## gsnea2021001 - INTRODUCTION

## Source details

**Canonical URL:** [gsnea2021001 - INTRODUCTION](https://www.imf.org/-/media/files/publications/gfs-notes/2021/english/gsnea2021001.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/gfs-notes/2021/english/gsnea2021001.pdf.md)
- [Structured JSON version](/-/media/files/publications/gfs-notes/2021/english/gsnea2021001.pdf.json)

---

### Overview
- The commercial real estate (CRE) sector in the United States was hit hard during the COVID-19 pandemic as demand for contact-intensive CRE spaces (traditional brick-and-mortar retail, restaurants, hotels, and offices) plunged in 2020.
- CRE transaction volumes and prices declined sharply, particularly in the hotel, retail, and office segments; the decline in the United States was more pronounced than in other regions.
- Preexisting structural trends—digitalization, e-commerce, and working-from-home—accelerated during the pandemic, creating uncertainty about the outlook and suggesting further price declines may be possible in some CRE segments.
- Banks are the largest providers of CRE debt financing in the US (share of total outstanding CRE debt held by banks is 54 percent in the US), while nonbank financial institutions hold 24 percent in the US.

### Research questions and methodology
- Investigates:
  - How does a decline in CRE prices affect banks’ profitability and solvency?
  - How large bank capital losses could be if CRE prices were to remain depressed permanently?
- Analysis steps:
  - Empirically identify effect of CRE loan exposure on bank performance using US bank-level data and location-specific CRE prices (2001–2020).
  - Use estimates to conduct a capital loss scenario assessing credit and revenue losses and potential bank capital losses based on a forward-looking provisioning rule.
- Data limitations: supervisory loan-level data and real-time appraisals are not publicly available; analysis relies on bank-level Call Reports and MSA-level CRE prices from MSCI.

### Key findings
- Banks with higher ex-ante CRE loan exposures perform considerably worse after a decline in CRE prices:
  - Significantly higher CRE non-performing loans ratio, higher CRE loan charge-off rates, lower revenues, and lower capital.
- Quantitative sensitivities (following a one standard deviation decline in local CRE prices over an eight quarter horizon—corresponding to a 16 percent cumulative decline in CRE prices):
  - Banks at the 75th percentile of CRE loan exposure experience an 8 percentage points higher non-performing CRE loan ratio and a 3 percentage points higher CRE loan charge-offs rate (compared to banks with no CRE loan exposure).
  - Their pre-provision net revenues and total regulatory capital are lower by 12 percent and 5 percent, respectively.
- Robustness: results broadly similar when focusing on small banks or a shorter (four quarter) horizon.

### Bank financing of the CRE sector
- CRE lending importance:
  - On average, CRE loans accounted for nearly half of total business loans made by banks over the last decade.
  - Total CRE loans surpassed $2.3 trillion prior to the pandemic.
- Distribution by bank size:
  - Small- and medium-sized banks (total assets < $100 billion) extended a disproportionately large share of CRE loans.
  - Small banks (total assets not exceeding $5 billion during the sample) collectively hold about two-thirds of overall CRE loans while accounting for about 30 percent of total banking sector assets.
  - CRE loans constitute about 6 percent of total assets for large banks and 30 percent for small banks.
- Historical context:
  - High CRE loan exposures were a key determinant of bank failures after the global financial crisis when overall CRE prices declined by 30 percent.

### CRE price changes and bank stability (empirical evidence)
- Declines in CRE prices lead to:
  - Higher cumulative CRE nonperforming loan rates and higher cumulative CRE net loan charge-off rates for banks with high CRE exposure (effects shown for 8 quarters ahead and 4 quarters ahead).
  - Lower net revenues before provisioning and lower total regulatory capital for banks with high CRE exposure.
- Definitions used:
  - High CRE exposure = 75th percentile CRE-loans-to-total-assets ratio (43 percentage points exposure).
  - Low CRE exposure = 25th percentile (16 percentage points exposure).
- Statistical significance: all estimated cumulative effects reported are statistically significant at 10 percent or less.

