## wpiea2024200-print-pdf - Introduction

## Source details

**Canonical URL:** [wpiea2024200-print-pdf - Introduction](https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024200-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2024/english/wpiea2024200-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2024/english/wpiea2024200-print-pdf.pdf.json)

---

### Overview and context
- The LG fiscal gap reflects a long-standing misalignment between LGs’ limited revenue sources and high expenditure responsibilities across China’s five tiers of governments.
- LGs fund expenditures from local tax and non-tax revenues, central government (CG) transfers, tax refund from shared tax revenues, LG bond issuance (with CG quotas), revenues from land sales, dividend transfers from local state-owned enterprises (LSOEs), and withdrawals of LG deposits.
- CG transfers are defined here as the net CG transfers excluding the tax refund (following Ma et al., 2016).
- Pre-pandemic trends and stresses:
  - LG fiscal gaps widened over time before the COVID-19 pandemic.
  - CG transfers are a major funding source, particularly for provinces with relatively large fiscal gaps.
  - Off-budget financing (e.g., debt issued by LGFVs) is excluded from the analysis.
  - Dividend transfers from LSOEs are unlikely to be a major countercyclical source because LSOE profits are procyclical and transfers function as self-insurance.

### COVID-19 and property-market related fiscal stress (selected facts)
- In 2020 the LG fiscal gap (in percent of GDP) widened in 23 out of 31 provinces compared to 2019.
- Examples of pandemic impacts:
  - Hubei and Heilongjiang: fiscal gaps widened significantly by over 3.5 percent of provincial GDP in 2020.
  - Tibet: experienced a significant decline in its fiscal gap in 2020, driven by continued GDP growth despite lower revenues and higher expenditures.
- CG mitigation measures in 2020:
  - Issued RMB2 trillion anti-epidemic special Treasury bonds with proceeds fully distributed directly to lower-level LGs.
  - Increased the quota for LGs’ own special bond issuance in 2020 by 74 percent (or RMB1.6 trillion).
- Fiscal outcomes in 2020:
  - Average expenditure growth among all provinces declined to 3 percent in 2020 from above-8 percent in 2018–19.
  - Four provinces had lower expenditures in 2020 than in 2019.
  - 14 provinces experienced lower fiscal revenues in 2020 (vs. 6 in 2019); average revenue growth among all provinces declined to -1.6 percent from 2.7 percent in 2019.
- Property-sector tightening since 2020 led to rapid slowdown in new housing starts and housing investment and a sharp decline in LG land revenues, increasing LG financing stress and potential fiscal risks for the CG.

### Fiscal transfers as redistribution and risk-sharing
- Functions:
  - Redistribution: transfers from richer to poorer provinces to alleviate poverty and reduce income inequality.
  - Risk sharing: channel more fiscal resources to LGs more affected by regional shocks.
- Institutional patterns:
  - General transfers (including tax refund, equalization transfer, and other general transfers) accounted for over 80 percent of total transfers in 2020 and are governed by an automatic rule since 2008 that equalizes fiscal resources but does not explicitly account for regional business cycles.
  - Specific-purpose transfers are subject to administrative measures and perceived as opaque and overly complex.
- Empirical 2020 patterns:
  - Transfers per capita remained negatively correlated with provincial GDP per capita (poorer provinces received higher transfers).
  - Transfers per capita were weakly positively, rather than negatively, correlated with the change in provincial GDP growth rate in 2020; provinces with larger declines in GDP growth tended to receive lower transfers per capita.

### Empirical strategy & data
- Primary estimation approach: two-step estimation following Bayoumi and Masson (1995) and Poghosyan et al. (2016):
  - Step 1: Cross-sectional regression on average levels to estimate redistribution; slope coefficient β_redist used to measure redistribution (1 − β_redist).
  - Step 2: Fixed-effects panel regression on changes to estimate risk sharing; slope coefficient β_risk−sharing used to measure risk sharing (1 − β_risk−sharing).
- Robustness checks:
  - Two-way cluster robust standard errors (clustering by province and year).
  - Driscoll-Kraay standard errors.
  - One-step PMG estimator per Poghosyan et al. (2016) as additional robustness.
- Data:
  - Annual provincial fiscal and macroeconomic data spanning 2006 to 2020.
  - Pre-transfer aggregate provincial income = household disposable income (excluding transfer income) + LG revenue + LG extrabudgetary revenue + CG tax refund (following Du et al., 2011).
  - Post-transfer disposable provincial income = pre-transfer income + net CG transfers.
  - Net CG transfers = gross CG transfers − tax refund (from CG to province) − transfers turned over (from that province) to the CG.
  - All variables in real per capita terms, deflated by provincial CPI.

