied energy weight is around 7%.

The Producer Price Index (PPI) measures the price of industrial products when they are sold for the first time after production.

We believe the PPI index is generally reliable. It is based on a large sample of 20,000+ products at factory gate prices. Both large and small enterprises, i.e., those above or below the standard minimum threshold, are sampled.

PPI is more sensitive to changes in investment demand, and to some extent export demand. Therefore, it contains more information on the state of the industrial cycle than the CPI. It can also be affected by controls on production in relevant sectors, such as the “supply-side reform” in upstream sectors (e.g., coal and steel) during 2015-2017.

Currently, more than 50,000 industrial enterprises report prices twice a month (the 5th and 20th day of the month). Weights of components are determined by the sales value. The basket is adjusted every five years, though adjustments can also be made during the interim if there are rapid changes in the production pattern. Similar to CPI, weights of the PPI are not disclosed and the NBS does not publish industrial sales value by sector either. We estimate the weighting scheme based on industrial revenues data, which serve as a proxy of sales value by sector.^[24] Our estimates suggest that computer, communication & other electronic equipment takes up the largest share (around 11%), followed by the automobile sector (about 8%).

There are sub-indices of the PPI for consumer goods and producer goods. Consumer goods can be further broken down into food, clothing, daily articles, and durables. Producer goods can be further broken down into mining, raw materials, and manufacturing goods. Producer goods account for around 75% of China’s PPI basket. The other breakdown available is by industry, i.e., metallurgical, power, coal, petroleum, chemicals, machinery, building materials, timber, food, textiles, sewing, leather, paper, cultural/education articles, and others.

There has been extensive commentary by media and analysts on the relationship between the PPI and the CPI. It is commonly assumed that the gap between the two represents the producer margin. While broadly true at a very high level, the linkage is more complicated when it comes to specific industries and companies. For instance, the transmission from PPI to CPI seems to vary for different components of the CPI. PPI inflation (particularly consumer goods) seems to matter for the CPI’s goods component excluding food, but not for services.

There is another price indicator named “input price for industrial enterprises” (also known as the “raw materials purchasing price index”), which measures the change in raw material input costs. The difference between this indicator and PPI can be informative for the margins of industrial enterprises, although other factors also affect margins.

Wholesale prices of agricultural products cover major food items, including pork, vegetables, eggs, fruits, etc. Retail prices of agricultural products measure the average retail prices of major food items in 36 cities. Both are reliable and informative for high-frequency food CPI tracking.

Raw material prices measure wholesale and retail prices, which include ex-factory prices, transportation fees, profits, and taxes. Although these prices could deviate from ex-factory prices, they are still informative for PPI tracking.

Wholesale prices are compiled by the Ministry of Agriculture and Rural Affairs based on a sample of 200+ markets. The prices of fresh vegetables and fruits are weighted average indices (Passche index) based on 28 major vegetables and 9 major fruits. The NDRC collects retail prices data from supermarkets and farmers’ markets on a daily basis.

The NBS collects raw material prices from around 2,000 distributors located across the country, covering 9 major categories, i.e., ferrous metals, non-ferrous metals, chemicals, petrol/gas, coal, non-metal minerals, agricultural products, fertilizer/pesticide, and forest products. The data are published on the 4th, 14th, and 24th of each month.

The Merchandise Trade Price Index measures changes in goods export/import prices, compiled by China Customs. The price indices are based on unit value indices from detailed goods trade data.

We believe the Merchandise Trade Price Index is generally reliable, as it is based on a bottom-up aggregation of detailed goods trade data. However, as the price index is not based on a fixed basket of products, the index may change notably due to changes in the product mix of goods trade.

Given exports remain an important growth engine for China, the Merchandise Trade Price Index is helpful to gauge trade growth in real terms and the price competitiveness of Chinese exports in the global market.

Export prices are calculated on an FOB (free on board) basis, which includes costs of delivering goods onto the vessels but not further costs, such as insurance or freight. Import prices are on a CIF (cost, insurance and freight) basis.

Price indices before 2014 were reported in USD terms, but they were subsequently changed to CNY terms. It is important to convert price indices into the same currency before applying them as deflators.

Four kinds of breakdown are available based on various sector classifications: the Harmonized System (HS), the Standard International Trade Classification (SITC), the Broad Economic Categories (BEC), and China’s industrial classification for national economic activities.

