## Banks’ Joint Exposure to Market and Run Risk

## Source details

**Canonical URL:** [Banks’ Joint Exposure to Market and Run Risk](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023200-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023200-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023200-print-pdf.pdf.json)

---

### Abstract and key findings
- Large liability withdrawals can damage capital positions when unrealized losses are present; recent US failures illustrate this interaction.
- A high-level global assessment covering 1,288 banks from 66 countries finds sizable exposure to joint market-and-run risk, with outcomes varying substantially across banks and highly concentrated in the most exposed institutions.
- Vulnerability is generally greater in advanced economies (AEs) than in emerging market and developing economies (EMDEs), reflecting lower cash buffers, securities holdings and capitalization in AEs.
- Within AEs—unlike in EMs—larger banks are most exposed due to greater wholesale funding and thinner capital buffers.
- Estimated aggregate losses are substantial in some countries, reflecting a range of recent shocks, including recent debt crises in some EMDEs that imply unrealized losses far larger than those caused by contractionary monetary policy alone.

### Approach and scenarios
- Goal: assess whether liquidity and capital cushions would be depleted by unrealized losses crystallized through hypothetical rapid liability withdrawals at the bank level.
- Coverage: banks that report cash and cash equivalents, total securities, total assets, total deposits, wholesale funding, total equity, and cost of funds in 2022Q4 — yielding 1,288 banks from 66 countries (62% in AEs and 38% in EMDEs). Total assets in the baseline scenario are USD 74 trillion, which is 73% of total assets across all banks in S&P Capital IQ Pro; 47 countries have at least five banks covered.
- Pricing window: bond price changes and durations measured from January 1, 2022 to July 7, 2023.
- Withdrawal scenarios:
  - Banks are split by above- or below-median cost of funds (COF).
  - Mild, moderate, severe scenarios assume S = 10%, 20% and 30% of deposits leaving in high-COF banks, and half as much in low-COF banks.
  - Wholesale funding outflows are modeled at 1.5×S for each bank.
  - A previously described “moderate scenario” considers outflows of 20 percent of deposits and 30 percent of wholesale funding for more vulnerable banks, consistent with stresses faced by Silicon Valley Bank, Signature, and First Republic.

### Data
- Primary bank dataset: S&P Capital IQ Pro (unbalanced quarterly panel, coverage through 2022Q4).
- Bond pricing and duration: Bloomberg Back Office (local currency sovereign bonds); country-specific spreads from the IMF’s internal Sovereign Debt Monitor (par-value-weighted average spreads across sovereign bonds with remaining maturity > one year).
- Where Sovereign Debt Monitor data cannot be matched, the median change in spread within the sample is used.
- For 9 EMDEs with good pricing data, using AE-style local-currency pricing yields similar market-value estimates.

### Methodology — marking assets to market
- Securities treatment:
  - Imposed held-to-maturity (HTM) share of 50% for all countries (benchmarked to the US and Granja (2023)).
  - For AEs: HTM securities are approximately marked to market using the value-weighted average percentage change in price observed for local currency bonds since start of 2022.
  - For AEs, banks’ portfolio duration is assumed to be 75 percent of the country-wise duration; the fraction δ[75]US = 0.81 is applied to all AEs to obtain the HTM discount factor.
  - For EMDEs: MTM discount = [∆7/7/23−1/1/22 rUST(Durationc) + ∆7/7/23−1/1/22 Spreadc] × Durationc × δ[75]US, where Durationc = Σj(wj,c × Durationj,c) and Spreadc is the par-value-weighted spread from the Sovereign Debt Monitor.
- Other assets:
  - Sales of other assets are imposed a larger discount equal to 1.25×Discountc to reflect higher risk premia on less liquid assets.
  - No additional fire-sale penalty beyond the 1.25× factor is imposed in the baseline.
- Discount calibration:
  - For AEs: Mark-to-Market DiscountAEc = Σj(wj,c × ∆7/7/23−1/1/22 Pricej,c) × δ[75]US.
  - For EMDEs: use formula above combining yield-curve changes and country spreads.
- The methodology focuses on existing unrealized losses by using observed price changes; additional unanticipated rate hikes could generate larger losses.