### Main empirical results (full sample 2006–20, two-step)
- Estimated β_risk−sharing = 0.83 (statistically significant at the one-percent level), implying a risk-sharing effect of 17 percent (1 − 0.83 = 0.17).
  - Interpretation: the disposable income of a province falls by 83 cents in response to a temporary 1-RMB decline in its pre-transfer income relative to the national average; 17 cents are smoothed by fiscal transfers.
- Estimated β_redist = 0.69 (statistically significant at the one-percent level), implying a redistribution effect of 31 percent (1 − 0.69 = 0.31).
  - Interpretation: a province with a 1-RMB permanently lower pre-transfer income relative to the national average would have a disposable income only 69 cents below the national average; 31 cents are smoothed by fiscal transfers.
- Robustness:
  - Two-way cluster robust and Driscoll-Kraay standard errors yield similar findings and significance.
  - One-step PMG estimator: 훽_risk-sharing = 0.82 (statistically significant), close to two-step baseline.
  - Two-step vs one-step redistribution: one-step produced redistribution effects of 28 percent (no trend) and 23 percent (with trend), average 26 percent (baseline two-step: 31 percent).
- Sub-period (2006–12 vs 2013–20):
  - Risk-sharing: 18 percent (2006–12) and 17 percent (2013–20).
  - Redistribution: 28 percent (2006–12) and 32 percent (2013–20).

### Subgroup heterogeneity: coastal vs inland regions
- Sample split: Coastal region (11 provinces) and Inland region (20 provinces).
- Group-average two-step findings:
  - Risk-sharing effects:
    - Inland region: about 35 percent.
    - Coastal region: about 3 percent.
  - Redistribution effects:
    - Inland region: 100 percent.
    - Coastal region: 11 percent.
- Robustness using national averages for each group (alternative specification):
  - Risk-sharing effects:
    - Inland region: 25 percent.
    - Coastal region: 15 percent.
  - Redistribution effects:
    - Inland region: 100 percent.
    - Coastal region: 21 percent.
- Interpretation:
  - Inland provinces are more dependent on fiscal transfers for risk sharing and redistribution; coastal provinces have stronger non-fiscal channels and self-insurance capacity.
  - Per capita fiscal transfers as percent of per capita pre-transfer provincial income increased rapidly in the inland region, median rising from 15 percent in 2006 to 29 percent in 2020.

### Comparison with other countries (selected literature comparisons)
- Cross-country ranges reported in literature (examples):
  - U.S. risk-sharing effect: 4 percent to 30 percent.
  - U.S. redistribution effect: 16 percent to 22 percent.
- China estimates from this study:
  - Fiscal risk-sharing effect: 17 percent.
  - Redistribution effect: 31 percent.
- Comparative fiscal-transfer shares (literature reports):
  - China’s fiscal transfers (including tax refund): 36 percent of the total fiscal revenues of the whole country in 2010.
  - OECD countries: roughly 10 percent in 2001.
  - India: 17 percent.
  - Argentina: 3 percent.
  - Brazil: 4 percent.

### Alternative transfer mechanism: specification and objectives
- Proposed automatic, non-regressive transfer rule (transfers only to provinces with negative idiosyncratic shocks):
  - Transfers to province i in year t, Tit:
    - Tit = 0, if εit ≥ 0
    - Tit = |εit| * GDPit / ∑ GDPit_i * τ * Tt_Total, if εit < 0
  - Where:
    - εit are the shocks for province i in year t (estimated province by province from regression ∆log(GDPit) = αi + ∑_{j=1}^2 βij ∆log(GDPi,t−j) + εit)
    - GDPit is provincial GDP
    - τ is a parameter constant across provinces (calibrates total size of alternative transfers)
    - Tt_Total is total transfers in year t
- For provinces with positive shocks in year t, transfers are zero in that year.
- Provincial shocks derived from a simple autoregressive growth regression (equation (4) in source).