“Rebound in PPI poses limited risk to China's CPI inflation”, Asia Economics Analyst, 15 January, 2017

“China: CPI inflation likely higher, but not a big constraint for monetary policy”, Asia in Focus, 20 May 2019

“China: Gauging the impact of imported inflation and supply chain disruptions on CPI and PPI inflation”, Asia in Focus, 8 April 2022

“China Post-Reopening Inflation Outlook: A Bottom-Up Approach”, Asia Economics Analyst, 3 February, 2023

Working-age population is related to the potential labor supply to the whole economy, and a shrinking working-age population (as in China in the coming years) will be a headwind to potential economic growth. The NBS also releases another set of population data called economically active population, which refers to the population aged 16 or over who are able and willing to participate in the labor market, including both employed and unemployed people in both urban and rural areas. Its historical readings are revised based on new information in population surveys. It differs from the commonly used working-age population because it does not have an upper limit on age, while the working-age population usually excludes the population aged above 64.

The size of migrant population can serve as a barometer of cyclical employment conditions given they are in general more sensitive to changes in urban labor market conditions. Migration to urban areas is more likely when urban labor market conditions are strong. New migration should slow and existing migrants may return to agricultural work in rural areas when urban employment conditions are weak.

Employment is defined as people who have worked for more than 1 hour during the surveyed week, as well as people who have positions but are on vacation/temporary leave. This series measures the overall employment of laborers 16 years of age or older in the economy based on sample surveys of the Chinese population. The NBS publishes the population and employment statistical yearbook since 1988. This yearbook provides data on employment/unemployment in urban areas with details about gender, age, industry, and education background.

There were 769mn people in China’s labor force in 2022 (including both employed and unemployed persons), accounting for 54.5% of China’s total population. Within the labor force, 60% is urban employment, 35% is rural employment, and the unemployment rate is around 5%. In the urban area, private enterprise (私营单位) workers constitute approximately two-thirds of total employment whereas non-private enterprise workers constitute the remaining one-third.

Flexible employment has seen strong expansion in China amid the development of the digital and platform economy. Flexible workers are not bound by formal employment contracts and enjoy fewer non-wage benefits compared to other workers. For example, many flexible workers are not included in public insurance schemes such as work-related injury insurance and unemployment insurance, and do not have paid maternity/paternity leaves. By the end of 2021, the number of flexible workers had reached 200 million in China, mainly in manufacturing, construction, delivery, platform livestreaming and ride-sharing companies.

This data series covers the number of persons employed in government agencies and non-private enterprises. It does not include private enterprises or self-employed individuals.

Details behind “employed persons in urban non-private units” include industry breakdown such as primary industry, manufacturing, construction, financial industry, property, wholesale and retail industry etc.

There is another set of data based on a subset of “employed persons in urban non-private units”, named “on the spot” staff (“在岗职工”). This “on the spot” dataset refers to persons who hold certain positions in the enterprise, and who still receive payments even though they are temporarily absent for reasons such as vacation, study or sickness. Compared with total employed persons in urban non-private units, this dataset excludes the retired but re-employed population, soldiers, and religious workers. There are also industry breakdowns available for “on the spot” staff, but the series is released with a significant lag (e.g., 2020 is the latest data available).

This indicator covers people who work in private enterprises/individual business, which have been registered at the local departments of industrial and commercial administration, including self-employed persons as well as helpers and hired laborers who work in individual households. There is industry breakdown information available, such as “employed persons in private enterprises and self-employed individuals” in manufacturing, construction, financial industry, wholesale and retail etc.

A subset of the “employed persons in private enterprises and self-employed individuals” dataset is “urban employed persons in private enterprises and self-employed individuals”. It covers only the portion of the employed persons in urban areas. Similar to all employed persons in private enterprises and self-employed individuals, this “urban employed persons in private enterprises and self-employed individuals” also has industry details available. In 2019, China had 405mn employed persons in private enterprises and self-employed individuals, of which 263mn were in urban areas.

This series reports the share of urban registered unemployed persons in the urban labor force. Urban registered unemployment measures the number of urban residents who are capable of working, but are out of work, want to work, and register themselves as such.

The urban registered unemployment rate series has not been updated since December 2021, although the number of urban registered unemployed individuals is still being updated annually. As of 2023, there were 10.74 million urban registered unemployed individuals.

