## _cr13277

## Source details

**Canonical URL:** [_cr13277](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/scr/2013/_cr13277.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/scr/2013/_cr13277.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/scr/2013/_cr13277.pdf.json)

---

### CHAPTER I. POTENTIAL OUTPUT AND UNEMPLOYMENT — The Great Recession in Sweden
- Real GDP and unemployment
  - Real GDP: contraction of 5 percent in 2009; rebound to above 3½ percent in 2010 and 2011 before slowing.
  - Unemployment: increased to 8 percent by 2012.
    - Note: classification of some full-time students as unemployed accounts for about 2 percentage points of overall unemployment.
- Sectoral and level effects
  - Manufacturing employment declined by close to 10 percent in 2009.
  - In 2012, output was 10 percent below the level implied by continuation of the 2000–07 average growth rate of 3¼ percent.
  - Forecast implication: the output loss relative to the 2007 trend continues to widen over the forecast horizon, implying a permanent output loss.

### CHAPTER I. POTENTIAL OUTPUT — Estimation and Results
- Estimation approaches used
  - Univariate HP filter with a range of smoothing parameters.
  - Production function (Cobb-Douglas) with TFP and unemployment determined by HP filter, combined with actual labor force and capital stock.
  - Multivariate filter (Benes et al. (2010) style) jointly determining potential output and the natural rate of unemployment.
  - To alleviate HP end-point problem, forecasts extended until 2025.
- Output gap estimates for 2013 (range from methods)
  - Multivariate filter: -1.6 percent of potential output.
  - HP filter and production function with low smoothing: -0.5 percent of potential output.
  - Average measure: -1 percent of potential output.
- Cross-institutional output gap excerpts (percent of potential output)
  - European Commission: 2010 -1.6; 2011 -0.1; 2012 -1.3; 2013 -1.6; 2014 -1.1.
  - Ministry of Finance: 2010 -3.1; 2011 -1.3; 2012 -2.2; 2013 -3.5; 2014 -3.3.
  - National Institute of Economic Research: 2010 -3.6; 2011 -1.5; 2012 -1.9; 2013 -2.1; 2014 -1.7.
  - Riksbank (average of quarterly estimates): 2010 -2.8; 2011 -0.5; 2012 -0.9; 2013 -1.3; 2014 -0.6.
  - Fund staff (World Economic Outlook): 2010 -1.0; 2011 0.8; 2012 -0.4; 2013 -1.0; 2014 -0.7.
- Potential growth and decomposition
  - 2012 potential growth estimates range from just below 2 to 2¼ percent.
  - 2000–07 average potential growth close to 3 percent (average estimate).
  - Production function decomposition (contributions to potential growth):
    - Employment contribution: 0.7 percentage points in 2005 → 0.5 percentage points in 2012.
    - TFP contribution: 0.2 percentage points in 2005 → -0.1 percentage points in 2012.
  - Projection: output gap to close in two to five years depending on model, but potential growth projected to remain below pre-crisis rates.
    - Average measure points to a ½ percentage point decline in potential growth over the medium term compared with 2000–07 average.
    - Production function indicates slowdown from close to 3 percent (2000–07) to just above 2 percent in the medium term, driven by subdued employment contribution.

### CHAPTER I. THE NATURAL RATE OF UNEMPLOYMENT
- Inflation vs. output gap signals
  - Year-on-year CPI inflation below zero suggests substantial cyclical slack.
  - Small output gap (around -1 percent of potential GDP in 2013) suggests much of actual unemployment is structural.
  - NIER (2013a) estimate: equilibrium unemployment just below 7 percent for 2013, gradually decreasing to 6.5 percent in 2017 as previously implemented reforms take effect.
- Drivers of low observed inflation
  - Low interest rates (notably mortgage rates) have a direct effect on Swedish CPI because CPI includes certain interest costs that HICP excludes.
    - Statistics Sweden computes CPIF and CPIX; CPIF and HICP evolve similarly.
  - Low imported inflation: non-energy industrial goods experienced deflation; services inflation was relatively constant at just below 2 percent until recently.
- Natural rate estimates and projections
  - Multivariate filter indicates the natural rate began to increase from the early 2000s and is projected to decline only gradually.
  - Projection: natural rate projected to remain around 6½ percent over the medium term; HP-filter approaches confirm elevated natural rate.

### CHAPTER I. CONCLUSION
- Sweden suffered a permanent output loss from the crisis; potential output growth will likely moderate relative to pre-crisis rates.
- The natural rate of unemployment has increased and is set to decline only gradually, remaining around 6½ percent according to model projections.

*Prepared by Lone Christiansen (EUR); August 2, 2013 — CHAPTER I. POTENTIAL OUTPUT AND UNEMPLOYMENT*

### Okun’s law: formulations, estimates, and interpretation
- Formulations used
  - Gap form: linear negative relationship between output gap Ygap and excess unemployment u − ū (as presented in the source).
  - Change form: relates GDP growth, ΔY/Y, to percentage point changes in unemployment, Δu (as presented in the source).
- Estimated parameter and interpretation
  - Linear relationship provides an estimate of parameter c of around 4.
    - Interpreted as: a 1 percentage point increase in the unemployment rate above the natural rate is associated with output which is 4 percent below potential (or output growth 4 percentage points lower).
- Natural rate estimates from Okun’s formulations
  - Gap form (production function approach with HP smoothing parameter of 100 for input variables): natural rate about 7½ percent on average over the sample.
  - Rolling regressions (quarterly data, 1996–2012) point to a natural rate just above 7 percent for the period ending 2012.
- Time variation and deviations
  - Both HP-filter and multivariate approach project a declining natural rate, but the ¾ percentage point decline in unemployment in 2011 was larger than Okun’s law suggested.
- Figures referenced in source (regression lines and statistics)
  - Figure 12: Change Form regression line y = -0.2884x + 0.8777, R² = 0.3591.
  - Figure 13: Gap Form regression line y = -0.2187x + 7.6403, R² = 0.1273.
  - Figure 14: Natural Rate of Unemployment — shows estimates for horizons 1996–2005 through 2003–2012 with confidence intervals.

### Methodological notes (filters and modeling assumptions)
- Production function approach
  - Cobb-Douglas; labor’s share α = 0.55.
  - TFP At backed out and HP-filtered; potential labor input from HP-filtered unemployment and actual labor force.
  - HP smoothing parameter used for input variables: 100.
  - End-point bias alleviated by extending variables with projections until 2025.
- Multivariate filter (Benes et al., 2010 implementation)
  - Jointly models output gap, unemployment gap, capacity utilization gap and links them to year-on-year core inflation π4t.
  - Okun’s law extended to include lag effects.
  - NAIRU modeled with transitory and more persistent shocks; potential output depends on trend growth and changes in NAIRU.
  - Effect of monetary policy captured via output gap response to inflation.
  - Estimation by Bayesian methods with priors:
    - Steady state unemployment rate set at 6 percent.
    - Steady state real GDP growth rate set at 2.4 percent (coinciding with the 2018 WEO projection).
    - Labor’s share set at 0.55.

### Policy implications from Okun’s law analysis
- With potential growth moderately weaker and the natural rate of unemployment elevated, policy focus should be on growth-enhancing reforms, especially labor market reforms.
- Suggested reforms and areas to address
  - Lifting labor market frictions.
  - Alleviating housing rigidities.
- Rationale: Tackling remaining structural weaknesses could raise medium-term growth and employment; Sweden already ranks high on many structural indicators but remaining weaknesses constrain outcomes.

### Appendix I — Measuring Banks’ Foreign Credit Exposures and Borrowers’ Reliance on Foreign Banking Credit
- Key exposure magnitudes
  - Swedish banks’ foreign credit exposures ≈ 150 percent of GDP and ≈ 1000 percent of Tier I capital.
  - Only Swiss banks are higher at 260 percent of GDP and almost 2000 percent of Tier I; UK banks are close at 145 percent of GDP and 925 percent of Tier I.
  - About 20 percent of Swedish banks’ foreign exposures originate from direct cross-border lending.
  - Non-Swedish Nordic and Baltic borrowers represent about 56 and 5 percent of total Swedish banks’ foreign credit exposures, respectively.
  - Swedish banks’ adjusted cross-border claims increased by about 30 percent since 2010.
  - Share of lending to Nordic countries increased from 61 percent in 2010 to 69 percent in 2012.
  - Lending to Baltic countries decreased from 11 percent in 2010 to 8 percent in 2012Q2.
  - “This increasing concentration to the Nordic countries (85 of total lending if we include lending to Swedish domestic borrowers) highlights the risks of shocks to the region.”
- Scenario analysis methodology (RES/MFU Bank Contagion Module)
  - Simulation rounds:
    - First round: losses on assets deplete bank capital based on assumed percentage losses by asset type.
    - Second round: banks restore capital adequacy (example threshold: 9 percent Core Tier I) via deleveraging.
    - Third round: funding shocks as banks reduce lending to other banks, possibly triggering fire sales and further losses.
    - Convergence when no further deleveraging occurs.
  - Aggregate results caveat: aggregate analysis may hide larger losses at individual bank level.