### Counterfactual simulation scenarios and quantitative thresholds
- Scenarios:
  - Scenario 1 (year-by-year constraint): total alternative transfers in any year cannot exceed total actual transfers in that year; CG chooses year-specific τ to equate alternative to actual transfers each year.
  - Scenario 2 (intertemporal common borrowing): only total alternative transfers over 2006–20 constrained by total actual transfers over the same period; CG chooses time-invariant τ across years.
- Key simulation outcomes:
  - Both scenarios can completely smooth idiosyncratic regional shocks and achieve full insurance (100% risk-sharing effect).
  - Redistribution effects under the alternative rule:
    - Scenario 1: 25 percent.
    - Scenario 2: 27 percent.
    - Actual redistribution effect: 31 percent.
  - Additional intertemporal borrowing in Scenario 2 is not necessary to reach 100% risk-sharing because horizontal redistribution across provinces is the main driver.
  - The alternative rule reaches full insurance when total transfers exceed a threshold of 2 percent of total GDP over the sample period.
  - Risk-sharing effect increases sharply with total transfers up to roughly 2 percent of GDP; beyond 2 percent, risk-sharing reaches 100 percent.
  - Relationship between total transfers and redistribution effect is approximately linear over the 0.5–20 percent of total GDP range simulated.
- Quantitative figures:
  - Total actual transfers over 2006–20: about 6½ percent of total GDP.
  - Threshold for full insurance under alternative rule: 2 percent of total GDP (over 2006–20).
  - Simulated rule-based redistribution effect at actual-transfer-equivalent (6½ percent of GDP): ≈ 27 percent; rule-based risk-sharing effect = 100 percent.

### Policy implications and recommended adaptation pathways
- There is scope to further enhance the risk-sharing effect of fiscal transfers in China to improve fiscal stabilization.
- Design recommendations:
  - Introduce an automatic, non-regressive transfer mechanism linking transfers to province’s share of national GDP and the size of idiosyncratic shocks (provincial GDP shock or local output gap).
  - Practical first step: incorporate provincial cyclical position into the existing automatic rule for general transfers by including size of provincial GDP shocks or local output gap as an additional driver of the transfer coefficient ρit in Tit = (Expit_standard − Revit_standard) * ρit.
  - Use intertemporal CG borrowing (e.g., common borrowing, special Treasury bonds) during periods when CG funding is constrained.
  - Complement transfer reforms with broader intergovernmental fiscal reforms:
    - Reassess expenditure assignment across government levels (referencing SC reform plan considerations).
    - Consider centralized pension and unemployment insurance systems.
    - Give LGs more authority over some tax rates and bases (e.g., personal income and property taxes) to strengthen fiscal accountability.
    - Align LG borrowing limits with expenditure responsibilities; contain off-budget investment; remove implicit guarantees; bring non-commercial investment on budget.
  - Reduce local protectionism and LG interventions to increase factor mobility and strengthen market/financial channels for risk sharing.
- Moral hazard and governance:
  - Higher fiscal risk sharing can create implicit bailout expectations and weaken LG incentives.
  - Policy response: remove explicit and implicit government guarantees for LSOEs and strengthen governance rules guiding LG fiscal decisions, including preventing use of transfers to protect weak LSOEs.

### Caveats, limitations, and robustness notes
- Results are from a partial equilibrium exercise and do not account for behavioral responses of households and LGs or potential aggregate price/borrowing-cost effects.
- Subject to Lucas’ Critique: implementing the alternative mechanism would change economic structure and could undermine robustness of estimates.
- Small sample-size concerns and potential small-sample bias; usual definition of full risk sharing (insignificant coefficient) may be misleading.
- One-step PMG estimator assumptions (large N, large T) limited application to sub-samples.

### Annex I — Data description (selected points)
- Dataset scope: annual provincial panel for the 31 provinces; sources span 2006-18 in Annex description and the main analysis covers 2006–2020.
- Main data sources: Ministry of Finance (MOF) of China, National Bureau of Statistics (NBS) of China, CEIC database.
- Construction of net CG transfers: gross transfers minus tax refund and transfers from LGs back to CG.
- Data adjustments and imputation:
  - Tax refund data available at provincial level only since 2015; provincial tax refund for 2006-14 estimated by assuming each province’s share remained the same as in 2015.
  - Household disposable income provincial series available only from 2013 onwards; earlier years estimated as weighted averages of urban and rural series using population shares.
  - Pre-transfer provincial income constructed by adding LG revenue, LG extrabudgetary revenue (before 2011), and CG tax refund to pre-transfer household disposable income; post-transfer provincial income adds net fiscal transfers from CG to LGs.
- Caveat: breakdown of disposable income between regular and transfer income is limited; removing total transfer income from disposable income to obtain pre-transfer household disposable income may underestimate true value.

_italic_Source: IMF Working Paper “How to Further Improve Fiscal Risk Sharing in China?” (provincial dataset 2006–2020; counterfactual simulations and robustness checks reported in Annex II)_

### Introduction ...........................................................................................................

### wpiea2024200-print-pdf - Introduction ...........................................................................................................