The usefulness of the “Urban Registered Unemployment Rate” is very limited because: (1) it covers only people who have an urban registration (Hukou) and excludes a large number of migrant workers who normally live in urban areas but do not have urban registrations, and (2) many unemployed people may not have registered with government agencies. As a result, the official urban unemployment rate has been very stable at around 4%. As more people seek jobs only through online platforms rather than through registration with local job market centers, this registered unemployment rate can be more biased. Having said that, the direction of changes may still be indicative, as small deviations from the stable trend level are typically counter-cyclical and lag activity growth as one would expect.

One useful source of information on the status of the labor market is the quarterly Labor Supply and Demand in Major Cities published by the MOHRSS (Ministry of Human Resources and Social Security). The report provides data on the number of jobs offered and the number of job seekers, as well as other information on the labor market. This report was collected together with the registered unemployment rate data through affiliated regional job centers. Although the indicator does not cover the whole labor market, as many employment activities do not take place at these employment centers, it nevertheless covers a significant portion of it, and can provide some useful color on changes in the labor market.

The surveyed unemployment rate is the ratio of urban surveyed unemployment to the sum of surveyed unemployment and employment. One would be defined as “unemployed” if he/she is 16 years or older, does not have a job currently, seeks for job actively within the three months prior to the survey, and can start working within two weeks if provided a job.

Prior to the Covid pandemic, large cities tended to have stronger labor markets and lower urban unemployment rates. However, Covid controls took a heavier toll on large cities where population is denser and mobility is higher, leading to sharper increases in the unemployment rate in large cities relative to smaller cities especially in 2022. By late 2023, the pattern had finally reversed, and urban unemployment rates in large cities dipped below the national average.

The NBS used to release urban surveyed unemployment rates by two broad age groups: 16-24 year-old, and 25-59 year-old. These by age decomposition data were suspended after June 2023. The NBS resumed the release of unemployment rate data for the 16-24 age group in December 2023, but the new series excludes students from the survey and resulted in a much lower youth unemployment rate (e.g., 14.9% in December 2023 under the new definition vs. 21.3% in June 2023 under the old definition). According to the NBS, over 60% of the 16-24 year-olds in China are students. Under the new definition, the NBS releases unemployment rates for 16-24, 25-29 and 30-59 year-olds separately, around three days later than general labor market statistics.

Since January 2021, the NBS releases survey-based urban unemployment rates for people with local Hukou and those without (i.e., migrant workers) separately each month. Although the history of these series is relatively short, they show migrant workers’ unemployment rate is more cyclical. The unemployment rate of people with local Hukou increased from 5.1% in 2021 to 5.4% in 2022 before falling back to 5.2% in 2023 after the end of China’s zero-Covid policy. In contrast, the unemployment rate of migrant workers jumped from 4.9% in 2021 to 5.7% in 2022 before dropping to 4.7% in 2023.

Urban surveyed unemployment rate data shed light on the labor market conditions. When activity growth is weak, surveyed unemployment rates tend to increase. Surveyed unemployment rates show bigger swings than the registered unemployment rate and this set of data is therefore the best official data gauging unemployment pressures. Having said that, the NBS releases surveyed unemployment rates without seasonal adjustments. Due to the short history of the survey (regular publication started in 2017) and Covid distortions over the past few years, seasonal adjustments can be challenging.

One main drawback of the urban surveyed unemployment rate is the lack of coverage of the migrant worker population if they move back to rural areas. By design, the urban unemployment rate is based on urban household surveys and thus only captures the unemployed population in the urban area. When the economy is weak and unemployment pressure increases, migrant workers might return to rural areas. The urban surveyed unemployment rate therefore tends to underestimate the unemployment pressure during economic downturns.

Compilation method: The registered unemployment rate is derived from official registration, while surveyed unemployment is based on labor surveys;

Definition of unemployment: Surveyed unemployment follows ILO standards, while the registered unemployment rate is based on administrative records;

Data coverage: Registered unemployment is based on population with local Hukou, while surveyed unemployment also covers migrants in theory, though the number of migrant workers in urban areas shifts with economic cycles as discussed above.

Average/total wages cover labor compensation to all employees, including wages, bonus, subsidies, etc. in urban non-private units. They are on a pre-tax basis and include rents, utility bills, etc. paid by employers to employees. In practice, when employees receive salaries, they are after-tax and social contribution deductions. The average wage is total wages divided by the average number of employees within the reporting period.