- Selected simulated outcomes (Table 2 entries preserved exactly)
  - Greece | 30 | 0.0 | 0.0 | 0.0
  - GIP 4/ | 30 | 0.0 | 0.1 | 0.1
  - Baltics Countries 5/ | 30 | 5.3 | 1.8 | 8.7
  - Denmark | 10 | 19.6 | 4.0 | 42.3
  - Finland | 10 | 10.6 | 2.5 | 17.9
  - Norway | 10 | 2.7 | 1.5 | 4.4
  - Nordic Countries 6/ | 10 | 57.2 | 8.1 | 104.5
  - Italy | 10 | 0.0 | 0.0 | 0.3
  - Spain | 10 | 0.0 | 0.0 | 0.4
  - France | 10 | 0.0 | 6.3 | 3.9
  - Germany | 10 | 0.0 | 1.3 | 5.4
  - Netherlands | 10 | 0.0 | 0.3 | 0.9
  - UK | 10 | 0.0 | 1.1 | 2.1
  - European Countries 7/ | 10 | 12.4 | 3.5 | 34.1
  - US | 10 | 4.0 | 1.8 | 8.2
  - European Countries & US 7/ | 20 | 64.6 | 9.0 | 127.7
- Interpretation of simulated outcomes
  - A 30 percent decline in Baltic asset values → Swedish bank losses of around 1¾ percent of GDP and slight deleveraging absent recapitalizations.
  - A 10 percent loss in Denmark or Finland → large Swedish bank losses that aggregate capital buffers could not offset, forcing double digit deleveraging and potentially severe second-round GDP effects.
  - A 10 percent loss on claims on Italy/Spain/France/Germany/Netherlands/UK → losses up to 1¼ of GDP; absorbable at aggregate level but simultaneous multi-country losses could trigger global deleveraging.
  - Indirect effects (confidence, asset prices, funding access) likely much larger than modeled direct effects.
- Policy implications and recommendations
  - Ensure adequate capital levels and macroprudential measures (e.g., minimum risk weights for mortgages, lower LTV ratios) in Sweden and across the Nordic region.
  - Regional coordination to avoid regulatory arbitrage within the Nordic region.
  - Address liability-side vulnerabilities: Swedish banks’ dependence on foreign external funding warrants ensuring Basel III Net Stable Funding Ratio targets are met in 2018 (or before); formal minimum intermediate targets desirable.
- Methodological adjustments to BIS data (Box 1)
  - Coverage break-in-series adjustments: BIS reports 84 series breaks during 2006–12 in consolidated banking statistics; 61 due to mergers and acquisitions; 23 due to expanded banking sector coverage.
  - Exchange rate variation adjustments: domestic-currency denominated local affiliates claims corrected using bilateral US dollar domestic currency exchange rate proxies and BIS locational statistics currency breakdown.
  - Magnitude: “almost up to 15 percent of total foreign claims in 2006.”
- Definitions and formulas (as presented)
  - Downstream exposure decomposition: Aij + Bij + Cij + Dij (definitions and formula text preserved as presented).
  - Downstream expected-loss indicator and upstream exposure formula preserved in form as given.
  - Note: upstream indicator is an upper bound given measurement limits.

### Appendix II — Drivers of Foreign Banking Exposures
- Creditor-side drivers (baseline and joint estimation)
  - Global risk aversion and presence of systemic bank crisis in creditor banking systems → reductions in banks’ foreign credit exposures.
  - Borrower countries’ GDP growth and deposit-to-loan ratio of creditor banking systems → positive relationship with foreign credit exposures.
  - TED spread and share of direct cross-border in total lending (Cred_CB_Share) not significant individually.
  - Joint estimation: systemic banking crisis in creditor system associated with ~3 percent decline in foreign credit exposures in a given quarter; interacting global financial variables with creditor systemic crisis intensifies decline:
    - At peak global financial variables, systemic crisis associated with ~8 percent decline (specification with risk aversion, column 9) and ~11 percent decline (specification with TED spreads, column 10).
    - One standard deviation increase in global financial variables with systemic crisis → decline ≈ 4½ percent.
- Sweden-specific creditor-slope findings (interacted Sweden dummy)
  - Global financial conditions less important for Sweden—Swedish banks increased cross-border exposures end-2011 and did not decrease much during 2008–09.
  - Swedish banks’ cross-border lending mainly through subsidiaries and heavy reliance on wholesale funding.
  - Swedish borrowers’ GDP growth may explain increase in Swedish banks’ foreign credit exposures, driven by Nordic performance.
  - Absence of a systemic banking crisis in Sweden contributed to increased foreign credit exposure.
- Borrower-side drivers
  - Borrowers exposed negatively when creditor banking systems faced systemic crises (up to -12 percent if all creditor systems in crisis).
  - Negative impact larger during high TED spreads; interaction TED * Borrower_CB_Share significant.
  - Borrower_CB_Share negative and significant: larger shares of cross-border lending → larger declines in borrowers’ exposures during stress.
  - One standard deviation increase in TED spreads reduced foreign banking exposures by 1¾ percentage points in baseline specification.
  - Borrower GDP growth coefficient positive: one percent GDP growth → up to 1/3 of a percent increase in foreign banking exposures (significance at 10 percent in some specs).
- Sweden-specific borrower-slope findings
  - Global financial conditions had an even larger impact (3 times larger) on Swedish borrowers’ foreign credit exposure evolution than average borrower.
  - Channels of from-whom and how countries borrow remain significant for Sweden.
  - Swedish GDP coefficient slightly lower than average borrower when interacted; borrower deposit funding coefficient reversal driven by Sweden’s high dependence on wholesale funding.

### Covered bonds in Sweden — usage, structure, and risks
- Use and size
  - Covered bonds increasingly used since 2006.
  - Total outstanding amount represents more than ½ of Swedish GDP.
  - First issued in 2006 by 3 institutions; since 2008, 7 approved banks have issued on Swedish market.
  - They represent more than EUR 200 billion as of 2011.
  - Only Denmark has larger exposure (covered bonds ≈ 150 percent of GDP).
- Currency and hedge structure
  - About one-quarter of covered bonds issued in foreign currency.
  - Banks hedge lending in Swedish kronor through currency swaps; swaps shorter duration than mortgages → rollover risk.
- Benefits (stability channels)
  - Longer funding duration (though below standard mortgage maturity).
  - Larger appeal to global investors; Basel III allows covered bonds as “Level 2” assets with 15 percent haircut for LCR calculation.
  - Better incentives for originators: issuer retains credit risk; LTV limits; external monitoring.
- Risks and potential harms to financial stability
  - Crowding out unsecured creditors.
  - Concentration risks within bank groups.
  - Increased liquidity risks due to large pledging of low-risk assets as collateral.
- Covered bond features and regulatory detail
  - Independent inspector must monitor cover pool and submit annual report and notify regulator of significant events.
  - Table: Mortgage Loans that can be Included in the Cover Pool (Percent)
    - Mortgage loans for housing purposes — Highest Loan-to-Value Ratio: 75 — Maximum Share of the Cover Pool: 100
    - Mortgage loans in property for agricultural purposes — Highest Loan-to-Value Ratio: 70 — Maximum Share of the Cover Pool: 100
    - Mortgage loans in property for commercial purposes — Highest Loan-to-Value Ratio: 60 — Maximum Share of the Cover Pool: 10
    - Public loans to local or central government — Highest Loan-to-Value Ratio: 100 — Maximum Share of the Cover Pool: 100
    - Complementary collateral (liquid claims on central and local government) — Highest Loan-to-Value Ratio: 100 — Maximum Share of the Cover Pool: 20

### Covered bonds — implications for a sharp house-price correction
- Direct effects on bank lending and buffers
  - Downward house-price correction → losses on bank lending potentially absorbable by existing buffers if historic parameters hold.
  - Mortgage lending historically had low default rates and low LGD: loans mostly for primary residences, full recourse, generous social benefits, historically stable/increasing prices.
- Four amplification channels tied to covered bonds
  - (i) Non-linear increases in overcollateralization needs: declines in house prices raise LTVs; banks must increase collateral; overcollateralization needs may rise faster than price falls.
  - (ii) Higher rollover risks: investor concern over covered bond liquidity/pricing could complicate rollovers of foreign-currency-denominated covered bonds and currency swaps.
  - (iii) Increasing risks for unsecured creditors: dynamic collateralization reduces assets available for unsecured creditors, potentially triggering higher funding costs or sudden stops.
  - (iv) Larger contingent liabilities for government: large house-price falls could increase contingent liabilities and sovereign risk premiums; deposits ≈ 100 percent of GDP.
- Box 1: Quantifying overcollateralization sensitivity
  - FSA 2013 Report: average LTV of mortgage portfolio ≈ 64 percent; LTV of new loans ≈ 70 percent.
  - Maximum LTV allowed in collateral pool for household mortgages: 75 percent.
  - Using midpoint LTVs and shocks of 10, 20, 30, 40 percent house-price falls:
    - Percentage of mortgage loans not meeting 75 percent threshold after shocks:
      - After 10 percent decline: 38
      - After 20 percent decline: 78
      - After 30 percent decline: 78
      - After 40 percent decline: 78
    - Reduction in eligible collateral is non-linear: 10 percent price reduction → eligible pool reduced by ≈ 5 percent; 30 percent drop → reduction ≈ 21 percent.
  - Caveats: midpoints used due to lack of full distribution; as of 2013Q1 average LTVs by bank: Swedbank 61 percent, SEB 59 percent, Nordea 55 percent, Handelsbanken 47 percent.
- Policy conclusions and recommendations
  - Covered bonds have advantages but can shift risks to unsecured creditors and taxpayers.
  - Consider issuance limits (examples cited) or linking deposit insurance premiums to size of collateral pools to internalize risks.
  - Reinforce monitoring of overcollateralization needs and rollover risks associated with foreign-currency issuance and currency swap hedges.