### Introduction
- Introduction ........................................................................................................................................................... 2

### Fiscal Transfers and Risk Sharing
- Fiscal Transfers and Risk Sharing ........................................................................................................................ 6

### Empirical Strategy & Data
- Empirical Strategy & Data ................................................................................................................................... 10

### Results
- Results ................................................................................................................................................................ 11

### An Alternative Transfer Mechanism & Counterfactual Simulations
- An Alternative Transfer Mechanism & Counterfactual Simulations ..................................................................... 16

### Concluding Remarks & Policy Considerations
- Concluding Remarks & Policy Considerations .................................................................................................... 20

### Annex I. Data Description
- Annex I. Data Description ................................................................................................................................... 23

### Annex II. Regression Results
- Annex II. Regression Results .............................................................................................................................. 24

### References
- References .......................................................................................................................................................... 28

- IMF WORKING PAPERS How to Further Improve Fiscal Risk Sharing in China?  
- INTERNATIONAL MONETARY FUND 2

*Source: wpiea2024200-print-pdf - Introduction (IMF PDF table of contents).*

### Introduction

### Introduction

### Overview
- The gap between local revenues and expenditures of China’s local governments (LGs) had been on an increasing trend in the last two decades before the COVID-19 pandemic. This gap is referred to as the LG fiscal gap.
- The LG fiscal gap mainly reflects a long-standing misalignment between LGs’ limited revenue sources and high expenditure responsibilities in China’s five tiers of governments.
- LGs rely heavily on funding sources other than local tax and non-tax revenues to cover expenditures, including central government (CG) transfers, tax refund from the shared tax revenues between CG and LGs, LG bond issuance (with quotas set by the CG), revenues from land sales, dividend transfers from local state-owned enterprises (LSOEs), and withdrawals of LG deposits.
- Following Ma et al. (2016), CG transfers are defined in this paper as the net CG transfers excluding the tax refund.

### LG fiscal gaps and key funding sources (pre-pandemic)
- LG fiscal gaps widened over time before the COVID-19 pandemic (figures and provincial rankings described in the source).
- CG transfers are a major source of funding, particularly for provinces with relatively large fiscal gaps.
- Off-budget financing (e.g., debt issued by local government financing vehicles, LGFVs) is excluded from the analysis here.
- On dividend transfers from LSOEs: they are unlikely to be a major source for LGs’ general budget financing during economic downturns because (i) LSOE profits tend to be procyclical and (ii) transfers are still a self-insurance scheme less effective than CG transfers from a risk-sharing perspective.

### COVID-19 pandemic effects on LG finances (2020)
- In 2020 the LG fiscal gap (in percent of GDP) widened in 23 out of 31 provinces compared to 2019.
- Provinces with more COVID-19 infections tended to have experienced higher increases in their fiscal gaps. Examples:
  - Hubei and Heilongjiang: fiscal gaps widened significantly by over 3.5 percent of provincial GDP.
  - Tibet: experienced a significant decline in its fiscal gap in 2020, driven by continued GDP growth despite lower revenues and higher expenditures.
- To mitigate the pandemic impact:
  - The CG issued RMB2 trillion anti-epidemic special Treasury bonds with proceeds fully distributed directly to lower-level LGs earlier than in normal years.
  - The CG increased the quota for LGs’ own special bond issuance in 2020 by 74 percent (or RMB1.6 trillion).
- Despite these measures, signs indicate some LGs’ ability to conduct countercyclical fiscal policy was constrained:
  - Average expenditure growth among all provinces declined to 3 percent in 2020 from above-8 percent in 2018–19.
  - Four provinces had lower expenditures in 2020 than in 2019.
  - 14 provinces experienced lower fiscal revenues in 2020 (vs. 6 in 2019); average revenue growth among all provinces declined to -1.6 percent from 2.7 percent in 2019.

### Property market stress and fiscal implications
- Regulatory tightening in the property sector since 2020 led to a rapid slowdown in new housing starts and housing investment and a sharp decline in LG land revenues.
- Falling land sale revenues reduced LG fiscal capacity while LGFVs significantly increased land purchases.
- A new funding mechanism for completion of troubled unfinished housing projects, although funded by the CG, requires LGs to backstop housing completion loans; several highly-indebted regions have large stocks of unfinished housing, adding financing pressure on LGs.
- LG financing stress can create significant fiscal risks for the CG by affecting service delivery and placing continuous pressure on central resources.

