## wpiea2020266-print-pdf

## Source details

**Canonical URL:** [wpiea2020266-print-pdf](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020266-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020266-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020266-print-pdf.pdf.json)

---

### I. Introduction — scale and fiscal implications of natural disasters in the ECCU
- Natural disasters (ND) cause human loss, infrastructure destruction, and fiscal costs in the Eastern Caribbean Currency Union (ECCU).
- Large economic disasters in advanced economies imply welfare costs equivalent to about 20 percent of annual GDP (Barro (2006, 2009)).
- Cantelmo, Melina and Papageorgiou (2019) estimate the welfare loss from natural disasters is equivalent to a permanent decline in consumption of 1.6 percent of GDP.
- Private-sector ND insurance is generally insufficient; governments act as the de-facto ultimate insurer covering public infrastructure damage and a significant share of private losses and social support.
- CCRIF provides quick parametric disbursements within two weeks but:
  - payouts are insufficient relative to needs in extreme events;
  - CCRIF payout is limited to up to about $100 million per event, insufficient to cover fiscal costs;
  - expected payouts are about half of insurance premia.
- Insurance multipliers (ratio of insurance premia to expected payout) for market instruments (including CAT bonds) are in the range of about 1.5-2.0; CCRIF insurance multipliers are around 2, and higher for extreme events.
- Parametric triggers are imperfectly correlated with actual ND damages; imperfect correlation is less likely at less frequent but more intense disasters.
- High cost and imperfect correlation imply these instruments worsen debt sustainability outlook in expected terms but help reduce debt volatility.

### II. Rationale for a Saving Fund (SF) as the first insurance layer
- A SF can provide public self-insurance for immediate expenditure needs, rehabilitation and reconstruction, strengthening fiscal sustainability.
- Practical challenges to relying on post-disaster financing:
  - Access to rapid financing is typically not immediate for small countries.
  - Official loan and grant disbursement processes are lengthy.
  - Domestic financing can be limited if NDs affect financial institutions’ asset quality and deposits.
- SFs with recurrent budget savings internalize expected periodic ND costs, reducing the need for debt issuance and supporting fiscal sustainability.
- Existing regional funds exist but none specifically target ND fiscal costs; regional experience shows risks from political influence and capture, highlighting the need for strong institutional design and oversight.
- Start-up financing assumptions:
  - Start-up cost of SFs is assumed financed with Citizenship by Investment (CBI) program revenues in all ECCU countries except St. Vincent and the Grenadines.
  - For St. Vincent and the Grenadines (no CBI), SF assumed built gradually with additional annual saving of 1 percent of GDP.
  - Alternatively, starting a SF for ND may require debt issuance.
- CBI revenues are significant but uncertain and volatile; saving CBI flows for SFs can reduce debt issuance after NDs and reduce use of an unreliable revenue source for current spending.

### III. Methodology — empirical model and Monte-Carlo simulation framework
- Empirical approach:
  - Unrestricted Vector Auto-regression Model (VAR) estimated for each country, including fiscal determinants of public debt dynamics.
  - Endogenous variables: cyclical components of GDP; government revenues excluding grants; grants; current primary expenditures; capital expenditures — each expressed as a share of its trend.
  - Cyclical components of GDP estimated using Hodrick-Prescott filter on 1990-2017 annual data. All variables in real terms using the GDP deflator.
  - Exogenous control variables: U.S. real effective exchange; oil price; cyclical component of U.S. output; dummy for September 11, 2001 terrorist attack.
  - Estimated residuals εt ~ N(0, Ω) are orthogonal to control variables; εt includes ND and “small” shocks; Ω captures contemporaneous disturbance properties.
- Monte-Carlo experiment:
  - Random vectors ε̂t+1,...,ε̂T generated with ετ = W vτ where vτ ~ N(0,1) and W is the Choleski factorization of Ω.
  - VAR used to generate 2000 forecasts Yt for each country using the randomly generated shocks.
  - Simulations produce joint dynamic responses for the five endogenous variables and compute probability density functions for 2017-2030.
- Debt dynamics and SF interactions:
  - Public debt identity: Dt+l+1 = (1 + it+l+1) Dt+l − PBt+l + (IFSS t+l − OFSS t+l); Dt+l > 0; l = t+1,...,T
    - where PBt is the primary balance from simulated revenue and primary expenditure series; it is the implicit interest rate; IFSS − OFSS are below-the-line inflows and outflows from the SF for NDs.
  - Different ND sequences across simulations produce different debt paths; SF flows replace debt issuance when used vis-à-vis the budget.
  - Debt stock projections are not affected in expected terms, provided savings into the fund are utilized in the long-term and across simulations in a given period.
- Converting deviations to percent of GDP:
  - Projections obtained as deviations from trend and calculated as percent of GDP.
  - Deterministic trends projected assuming all trends grow at the same constant rate g starting from the end point of the estimated trend in the sample period.
  - Endogenous variables in real-term levels expressed as percent of GDP by dividing by the GDP projection in each vector simulation.
  - Starting points of trend projections set at constant prices of the last year of the sample.