### Methodology — modeling withdrawals and asset sales
- Excess withdrawals for bank b under scenario S:
  - EWithdrawalsS_b = TotalDepositb × S_b + Wholesaleb × 1.5 × S_b − Cashb − SecuritiesMTM_b, constrained that EWithdrawalsS_b ≥ 0.
- Asset sale rules (sell securities first, then other assets if needed):
  - SecSalesS_b = 
    - EWithdrawalsS_b [1−Discountc] if EWithdrawalsS_b [1−Discountc] < SecuritiesHTM_b
    - SecuritiesHTM_b if EWithdrawalsS_b [1−Discountc] ≥ SecuritiesHTM_b
  - OtherAssetSalesS_b =
    - 0 if EWithdrawalsS_b [1−Discountc] < SecuritiesHTM_b
    - (EWithdrawalsS_b − SecuritiesHTM_b × [1−Discountc]) / [1−1.25×Discountc] if EWithdrawalsS_b [1−Discountc] ≥ SecuritiesHTM_b
- Post-sale asset value when EWithdrawalsS_b > 0:
  - AssetsS_b = SecuritiesHTM_b − SecSalesS_b + OtherAssets2022_b − OtherAssetSalesS_b − Cashb − SecuritiesMTM_b

### Assessing bank losses and concentration
- Losses realized on forced asset sales are compared to equity to identify banks for which solvency would be threatened in the current market environment if a run occurs.
- The approach:
  - Focuses on losses realized only on assets sold in the simulated scenarios, not on all assets held.
  - Uses sovereign bond durations and price changes as proxies for bank asset exposures due to data constraints.
  - Assumes sales occur at current market prices; does not assume withdrawals are redeposited elsewhere or that policy support is provided.
  - Does not account for changes in deposit franchise values as interest rates change.
- Sensitivity analysis:
  - Explores alternative values for HTM share, portfolio duration assumptions, wholesale funding runoff rates, and additional assumptions on already-recognized mark-to-market losses.
  - Notes that lack of granular data may lead to overestimation of losses in some cases.

### Losses definition, scope, and main results
- Bank-level losses are defined as:
  - Losses S_b = SecSales S_b ×Discount_c + OtherAssetSales S_b ×1.25×Discount_c (Equation 9)
- Baseline central scenario uses 20% deposit withdrawals at vulnerable banks; mild/severe scenarios reflect 10%/30% withdrawals, respectively.
- Aggregate and tail vulnerability:
  - Estimated losses are above 20 percent of equity for about 6 percent of banks in a moderate scenario, and for nearly 20 percent of banks in a severe scenario.
  - For the most exposed decile of banks, estimated losses are almost half of total equity in the severe scenario.
- AE vs EMDE patterns:
  - Larger aggregate exposure of banks in AEs relative to EMDEs is driven by greater use of wholesale funding, tighter cash and capital buffers, and lower securities holdings.
  - Projected losses vary significantly across banks but are substantial for those most exposed.

### Drivers of losses (Shapley-style decomposition)
- Procedure (preserved exactly):
  - Construct S of 1000 hypothetical banks by repeatedly sampling from the relevant group (AEs, EMs).
  - For each bank b and variable i: take x_bi, apply it to all banks in S, calculate change in losses (holding other variables fixed), average change across S.
  - Relevant variables: share of liabilities that are potentially runnable; share of these that are wholesale funds; equity as a share of total assets; cash as a share of total assets; securities as a share of non-cash assets; vector of discounts on securities and other assets.
  - Potentially runnable liabilities are defined as deposits plus 1.5 times wholesale funds (see Equation 5).
- Key findings:
  - Differences across banks are primarily driven by cash holdings, capitalization, securities holdings and the mark-to-market discount.
  - AEs are primarily more exposed because they have lower cash and securities holdings and smaller capital buffers.
  - Within most advanced economies, lower capitalization and greater wholesale funding drive higher exposure for larger banks (though not for the largest banks).
  - In EMDEs, smaller banks are more exposed, reflecting higher costs of funds, less sticky liabilities, and lower levels of security holdings.