### Bank capital adequacy under CRE sector stress — scenarios and results
- Two stressed CRE price scenarios (based on IMF (2021) valuation model):
  - Mild scenario: cumulative one-standard-deviation decline in CRE prices after eight quarters; corresponds to a 1¼ percent permanent increase in the CRE vacancy rate.
  - Severe scenario: permanent increase in CRE vacancy rate by 5 percentage points (rate observed during the global financial crisis); implies a decline in CRE prices by about 30 percent in the long-run.
- Marginal-impact (no other macro factors):
  - Average projected drop in capital adequacy is 0.14 percentage point on average.
  - Projected credit and revenue losses could reach more than 1 percent of pre-shock total risk-weighted assets for banks with very high CRE loan exposures (top 3 percent of CRE loan exposure).
  - Nearly all banks in the left tail (5th percentile of the distribution of loss ratio) are small or community banks.
  - Under the severe scenario, average capital losses are almost twice as large as under the mild scenario.
- Incorporating macroeconomic/location-specific effects:
  - Average decline in capital adequacy increases to 1.4 percentage points.
  - Banks in the left tail of the loss distribution are predominantly small or community banks; their losses reach 1.5 percentage points in this scenario.

### Structural shifts in the CRE sector and bank provisioning
- Structural shifts (rise of e-commerce and teleworking) can have stronger valuation effects in:
  - Densely populated areas.
  - Areas with a higher share of less contact-intensive jobs that can be performed remotely.
- Evidence:
  - Community banks in very densely populated areas or areas with a higher share of teleworkable jobs increased their provisions more strongly during 2020 than other banks.

### Policy implications and recommendations
- Average bank capital losses are modest across scenarios, but:
  - Losses are significantly larger for small and geographically concentrated community banks, especially if CRE price drops are persistent.
- Recommended policy actions:
  - Policymakers should remain vigilant amidst ongoing structural shifts in the CRE sector.
  - Maintaining the frequency of stress tests for small/community banks, particularly for those with high levels of exposures to riskier CRE segments, would be beneficial to assess the risks to financial stability in a timely fashion.

### Distribution of projected capital losses under the mild adverse scenario
- Baseline projection horizon: k = 8 quarters.
- Projected provisioning rule: provision expenses = Allowance for Loan and Lease Losses (ALLL) due to CRE loan exposure equal to four quarters of projected net CRE loan charge-offs.
- Assumed decline in total CRE loan balances in the forecast path: 2 percent quarterly.
- Total risk-weighted assets for capital impact calculations are held equal to the last historical observation (2020:Q3).
- Distributional finding: projected 8-quarter-ahead capital losses (percent of pre-shock risk-weighted assets) are concentrated among smaller and community banks with higher CRE loan exposures; large banks have lower average CRE loan exposures and have been more prudent in CRE lending compared to the pre-global financial crisis period.

### Sensitivity of provisions in 2020 to local population density and teleworkability (qualitative)
- Regression links ratio of total provisions during 2020:Q1–2020:Q3 to pre-provision net revenues in 2019, interacted with 2019 CRE loans-to-total assets and location characteristics.
- Definitions:
  - "Very Densely Populated" = community banks headquartered in zip codes with population densities above the 99th percentile.
  - "Densely Populated" = community banks above the 95th percentile and less than the 99th percentile.
  - "Teleworkable (>75th percentile)" = community banks headquartered in MSAs with share of jobs that can be done at home greater than the 75th percentile.
- Statistically significant relations shown by solid bars in figure at the 5 percent level.

### Data, empirical strategy, and limitations (Annex summary)
- Data sources and sample:
  - Call Reports (FFIEC 031 and 051), MSCI Real Estate CRE prices for MSAs, FDIC Deposit Survey.
  - Sample period: 2001:Q1–2020:Q3.
  - Sample size: total of 10,796 reporting banks (5,116 banks in 2020:Q3).
  - CRE price indices available for 69 major MSAs; baseline covers more than 50 percent of total banking sector assets on average.
- Main estimation model:
  - Outcomes Y_{b,t}^l: (i) nonperforming CRE loan ratio at t; (ii) net CRE loan charge-off rate at t; (iii) log change in pre-provision net revenues from t-k to t; (iv) log change in total regulatory capital from t-k to t.
  - Key explanatory variables: CRE loans / total assets at t-k and ∆P_{t,t−k}^l (log change in MSA CRE prices), with indicator I(∆P_{t,t−k}^l<0).
  - Parameters α_k and β_k: α_k + β_k measures differential outcome for banks with different ex-ante CRE exposures after price declines over k quarters.
  - Controls: bank capital ratio, liquidity ratio, (log) total asset size at t-k and interactions, bank fixed effects μ_b, location (MSA) × quarter fixed effects η_{l,t}.
  - Estimation: weighted least squares (weights proportional to log of bank total assets); standard errors double clustered at bank and quarter levels.
- Limitations:
  - CRE price indices only for 69 MSAs.
  - Call Reports lack CRE loans by segment and property locations; MSA-level average prices used and exposures assumed to local MSA of bank headquarters.
  - Mitigating factor: most US banks are small and geographically concentrated.