### Government contingent liabilities from Nordic-6 banks — scope, scenarios, and key estimates
- Nordic-6 overview (selected entries)
  - Nordea Bank AB (Sweden): 677.4 (EUR bil.) ; Share GDP in Host Country: 0.30 ; Assets/Host GDP: 410.9 ; 1.6
  - Danske Bank A/S (Denmark): 482.8 ; 0.21 ; 244.4 ; 2.0
  - DNB ASA (Norway): 321.8 ; 0.14 ; 394.7 ; 0.8
  - Swedbank AB (Sweden): 215.0 ; 0.09 ; 410.9 ; 0.5
  - Skandinaviska Enskilda Banken AB (Sweden): 285.7 ; 0.13 ; 410.9 ; 0.7
  - Svenska Handelsbanken AB (Sweden): 297.7 ; 0.13 ; 410.9 ; 0.7
  - Total: 2,280.4 ; 1.00
- Balance sheet method and bailout scenarios
  - Assets placed into Bankruptcy Estate (BE); depositors paid by Deposit Insurance Fund (DIF).
  - Net government loss = liquidation shortfall considering government claim priority.
  - Three scenarios:
    - Scenario A: insured depositors bailed out; uninsured depositors bailed in (100% haircut assumed for uninsured in Scenario A).
    - Scenario B: insured depositors bailed out; uninsured depositors receive 20% haircut (assumption).
    - Scenario C: insured depositors bailed out; uninsured depositors kept whole; senior unsecured creditors also held whole.
- Key assumptions (Table 2 highlights)
  - Fraction of deposits insured (Sweden, Denmark, Finland): 70%; Deposit Insurance Coverage = €100,000.
  - Fraction of deposits insured (Norway): 56%; Deposit Insurance Coverage ≈ €264,000.
  - Fraction recovered by DIF from BE of Synthetic Bank: 50% (relatively high due to senior secured liabilities).
  - Scenario A levy on uninsured depositors: 100% (assumption).
  - Scenario B levy on uninsured depositors: 20% (assumption).
- Selected fiscal-cost estimates (Table 4 — Sweden examples)
  - Sweden (2012 National GDP): Total Assets of Big 6 Banks (consolidated basis): 409.2 (EUR bil.)
    - Estimated Fiscal Costs by Depositor Base: Scenario A: 14.7 ; Scenario B: 5.9 ; Scenario C: 24.4
    - Estimated Fiscal Costs by Location of Parent: Scenario A: 69.1 ; Scenario B: 5.9 ; Scenario C: 1.5
  - Denmark (2012 National GDP): Total Assets 482.8 (EUR bil.)
    - Estimated Fiscal Costs by Depositor Base: Scenario A: 11.2 ; Scenario B: 19.1 ; Scenario C: 39.2
    - By Location of Parent: Scenario A: 13.3 ; Scenario B: 24.3 ; Scenario C: 50.0
  - Finland and Norway entries also reported in table; Nordic aggregates provided.
- Main quantitative findings on contingent liabilities
  - Swedish government contingent liabilities from supporting depositors of the six largest Nordic banks range from just below 20 percent of GDP to 90 percent of GDP, depending on bailout scope and burden sharing rule.
  - Under deposit-base burden sharing, losses vary from 17–60 percent of GDP.
  - Under location-of-parent burden sharing, losses vary from 26–90 percent of GDP.
  - Mid-range across burden sharing rules suggests contingent liabilities amounting to 30–45 percent of GDP.
  - If government only supports insured depositors, estimate drops to between 20 and 30 percent of GDP.
  - Estimates are large relative to ex-post cost of 1990s Swedish bank bailout (less than 5 percent of GDP).
- Policy implications and optimal fiscal buffer considerations
  - Contingent liabilities potentially large; policy responses include strengthening bank resilience, reducing household credit growth risks, improving internal and Nordic macroprudential coordination.
  - Fiscal buffers could be built by limiting gross debt levels and/or building liquid reserves in a dedicated fund.
  - Trade-offs: dedicated funds may exacerbate moral hazard but support an active sovereign debt market.
  - Uncertainties: approach to uninsured depositors and bond holders; level of bank losses relative to end-2012 data.

### Fiscal-buffer model — objectives, setup, and core results
- Model setup and risk structure
  - Government smooths spending subject to dynamic budget constraint; fixed tax revenue each period.
  - Single risk: one-time contingent liability Z (>0) realized with constant probability φ each period; once realized, no further risks.
  - Preferences: household lifetime utility additively depends on private consumption C_t and government expenditure G_t; discount factor β ∈ (0,1); CRRA parameter ρ.
- Main analytical and numerical results
  - Long-run fiscal buffer target should approximately match the size of the contingent liability Z.
  - Optimal policy slightly “over-buffers”: target fiscal buffer somewhat larger than Z to smooth spending post-shock because higher post-shock debt raises interest payments.
  - Speed of accumulation: front-loaded in initial years, then tapers as buffer approaches target.
- Numerical illustration (parameter example)
  - Contingent liabilities Z/GDP = 0.3 (30 percent of GDP) → model-implied target fiscal buffer ≈ 34 percent of GDP.
  - Government’s maximum acceptable level of gross debt after contingent liabilities realized: 60 percent of GDP.
  - Implied ex-ante target gross debt level: around 26 percent of GDP, to be reached gradually.
- Sensitivity and intuition
  - With constant per-period crisis probability φ, long-horizon likelihood of crisis →1, so buffer ≈ Z.
  - Derivative dB/dZ > 1: increase in Z by one unit leads to >1 unit increase in target buffer due to post-shock interest income loss.
  - Higher φ increases precautionary motive; target buffer wealth increases with φ.
- Parameters used in example (Table A1)
  - Gross Return (R): 1.01
  - Discount Factor (β): 0.97
  - CRRA parameter (ρ): 1
  - Probability of Shock (φ): 0.01
  - Tax Revenue to Structural GDP (T/GDP): 0.52
  - Contingent Liability to Structural GDP (Z/GDP): 0.3
- Policy implications tied to contingent-liability estimates
  - Swedish contingent liabilities from financial sector estimated in a range just below 20 to 90 percent of GDP with large uncertainty.
  - Policy emphasis: fast financial reforms, measures to cool household credit growth, strengthen banks’ liquidity and capital to reduce systemic risk and limit contingent liability size.

*Prepared by Lone Christiansen (EUR); August 2, 2013 — CHAPTER I. POTENTIAL OUTPUT AND UNEMPLOYMENT*

### CHAPTER I. POTENTIAL OUTPUT AND UNEMPLOYMENT _____________________________ 5

### CHAPTER I. POTENTIAL OUTPUT AND UNEMPLOYMENT

### A. The Great Recession in Sweden
- Sweden experienced a strong hit from the global financial crisis of 2008–09 and moderated growth after an initial rebound.
- Real GDP: contraction of 5 percent in 2009; rebound to above 3½ percent in 2010 and 2011 before slowing.
- Unemployment: increased to 8 percent by 2012.
  - Note: classification of some full-time students as unemployed accounts for about 2 percentage points of overall unemployment.
- Sectoral impact: manufacturing employment declined by close to 10 percent in 2009.
- Output level: In 2012, output was 10 percent below the level that would have been reached had output continued to grow at the average 2000–07 growth rate of 3¼ percent.
- Forecast implication: the output loss relative to the 2007 trend continues to widen over the forecast horizon, implying a permanent output loss.

### B. Potential Output
- Estimation approaches:
  - Univariate HP filter with a range of smoothing parameters.
  - Production function approach (Cobb-Douglas) with total factor productivity (TFP) and unemployment determined by HP filter, combined with actual labor force and capital stock.
  - Multivariate filter (Benes et al. (2010) style) that jointly determines potential output and the natural rate of unemployment.
  - To alleviate end-point problem from HP-filter, forecasts extended until 2025.
- Output gap estimates (range and central tendencies):
  - Fund staff (World Economic Outlook): output gap estimates shown in Table 1 and Figure 4.
  - For 2013, methods point to a negative output gap in the range of -1.6 (multivariate filter) to -0.5 (HP filter and production function with low smoothing) percent of potential output.
  - The average measure stands at a negative gap of 1 percent.
- Cross-institutional output gap estimates (Table 1 excerpted values by year):
  - European Commission: 2010 -1.6; 2011 -0.1; 2012 -1.3; 2013 -1.6; 2014 -1.1 (percent of potential output).
  - Ministry of Finance: 2010 -3.1; 2011 -1.3; 2012 -2.2; 2013 -3.5; 2014 -3.3.
  - National Institute of Economic Research: 2010 -3.6; 2011 -1.5; 2012 -1.9; 2013 -2.1; 2014 -1.7.
  - Riksbank (average of quarterly estimates): 2010 -2.8; 2011 -0.5; 2012 -0.9; 2013 -1.3; 2014 -0.6.
  - Fund staff (World Economic Outlook): 2010 -1.0; 2011 0.8; 2012 -0.4; 2013 -1.0; 2014 -0.7.
- Potential growth:
  - 2012 potential growth estimates range from just below 2 to 2¼ percent.
  - 2000–07 average potential growth close to 3 percent (average estimate).
  - Decomposition (production function): contributions to potential growth declined from 2005 to 2012:
    - Employment contribution: from 0.7 percentage points in 2005 to 0.5 percentage points in 2012.
    - Total factor productivity (TFP) contribution: from 0.2 percentage points in 2005 to -0.1 percentage points in 2012.
  - Projection: output gap will close in two to five years depending on model, but potential growth is projected to remain below pre-crisis rates.
    - Average measure points to a ½ percentage point decline in potential growth over the medium term compared with 2000–07 average.
    - Production function indicates slowdown from close to 3 percent (2000–07) to just above 2 percent in the medium term, driven by subdued contribution from employment.