### Sensitivity analysis
- Three avenues tested: market values (HTM share and duration), withdrawal scenario parameters (wholesale runoff multiplier and extra discount on other assets), and an alternative more granular methodology.

- Market values (HTM share and duration)
  - HTM share of securities: alternative assumptions from 25% to 75%.
  - Average duration of bank portfolios: consider between 50% and 100% of full portfolio duration in each country.
  - Example δ values for the US applied to all countries:
    - δ[50]_US = 0.65
    - δ[62.5]_US = 0.70
    - δ[75]_US = 0.81
    - δ[87.5]_US = 0.91
    - δ[100]_US = 1.
  - Results:
    - Losses as a share of equity remain higher in AEs in all cases, and rise particularly rapidly with the HTM share.
    - Median bank does not face losses in EMs, and likewise in AEs if the HTM share is less than 70%.
    - The 90th percentile of AE banks by losses experience significant losses even with 50% of the average duration and a 25% HTM share.

- Withdrawal scenarios (wholesale runoff and other-asset discount)
  - Wholesale funding extra runoff multipliers examined: 1.25 to 1.75 (baseline 1.5).
  - Extra discount on forced sales of other assets relative to securities: 1.0 to 1.5 (baseline 1.25).
  - Results:
    - Losses in AEs remain larger than those in EMs in all cases.
    - Total losses are relatively insensitive to the values of the multipliers, particularly in EMs.
    - Even for the maximum wholesale outflow differential of 75%, there are no losses in the 75th percentile bank in EMs or in the 50th percentile bank in AEs.
    - For the 90th percentile of banks by losses, a high-high (low-low) combination of the two multipliers can generate losses in AEs approximately 5% larger (smaller) than the baseline of approximately 20%.
    - For EMs, 90th percentile losses are relatively unaffected by the additional discount applied to other assets because few banks sell all of their securities.

- More granular methodology (where data permit)
  - Distinguish sovereign bonds and other securities; apply an extra discount to other securities based on change in spreads of BBB-rated US corporate bonds:
    - Extra Discount_c = ∆_7/7/23−1/1/22 Spread_USCorpBonds × Duration_c (Equation 10)
    - Spread_USCorpBonds from ICE BofA option-adjusted spread index.
  - Compare losses to regulatory capital rather than equity in this version.
  - Data coverage and totals:
    - Total assets in the baseline scenario are USD 31 trillion, which is 31% of total assets across all banks in S&P Capital IQ Pro.
    - Of these banks 50% are in AEs and 50% are in EMDEs; there are 33 countries with at least five banks covered.
  - Results:
    - Main results are similar: losses remain substantially higher in AEs than EMs and are highly concentrated in the most vulnerable quartile of banks.
    - For the granular method, larger losses in AEs are primarily driven by lower securities holdings as a share of assets, greater market risk exposure and lower capitalization.

### Stylized quantitative summaries and key statistics
- Scenario withdrawal rates used:
  - Central scenario: 20% deposit withdrawals at vulnerable banks.
  - Mild scenario: 10% withdrawals.
  - Severe scenario: 30% withdrawals.
- Wholesale-runoff multiplier baseline: 1.5.
- Extra discount on other assets baseline: 1.25.
- Alternative multiplier ranges tested:
  - Wholesale funding extra runoff: 1.25 to 1.75.
  - Extra discount on other assets: 1.0 to 1.5.
- HTM share range tested: 25% to 75%.
- Duration range tested: 50% to 100% of full portfolio duration.
- Granular-method baseline totals:
  - Total assets = USD 31 trillion (31% of total assets across all banks in S&P Capital IQ Pro).
- Table A.1 raw sequences (preserved exactly as in source):
  - Cash and equivalents: 12.667.7187752.636.8129463
  - Securities: 19.9105187755.2515.929463
  - Wholesale funding: 20.084.8187753.8013.029463
  - Deposits: 42.71761877512.128.429463
  - Equity: 5.2721.6187752.195.2229463
  - Cost of funds (%): 0.661.28187755.355.9529463
  - MTM discount (%): 11.03.351877510.67.1029463
  - ∆ Bond prices (%): -13.64.1318775
  - ∆ SDM spreads (bps): 86.843129463