Similar to the employment data, quarterly data were suspended after 2014 and only annual data are available since then. Note that foreign companies and large domestic companies listed overseas are included in “urban non-private” enterprises, raising the average wage of urban non-private units. In 2023, the average wage of urban non-private enterprises was RMB121,000.

As this set of data only captures wage growth for non-private units, it may fail to capture the genuine situation of the labor market. Wages in non-private units are generally viewed as more sticky and less sensitive to economic growth/labor market developments.

“Compensation to Laborers” in input-output data does not suffer from the problem of narrow coverage like the few series mentioned above, as it covers all employed persons. Furthermore, it has the advantage of including non-monetary income, such as social security contributions. However, this measure has two limitations: (1) the data are reported with a significant delay (the last reported data were for 2020), and (2) many forms of non-monetary compensations have to be estimated roughly, which affects the signal to noise ratio of the data.

The issue of under-reporting of income data also likely affects wage data, especially in terms of state sectors. In these sectors, official wages can be restricted by regulations, so there is a tendency to compensate employees via other types of payments. Furthermore, some private company owners can choose to forgo cash wages to help with the development of their own companies. This may also bias wage data to the downside.

GS China wage tracker gauges nominal wage growth based on the average reading across different wage indicators. Currently, components of this wage tracker include NBS data on urban disposable income – wage income, migrant worker income, income sentiment index from the quarterly urban depositor survey conducted by the PBOC, and hiring wage from the online recruitment company Zhaopin.com.

We first transform all input series into year-over-year growth rates and then calculate the average growth rate of these indicators. As we incorporate survey-based information as well as third-party data from the online recruitment company, our wage tracker is arguably a more comprehensive gauge of wage growth trend in China than official indicators such as the urban wage income reported by the NBS quarterly household income survey alone. Our wage tracker shows that wage growth has been decelerating from around 10% during the pre-pandemic period to around 4% in recent years, more consistent with listed companies’ reported wage growth than the official series.

“China: Labor market consequences of the virus outbreak”, Asia Economics Analyst, 6 March 2020.

“Tracking the labor market post the coronavirus”, China Data Insights, 14 June 2020.

“The implications of China's shrinking working-age population”, Asia Economics Analyst, 25 May 2021.

“Population Aging, Pension System, and Individual Retirement Savings in China”, Asia Economics Analyst, 10 February 2023.

“Why has youth unemployment risen so much in China?”, China Data Insights, 22 May 2023.

“Tracking wage growth in China”, China Data Insights, 6 June 2024.

In China, the national fiscal budget is composed of four accounts: the General Public Budget Account, the Government Managed Fund Account (GMF), the State Capital Operation Account, and the Social Insurance Fund Account. Fiscal data for the former two accounts are released monthly by the Ministry of Finance (MOF),^[28] while those for the remaining two accounts are released on an annual basis.

The fiscal balance that the government targets each year is the difference between revenue and expenditure in the General Public Budget Account.

1. On-budget fiscal revenue comes from various taxes and charges the government imposes, categorized as tax and non-tax revenue, in the MOF budget. Government borrowings (via bond issuance) were included in the revenue statistics before 1994 but have been excluded since then.

2. On-budget fiscal expenditure includes funds the government spends on goods and services that it provides, as well as on interest payments. These funds come from on-budget fiscal revenue and the proceeds raised through central government general bonds and local government general bonds.^[29] Conceptually, the fiscal balance is the difference between on-budget fiscal revenue and expenditure, although in China’s case there are some intricacies related to the fiscal deficit, especially the fiscal stabilization fund at the central government level and carryover/surplus funds at the local government level.

Top policymakers set the target for China’s official fiscal deficit at the “Two Sessions” every year, and this is directly linked to the quota for government bond net issuance under the General Public Budget Account. In other words, the official fiscal deficit (target) in value terms refers to the combined quota for the net issuance of central government general bonds (CGGB) and local government general bonds (LGGB). Unlike the official fiscal deficit, the effective fiscal deficit refers to the gap between on-budget fiscal revenue and expenditure, which equals the official fiscal deficit plus net drawdown of fiscal deposits and transfer from other fiscal accounts (i.e., the adjustment by central fiscal stabilization fund and carryover/surplus funds).