### C. The Natural Rate of Unemployment
- Tension between inflation and output gap signals:
  - Year-on-year CPI inflation below zero suggests substantial cyclical slack.
  - Small output gap (around -1 percent of potential GDP in 2013) suggests much of actual unemployment is structural.
  - NIER (2013a) estimate: equilibrium level of unemployment just below 7 percent for 2013, gradually decreasing to 6.5 percent in 2017 as previously implemented reforms take effect.
- Drivers of low observed inflation:
  - Low interest rates (notably mortgage rates) have a direct effect on Swedish CPI because CPI includes certain interest costs that HICP excludes.
    - Statistics Sweden computes CPIF (CPI at fixed interest rates) and CPIX (CPI excluding all interest costs for owner-occupied dwellings) to support monetary policy.
    - CPIF and HICP evolve similarly, underlining importance of interest rate effects.
  - Low imported inflation: non-energy industrial goods (high import content) have experienced deflation; services inflation (mainly domestic) has been relatively constant at just below 2 percent until recently.
- Estimating the natural rate:
  - Multivariate filter indicates the natural rate of unemployment started to increase from the early 2000s.
  - Model suggests the natural rate will begin to decline over the forecast horizon but remain elevated.
  - Projection: natural rate of unemployment projected to remain around 6½ percent over the medium term; HP-filter approaches confirm elevated natural rate.

### D. Conclusion
- Sweden suffered a permanent output loss from the crisis; potential output growth will likely moderate relative to pre-crisis rates.
- The natural rate of unemployment has increased and is set to decline only gradually over the medium term, remaining around 6½ percent according to model projections.

*Prepared by Lone Christiansen (EUR); August 2, 2013 — CHAPTER I. POTENTIAL OUTPUT AND UNEMPLOYMENT*

### 11.      Okun’s law can be used to generate an

### 11.      Okun’s law can be used to generate an

### Okun’s law: formulations and interpretation
- Two forms of Okun’s law are used:
  - Gap form (linear negative relationship between the output gap, Ygap, and excess unemployment, u − ū): (1) (u − ū) c − − = Ygap  or  Ygap = c (1 − u/ū)  (as presented in the source).
  - Change form (relates GDP growth, ΔY/Y, to percentage point changes in the unemployment rate, Δu): (2) c k − Δ = ΔY/Y  or  Δu = c (1 − ΔY/Y)  (as presented in the source).
- Parameters:
  - c indicates the sensitivity (regression slope) linking unemployment deviations to output gap or output growth.
  - k indicates the average growth rate of potential output.
- Econometric interpretation:
  - The intercept in a regression of the unemployment rate on the output gap can be interpreted as the (time-invariant) natural rate of unemployment.
- Data and sample:
  - Gap and change forms estimated over 1993–2012 (figures reference 1993–2012).
  - Rolling regressions use quarterly data during 1996–2012; output gap estimates for rolling regressions come from the multivariate filter approach.

### Empirical findings from the estimates
- Parameter estimates and implications:
  - The linear relationship provides an estimate of the parameter c of around 4.
    - Interpreted as: a 1 percentage point increase in the unemployment rate above the natural rate (or a 1 percentage point increase in the actual rate) is associated with output, which is 4 percent below potential (or output growth, which is 4 percentage points lower).
- Natural rate estimates:
  - From the gap form (production function approach with smoothing parameter of 100 for HP-filtered input variables), the natural rate of unemployment is about 7½ percent on average over the sample.
  - This estimate is similar to the smallest value for 2012 produced by the filtering methods but above the low rates of the early 2000s.
- Time variation and deviations:
  - Any linear relationship among these variables likely changes over time; both HP-filter and multivariate approach project a declining natural rate of unemployment.
  - The ¾ percentage point decline in the unemployment rate in 2011 was somewhat larger than that suggested by Okun’s law.

### Rolling regressions and time variation
- Rolling regression approach:
  - Regressions of equation (1) (gap form) run over varying time periods with the actual unemployment rate as the left-hand-side variable; output gap from the multivariate filter is used.
  - Each rolling estimate is based on Okun’s law in gap form over various time horizons, with data quarterly.
- Main rolling-regression result:
  - Rolling regressions confirm an increase in the natural rate of unemployment in the second half of the 2000s.
  - The most recent rolling-regression period (ending 2012) points to a natural rate of just above 7 percent.
- Visualization references (figures cited in source):
  - Figure 12: Okun’s Law: Change Form, 1993–2012 (regression line y = -0.2884x + 0.8777, R² = 0.3591; axes: Change in unemployment rate and Real GDP growth (percent)).
  - Figure 13: Okun’s Law: Gap Form, 1993–2012 (regression line y = -0.2187x + 7.6403, R² = 0.1273; axes: Unemployment rate (percent) and Output gap (percent of potential)).
  - Figure 14: Okun’s Law: Natural Rate of Unemployment (Percent) — shows estimates for horizons 1996–2005 through 2003–2012 with confidence intervals.

### Methodological notes (filters and model assumptions)
- Production function approach:
  - Assumes Cobb-Douglas production function; total factor productivity At backed out and HP-filtered.
  - Labor’s share in income, α, is set at 0.55.
  - Potential labor input determined through an HP-filtered unemployment rate and the actual labor force.
  - To alleviate HP end-point bias, variables are extended with projections until 2025.
- Multivariate filter (Benes et al., 2010 approach as implemented):
  - Output gap, unemployment gap, and capacity utilization gap are modeled jointly and linked to year-on-year core inflation, 4t, with an inflation equation.
  - Okun’s law is extended to include lag effects, relating unemployment and output gaps.
  - Laws of motion for equilibrium variables include transitory and more persistent shocks; NAIRU is modeled as affected by transitory and more persistent shocks.
  - Potential output depends on trend growth of potential GDP and changes in the NAIRU.
  - The effect of monetary policy is captured through the output gap responding negatively when inflation is above long-term inflation expectations.
  - Estimation by Bayesian methods with priors to ensure reasonable parameter values.
  - Specific prior/steady-state settings used:
    - Steady state unemployment rate set at 6 percent.
    - Steady state real GDP growth rate set at 2.4 percent (coinciding with the 2018 WEO projection).
    - Labor’s share in income set at 0.55.
  - Smoothing parameter: production function approach used smoothing parameter of 100 for HP-filtered input variables (as noted in the source).

### Policy implications and conclusion
- Main policy implication:
  - With potential growth moderately weaker and the natural rate of unemployment to remain elevated, policies should focus on growth-enhancing reforms, especially in the labor market.
- Suggested reforms and areas to address:
  - Lifting labor market frictions.
  - Alleviating housing rigidities.
- Rationale:
  - Tackling remaining areas of structural weakness could help raise the medium-term growth and employment outlook.
  - Sweden already ranks high in most comparisons of structural indicators owing to extensive reforms in the 1990s, but remaining weaknesses still constrain medium-term outcomes.

*Source: IMF staff analysis as presented in the chapter titled “Okun’s law can be used to generate an alternative estimate of the natural rate of unemployment” (excerpts from the Sweden staff report).*

### Appendix I for more methodological details).

### Appendix I. Measuring Banks’ Foreign Credit Exposures and Borrowers’ Reliance on Foreign Banking Credit

### Key findings on Swedish banks' foreign exposures
- Swedish banks’ foreign credit exposures represent about 150 percent of GDP and about 1000 percent of Tier I capital buffers.
- These figures are only surpassed by Swiss banks (260 percent of GDP and almost 2000 percent of banks’ Tier I capital) and are very close to UK banks (145 percent of GDP and 925 percent of Tier I).
- Only about 20 percent of Swedish banks’ foreign exposures originate from direct cross-border lending.
- Non-Swedish Nordic and Baltic borrowers represent about 56 and 5 percent of total Swedish banks’ foreign credit exposures, respectively.
- Swedish banks’ adjusted cross-border claims increased by about 30 percent since 2010.
- Share of lending to Nordic countries increased from 61 percent of the foreign loan portfolio in 2010 to 69 percent in 2012.
- Lending to Baltic countries decreased from 11 percent in 2010 to 8 percent in 2012Q2.
- “This increasing concentration to the Nordic countries (85 of total lending if we include lending to Swedish domestic borrowers) highlights the risks of shocks to the region.”