### Selected exact summary statistics (Table A1)
- CRE NPL Ratio (in ppt): Mean 1.56, Std 3.45, Obs. 94521
- CRE Charge-Off Rate (net of recovery rate): Mean 0.00, Std 0.01, Obs. 90447
- Change in Pre-Provision Net Revenues (8-quarter log change): Mean 0.15, Std 0.75, Obs. 94521
- Change in Regulatory Capital (8-quarter log change): Mean 0.17, Std 0.24, Obs. 93462
- CRE Loans to Total Assets: Mean 0.30, Std 0.19, Obs. 94521
- Bank Total Assets (in logs): Mean 12.73, Std 1.56, Obs. 94521
- Bank Liquid Assets-to-Total Assets Ratio: Mean 0.26, Std 0.17, Obs. 94521
- Bank Equity Capital-to-Total Assets: Mean 0.13, Std 0.13, Obs. 94521
- Change in CRE Prices (8-quarter log change in MSA-level average CRE prices): Mean 0.09, Std 0.16, Obs. 121187

### Scenario construction and provisioning mechanics
- Forecasted outcomes Y_{b,t+k}^l computed as (α_k + β_k) * CRE_exposure_{b,t}^l * ∆P_{t+k,t}^l, where ∆P_{t+k,t}^l is the CRE price path used.
- Forward-looking provisioning rule: ALLL due to CRE loan exposure = 4 quarters of projected net CRE charge-offs.
- Total CRE loan balances assumed to decline by 2 percent quarterly (consistent with CRE charge-off rates during the global financial crisis).
- Projected provision expenses and PPNR losses divided by total risk-weighted assets (held equal to 2020:Q3) to compute capital impact.
- Macroeconomic/location-specific effects incorporated via MSA × time fixed effects following the path observed during the global financial crisis; MSA × time fixed effects significantly correlated with MSA-level change in CRE prices (correlation coefficients statistically significant at 10 percent or lower except horizon 6 for the PPNR specification).

*Source: gsnea2021001 — INTRODUCTION — https://www.imf.org/-/media/files/publications/gfs-notes/2021/english/gsnea2021001.pdf*

### INTRODUCTION

### gsnea2021001 - INTRODUCTION

### Overview
- The commercial real estate (CRE) sector in the United States was hit hard during the COVID-19 pandemic as demand for contact-intensive CRE spaces (traditional brick-and-mortar retail, restaurants, hotels, and offices) plunged in 2020.
- CRE transaction volumes and prices declined sharply, particularly in the hotel, retail, and office segments; the decline in the United States was more pronounced than in other regions.
- Preexisting structural trends—digitalization, e-commerce, and working-from-home—accelerated during the pandemic, creating uncertainty about the outlook and suggesting further price declines may be possible in some CRE segments.
- Banks are the largest providers of CRE debt financing in the US (share of total outstanding CRE debt held by banks is 54 percent in the US), while nonbank financial institutions hold 24 percent in the US.

### Research questions and methodology
- This note investigates:
  - How does a decline in CRE prices affect banks’ profitability and solvency?
  - How large bank capital losses could be if CRE prices were to remain depressed permanently?
- Analysis steps:
  - Empirically identify effect of CRE loan exposure on bank performance using US bank-level data and location-specific CRE prices (2001–2020).
  - Use estimates to conduct a capital loss scenario assessing credit and revenue losses and potential bank capital losses based on a forward-looking provisioning rule.
- Data limitations noted: supervisory loan-level data and real-time appraisals are not publicly available; analysis relies on bank-level Call Reports and MSA-level CRE prices from MSCI.