The Government Managed Fund (GMF) Account is comprised of around 20 sub-funds or sub-accounts (e.g., the railway construction fund, the national major water conservancy project construction fund), each of which is designed for a specific area. Land sales proceeds account for 82% of national GMF revenue in 2023 (vs. 86% in 2019). Moreover, land sales-related spending, of which more than 70% is for land acquisition and re-development, accounts for around 55% of national GMF expenditure, based on 2023 data (vs. 90% in 2019). Land sales revenue and expenditure are collected and spent by local governments, subject to approval by the central government.

In addition to the General Public Budget Account and GMF Account, there are two other accounts within the National Fiscal Budget – the State Capital Operation Account, and the Social Insurance Fund Account, although the operation of the latter two is less connected with the former two, and less relevant to the government’s fiscal stance.

Ranking these four Budget Accounts by revenue size based on 2023 data, the General Public Budget Account remains the largest (RMB21.7 trillion; before the adjustment by central fiscal stabilization fund and carryover/surplus funds), followed by the Social Insurance Fund Account (RMB11.1 trillion), the GMF Account (RMB7.1 trillion), and the State Capital Operation Account (RMB0.7 trillion). By expenditure size based on 2023 data, the General Public Budget Account (RMB27.5 trillion) is larger than the GMF Account (RMB10.1 trillion), followed by the Social Insurance Fund Account (RMB9.9 trillion) and the State Capital Operation Account (RMB0.3 trillion).

Fiscal data are usually considered to be among the most reliable data since mis-reporting can be financially and politically costly. However, this may not always be the case. For example, the government disclosed cases of over-reporting of fiscal revenue by Liaoning Province in 2011-2014, and by Inner Mongolia in 2016. The MOF listed four methods used: (1) fake tax collection from corporates that is refunded back; (2) over-reporting of tax revenue from state asset sales or the right to use state assets, offset by over-reporting of government expenditure, often related to these transactions; (3) over-reporting of non-tax revenue such as revenue from the rights to use state resources; and (4) outright number cooking.^[30] In recent years, the government has strengthened its scrutiny over central and local statistics agencies to protect against data fraud and falsification.

An important definition-related issue is that on-budget fiscal balance does not cover the GMF, most of which is based on land sales-related revenue and expenditure, and is important to local governments.

As China’s economic growth trended down over the past decade on multiple headwinds, the government turned to more proactive fiscal policy, through both on-budget fiscal policy and quasi-fiscal spending in areas such as infrastructure. This increased the importance of tracking off-budget spending to assess the fiscal deficit and fiscal impulse. We have tried to “augment” the official fiscal policy measures by incorporating off-budget quasi-fiscal policy to obtain a comprehensive picture of the stance of China’s fiscal authority, as described in the “GS China Augmented Fiscal Deficit (AFD)” sub-section.

Fiscal revenue growth is also affected by the degree of effort in collecting taxes, which is often counter-cyclical (e.g., local governments may strengthen tax collection efforts amid economic downturns), and includes scheduled tax cuts, rebates and deferrals, as well as profit transfer from the operation of state-owned capital.

The government has some flexibility in controlling the pace of scheduled fiscal measures within a year, especially on government bond issuance and fiscal spending, resulting in “frontloaded” and “backloaded” fiscal patterns.

Reported fiscal revenues and expenditures cover both central and local governments.

Besides government agencies, a portion of state-controlled non-governmental institutions (事业单位) are also included. These institutions are set up, owned, and fully or partly funded by the government. However, as economic reforms progressed, many of these institutions changed in nature and became self-funded and profit-making.

China’s fiscal year is the same as the calendar year. Revenue and expenditure data are currently released by the MOF monthly. Complete annual data including additional adjustments, such as those related to the fiscal stabilization fund, only become available when the Ministry of Finance reports to the National People’s Congress in March of the following year and are then released in the Statistical Yearbook and Fiscal Statistical Yearbook.

State-owned enterprises (SOEs) were an integral part of the public sector and were previously included in government finance statistics. With progress in economic reforms, SOEs have increasingly become responsible for their own financing and are now mostly excluded from government financial accounts (e.g., the operation of state sole proprietors (国有独资企业/公司) is under the management of the State Capital Operation Account).

Major tax reforms in recent decades also had a meaningful impact on on-budget fiscal revenue, including several rounds of VAT reforms that have replaced the former business tax with VAT, and have lowered the effective VAT rate for the real economy. Individual income tax (IIT) reforms have relieved tax burdens on households, by raising the threshold for IIT exemption, and introducing IIT reduction items.