### Conclusions and policy implications
- Exposure to the combination of market and run risk is relevant for banks beyond the US and is driven by a wider range of shocks than seen in recent US failures.
- Exposure is heavily skewed toward the most vulnerable banks and generally greater in AEs, reflecting lower cash buffers, securities holdings and capitalization.
- Policy recommendations and implications:
  - Policymakers should be attentive to liquidity-solvency interaction risks revealed by recent US banking turmoil.
  - Banks should actively manage cash and capital buffers.
  - Regulators should prepare or update contingency plans—for example, preparing to rapidly introduce emergency liquidity facilities as deployed in the US following the SVB crisis if needed—and improve supervision going forward.
  - Country authorities with access to detailed balance sheet data for domestic banks could improve upon this methodology with more granular data.

*Source: wpiea2023200-print-pdf - Section 1.*

### Section 1

### Banks’ Joint Exposure to Market and Run Risk

### Abstract and key findings
- Large liability withdrawals can damage capital positions when unrealized losses are present; recent US failures illustrate this interaction.
- A high-level global assessment covering 1,288 banks from 66 countries finds sizable exposure to joint market-and-run risk, with outcomes varying substantially across banks and highly concentrated in the most exposed institutions.
- Vulnerability is generally greater in advanced economies (AEs) than in emerging market and developing economies (EMDEs), reflecting lower cash buffers, securities holdings and capitalization in AEs.
- Within AEs—unlike in EMs—larger banks are most exposed due to greater wholesale funding and thinner capital buffers.
- Estimated aggregate losses are substantial in some countries, reflecting a range of recent shocks, including recent debt crises in some EMDEs that imply unrealized losses far larger than those caused by contractionary monetary policy alone.

### Approach and scenarios
- Goal: assess whether liquidity and capital cushions would be depleted by unrealized losses crystallized through hypothetical rapid liability withdrawals at the bank level.
- Coverage: banks that report cash and cash equivalents, total securities, total assets, total deposits, wholesale funding, total equity, and cost of funds in 2022Q4 — yielding 1,288 banks from 66 countries (62% in AEs and 38% in EMDEs). Total assets in the baseline scenario are USD 74 trillion, which is 73% of total assets across all banks in S&P Capital IQ Pro; 47 countries have at least five banks covered.
- Pricing window: bond price changes and durations measured from January 1, 2022 to July 7, 2023.
- Withdrawal scenarios:
  - Banks are split by above- or below-median cost of funds (COF).
  - Mild, moderate, severe scenarios assume S = 10%, 20% and 30% of deposits leaving in high-COF banks, and half as much in low-COF banks.
  - Wholesale funding outflows are modeled at 1.5×S for each bank.
  - A previously described “moderate scenario” considers outflows of 20 percent of deposits and 30 percent of wholesale funding for more vulnerable banks, consistent with stresses faced by Silicon Valley Bank, Signature, and First Republic.

### Data
- Primary bank dataset: S&P Capital IQ Pro (unbalanced quarterly panel, coverage through 2022Q4).
- Bond pricing and duration: Bloomberg Back Office (local currency sovereign bonds); country-specific spreads from the IMF’s internal Sovereign Debt Monitor (par-value-weighted average spreads across sovereign bonds with remaining maturity > one year).
- Where Sovereign Debt Monitor data cannot be matched, the median change in spread within the sample is used.
- For 9 EMDEs with good pricing data, using AE-style local-currency pricing yields similar market-value estimates.