### Scenario analysis and simulated spillovers (method: RES/MFU Bank Contagion Module)
- Simulation rounds:
  - First round: losses on assets that deplete bank capital partially or fully, based on assumed percentage losses on asset types (public sector, banking sector, non-bank private sector).
  - Second round: banks restore capital adequacy to at least a threshold (example: 9 percent Core Tier I for European banks) through deleveraging (asset sales, refusal to roll-over loans).
  - Third round: banks reduce lending to other banks (funding shocks), potentially triggering fire sales, further deleveraging, and additional losses.
  - Convergence when no further deleveraging occurs.

- Aggregate results caveat:
  - Analysis performed at aggregate level and may hide larger losses for individual banks; aggregate results should be interpreted with care.

### Selected simulated outcomes (from Table 2)
- Table column headings: Shock Originating From | Magnitude 1/ | Deleveraging Need 2/ | Swedish Lenders' Losses (percent GDP) | Impact on Credit Availability (percent of GDP) 3/
- Selected entries (preserve exact values):
  - Greece | 30 | 0.0 | 0.0 | 0.0
  - GIP 4/ | 30 | 0.0 | 0.1 | 0.1
  - Baltics Countries 5/ | 30 | 5.3 | 1.8 | 8.7
  - Denmark | 10 | 19.6 | 4.0 | 42.3
  - Finland | 10 | 10.6 | 2.5 | 17.9
  - Norway | 10 | 2.7 | 1.5 | 4.4
  - Nordic Countries 6/ | 10 | 57.2 | 8.1 | 104.5
  - Italy | 10 | 0.0 | 0.0 | 0.3
  - Spain | 10 | 0.0 | 0.0 | 0.4
  - France | 10 | 0.0 | 6.3 | 3.9
  - Germany | 10 | 0.0 | 1.3 | 5.4
  - Netherlands | 10 | 0.0 | 0.3 | 0.9
  - UK | 10 | 0.0 | 1.1 | 2.1
  - European Countries 7/ | 10 | 12.4 | 3.5 | 34.1
  - US | 10 | 4.0 | 1.8 | 8.2
  - European Countries & US 7/ | 20 | 64.6 | 9.0 | 127.7

- Interpretation of simulated outcomes:
  - A 30 percent decline in Baltic asset values could result in Swedish bank losses of around 1¾ percent of GDP and, absent recapitalizations, would require slight deleveraging.
  - A 10 percent loss in Denmark or Finland would trigger large Swedish bank losses that aggregate capital buffers could not offset, forcing double digit bank deleveraging and potentially severe second-round effects on overall GDP growth.
  - A 10 percent loss on claims on either Italy, Spain, France, Germany, Netherlands, or UK borrowers could trigger losses up to 1¼ of GDP; these could be absorbed at aggregate level but simultaneous losses across countries could trigger global deleveraging.
  - Indirect effects (confidence, asset prices, funding access) are likely to be much larger than modeled direct effects, particularly if access to wholesale funding is impaired.

### Policy implications and recommendations
- The large cross-border exposures of Swedish banks, particularly to Nordic markets, strengthen the case for:
  - Adequate capital levels.
  - Macroprudential measures such as minimum risk weights for mortgages and lower LTV ratios in Sweden and across the Nordic region to reduce the likelihood and impact of house price corrections.
  - Regional coordination to avoid regulatory arbitrage within the Nordic region.
- Addressing liability-side vulnerabilities:
  - Swedish banks are dependent on foreign external funding sensitive to global financial conditions; a re-emergence of global risk aversion would impact Sweden beyond direct asset exposures.
  - A regulatory measure: ensure Basel III Net Stable Funding Ratio targets are met in 2018 (or before) by all banks; formal minimum intermediate targets would be desirable.

### Methodological adjustments to BIS data (Box 1)
- Two main adjustments required for BIS CBS time series analysis:
  - Coverage break-in-series adjustments:
    - BIS reports 84 series breaks during 2006–12 in BIS consolidated banking statistics at ultimate borrower risk basis.
    - About 61 breaks are due to mergers and acquisitions among foreign banks.
    - Other 23 coverage break-in-series are driven by an expansion of banking sector coverage (e.g., inclusion of former investment banks).
    - Example: US 2009Q1 USD 1,334 billion break-in-series when former investment banks became banks (e.g., Goldman Sach’s foreign claims existed before 2009Q1).
  - Exchange rate variation adjustments:
    - Domestic-currency denominated local affiliates claims are corrected following the bilateral US dollar domestic currency exchange rate, proxied by their share of total BIS CBS foreign claims at immediate borrower basis.
    - This procedure identifies foreign-currency denominated local affiliates’ claims, assumed to be in Euros in Europe and US dollars in the rest of countries.
    - Bilateral CBS cross-border claims are adjusted using the currency breakdown proxy available from BIS locational banking statistics (among US. Dollar, Euro, British Pound, Japanese Yen, and Swiss Francs).
- Magnitude of adjustments: “The magnitude of these adjustments is important (almost up to 15 percent of total foreign claims in 2006).”

### Definitions and formulas (as presented)
- Downstream exposure decomposition:
  - A creditor country i downstream exposure would be equal to Aij + Bij + Cij + Dij where:
    - Aij = captures the direct cross-border exposure from creditor country i on debtor country j (ij claimsborderCrossA).
    - Bij = captures the exposure to subsidiaries and branches, taking into account the legal differences between them (branch_ij subs_ij subs_ij assetstotaldepositsassetstotalB__).
    - Cij = represents the non-identified exposure by bank level data with respect to BIS reported affiliates claims ( branchsubs ijijij assetstotalclaimslocal_&).
    - Dij = capture off-balance sheet exposure from country i banks on country j based on BIS data (ijijijij scommitmentcreditguaranteessderivativeD_).
- Downstream expected-loss indicator:
  - Di = sum over j=1 to N of ( (Aij + Bij + Cij + Dij) * Vj / Zi ), where Zi is a scaling factor (GDP or total banking sector assets in country i) and Vj = Pr(crisis_j) * LGD_j.
- Upstream exposure (borrowers’ reliance on foreign banking credit) formula (as given):
  - )1,_(1( * ijijijj ratioloandepositMinclaimsLocalclaimsborderCrossExposureUpstream
    - where ij claimsrCrossborde captures direct cross-border claims from country i on country j; ij claimsLocal captures affiliates claims; and )1,_(1 ij ratioloandepositMin is a proxy for the proportion of loans not financed by local consumer deposits.
- Notes on upstream indicator:
  - The upstream indicator could be considered an upper bound because the amount of lending by affiliates funded by their parent banks cannot be fully measured with available bank-level data.

*Italic — Appendix I, “Measuring Banks’ Foreign Credit Exposures and Borrowers’ Reliance on Foreign Banking Credit,” IMF staff (extracted from the provided content).*

### Appendix II. Drivers of Foreign Banking Exposures

### Appendix II. Drivers of Foreign Banking Exposures

### Drivers of Creditor Banks’ Foreign Credit Exposures
- Baseline findings (individual variable estimations):
  - Higher global risk aversion and the presence of systemic bank crisis in creditor banking systems are linked with a reduction in banks’ foreign credit exposures.
  - An increase in borrower countries’ GDP growth or in the deposit to loan ratio of the creditor banking systems display a positive significant relationship with variations in foreign credit exposures.
  - The TED spread and the share of direct cross-border in total lending (Cred_CB_Share) do not display statistically significant correlations when considered individually.
- Joint estimation (columns 7–10):
  - The presence of a systemic banking crisis in the creditor banking system is associated with a decline in foreign credit exposures of about 3 percent in a given quarter.
  - This result is robust to the introduction of time fixed effects.
  - Interactions: when global financial variables are interacted with creditor systemic crisis, both an increase in global risk aversion or funding spreads reinforce the fall in foreign credit exposures.
    - At the peak of the global financial variables in the sample, the presence of a systemic banking crisis would be associated with a decline in foreign credit exposures of about 8 percent (specification with risk aversion, column 9) and about 11 percent (specification with TED spreads, column 10).
  - A one standard deviation increase in global financial variables, together with the presence of a systemic banking crisis, would be associated with a decline in foreign credit exposures of about 4½ percent.
- Overall interpretation:
  - Creditor banking systems’ foreign exposures were driven by the presence of bank systemic crisis and global financial conditions.
  - The characteristic of lending—through either direct cross-border or affiliate lending—does not seem as relevant.
  - Demand factors in borrowing countries proxied by borrowers GDP growth and creditor banks’ funding structure characteristics do not seem to be statistically significant drivers in the full-sample creditor regressions.

### Allowing Different Coefficient Slopes for Sweden (creditors)
- Method: interacted variables capturing each explanatory variable and a Sweden dummy were introduced (short time series: 24 quarters).
- Key indications:
  - Global financial conditions were not as important for Sweden—Swedish banks increased cross-border exposures at the end of 2011 and did not decrease much during 2008–09.
  - Evidence is mixed for other factors; reversals in signs suggest characteristics of lending and parent funding coefficients are influenced by Swedish banks lending cross-border mostly through subsidiaries and heavy reliance on wholesale funding.
  - Swedish borrowers’ GDP growth (proxy of demand) might explain the increase in Swedish banks’ foreign credit exposures, likely driven by Nordic country performance in recent years.
  - The fact that Sweden did not experience a systemic banking crisis played a role in their increase in foreign credit exposure.