Government expenditure is closely related to government consumption expenditure in GDP accounting. However, not all government expenditures are government consumption. Most notably, expenditures on capital construction, mine exploration, and new product R&D costs are counted as gross fixed capital formation (GFCF), as part of investment.

Gross local government revenue through General Public Budget and GMF accounts include three major sources, i.e., the on-budget fiscal revenue, GMF revenue (mostly land sales revenue), and transfer from the central government. Based on our estimates, the share of land sales revenue in gross local government revenue fell sharply in 2022-23 amid the prolonged property downturn and may trend lower in coming years.

In China, local governments had been prohibited to borrow directly according to the budget law until 2009, when pilot programs were initiated to allow some local governments to issue bonds. At the beginning, the central government continued to help with issuance and repayment on behalf of local governments, and the quota of issuance was limited. In late 2014, the budget law was revised to allow all local government to issue government bonds, subject to quota restrictions.

Given their limited on-budget revenue space and prohibition by law to borrow from banks, local governments turned to off-balance sheet entities (e.g., local government financing vehicles or LGFVs) as a financing and spending platform, but they had repayment obligations for a large part of those off-budget borrowings, which was technically not in line with the budget law.

In late 2014, the central government started to implement local government debt reform, and a series of documents were issued, including Document #43, which required local governments to classify proper local government debt from existing LGFV debt and also reinforced the policy of no new borrowing through LGFVs. From 2015, local governments, in theory, could only borrow by issuing bonds, and local government debts accrued before 2015 (excluding outstanding local government bonds) started to be swapped into bonds through the debt swap program. Since 2017, the government had launched several rounds of regulation on local government borrowing. These measures put additional caps on how and how much local governments could borrow apart from bonds, but local government debt continued to grow. Furthermore, the rapidly expanding Public Private Partnership (PPP) projects in 2016-17 became the substitute for financing functions that LGFVs had played, although the government has made continued efforts in containing the debt risks and improving the project implementation. There is so far no public information on the size of potential local government liabilities related to PPPs.

As of the end of 2023, outstanding (official) local government debt stood at around RMB41 trillion (32% of GDP), including RMB16 trillion in local government general bonds (LGGB), RMB25 trillion in local government special bonds (LGSB), and RMB0.2 trillion in non-bond debt which local governments are obligated to repay^[31]. Compared to LGGB, LGSB are under the management of the GMF account, and thus their issuance quota is not subject to the restrictions from the official fiscal deficit. For LGSB-funded projects, there are prerequisites for the return on capital and thus require the approval of the National Development and Reform Commission and MOF, while LGGB-funded projects do not have such restrictions and thus cover more projects related to people’s livelihood. LGSB usually has had longer tenors than LGGB in recent years.

In recent years, the central government has become more cautious on local government hidden debt and strengthened related regulations, but local government implicit debt is unlikely to be eliminated anytime soon.

Unspent local government bond (LGB) quota refers to the difference between the official limit for local government debt and the outstanding amount of LGBs, if any. In theory, there are two sources of unspent LGB quota available for use, including the quota never used before, and that used previously but renewed after debt repayment (without rolling over).

Based on our estimates, as of end-2023, the unspent LGB quota was RMB1.4 trillion (vs. RMB2.6 trillion at end-2022), including RMB680 billion for local government general bonds (LGGB) and RMB750 billion for local government special bonds (LGSB). Policymakers can utilize this buffer to raise extra-budget funding if necessary and upon approval. For example, in August 2022, the then-Premier Li Keqiang approved the use of RMB500 billion unspent LGSB quota and required local governments to fulfill the quota by end-October 2022, in an effort to support fiscal spending and offset strong growth headwinds.

However, unspent LGSB quota has been unevenly distributed across provinces. Some provinces with significant funding challenges do not have sufficient LGSB quota (e.g., Qinghai, Gansu, Heilongjiang and Jilin), while those with more quotas may not have an immediate need to raise funds (e.g., Beijing and Shanghai). As such, in 2022, policymakers assigned 30% of the RMB500 billion additional quota for regional redistribution, in favor of provinces with more ready-to-go projects.