### Methodology — marking assets to market
- Securities treatment:
  - Imposed held-to-maturity (HTM) share of 50% for all countries (benchmarked to the US and Granja (2023)).
  - For AEs: HTM securities are approximately marked to market using the value-weighted average percentage change in price observed for local currency bonds since start of 2022.
  - For AEs, banks’ portfolio duration is assumed to be 75 percent of the country-wise duration; the fraction δ[75]US = 0.81 is applied to all AEs to obtain the HTM discount factor.
  - For EMDEs: MTM discount = [∆7/7/23−1/1/22 rUST(Durationc) + ∆7/7/23−1/1/22 Spreadc] × Durationc × δ[75]US, where Durationc = Σj(wj,c × Durationj,c) and Spreadc is the par-value-weighted spread from the Sovereign Debt Monitor.
- Other assets:
  - Sales of other assets are imposed a larger discount equal to 1.25×Discountc to reflect higher risk premia on less liquid assets.
  - No additional fire-sale penalty beyond the 1.25× factor is imposed in the baseline.
- Discount calibration:
  - For AEs: Mark-to-Market DiscountAEc = Σj(wj,c × ∆7/7/23−1/1/22 Pricej,c) × δ[75]US.
  - For EMDEs: use formula above combining yield-curve changes and country spreads.
- The methodology focuses on existing unrealized losses by using observed price changes; additional unanticipated rate hikes could generate larger losses.

### Methodology — modeling withdrawals and asset sales
- Excess withdrawals for bank b under scenario S:
  - EWithdrawalsS_b = TotalDepositb × S_b + Wholesaleb × 1.5 × S_b − Cashb − SecuritiesMTM_b, constrained that EWithdrawalsS_b ≥ 0.
- Asset sale rules (sell securities first, then other assets if needed):
  - SecSalesS_b = 
    - EWithdrawalsS_b [1−Discountc] if EWithdrawalsS_b [1−Discountc] < SecuritiesHTM_b
    - SecuritiesHTM_b if EWithdrawalsS_b [1−Discountc] ≥ SecuritiesHTM_b
  - OtherAssetSalesS_b =
    - 0 if EWithdrawalsS_b [1−Discountc] < SecuritiesHTM_b
    - (EWithdrawalsS_b − SecuritiesHTM_b × [1−Discountc]) / [1−1.25×Discountc] if EWithdrawalsS_b [1−Discountc] ≥ SecuritiesHTM_b
- Post-sale asset value when EWithdrawalsS_b > 0:
  - AssetsS_b = SecuritiesHTM_b − SecSalesS_b + OtherAssets2022_b − OtherAssetSalesS_b − Cashb − SecuritiesMTM_b

### Assessing bank losses and concentration
- Losses realized on forced asset sales are compared to equity to identify banks for which solvency would be threatened in the current market environment if a run occurs.
- The approach:
  - Focuses on losses realized only on assets sold in the simulated scenarios, not on all assets held.
  - Uses sovereign bond durations and price changes as proxies for bank asset exposures due to data constraints.
  - Assumes sales occur at current market prices; does not assume withdrawals are redeposited elsewhere or that policy support is provided.
  - Does not account for changes in deposit franchise values as interest rates change.
- Sensitivity analysis:
  - Explores alternative values for HTM share, portfolio duration assumptions, wholesale funding runoff rates, and additional assumptions on already-recognized mark-to-market losses.
  - Notes that lack of granular data may lead to overestimation of losses in some cases.

### Conceptual contribution and relation to literature
- The paper’s conceptual contribution is to jointly consider how market-risk-driven unrealized losses and liquidity-driven asset sales (runs) can threaten solvency—bringing together liquidity and solvency analysis in a single framework.
- Methodology differs from Jiang, Matvos, Piskorski & Seru (2023) by focusing on market-value losses only for assets sold in scenarios, using sovereign bond proxies for bank assets, and comparing losses as a share of equity across withdrawal scenarios.