### Key findings
- Banks with higher ex-ante CRE loan exposures perform considerably worse after a decline in CRE prices:
  - They experience significantly higher CRE non-performing loans ratio, higher CRE loan charge-off rates, lower revenues, and lower capital.
- Quantitative sensitivities (following a one standard deviation decline in local CRE prices over an eight quarter horizon—corresponding to a 16 percent cumulative decline in CRE prices):
  - Banks at the 75th percentile of CRE loan exposure experience an 8 percentage points higher non-performing CRE loan ratio and a 3 percentage points higher CRE loan charge-offs rate (compared to banks with no CRE loan exposure).
  - Their pre-provision net revenues and total regulatory capital are lower by 12 percent and 5 percent, respectively.
- Results are broadly similar when focusing on small banks or a shorter (four quarter) horizon.

### Bank financing of the CRE sector
- CRE lending is a significant component of bank business lending:
  - On average, CRE loans accounted for nearly half of total business loans made by banks over the last decade.
  - Total CRE loans surpassed $2.3 trillion prior to the pandemic.
- Distribution by bank size:
  - Small- and medium-sized banks (defined as banks with total assets less than $100 billion) extended a disproportionately large share of CRE loans.
  - Small banks (total assets not exceeding $5 billion during the sample) collectively hold about two-thirds of overall CRE loans while accounting for about 30 percent of total banking sector assets.
  - CRE loans constitute about 6 percent of total assets for large banks and 30 percent for small banks.
- Historical context:
  - High CRE loan exposures were a key determinant of bank failures after the global financial crisis when overall CRE prices declined by 30 percent.

### CRE price changes and bank stability
- Empirical evidence shows a decline in CRE prices leads to:
  - Higher cumulative CRE nonperforming loan rates and higher cumulative CRE net loan charge-off rates for banks with high CRE exposure (effect shown for 8 quarters ahead and 4 quarters ahead).
  - Lower net revenues before provisioning and lower total regulatory capital for banks with high CRE exposure.
- Definition used in empirical comparisons:
  - High CRE exposure = 75th percentile CRE-loans-to-total-assets ratio (43 percentage points exposure).
  - Low CRE exposure = 25th percentile (16 percentage points exposure).
- All estimated effects reported are statistically significant at 10 percent or less for cumulative effects.

### Bank capital adequacy under CRE sector stress
- Two stressed CRE price scenarios (based on IMF (2021) valuation model):
  - Mild scenario: cumulative one-standard-deviation decline in CRE prices after eight quarters; corresponds to a 1¼ percent permanent increase in the CRE vacancy rate.
  - Severe scenario: permanent increase in CRE vacancy rate by 5 percentage points (rate observed during the global financial crisis); implies a decline in CRE prices by about 30 percent in the long-run.
- Marginal-impact results (without other macroeconomic factors):
  - Average projected drop in capital adequacy is mild: 0.14 percentage point on average.
  - Projected credit and revenue losses could reach more than 1 percent of pre-shock total risk-weighted assets for banks with very high CRE loan exposures (top 3 percent of CRE loan exposure).
  - Nearly all banks in the left tail (5th percentile of the distribution of loss ratio) are small or community banks.
  - Under the permanent high vacancy (severe) scenario, average capital losses are almost twice as large as under the mild scenario.
- Results incorporating macroeconomic/location-specific effects:
  - Average decline in capital adequacy increases to 1.4 percentage points.
  - Banks in the left tail of the loss distribution are predominantly small or community banks; their losses reach 1.5 percentage points in this scenario.

### Structural shifts in the CRE sector and bank provisioning
- Structural shifts (rise of e-commerce and teleworking) can have stronger valuation effects in:
  - Densely populated areas.
  - Areas with a higher share of less contact-intensive jobs that can be performed remotely.
- Evidence from correlations:
  - Community banks in very densely populated areas or areas with a higher share of teleworkable jobs increased their provisions more strongly during 2020 than other banks.

### Policy implications and recommendations
- Bank capital losses are estimated to be modest on average across scenarios, but:
  - Losses are significantly larger for small and geographically concentrated community banks, especially if CRE price drops are persistent.
- Policy recommendations highlighted:
  - Policymakers should remain vigilant amidst ongoing structural shifts in the CRE sector.
  - Policymakers may benefit from maintaining the frequency of stress tests for small/community banks to assess financial stability risks in a timely way.