### CG transfers as redistribution and risk-sharing mechanism
- CG transfers perform redistributive and risk-sharing roles:
  - Redistribution: transfers from richer to poorer provinces to alleviate poverty and reduce income inequality.
  - Risk sharing: channel more fiscal resources to LGs more affected by regional shocks from those less affected.
- In China, the fiscal transfer system has primarily focused on redistribution.
  - General transfers (including tax refund, equalization transfer, and other general transfers) accounted for over 80 percent of total transfers in 2020 and are governed by an automatic rule since 2008 designed to equalize fiscal resources among regions; the rule does not explicitly or adequately account for regional business cycles for risk-sharing.
  - Specific-purpose transfers (conditional transfers) are subject to administrative measures and perceived as opaque and overly complex.
- Empirical patterns in 2020:
  - Fiscal transfers per capita remained negatively correlated with provincial GDP per capita (poorer provinces received higher transfers).
  - Fiscal transfers per capita in 2020 were weakly positively, rather than negatively, correlated with the change in provincial GDP growth rate in 2020; provinces with larger declines in GDP growth in 2020 tended to receive lower transfers per capita, indicating a limited countercyclical/risk-sharing role in that year.

### Empirical strategy & data
- Primary estimation approach: two-step estimation following Bayoumi and Masson (1995) and Poghosyan et al. (2016):
  - Step 1: Cross-sectional regression on average levels of regional variables to estimate redistribution effect; slope coefficient β_redist used to measure redistribution (1 − β_redist).
  - Step 2: Fixed-effects panel regression on changes of regional variables to estimate risk-sharing effect; slope coefficient β_risk−sharing used to measure risk sharing (1 − β_risk−sharing).
- Robustness checks:
  - Two-way cluster robust standard errors (clustering by province and year).
  - Driscoll-Kraay standard errors (robust to heteroskedasticity, autocorrelation, cross-sectional dependence).
  - One-step approach (PMG estimator) per Poghosyan et al. (2016) as an additional robustness check that estimates redistribution and risk sharing jointly.
- Data:
  - Annual provincial fiscal and macroeconomic data spanning 2006 to 2020.
  - Pre-transfer aggregate provincial income defined as sum of household disposable income (excluding transfer income), LG revenue, LG extrabudgetary revenue, and CG tax refund (following Du et al., 2011).
  - Post-transfer disposable provincial income equals pre-transfer income plus net CG transfers.
  - Net CG transfers calculated by subtracting tax refund (from CG to province) and transfers turned over (from that province) to the CG from gross CG transfers.
  - All variables in real per capita terms, deflated by provincial CPI.

### Main empirical results
- Baseline (two-step approach, full sample 2006–20):
  - Estimated β_risk−sharing = 0.83 (statistically significant at the one-percent level), implying a risk-sharing effect of 17 percent (1 − 0.83 = 0.17).
    - Interpretation: the disposable income of a province falls by 83 cents in response to a temporary 1-RMB decline in its pre-transfer income relative to the national average; 17 cents are smoothed by fiscal transfers.
  - Estimated β_redist = 0.69 (statistically significant at the one-percent level), implying a redistribution effect of 31 percent (1 − 0.69 = 0.31).
    - Interpretation: a province with a 1-RMB permanently lower pre-transfer income relative to the national average would have a disposable income only 69 cents below the national average; 31 cents are smoothed by fiscal transfers.
- Robustness:
  - Two-way cluster robust and Driscoll-Kraay standard errors yield similar findings and levels of statistical significance.
- Comparison to earlier literature:
  - Du et al. (2011) found an effect of about 9 percent (on consumption risk sharing for 1980–2007); differences likely due to focus on income risk sharing and a more updated sample (2006–20).
- Sub-period analysis (two-step approach; sub-samples 2006–12 and 2013–20):
  - Risk-sharing effect: 18 percent during 2006–12 and 17 percent during 2013–20 — broadly stable over time.
  - Redistribution effect: increased from 28 percent in 2006–12 to 32 percent in 2013–20 — suggesting an increased emphasis on equalizing fiscal resources across regions over time.

### Proposed alternative transfer mechanism and simulations
- The paper proposes an alternative transfer mechanism that is automatic and non-regressive with transfers proportional to the size of idiosyncratic provincial shocks (similar to rule proposed by Furceri and Zdzienicka, 2015, for the euro area).
- Counterfactual simulations show such a mechanism could:
  - Significantly increase interprovincial risk sharing.
  - Keep the redistribution effect close to the actual level observed under the existing system.