To reduce the servicing costs for outstanding debt and restructure some non-performing debt (especially for local government implicit debt), debt swap programs could be one option. There was a large-scale local government debt swap program during 2015-18. The then-Finance Minister Lou Jiwei in August 2015 indicated that the outstanding amount of non-bond liabilities that local governments had obligations to repay was around RMB14.2 trillion. According to the National Audit Office, local governments issued a total of RMB12.2 trillion bonds (with low interest rates) to replace these non-bond debts (with high interest rates).

There was another, albeit smaller, round of local government debt swap program more recently. At the July 2023 Politburo meeting, President Xi pledged to launch “a basket of local government debt resolution plans”. Subsequently, media reports suggested that provincial governments were allowed to raise RMB1.5 trillion via refinancing bond sales, utilizing previously unspent LGGB and LGSB quota, to repay local government implicit debt. Additionally, banks were also asked to roll over part of local government implicit debt to manage potential risks.

The stance of fiscal policy is typically measured by the fiscal balance to GDP ratio, after seasonal adjustments. An increase in this ratio (i.e., a narrowing in the deficit) means a contractionary fiscal policy, and a decrease in the ratio (i.e., a widening in the deficit) indicates an expansionary fiscal policy.

In addition to on-budget fiscal tools, the government can affect the economy through off-budget activities (which are also referred to as “quasi-fiscal policy”). The most important and frequently used measure is the infrastructure investment funded by LGSB, policy banks, LGFVs, other state-owned enterprises (e.g., China Railway Corporation) and Public-Private Partnership (PPP) programs.

We “augment” the official fiscal policy measures by incorporating off-budget quasi-fiscal policy to obtain a comprehensive picture of the stance of China’s fiscal authority, and to examine its implications for the growth of the Chinese economy. Specifically, our measure of AFD is a sum of effective on-budget and off-budget fiscal deficits. We estimate the off-budget spending by major channels that finance quasi-fiscal activities. This further includes central government special bonds (CGSB), LGSB, net land sales revenue (excluding the costs of land acquisition and redevelopment), LGFV bonds, railway construction bonds, policy banks support (mostly via policy bank bonds and PBOC’s PSL), shadow banking loans, etc. As in some years there could be a significant time lag between LGSB issuance and proceeds spending (e.g., in 2021), we use a projected pace of LGSB proceeds spending as the input for LGSB in AFD estimates.^[32]

Our AFD metric is constructed with monthly frequency, in order to monitor the government’s overall fiscal stance in a timely manner and accordingly serve as a relatively reliable policy parameter input for our forecasting of the real economy. By interpolating GDP data, we are able to derive a monthly AFD-to-GDP ratio. To mitigate potential distortions from residual seasonality, we also focus on 3-month and 12-month moving averages of AFD-to-GDP ratio.

We note several caveats to this approach: First, some off-budget government financing channels may not be fully captured due to the availability of monthly data, such as LGFV loans and PPP projects. Second, the government’s off-budget activities might not be confined to infrastructure spending and the sectors we choose, and the share of government spending on debt repayment/servicing has been trending up over the past decade.

In the Government Work Report and budget report released during the “Two Sessions” every year, the government usually unveils the annual target for some key fiscal statistics, including official fiscal deficits (in both value terms and percentage of GDP terms) and LGSB net issuance quota. Official fiscal deficits are equivalent to the sum of government target for the net issuance of central government general bonds (CGGB) and LGGB.

On very rare occasions, the government may consider approving a central government special bond (CGSB) quota. This happened for the first time in 1998 to replenish the equity capital of big four banks, the second time in 2007 to establish China Investment Corp, the third time in 2020 to counteract the initial Covid outbreak, and the fourth time in 2024 for key projects and strategically important initiatives (e.g., high-tech manufacturing, urbanization, food and energy supply chains, green industries, equipment upgrade). CGSB are part of the central government official debt, but different from CGGB. Similar to local government special bonds, CGSB are managed under the Government Managed Fund (GMF) Account, outside the General Public Budget Account, and therefore their issuance does not lead to a higher official fiscal deficit.

Our AFD is similar in spirit to the IMF’s “augmented government deficit” metric.^[33] However, in recent years IMF’s measure has been larger than our estimate. Our AFD differs from the IMF measure on two major dimensions: 1) on the local government implicit debt financing (mostly through LGFVs), our measure captures LGFV bonds, net land sales revenue, and part of trust loans, to make sure our data series can be updated on a timely (monthly) manner. By comparison, the IMF measure leverages their own channel checks and projections to gauge LGFV loan financing and shadow banking financing (including trust loans, entrusted loans and PPP related debt), but this is only reported on an annual basis. 2) On the central govt implicit debt financing, our measure covers debt financing by policy banks and China Railway Corp., mainly via bond financing channels, e.g., policy bank bond net issuance, China Railway construction bond net issuance, and PBOC's pledged supplementary lending (PSL). By comparison, the IMF measure does not fully include all these funding channels.