*Source: gsnea2021001 - INTRODUCTION — https://www.imf.org/-/media/files/publications/gfs-notes/2021/english/gsnea2021001.pdf*

### 1. Distribution  of  Projected Capital Losses under Mild Adverse

### 1. Distribution  of  Projected Capital Losses under Mild Adverse Scenario

### Distributional findings
- Panels show the distribution of 8-quarter-ahead projected capital losses (percent of pre-shock risk-weighted assets) using an alternative specification that incorporates macroeconomic effects by including location interacted with time fixed effects over the forecast horizon.
- Results indicate potential stress from CRE price declines is concentrated in smaller and community banks with higher CRE loan exposures.
- Large banks have lower average CRE loan exposures compared to smaller banks (as noted in the Annex discussion), and have been more prudent in CRE lending compared to the pre-global financial crisis period.

### Key numeric and categorical points from figures and notes
- Baseline projection horizon: k = 8 quarters.
- Projected provisioning rule: provision expenses = Allowance for Loan and Lease Losses (ALLL) due to CRE loan exposure equal to four quarters of projected net CRE loan charge-offs.
- Assumed decline in total CRE loan balances in the forecast path: 2 percent quarterly.
- Total risk-weighted assets for capital impact calculations are held equal to the last historical observation (2020:Q3).

### Policy recommendation (from concluding remarks)
- "Maintaining the frequency of stress tests for small/community banks, particularly for those with high levels of exposures to riskier CRE segments, would be beneficial to assess the risks to financial stability in a timely fashion."

---

### Concluding remarks (sector context and resilience)
- "The COVID-19 crisis has hit the commercial real estate sector hard in the US. The structural shifts in CRE demand pose considerable uncertainly around the outlook for the sector and suggest that further price declines may be possible, at least in some CRE segments."
- "This note shows that such price declines may potentially be a source of stress for banks, especially for smaller and community banks with higher CRE loan exposures."
- "So far, the banking sector in the US has been largely resilient to the developments in the CRE sector during the COVID-19 crisis because of strong bank capital buffers and the unprecedented policy support and regulatory responses during the crisis."
- "Large banks have also been more prudent in their CRE lending compared to the pre-global financial crisis period, with their exposures declining notably."

---

### Sensitivity of Provisions in 2020 to Local Population Density and Teleworkability (Figure 7 summary)

### Empirical sensitivity results (qualitative)
- Regression relates the ratio of total provisions during 2020:Q1–2020:Q3 to pre-provision net revenues in 2019 on total CRE loans-to-total assets ratio in 2019:Q4, interacted with location characteristics.
- "Very Densely Populated" corresponds to community banks headquartered in zip codes with population densities above the 99th percentile.
- "Densely Populated" corresponds to community banks above the 95th percentile and less than the 99th percentile.
- "Teleworkable (>75th percentile)" corresponds to community banks headquartered in MSAs with a share of jobs that can be done at home greater than the 75th percentile (similarly defined for other percentiles).
- Solid bars in the figure indicate estimated relations statistically significant at the 5 percent level.

---

### Annex: Data and Empirical Strategy

### Data sources and sample
- Data sources:
  - Federal Financial Institutions Examination Council 031 and 051 Report Forms (Call Reports) for detailed bank-level data at a quarterly frequency on CRE loans, outcome indicators, bank controls, and zip-code-level bank headquarters.
  - MSCI Real Estate for average quarterly CRE prices in the MSA where the bank headquarters are located.
  - Federal Deposit Insurance Corporation (FDIC) Deposit Survey to measure geographical concentration of banks.
- Sample period: 2001:Q1–2020:Q3.
- Sample size: total of 10,796 reporting banks (5,116 banks in 2020:Q3).
- CRE price indices available for 69 major MSAs; baseline estimations use banks located in these MSAs, which on average cover more than 50 percent of total banking sector assets.
- Results are robust to expanded samples using state- or Census Bureau region averages of MSA-level CRE prices.