### Policy implications highlighted
- There is room to further enhance the risk-sharing effect of fiscal transfers in China to improve the effectiveness of fiscal policy as a stabilization tool.
- Increasing the risk-sharing component of transfers would help mitigate idiosyncratic regional shocks, stabilize local economies, and improve national-level fiscal stabilization (especially where output gaps are largest and multipliers higher).
- A well-designed, automatic, and transparent transfer rule that better accounts for regional business cycles could increase countercyclical support without materially undermining redistribution objectives.
- Consideration of moral hazard: transfer-based risk sharing can create implicit bailout expectations; transfer design should balance risk-sharing gains with incentives for sound local policies.

*IMF Working Paper — Introduction (wpiea2024200-print-pdf)*

### 3. This confirms our earlier observation that there could be room to increase the fiscal risk-sharing effect during

### 3. This confirms our earlier observation that there could be room to increase the fiscal risk-sharing effect during the pandemic.

### Results for sub-groups
- Sample split: 31 provinces divided into two groups following Du et al. (2011):
  - Coastal region (11 provinces): Beijing, Fujian, Guangdong, Hainan, Hebei, Jiangsu, Liaoning, Shandong, Shanghai, Tianjin, and Zhejiang.
  - Inland region (20 provinces): the remaining 20 provinces (less developed, lagging in marketization and economic development).
- Estimation approach: two-step regressions using group averages for 푌퐷_t̅̅̅̅̅, 푌_t̅, 푌퐷̿̿̿̿, and 푌̿.
- Core findings (group-average specification):
  - Risk-sharing effects:
    - Inland region: about 35 percent.
    - Coastal region: about 3 percent.
  - Redistribution effects:
    - Inland region: 100 percent (permanent provincial shocks completely smoothed out by fiscal transfers).
    - Coastal region: 11 percent.
- Robustness using national averages for each group (alternative specification) — qualitative pattern holds but differences shrink:
  - Risk-sharing effects:
    - Inland region: 25 percent.
    - Coastal region: 15 percent.
  - Redistribution effects:
    - Inland region: 100 percent.
    - Coastal region: 21 percent.
- Interpretation:
  - Less developed inland provinces are more dependent on fiscal transfers for both risk sharing and redistribution.
  - Coastal provinces have better-functioning non-fiscal channels (e.g., more developed financial intermediaries and markets) and better self-insurance capacity on average, reducing reliance on fiscal channels.
  - Coastal provinces are major sources of tax revenues for the CG (Du et al., 2011).
- Contrast with earlier literature:
  - Du et al. (2011) found larger consumption risk sharing through fiscal transfers in the coastal region during 1980-2007, attributing inland region’s smaller effect to small fiscal transfers in that period.
  - In the more recent sample period used here, per capita fiscal transfers as percent of per capita pre-transfer provincial income increased rapidly in the inland region and remained much higher than in the coastal region, with the median rising from 15 percent in 2006 to 29 percent in 2020.

### Comparison with other countries
- Cross-country evidence shows wide variation in estimated fiscal risk-sharing and redistribution effects, even among advanced economies (AEs).
- Example ranges from literature for the U.S.:
  - Risk-sharing effect: 4 percent to 30 percent.
  - Redistribution effect: 16 percent to 22 percent.
- China estimates from this study:
  - Fiscal risk-sharing effect: 17 percent.
  - Redistribution effect: 31 percent.
- Interpretation and caveat:
  - China’s 17 percent risk-sharing effect lies within the AE range; the 31 percent redistribution effect is on the high end.
  - High redistribution does not imply equivalence in efficiency to AEs because China’s fiscal transfers (relative to total fiscal revenues) are significantly higher than in international peers.
- Comparative fiscal-transfer shares (reported in literature):
  - China’s fiscal transfers (including tax refund) accounted for 36 percent of the total fiscal revenues of the whole country in 2010 (Zhang, 2012; Joumard and Kongsrud, 2003).
  - OECD countries: roughly 10 percent in 2001.
  - India: 17 percent.
  - Argentina: 3 percent.
  - Brazil: 4 percent.