### Drivers of Borrowers’ Foreign Banking Credit
- Broader set of significant factors compared with creditor-side analysis.
- From whom a country borrows:
  - Borrowing countries operating with creditor banking systems experiencing systemic banking crisis suffered a negative change in borrowers’ foreign exposures (up to -12 percent if all creditor banking systems were through systemic banking crisis).
  - This negative impact was larger during high TED spreads (interactive coefficient column 8), indicating limited substitution between creditor banking systems.
  - The negative impact of systemic banking crisis in creditor systems was lower the higher the borrower deposit to loan ratio (interaction coefficient column 9), suggesting domestic banking systems less dependent on non-deposit funding insulated themselves better.
- How a country borrows:
  - A larger share of cross-border lending on total borrower foreign claims is associated with a larger decline in borrowers’ exposures (Borrower_CB_Share negative and significant).
  - Interaction TED * Borrower_CB_Share was statistically significant, showing deterioration during crisis peaks was higher with larger direct cross-border shares.
  - Interaction Cred_Syst_Crisis * Borrower_CB_Share was negative and significant, highlighting that creditor systemic banking crises amplified the negative effect of large direct cross-border shares.
- International financial conditions:
  - In the baseline specification (column 6), a one standard deviation increase in TED spreads reduced foreign banking exposures by 1¾ percentage points.
  - TED spreads interacted with other borrower variables increased their negative impact.
- Domestic demand and other factors:
  - Borrower GDP growth coefficient is positive (significant at 10 percent in some specifications): a one percent increase in GDP growth increases foreign banking exposures by up to 1/3 of a percent.
- Overall interpretation:
  - Both creditor-side distress and the composition of borrower liabilities (cross-border vs. local affiliate claims), together with global funding conditions, were key drivers of changes in borrowers’ foreign banking exposures.

### Allowing Different Coefficient Slopes for Sweden (borrowers)
- Interacted Sweden dummy results:
  - Global financial conditions had an even larger impact (3 times larger) on the evolution of Swedish borrowers’ foreign credit exposure than for the average borrower in the panel.
  - From whom countries borrowed and how they borrowed remain significant channels for Sweden as for other borrowers.
  - The coefficient for Swedish GDP (sum of base and Sweden interaction) is slightly lower than for the average borrower when GDP is considered alone, and may be negative with other controls included.
  - Reversal in borrower deposit funding coefficient is driven by Sweden’s high dependence on wholesale funding.

### Covered Bonds and Financial Stability — A. The Use of Covered Bonds in Sweden
- Covered bonds usage:
  - Covered bonds have been increasingly used by Swedish banks since 2006.
  - Total outstanding amount currently represents more than ½ of Swedish GDP.
- Effects on financial stability:
  - Enhanced stability channels: longer funding maturity, better incentives to bank issuers, broader appeal to global investors.
  - Potential risks: extended use could increase liquidity risks and government contingent liabilities.
- Covered bonds features (contrast with traditional mortgage bonds):
  - (i) Governed by a well-defined regulatory framework ensuring cover pool quality (e.g., maximum Loan to Value (LVT) ratios and maximum share of riskier assets).
  - (ii) Cover pool is dynamic.
  - (iii) Holder of covered bonds has seniority on the cover pool if the issuing institution suspends payments.
  - (iv) Credit risk remains on the balance sheet of the issuer.
- Swedish regulatory detail:
  - An independent inspector must monitor the cover pool and submit an annual report and notify the regulator of significant events.
- Table 1 — Mortgage Loans that can be Included in the Cover Pool (Percent):
  - Mortgage loans for housing purposes — Highest Loan-to-Value Ratio: 75 — Maximum Share of the Cover Pool: 100
  - Mortgage loans in property for agricultural purposes — Highest Loan-to-Value Ratio: 70 — Maximum Share of the Cover Pool: 100
  - Mortgage loans in property for commercial purposes — Highest Loan-to-Value Ratio: 60 — Maximum Share of the Cover Pool: 10
  - Public loans to local or central government — Highest Loan-to-Value Ratio: 100 — Maximum Share of the Cover Pool: 100
  - Complementary collateral, such as liquid claims on central and local government — Highest Loan-to-Value Ratio: 100 — Maximum Share of the Cover Pool: 20

*Source: _cr13277 - Appendix II. Drivers of Foreign Banking Exposures*

### 2.      Swedish banks started issuing covered bonds in 2006 and they now represent more

### 2.      Swedish banks started issuing covered bonds in 2006 and they now represent more than ½ of Swedish GDP

### B. Implications for Financial Stability
- Covered bond market growth and size
  - The first Swedish covered bonds were issued in 2006 by 3 institutions; since 2008, 7 approved banks have issued them on the Swedish market.
  - They represent more than EUR 200 billion as of 2011.
  - Only Denmark, where covered bonds represent about 150 percent of GDP, has a larger exposure to this type of funding.
- Currency and hedge structure
  - About one-quarter of covered bonds are issued in foreign currency.
  - Banks hedge lending in Swedish kronor through currency swaps to avoid exchange rate fluctuations, which avoids foreign currency liquidity risks but introduces rollover risks because these swaps have shorter duration than mortgages.
- Channels through which covered bonds could increase financial stability
  - Longer funding duration: issued at fixed longer-term periods than traditional bonds (but still below a standard mortgage in Sweden), improving asset-liability management and funding certainty.
  - Larger appeal for bond holders: higher demand due to secured nature and preferential weighting for banks’ liquidity requirements (Basel III allows covered bonds as “Level 2” assets; they have only 15 percent haircut for Liquidity Coverage Ratio calculation). Higher demand typically lowers spreads and funding costs; in distressed markets covered bonds can offer a safe alternative to government bonds.
  - Better incentives for originators: issuer retains the credit risk of assets in the cover pool (bank issuer retains a 100 percent interest in the asset pool), incentivizing credit quality; additional measures include loan-to-value limits, credit risk assessment of pools, and external monitoring of loan performance.
- Channels through which covered bonds could harm financial stability
  - Crowding out of unsecured creditors: uninsured depositors and other creditors may be less willing to make unsecured loans as assets are pledged as collateral.
  - Concentration risks: banks could replace third-party bonds they hold with mortgage covered bonds issued by other parts of their group, increasing exposures to the same clients.
  - Increased liquidity risks: large portions of low-risk assets pledged as collateral reduce banks’ future access to liquidity; uncollateralized assets serve as unused liquidity buffers for unexpected needs (committed credit lines, margin calls).

### C. Potential Impact of a Sharp Reduction in House Prices
- Direct effect on bank lending and buffers
  - A downward correction of house prices would trigger losses on bank lending, but these could be absorbed by existing bank buffers assuming historic parameters remain stable.
  - Mortgage lending in Sweden has historically had low default rates and low loss–given-default rates because loans are mostly for primary residences with full recourse, generous and reliable social benefits, and historically stable/increasing housing prices.
- Four channels related to covered bonds that could amplify stress
  - (i) Non-linear increases in overcollateralization needs
    - Declines in house prices increase loan-to-value ratios of existing mortgages; banks may need to increase collateral to meet cover pool overcollateralization requirements.
    - The amount of collateral required to replenish cover pools might increase faster, and at increasing speeds, than the decrease in house prices given the current LTV distribution of mortgages.
  - (ii) Higher rollover risks
    - A sharp fall in house prices could lead investors to question covered bond liquidity and pricing, complicating rollovers of foreign-currency-denominated covered bonds and associated currency swaps.
    - Rollover risk could extend to currency swaps used to hedge domestic-currency lending funded by foreign-currency borrowing.
  - (iii) Increasing risks for unsecured creditors
    - Lower house prices reduce assets available to unsecured creditors in bankruptcy due to dynamic collateralization and effective subordination of unsecured creditors.
    - This could trigger sharp increases in funding costs or sudden stops in unsecured funding flows, particularly if perceived implicit government support is low.
  - (iv) Impact of larger contingent liabilities on government funding costs
    - Very large falls in house prices could increase contingent government liabilities and drive up sovereign risk premiums.
    - Depositors and banks may not price increased default risk, exposing deposit insurance schemes (and the government) to large costs in bank failures.
    - Note: the level of deposits in Swedish banks is about 100 percent of GDP.