*Source: wpiea2023200-print-pdf - Section 1.*

### Section 2

### Section 2

### Losses definition and scope
- Bank-level losses are defined as:
  - Losses S_b = SecSales S_b ×Discount_c + OtherAssetSales S_b ×1.25×Discount_c (Equation 9)
- Analysis examines both bank-level losses and combined losses of all banks in each country.
- Baseline central scenario uses 20% deposit withdrawals at vulnerable banks; mild/severe scenarios reflect 10%/30% withdrawals, respectively.

### Methodology caveats
- Key simplifications:
  - In absence of consistent bank-specific data on market values of assets or risk exposures, risk exposures are assumed to mirror those of outstanding domestic sovereign bonds.
  - Hedging is not accounted for; no differentiation among types of deposits or sources of wholesale funding.
  - Factors abstracted from include geographic diversification, investor sophistication, digitalization, improvements in net interest income from deposit franchise, redepositing of simulated withdrawals at other banks, and access to central bank liquidity facilities.
  - No assumption of policy support for banks in the baseline assessment.
- Note on hedging evidence:
  - Recent ECB exercise: interest rate hedging reduced potential losses on held-to-maturity bond portfolios by 22% under the adverse market risk scenario (ECB 2023).
  - Recent granular work: large gross positions in interest rate swaps translate to essentially no meaningful change in interest-rate exposure for the average US bank (McPhail et al. 2023).

### Main results and distributions
- Aggregate and tail vulnerability:
  - Estimated losses are above 20 percent of equity for about 6 percent of banks in a moderate scenario, and for nearly 20 percent of banks in a severe scenario.
  - For the most exposed decile of banks, estimated losses are almost half of total equity in the severe scenario.
- Advanced economy (AE) vs EMDE patterns:
  - Larger aggregate exposure of banks in AEs relative to EMDEs is driven by greater use of wholesale funding, tighter cash and capital buffers, and lower securities holdings.
  - Projected losses vary significantly across banks but are substantial for those most exposed.
- Figures referenced (descriptive):
  - Figure 1: Distribution of estimated losses across banks (moderate scenario bars; mild/severe error bars).
  - Figure 2: Breakdown of drivers of losses in the central scenario (20% withdrawals).

### Drivers of losses (Shapley-style decomposition)
- Approach:
  - Construct S of 1000 hypothetical banks by repeatedly sampling from the relevant group (AEs, EMs).
  - For each bank b and variable i: take x_bi, apply it to all banks in S, calculate change in losses (holding other variables fixed), average change across S.
  - Relevant variables: share of liabilities that are potentially runnable; share of these that are wholesale funds; equity as a share of total assets; cash as a share of total assets; securities as a share of non-cash assets; vector of discounts on securities and other assets.
  - Potentially runnable liabilities are defined as deposits plus 1.5 times wholesale funds (see Equation 5).
- Key findings:
  - Differences across banks are primarily driven by cash holdings, capitalization, securities holdings and the mark-to-market discount.
  - AEs are primarily more exposed because they have lower cash and securities holdings and smaller capital buffers.
  - Within most advanced economies, lower capitalization and greater wholesale funding drive higher exposure for larger banks (though not for the largest banks).
  - In EMDEs, smaller banks are more exposed, reflecting higher costs of funds, less sticky liabilities, and lower levels of security holdings.
- Figures referenced:
  - Figure 3: Drivers of losses across banks (central scenario).
  - Figure 4: Losses by bank size quartile (central scenario).

### Sensitivity analysis
- Three avenues tested: market values (HTM share and duration), withdrawal scenario parameters (wholesale runoff multiplier and extra discount on other assets), and an alternative more granular methodology.