*IMF WORKING PAPERS — How to Further Improve Fiscal Risk Sharing in China? — INTERNATIONAL MONETARY FUND*

### 1997. This suggests that, despite having a similar overall fiscal risk-sharing effect, the efficiency of the risk-

### wpiea2024200-print-pdf - 1997. This suggests that, despite having a similar overall fiscal risk-sharing effect, the efficiency of the risk-

### Key empirical findings
- Using provincial-level data for 2006–2020, estimated effects of actual fiscal transfers in China:
  - Risk-sharing effect: 17%
  - Redistribution effect: 31%
- The total actual transfers over 2006–20 are equivalent to about 6½ percent of total GDP.
- The fiscal risk-sharing effect in China has remained broadly stable between sub-periods 2006–12 and 2013–20, unlike many AEs where the effect generally declined over time.
- The fiscal risk-sharing effect is more prominent among inland provinces and less significant in coastal provinces.
- Comparison with selected estimates in advanced economies (literature-based):
  - U.S. estimates vary (e.g., 30% risk-sharing effect in Bayoumi and Masson (1995); 4% in Furceri et al. (2020); 8% average in Poghosyan et al. (2016))
  - Canada and U.K. estimates shown with varying risk-sharing and redistribution effects in the table.

### Robustness checks and estimation notes
- One-step approach (following Poghosyan et al. (2016)) run for full sample produced 훽_risk-sharing = 0.82 (statistically significant) in both regressions (without and with trend), close to two-step baseline estimate of 0.83.
- Two-step approach (one-step applied as robustness) produced redistribution effects of 28 percent (no trend) and 23 percent (with trend), average 26 percent (baseline two-step: 31 percent).
- Pre-COVID sample (2006–19) results: risk-sharing effect = 14 percent (slightly below 17 percent full sample); redistribution effect = 31 percent (same as full sample).
- The one-step approach was not applied to sub-samples due to large N and large T assumptions of the PMG estimator.

### Alternative transfer mechanism — specification
- Proposed non-regressive, automatic transfer rule (transfers only to provinces with negative idiosyncratic shocks):
  - Transfers to province i in year t, Tit:
    - Tit = 0, if εit ≥ 0
    - Tit = |εit| * GDPit / ∑ GDPit_i * τ * Tt_Total, if εit < 0
  - Where:
    - εit are the shocks for province i in year t (estimated province by province from regression ∆log(GDPit) = αi + ∑_{j=1}^2 βij ∆log(GDPi,t−j) + εit)
    - GDPit is provincial GDP
    - τ is a parameter constant across provinces (calibrates total size of alternative transfers)
    - Tt_Total is total transfers in year t
- For provinces with positive shocks in year t, transfers are zero in that year.
- Provincial shocks derived from simple autoregressive growth regression (equation (4) in source).

### Counterfactual scenarios simulated
- Scenario 1: Year-by-year constraint — total alternative transfers in any year cannot exceed total actual transfers in that year. The CG can choose a year-specific τ to equate alternative to actual transfers each year.
- Scenario 2: Intertemporal common borrowing — only the total size of alternative transfers over 2006–20 is constrained by total actual transfers over the same period. The CG can choose a time-invariant τ across years.
- Key simulation outcomes:
  - Both scenarios can completely smooth idiosyncratic regional shocks and achieve full insurance (100% risk-sharing effect).
  - Redistribution effects under the alternative rule:
    - Scenario 1: 25 percent
    - Scenario 2: 27 percent
    - Actual redistribution effect: 31 percent
  - Additional intertemporal borrowing in Scenario 2 is not necessary to reach 100% risk-sharing because horizontal redistribution across provinces is the main driver.
  - The alternative rule reaches full insurance when total transfers exceed a threshold of 2 percent of total GDP over the sample period.
  - Risk-sharing effect increases sharply with total transfers up to roughly 2 percent of GDP; beyond 2 percent, risk-sharing effect reaches 100 percent.
  - Relationship between total transfers and redistribution effect is approximately linear over the 0.5–20 percent of total GDP range simulated.

### Quantitative thresholds and illustrative numbers
- Total actual transfers over 2006–20: about 6½ percent of total GDP.
- Threshold total transfers to achieve full insurance under alternative rule: 2 percent of total GDP (over 2006–20).
- Example alternative totals simulated: range from 0.5 to 20 percent of total GDP (over 2006–20).
- Simulated redistribution effects at actual-transfer-equivalent (6½ percent of GDP): rule-based redistribution effect ≈ 27 percent; rule-based risk-sharing effect = 100 percent.

### Policy recommendations and adaptation pathways
- Introduce an automatic, non-regressive transfer mechanism linking transfers to:
  - Province’s share of national GDP
  - Size of idiosyncratic shocks (provincial GDP shock or local output gap)
- Practical adaptation approach:
  - Incorporate provincial cyclical position into existing automatic rule for general transfers by including size of provincial GDP shocks or local output gap as an additional driver of the transfer coefficient ρit in Tit = (Expit_standard − Revit_standard) * ρit.
- Use intertemporal CG borrowing (e.g., common borrowing, special Treasury bonds) during periods of stress when CG funding is constrained.
- Complement transfer reforms with broader intergovernmental fiscal reforms:
  - Reassess assignment of expenditures to appropriate government levels (SC reform plan considerations).
  - Consider centralized pension and unemployment insurance systems.
  - Provide LGs with more authority over some tax rates and bases (e.g., personal income and property taxes) to strengthen fiscal accountability.
  - Align LG borrowing limits with expenditure responsibilities; contain off-budget investment; remove implicit guarantees; bring non-commercial investment on budget.
- Reduce local protectionism and LG interventions to increase factor mobility and market/financial channels for risk sharing.