- Box 1: Quantifying overcollateralization sensitivity to house price declines
  - Key parameters and data:
    - FSA 2013 Report: average LTV ratio of banks’ mortgage loan portfolio is approximately 64 percent; LTV ratio of new loans is about 70 percent.
    - Maximum LTV ratio allowed in the collateral pool for household mortgages: 75 percent.
  - Illustration using midpoint LTVs of 2012 distribution and shocks of 10, 20, 30, and 40 percent house price falls:
    - Example: 38 percent of mortgage loans will not meet the 75 percent threshold for full inclusion in cover pools if house prices fall by 10 percent; this ratio increases to 78 percent with a 20 percent house price drop.
  - Aggregate results (as presented)
    - Average LTV ratio of mortgage portfolio: approximately 64 percent.
    - LTV ratio of new loans: about 70 percent.
    - Percentage of Mortgage Loans not meeting 75 percent LTV threshold by shock:
      - After 10 percent decline in house prices: 38
      - After 20 percent decline in house prices: 78
      - After 30 percent decline in house prices: 78
      - After 40 percent decline in house prices: 78
    - Reduction in eligible collateral is non-linear: a 10 percent reduction in house prices reduces the eligible pool by about 5 percent; a 30 percent drop triggers a reduction in eligible collateral of about 21 percent.
  - Caveats:
    - Calculations use midpoints of LTV categories due to lack of full mortgage distribution data.
    - Stock of mortgage loans used as proxy for cover pool composition; as of 2013Q1, average LTVs of specific banks: Swedbank 61 percent, SEB 59 percent, Nordea 55 percent, Handelsbanken 47 percent.

### D. Conclusion and policy implications
- Net assessment
  - Covered bonds have many advantages but can shift risks to unsecured creditors and taxpayers via government contingent liabilities.
  - The level of covered bonds could reach levels where even small declines in house prices produce important non-linear effects.
- Policy options and recommendations mentioned
  - Some countries establish covered bond issuance limits as percentages of total assets/liabilities (examples cited: Canada, Australia, New Zealand, US) or as a function of the capital ratio of each bank (example cited: Italy).
  - Consider making deposit insurance premiums charged to banks a function of the size of collateral pools in bank balance sheets to incentivize banks to internalize risks associated with larger covered bond volumes.
  - Reinforce monitoring of overcollateralization needs and rollover risks associated with foreign-currency issuance and currency swap hedges.

*International Monetary Fund — excerpt from Sweden country report chapter on covered bonds and fiscal policy.*

### References

### _cr13277 - References

### References cited
- Bohn, Henning, 1998, "The Behavior of U.S. Public Debt And Deficits," The Quarterly Journal of Economics, MIT Press, vol. 113(3), pages 949–963, August.
- Cottarelli, Carlo and Annalisa Fedelino, 2010, “Automatic Stabilizers and the Size of Government: Correcting a Common Misunderstanding” IMF Working Paper 10/155
- Fedelino, Annalisa, Anna Ivanova and Mark Horton, 2009, “Cyclically-Adjusted Balances and Automatic Fiscal Stabilizers: Some Computational and Interpretation Issues” IMF Technical Guidance Note No. 5.
- GalÌ, Jordi and Roberto Perotti, 2003, "Fiscal policy and monetary integration in Europe," Economic Policy, CEPR & CES & MSH, vol. 18(37), pages 533–572, October.
- National Institute of Economic Research, 2013, “The Swedish Economy. Summary,” March.
- Wyplosz, Charles, 2002, "Fiscal Policy: Institutions versus Rules," CEPR Discussion Papers 3238, C.E.P.R. Discussion Papers.

### Appendix. Econometric methodology — setup and identification
- Cyclically adjusted primary balance (CAPB) is used as a measure of discretionary fiscal policy; the residual (actual primary balance minus CAPB) measures automatic stabilizers.
- Sample period: 1970–2012.
- Fiscal rule estimated (following Gali and Perotti (2003)) with dependent variable CAPB as a percentage of potential output.
- Key regressors:
  - Output gap (deviation of actual from potential output in percent of potential output).
  - Gross government debt (percent of GDP).
  - One-period lag of the dependent variable.
- Endogeneity treatment: instrumental variable (IV) estimates using instruments for the output gap:
  - Its own lag.
  - Output gap of the US.
  - Average output gap of the three other Nordic countries (Denmark, Norway, and Finland).
- Tests for time variation/structural break after 1997:
  - Model (2) interacts the output gap with an indicator equal to one for post 1997.
  - Rolling IV regressions with a fixed twenty-year (20 observations) window to examine stability of coefficients over time.
- Methodological caveats noted:
  - Assumes constant cyclicality of fiscal revenues and spending items over the business cycle.
  - Does not take into account effects of asset price fluctuations (e.g., housing prices) on the fiscal budget.

### Econometric estimation results (Tables A1 and A2) — key statistics and interpretation
- Table A1: Estimation Results for Model 1
  - Dependent Variable: CAPB
  - Columns: (1) OLS; (2) IV (using all three instruments)
  - Coefficients (with robust standard errors in parentheses):
    - One Period Lagged CAPB: 0.76*** (0.12) ; 0.75*** (0.11)
    - Output Gap (percent of potential GDP): 0.32** (0.15) ; 0.38** (0.18)
    - Gross Government Debt (percent of GDP): 0.05* (0.03) ; 0.06** (0.03)
  - Observations: 39 39
  - R-squared: 0.681 0.679
  - AIC: 173.5 173.7
  - Weak Identification Test: 6.608
  - Test for Over-Identification: 3.976
  - Significance notation: *** p<0.01, ** p<0.05, * p<0.1
  - Interpretation: Overall fiscal policy has been countercyclical over the period 1970 to 2012 based on these estimates.

- Table A2: Estimation Results for Model 2 (interaction with post-1997 indicator)
  - Dependent Variable: CAPB
  - Columns: (1) OLS; (2) IV (using all three instruments)
  - Coefficients (with robust standard errors in parentheses):
    - Indicator Variable: 3.69 (3.01) ; 4.28 (2.86)
    - One Period Lagged CAPB: 0.71*** (0.15) ; 0.67*** (0.13)
    - One Period Lagged CAPB X Indicator Variable: -0.48** (0.23) ; -0.45** (0.21)
    - Output Gap (percent of potential GDP): 0.51** (0.19) ; 0.71** (0.29)
    - Output Gap X Indicator Variable: -0.44** (0.21) ; -0.62** (0.30)
    - Gross Government Debt (percent of GDP): 0.07** (0.03) ; 0.08** (0.03)
    - Gross Government Debt X Indicator Variable: -0.04 (0.06) ; -0.05 (0.05)
  - Observations: 39 39
  - R-squared: 0.715 0.708
  - AIC: 177.0 178.1
  - Weak Identification Test: 4.119
  - Test for Over-Identification: 8.735
  - Interpretation: Results suggest lower countercyclicality in the second part of the sample; fiscal policy appears acyclical after 1997. Rolling IV regression and Figure 4 in the main text track time-variation of the output gap coefficient (first point covers 1973–1992, next covers 1974–1993, etc.).

### Government contingent liabilities from Nordic-6 banks — scope, scenarios, and key estimates
- Focus: Six largest Nordic banks (“Nordic-6”), four headquartered in Sweden.
- Table 1 (largest Nordic banks, EUR billions, most recent quarter) — key entries (Total Assets; Share GDP in Host Country; Assets/Host GDP):
  - Nordea Bank AB (Sweden): 677.4 ; 0.30 ; 410.9 ; 1.6
  - Danske Bank A/S (Denmark): 482.8 ; 0.21 ; 244.4 ; 2.0
  - DNB ASA (Norway): 321.8 ; 0.14 ; 394.7 ; 0.8
  - Swedbank AB (Sweden): 215.0 ; 0.09 ; 410.9 ; 0.5
  - Skandinaviska Enskilda Banken AB (Sweden): 285.7 ; 0.13 ; 410.9 ; 0.7
  - Svenska Handelsbanken AB (Sweden): 297.7 ; 0.13 ; 410.9 ; 0.7
  - Total: 2,280.4 ; 1.00
- Balance sheet method overview:
  - Assumes bank assets placed into Bankruptcy Estate (BE) upon failure; depositors paid by government via Deposit Insurance Fund (DIF).
  - Net government loss = liquidation value shortfall after considering priority ranking of government claim against the BE.
  - Three bailout scenarios considered:
    - Scenario A: Insured depositors bailed out; uninsured depositors bailed in (100% haircut assumed for uninsured in Scenario A).
    - Scenario B: Insured depositors bailed out; uninsured depositors kept whole but levied 20% haircut (per Table 2 assumptions).
    - Scenario C: Insured depositors bailed out; uninsured depositors kept whole; senior unsecured creditors also held without loss.
- Key assumptions (Table 2):
  - Fraction of deposits that is insured (Sweden, Denmark, Finland): 70% ; Deposit Insurance Coverage is €100,000
  - Fraction of deposits that is insured (Norway): 56% ; Deposit Insurance Coverage is approx. €264,000
  - Fraction eventually recovered by DIF from BE of Synthetic Bank: 50% ; Relatively high due to senior secured liabilities
  - Levy on uninsured depositors: 100% (by assumption) for Scenario A
  - In Scenario B, fraction recovered by DIF from BE of Synthetic Bank: 1/40% ; Levy on uninsured depositors: 20% (by assumption)
- Timeline of scenario analysis (Table 3 — condensed steps):
  - t = 1: All Banks fail
  - t = 2: Assets put into Bankruptcy Estate (BE)
  - t = 3: Bailed out depositor claims moved to health bank; financed by DIF/State ("Liquidity Payout")
  - t = 4: DIF/State is senior-most claimant against the BE (after secured creditors claimed collateral)
  - t = 5: Payout by BE to DIF/State (Liquidity Payout minus this payment determines "Eventual Payout")
- Table 4: Estimated Fiscal Costs from the Failure of Big 6 Banks under Different Scenarios (EUR bil.) and Percent of GDP — selected country-level results (by scenario and burden sharing rule)
  - Sweden (2012 National GDP): Total Assets of Big 6 Banks (consolidated basis): 409.2 ; Estimated Fiscal Costs:
    - By Depositor Base: Scenario A: 14.7 ; Scenario B: 5.9 ; Scenario C: 24.4
    - By Location of Parent: Scenario A: 69.1 ; Scenario B: 5.9 ; Scenario C: 1.5
  - Denmark (2012 National GDP): 245.0 ; Total Assets 482.8 ; Estimated Fiscal Costs:
    - By Depositor Base: Scenario A: 11.2 ; Scenario B: 19.1 ; Scenario C: 39.2
    - By Location of Parent: Scenario A: 13.3 ; Scenario B: 24.3 ; Scenario C: 50.0
  - Finland: 194.5 ; Total Assets 0.0 ; Estimated Fiscal Costs (By Depositor Base): Scenario A: 9.3 ; Scenario B: 15.9 ; Scenario C: 32.6
  - Norway: 390.0 ; Total Assets 321.8 ; Estimated Fiscal Costs (By Depositor Base): Scenario A: 10.2 ; Scenario B: 21.8 ; Scenario C: 44.7
  - Nordic-4 / Nordic-6 aggregates also reported in table (EUR bil. and Percent of GDP).
- Main quantitative findings (Summing Up):
  - Swedish government contingent liabilities from supporting depositors of the six largest Nordic banks range from just below 20 percent of GDP to 90 percent of GDP, depending on bailout scope and burden sharing rule.
  - Under a deposit base burden sharing approach, losses vary from 17–60 percent of GDP.
  - Under a location-of-parent burden sharing approach, losses vary from 26–90 percent of GDP.
  - Taking a mid-range across burden sharing rules suggests contingent liabilities amounting to 30–45 percent of GDP.
  - If the government only supports insured depositors, the estimate would drop to between 20 and 30 percent of GDP.
  - Estimates are large relative to the ex-post cost of the 1990s Swedish bank bailout, estimated at less than 5 percent of GDP.
  - Sources of uncertainty include: government’s approach to uninsured depositors and bond holders; level of bank losses on balance sheets relative to end-2012 data used.