### Main estimation model (equation 1) — structure and identification
- Outcome variables Y_{b,t}^l are:
  - (i) the nonperforming CRE loan ratio at t,
  - (ii) the net CRE loan charge-off rate at t,
  - (iii) the log change in pre-provision net revenues from t-k to t,
  - (iv) the log change in total regulatory capital from t-k to t.
- Key explanatory variables include bank b’s exposure to CRE loans at t-k (CRE loans / total assets) and ∆P_{t,t−k}^l, the (log) change in average CRE price index in MSA l from t-k to t, with an indicator I(∆P_{t,t−k}^l<0) to capture asymmetry for price declines.
- Parameters of interest: α_k and β_k, where α_k + β_k reflects how the outcome differs across banks with different ex-ante CRE exposures following a decline in CRE prices over k quarters.
- Controls: bank capital ratio, liquidity ratio, (log) total asset size measured at t-k, interactions of these with ∆P, and level of CRE exposure.
- Fixed effects: bank fixed effects μ_b and location (MSA) × quarter fixed effects η_{l,t}.
- Estimation method: weighted least squares (weights proportional to log of bank total assets); standard errors double clustered at the bank and quarter levels.
- Identification relies on within-MSA variation in bank ex-ante CRE loan exposures; sample restricted to MSAs with multiple banks (results robust to including single-bank MSAs).

### Limitations noted
- CRE price indices only available for 69 major MSAs.
- Bank CRE loans by segment (office, retail, industrial, lodging) are not available in Call Reports; average MSCI CRE prices per MSA are used instead.
- Locations of underlying properties are not observed; bank exposures are assumed to be to CRE price changes in the MSA of bank headquarters.
- These limitations are mitigated by the fact that most US banks are small and have geographically concentrated portfolios.

---

### Table A1 — Definitions and summary statistics (selected exact values)
- Sample observations reported for variables (obs counts match those in table A1).
- Dependent variables:
  - CRE NPL Ratio (in ppt): Mean 1.56, Std 3.45, Obs. 94521
  - CRE Charge-Off Rate (net of recovery rate): Mean 0.00, Std 0.01, Obs. 90447
  - Change in Pre-Provision Net Revenues (8-quarter log change): Mean 0.15, Std 0.75, Obs. 94521
  - Change in Regulatory Capital (8-quarter log change): Mean 0.17, Std 0.24, Obs. 93462
- Independent variables / bank controls:
  - CRE Loans to Total Assets: Mean 0.30, Std 0.19, Obs. 94521
  - Bank Total Assets (in logs): Mean 12.73, Std 1.56, Obs. 94521
  - Bank Liquid Assets-to-Total Assets Ratio: Mean 0.26, Std 0.17, Obs. 94521
  - Bank Equity Capital-to-Total Assets: Mean 0.13, Std 0.13, Obs. 94521
  - Change in CRE Prices (8-quarter log change in MSA-level average CRE prices): Mean 0.09, Std 0.16, Obs. 121187

---

### Scenario analysis and incorporation of macroeconomic effects

### Scenario construction
- Using estimated α_k and β_k from equation (1), forecasted outcomes Y_{b,t+k}^l are calculated as (α_k + β_k) * CRE_exposure_{b,t}^l * ∆P_{t+k,t}^l, where Y_{b,t+k}^l is either k-quarters-ahead net CRE charge-off rate or change in PPNR over k quarters, and ∆P_{t+k,t}^l is the CRE price path given by panel 1 of Figure 3.

### Provisions and capital-impact computation
- Forward-looking provisioning rule: provisions = four quarters of projected net CRE loan charge-offs (ALLL due to CRE loan exposure = 4 quarters of projected net CRE charge-offs).
- Total CRE loan balances assumed to decline by 2 percent quarterly (consistent with CRE charge-off rates during the global financial crisis).
- Projected provision expenses and losses in PPNR are divided by total risk-weighted assets (assumed equal to last historical observation, 2020:Q3) to compute capital impact.

### Incorporating macroeconomic effects
- Macroeconomic effects incorporated by including location-specific factors proxied by MSA × time fixed effects in the forecast horizon; these fixed effects are assumed to follow a path observed during the global financial crisis.
- Figure A1 shows the MSA × time fixed effects are significantly correlated with MSA-level change in CRE prices (correlation coefficients statistically significant at 10 percent or lower except horizon 6 for the PPNR specification).
- Correlations are stronger for the CRE charge-off rate specification than for the PPNR specification.

---

*IMF | Monetary and Capital Markets — gsnea2021001, "1. Distribution of Projected Capital Losses under Mild Adverse Scenario"*

---


_Source: https://www.imf.org/-/media/files/publications/gfs-notes/2021/english/gsnea2021001.pdf_