### Caveats, risks, and governance considerations
- Results are from a partial equilibrium exercise and do not account for behavioral responses of households and LGs or potential aggregate price/borrowing-cost effects.
- Subject to Lucas’ Critique: implementing the alternative mechanism would change economic structure and could undermine robustness of estimates.
- Small sample-size concerns and potential small-sample bias; usual definition of full risk sharing (insignificant coefficient) may be misleading.
- Moral hazard risks:
  - Higher fiscal risk sharing may weaken LG incentives and worsen LG fiscal positions over time.
  - Bailout guarantees could create moral hazard for LSOEs and local state-owned banks by reducing borrowing costs and encouraging risk-taking.
  - Policy response: remove explicit and implicit government guarantees for LSOEs and strengthen governance rules guiding LG fiscal decisions, including preventing use of transfers to protect weak LSOEs.

_italic_Source: IMF Working Paper “How to Further Improve Fiscal Risk Sharing in China?” (provincial dataset 2006–2020; counterfactual simulations and robustness checks reported in Annex II)

### Annex I. Data Description

### Annex I. Data Description

### Dataset scope
- Annual provincial panel data for the 31 provinces of China covering 2006-18.
- Main set of variables:
  - fiscal transfers from CG to each province,
  - fiscal revenues and expenditures of each province,
  - provincial macroeconomic variables including nominal GDP and inflation,
  - aggregate provincial income per capita including both pre- and post-transfer income per capita.

### Main data sources
- Ministry of Finance (MOF) of China
- National Bureau of Statistics (NBS) of China
- CEIC database

### Construction of fiscal transfers from CG to provinces
- Fiscal transfers from CG to each province are calculated by taking out the tax refund and the transfers from LGs back to CG from the gross transfer data.
- The gross transfer data include three components: transfers from CG to LGs, tax refund from CG to LGs on shared taxes, and transfers from LGs back to CG.
- The first two components are removed from the gross transfer data to calculate the net transfers from CG to LGs.
- Data availability note for tax refunds:
  - Tax refund data are only available at the provincial level since 2015.
  - Since the total tax refund for all provinces are available since 2006, the provincial tax refund for 2006-14 is estimated by assuming that the share of each province’s tax refund remained the same during 2006-14 as in 2015.

### Provincial fiscal revenue and expenditure
- Provincial fiscal revenue and expenditure data, including both LG revenue and LG extrabudgetary revenue (before 2011), are obtained from the MOF and CEIC database.

### Provincial macroeconomic and demographic variables
- Provincial nominal GDP, inflation, and urban and rural population are obtained from the CEIC database.

### Provincial income per capita: construction and caveats
- Provincial income per capita data include both pre- and post-transfer income per capita.
- All income variables are calculated in per capita terms using the provincial population.
- Breakdown of disposable income between regular and transfer income is available and collected from NBS’ yearbooks during 2006-20.
- The transfer income includes both transfer income from fiscal transfers to individuals and transfer income from remittances from migrant workers.
- Given the lack of the breakdown between the two types of transfer incomes, pre-transfer household disposable income is calculated by removing the total transfer income from the disposable income; this procedure may however underestimate the true value.
- Following Du et al. (2011), the pre-transfer aggregate provincial income is calculated by adding the LG revenue, LG extrabudgetary revenue (before 2011), and CG tax refund to the pre-transfer household disposable income.
- The post-transfer aggregate provincial income is calculated as the sum of the pre-transfer aggregate provincial income and the net fiscal transfers from CG to LGs.

### Additional data imputation for household disposable income (note 24)
- Provincial-level data for household disposable income are only available from 2013 onwards.
- Household disposable income is available separately for urban and rural households for years before 2013.
- Household disposable income for earlier years is estimated as a weighted average of the urban and rural household disposable incomes using the share of urban and rural population as the weights, following Hao and Wei (2009).

*Source: Annex I. Data Description, IMF Working Paper "How to Further Improve Fiscal Risk Sharing in China?"*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024200-print-pdf.pdf_