### Policy implications and optimal fiscal buffer considerations
- Key points:
  - Government contingent liabilities from the financial sector are potentially large; policy responses should include:
    - Strengthening bank resilience.
    - Reducing household credit growth risks.
    - Improving internal and Nordic macroprudential coordination.
  - Fiscal buffers could be built by:
    - Limiting gross debt levels so contingent liabilities could be covered by issuing new debt if necessary.
    - Building up sufficiently liquid reserves in a dedicated fund to be deployed when contingencies materialize.
- Discussion notes:
  - Dedicated funds have trade-offs:
    - Downside: may exacerbate moral hazard by pre-committing funds for banking crises.
    - Upside: allow Sweden to maintain an active and liquid sovereign debt market without limiting the size of debt levels.
  - A conceptual exploration is provided of building buffers if none existed at the outset.
  - A separate question remains whether existing debt levels or funds in the Swedish Stability Fund already provide the required buffer.

*Source: _cr13277 - References (IMF).*

### 13.      In order to determine the optimal size of a fiscal buffer given contingent liabilities and

### _cr13277 - 13.      In order to determine the optimal size of a fiscal buffer given contingent liabilities and

### Model setup and objectives
- Government objective: smooth government spending subject to a dynamic budget constraint; each period the government raises a fixed quantity of tax revenue and chooses spending.
- Risk structure:
  - Only risk: one-time contingent liability Z (>0) realized with constant probability φ in each period; once realized, no further risks.
  - No other fiscal risks (longevity, cyclical fluctuations) are modeled; analysis pertains to the permanent (detrended) component of government spending.
- Key definitions and constraints (as in the Appendix):
  - Government resources M_t, end-of-period assets A_t, balance before receipt of current income B_t.
  - Budget constraint decomposed as in equation (2) with contingent liability Z and realization indicator ξ_{t+1}.
  - No-Ponzi condition given by equation (3).
  - Preferences: household lifetime utility additively depends on private consumption C_t and government expenditure G_t with discount factor β ∈ (0,1). Government chooses G_t to maximize (0).
  - Post-shock marginal propensity to spend out of total assets: parameter defined by (7) as (1 - β R^{1/ρ}) / (1 - R^{1/ρ}), with CRRA parameter ρ and gross return R.
  - Assumption 1: 0 > ? (condition imposed to guarantee finite PDV of government spending).
  - Assumption 2: (1 - R^{-1}) < β^{-1} (?) (text states condition (15) 1/(1R)β^{-1} form; summarized purpose: consumer sufficiently impatient so steady state consumption positive).

### Main results (key findings)
- Long-run fiscal buffer target:
  - The long run fiscal buffer target should approximately match the size of the contingent liability.
  - Over several decades the government should target building a fiscal buffer almost as large as the size of the contingent liability.
- Slight over-buffering:
  - Optimal solution requires the government to slightly “over-buffer” so the target fiscal buffer is somewhat larger than the contingent liability.
  - Rationale: to smooth spending when liability is realized because higher post-shock debt raises interest payments and spending; ex-ante over-buffering keeps debt low and limits spending increases.
- Speed of accumulation:
  - The speed of accumulation of the fiscal buffer is high (front-loaded) in the first few years, then gradually declines as the buffer approaches the target.
  - Front-loading ensures partial insurance in case contingent liabilities materialize immediately.

### Numerical illustration and transition path
- Illustrative example parameters and implications:
  - Contingent liabilities amount to 30 percent of GDP (Z/GDP = 0.3).
  - Model-implied target fiscal buffer: roughly 34 percent of GDP.
  - Government’s maximum acceptable level of gross debt after contingent liabilities realized: 60 percent of GDP.
  - Implied ex-ante target gross debt level: around 26 percent of GDP, to be reached gradually.
- Transition path properties (Figure 5 description):
  - Buffer building occurs over several years.
  - Optimal path: high accumulation speed initially, tapering off as target nears.
  - Alternative contingent liability sizes (20 percent, 30 percent, 40 percent of GDP) yield different transition paths but same qualitative front-loading.

### Intuition on buffer sizing and sensitivity
- Long-horizon interpretation:
  - With constant per-period crisis probability φ, as horizon grows the likelihood of a crisis converges to one; contingent liability approximates a real liability in long run, motivating buffer ~ size of Z.
- Marginal sensitivity:
  - Under model assumptions, derivative dB/dZ > 1: an increase in contingent liability by one unit leads to a greater-than-one unit increase in target buffer wealth.
  - Explanation: post-shock interest income falls by Z when buffer is depleted; to smooth pre- and post-shock spending gap (G^b/G^a), the government targets buffer slightly larger than Z.
- Role of φ:
  - Higher φ strengthens precautionary motive; target buffer wealth increases with φ (derivative ∂B/∂Z increases with φ).
  - Figure A2: when probability of shock is higher, the drop in government spending after the shock is lower because pre-period buffer saving is larger.

### Steady state and transitional system
- Steady-state characterization:
  - Steady-state conditions derived from Euler equations (equations (10)–(18)).
  - Steady-state before-shock asset target B^b and spending G^b given by closed-form expressions (equations (17) and (18)) as functions of R, β, ρ, φ, T, and Z.
  - Government spending function is concave in assets: marginal propensity to spend is higher at low asset levels (precautionary motive larger when resources decline).
- Transition system:
  - Two difference equations pin down transition path: budget constraint (19) and modified Euler condition (20).
  - No closed-form transition solution; numerical methods used to compute paths for given parameters.

### Policy implications and recommendations (from conclusions)
- Quantitative range and uncertainty:
  - Swedish government could face contingent liabilities from the financial sector in the range of just below 20 to 90 percent of GDP, with large uncertainties around these estimates.
  - Realized magnitude depends on bank loss estimates, whether government bails out unsecured depositors and unsecured bondholders (in addition to insured depositors), burden-sharing rules between countries, and exposure of bank assets to euro area risks.
- Policy emphasis:
  - Fast progress on financial reforms is emphasized.
  - Measures to further cool household credit growth and strengthen banks’ liquidity and capital positions will help reduce systemic banking risk and help limit the size of the contingent liability (see Policy Agenda section of the 2013 staff report for Sweden).

### Appendix: methodology inputs and parameter values
- Environment and solution approach:
  - Government chooses expenditure path to maximize household utility with CRRA preferences v(G) = (G^{1-ρ} - 1)/(1 - ρ).
  - Post- and pre-shock Euler equations and steady-state system derived analytically; transition path solved numerically.
- Parameters used (Table A1):
  - Gross Return (R): 1.01
  - Discount Factor (β): 0.97
  - CRRA parameter (ρ): 1
  - Probability of Shock (φ): 0.01
  - Tax Revenue to Structural GDP (T/GDP): 0.52
  - Contingent Liability to Structural GDP (Z/GDP): 0.3

*Source: Fund staff calculations.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/scr/2013/_cr13277.pdf_