Our proprietary Augmented Government Debt (AGD) measure aims to capture all debt raised by the broad government sector (including its agents).

We divide the AGD into two categories based on issuers: official debt and implicit debt. Official government debt includes debt directly borrowed by central and local governments, such as CGGB, CGSB, LGGB and LGSB. Implicit government debt is off the government budget, including debt of government agents such as policy banks, China Railway and LGFVs, which has implicit guarantees from either central or local governments, or at least perceived by markets as having such guarantees. Of course, some of these implicit debts are backed by high-quality assets such as railway and land, while others might not. Within the AGD, there is a very small proportion raised through offshore dollar bond markets, mainly by policy banks and LGFVs, but it has been counted into the balance sheet of these entities.

There could be some double counting issues between the above-mentioned sub-categories, as policy banks may hold some liabilities of the government, LGFVs and China Railway. To adjust for the distortion, we exclude PSL and a fraction of remaining liabilities from policy bank debts.

Based on our estimates, China’s AGD reached RMB166 trillion in 2023 (or 131% of GDP), marking a more than ten-fold increase from its 2008 level in RMB terms (RMB14 trillion, or 43% of GDP). The ratio of central to local government debts, and that of official to implicit government debts, were both close to 40% vs. 60% in 2022-23. Offshore government debt (mostly raised through offshore dollar bonds) accounted for only 1% of AGD. China’s AGD has been rising rapidly over the past decades, led mainly by implicit and local government debts.

Our estimates for implicit local government debt (or LGFV debt) are based on a bottom-up approach using firm-level balance sheet data of LGFV companies, plus some conservative assumptions. We estimate total LGFV debt reached ~RMB62 trillion at end-2023, and our estimates for historical data are broadly in line with academic and policy studies. For example, according to a team led by Mr. Zhang Xiaojing at the China Academy of Social Science, total LGFV debt ranged from RMB30 trillion to RMB50 trillion at end-2016, hinging on different definitions used. Professor Bai Chong’en at Tsinghua University estimated total LGFV debt at RMB47 trillion as of mid-2017. In its 2018 China Financial Stability Report, the PBOC cited an unnamed province as an example in highlighting local governments’ surging implicit debt: the unnamed province’s implicit debt was 80% higher than its official debt.^[34]

In general, government financing through official debt usually has lower costs and longer tenors than that through implicit debt, for both central and local governments. The financing cost for central government debt by nature is also lower than that for local government debt. Among major financing channels of local governments, LGSB has become an increasingly important funding source. The financing cost of LGSB is higher than central government bonds (CGB), policy bank bonds and LGGB, but much lower than implicit financing channels such as LGFV bonds and railway construction bonds. Moreover, CGSB and LGSB usually have a longer tenor due to their focus to fund major infrastructure projects.

“Tracking China’s fiscal stance: Beyond the official fiscal balance”, Asia Economics Analyst, 18 January 2016

“China: Fiscal stimulus a potent policy lever early this year, but moderate headwinds building in H2”, Asia in Focus, 20 June 2018

“Can Chinese fiscal policy be “more proactive” in H2?”, Asia in Focus, 29 July 2018

“China: Navigating regions with high reliance on the property sector”, Asia Economics Analyst, 16 June 2022

“China fiscal update: Borrowing more from the future to offset increased headwinds”, Asia in Focus, 19 September 2022

“China: Local government special bond, an increasingly important source of government financing”, Asia in Focus, 9 February 2023

“Population Aging, Pension System, and Individual Retirement Savings in China”, Asia Economics Analyst, 10 February 2023

“The Size, Form and Implications of China’s Growing Government Debt”, Asia Economics Analyst, 2 April 2023

“China: A likely return of PSL-backed property easing, in a different way”, Asia in Focus, 3 January 2024

“China: Beijing’s Balancing Act between Infrastructure Stimulus and LGFV Deleveraging”, Asia in Focus, 6 March 2024

“Potential Reform Remedy for China’s Fiscal Challenges”, Asia Economics Analyst, 9 June 2024