- Market values (HTM share and duration)
  - HTM share of securities: alternative assumptions from 25% to 75%.
  - Average duration of bank portfolios: consider between 50% and 100% of full portfolio duration in each country.
  - Calculate discount adjustment factor δ for the US and apply to all countries. Examples provided:
    - δ[50]_US = 0.65
    - δ[62.5]_US = 0.70
    - δ[75]_US = 0.81
    - δ[87.5]_US = 0.91
    - δ[100]_US = 1.
  - Results:
    - Losses as a share of equity remain higher in AEs in all cases, and rise particularly rapidly with the HTM share.
    - Median bank does not face losses in EMs, and likewise in AEs if the HTM share is less than 70%.
    - The 90th percentile of AE banks by losses experience significant losses even with 50% of the average duration and a 25% HTM share.
  - Figure referenced: Figure 5 (losses under alternative HTM and duration assumptions); Figures A.1–A.3 in the Appendix (50th, 75th, 90th percentiles).

- Withdrawal scenarios (wholesale runoff and other-asset discount)
  - Wholesale funding extra runoff multipliers examined: 1.25 to 1.75 (baseline 1.5).
  - Extra discount on forced sales of other assets relative to securities: 1.0 to 1.5 (baseline 1.25).
  - Results:
    - Losses in AEs remain larger than those in EMs in all cases.
    - Total losses are relatively insensitive to the values of the multipliers, particularly in EMs.
    - Even for the maximum wholesale outflow differential of 75%, there are no losses in the 75th percentile bank in EMs or in the 50th percentile bank in AEs.
    - For the 90th percentile of banks by losses, a high-high (low-low) combination of the two multipliers can generate losses in AEs approximately 5% larger (smaller) than the baseline of approximately 20% (shown in Figure 1).
    - For EMs, 90th percentile losses are relatively unaffected by the additional discount applied to other assets because few banks sell all of their securities (a precondition for any sales of other assets in the model).
  - Figure referenced: Figure 6 (losses under alternative assumptions for wholesale outflows and other assets); Figures A.4–A.6 in the Appendix.

- More granular methodology (where data permit)
  - Distinguish sovereign bonds and other securities; apply an extra discount to other securities based on change in spreads of BBB-rated US corporate bonds:
    - Extra Discount_c = ∆_7/7/23−1/1/22 Spread_USCorpBonds × Duration_c (Equation 10)
    - Spread_USCorpBonds from ICE BofA option-adjusted spread index.
  - Compare losses to regulatory capital rather than equity in this version.
  - Data coverage and totals:
    - Total assets in the baseline scenario are USD 31 trillion, which is 31% of total assets across all banks in S&P Capital IQ Pro.
    - Of these banks 50% are in AEs and 50% are in EMDEs; there are 33 countries with at least five banks covered.
  - Results:
    - Main results are similar: losses remain substantially higher in AEs than EMs and are highly concentrated in the most vulnerable quartile of banks.
    - For the granular method, larger losses in AEs are primarily driven by lower securities holdings as a share of assets, greater market risk exposure and lower capitalization.
  - Figures referenced: Figures 7 and 8; Figure A.7 in the Appendix.

### Stylized quantitative summaries and key statistics
- Scenario withdrawal rates used:
  - Central scenario: 20% deposit withdrawals at vulnerable banks.
  - Mild scenario: 10% withdrawals.
  - Severe scenario: 30% withdrawals.
- Wholesale-runoff multiplier baseline: 1.5.
- Extra discount on other assets baseline: 1.25.
- Alternative multiplier ranges tested:
  - Wholesale funding extra runoff: 1.25 to 1.75.
  - Extra discount on other assets: 1.0 to 1.5.
- HTM share range tested: 25% to 75%.
- Duration range tested: 50% to 100% of full portfolio duration.
- Example δ values for the US applied to all countries:
  - δ[50]_US = 0.65
  - δ[62.5]_US = 0.70
  - δ[75]_US = 0.81
  - δ[87.5]_US = 0.91
  - δ[100]_US = 1.
- Granular-method baseline totals:
  - Total assets = USD 31 trillion (31% of total assets across all banks in S&P Capital IQ Pro).

### Table A.1 summary statistics (raw sequences from source)
- Cash and equivalents: 12.667.7187752.636.8129463
- Securities: 19.9105187755.2515.929463
- Wholesale funding: 20.084.8187753.8013.029463
- Deposits: 42.71761877512.128.429463
- Equity: 5.2721.6187752.195.2229463
- Cost of funds (%): 0.661.28187755.355.9529463
- MTM discount (%): 11.03.351877510.67.1029463
- ∆ Bond prices (%): -13.64.1318775
- ∆ SDM spreads (bps): 86.843129463

### Conclusions and policy implications
- Exposure to the combination of market and run risk is relevant for banks beyond the US and is driven by a wider range of shocks than seen in recent US failures.
- Exposure is heavily skewed toward the most vulnerable banks and generally greater in AEs, reflecting lower cash buffers, securities holdings and capitalization.
- Policy recommendations and implications:
  - Policymakers should be attentive to liquidity-solvency interaction risks revealed by recent US banking turmoil.
  - Banks should actively manage cash and capital buffers.
  - Regulators should prepare or update contingency plans—for example, preparing to rapidly introduce emergency liquidity facilities as deployed in the US following the SVB crisis if needed—and improve supervision going forward.
  - Country authorities with access to detailed balance sheet data for domestic banks could improve upon this methodology with more granular data.

*Source: wpiea2023200-print-pdf - Section 2*

### Section 3

### Section 3

### Losses under alternative assumptions on HTM assets and duration
- Figures reported: A.3 (90th percentile) and earlier figure showing the 75th percentile (note text).
- Scenario: central scenario (20% deposit withdrawals at vulnerable banks).
- Data sources: S&P Capital IQ Pro, Bloomberg Back Office, IMF Sovereign Debt Monitor & IMF staff estimates.
- Notes: Graphs show how the 75th percentile and 90th percentile of bank losses vary under alternative assumptions on HTM assets and duration in the central scenario (20% deposit withdrawals at vulnerable banks).

### Losses under alternative assumptions for wholesale outflows and other assets
- Figures reported: A.4 (50th percentile), A.5 (75th percentile), A.6 (90th percentile).
- Scenario: central scenario (20% deposit withdrawals at vulnerable banks).
- Data sources: S&P Capital IQ Pro, Bloomberg Back Office, IMF Sovereign Debt Monitor & IMF staff estimates.
- Notes:
  - Graphs show how the 50th, 75th, and 90th percentiles of bank losses in AEs/EMs vary under alternative assumptions on:
    - the extra runoff for wholesale funding, relative to deposits, and
    - the extra discount applied to any forced sales of other assets, relative to securities.
  - Values are aggregates of bank-level results across all banks within AEs or EMs.
  - All estimates are for the central scenario (20% deposit withdrawals at vulnerable banks).

### Drivers of losses across banks (granular method)
- Figure reported: A.7.
- Scenario: central scenario (20% withdrawals in vulnerable banks).
- Data sources: S&P Capital IQ Pro, Bloomberg Back Office, IMF Sovereign Debt Monitor & IMF staff estimates.
- Presentation: Bars show the 25th, 50th and 75th percentiles of banks; dots show outliers. Outliers are values more than 1.5 times the interquartile range above (below) the 75th (25th) percentile. Whiskers extend to the maximum (minimum) value that is not an outlier.
- Relative importance score construction (process preserved exactly):
  - (i) construct a set S of 1000 hypothetical banks, with features representative of the population, by repeatedly sampling from the relevant group (i.e., AEs, EMs);
  - (ii) take the value x_bi of variable i for bank b, and apply it to all the banks in S;
  - (iii) calculate the change in losses for those banks, assuming that all variables except i stay the same;
  - (iv) average this change in losses across all 1000 banks in S.
  - Values are multiplied by -1 so that a more negative number corresponds to larger losses.

*Source: Banks’ Joint Exposure to Market and Run Risk — Working Paper No. WP/2023/200 (Section 3 figures and notes).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023200-print-pdf.pdf_