### Identification of Natural Disasters (NDs) in simulations
- Simulations identify as a ND the largest dd percent fiscal deteriorations across draws.
- Fiscal deteriorations computed as sum of year-on-year changes of:
  - (i) non-grant revenue NNRt (with a negative sign);
  - (ii) grant revenues NNGt;
  - (iii) current primary expenditure Nt;
  - (iv) capital expenditure KGt.
- Fiscal aggregate used: zt = △NGt − △NNRt + △NGGGt + △NKG Gt (notation preserved from source).
- Simulated NDs are realizations of zt falling in the highest dd percent tail of the pdf of this sum.
- ND identification assumes major other large shocks controlled for in VARs; remaining negative large shocks treated as NDs.

### A. Saving Funds for Self-insurance (SFs): design and mechanics
- Budget contributions in no-ND years:
  - Modeled as fixed parameter θ as a percent of previous-year GDP, calibrated to achieve financial sustainability with a low probability of depletion.
  - If simulations result in a fiscal deficit, issuing public debt may be needed to finance the required contribution to the Fund.
- SF → budget financing when ND occurs equals the sum of four components:
  - + Gap of non-grant revenues below trend.
  - − Gap of grant revenues above trend (donor support reducing need for SF financing).
  - + Gap of current primary expenditure above trend minus a fixed reprioritization term ρG with ρG ∈(−1,1).
  - + Gap of capital expenditure above trend minus a fixed reprioritization term ρK with ρK ∈(−1,1).
- Denote the sum as σt; contributions continue until each indicator returns below the pre-ND year value; the SF finances the “hump”.
- Re-prioritization is key: reconstruction expenditures often obtained by reallocations and postponed or cancelled pre-ND allocations (illustrated by Dominica after Tropical storm Erika, August 2015).

### SF stock evolution and operational assumptions
- SF stock evolution per simulation:
  - SSt+1 = SSt (1 + Gt) + It+1SSS − Ot+1SSS
  - Equivalently: SSt (1 + Gt) + (1 − NDt) θ NDPt−1 − NDt σt.
- Definitions and calibration links:
  - θ NDPt+ll−1 = 0 in years when algorithm identifies a ND (NDt =1).
  - Probability P[d] = P[σt ≥ σ̄] = p, where σ̄ set to match historical ND frequency (example: if ND every 5 years on average, set p=0.2).
  - Gt is rate of return of assets in the SF.
  - If SF is depleted (SSt = 0), remaining financing covered with public debt issuance.
- Initial stock SSt0 > 0 affects probability of depletion; start-up cost assumed funded with existing CBI assets and not added to debt stock.

### Calibration of simulation parameters
- Macroeconomic/fiscal calibration consistent with WEO database: potential GDP growth rate; implicit interest rate on public debt; fiscal consolidation targets in percent of GDP.
- SF parameters calibrated to cover full fiscal deteriorations after NDs in 95 percent of simulations (i.e., SFs depleted in largest 5 percent of simulated ND shocks).
- Illustrative Dominica calibration:
  - P[d] = 0.2 (ND every 5 years on average).
  - Initial Fund stock = 10 percent of GDP → probability of depletion within next ten years = 0.08.
  - Simulation target probability of depletion = 5 percent.
  - Budget saving flows in non-ND years = 1.5 percent of previous-year’s GDP.
  - Rate of return Gt = 2 percent.
  - ρG and ρK set to capture realistic expenditure reprioritization (base levels defined as spending in years without ND-related spending).

### B. Parametric Insurance Instruments (Layers 2 and 3)
- CCRIF:
  - Premium PCCRC = a AAL + b SD(AAL); a and b calibrated to match observed premia; insurance multipliers in range 1.5-2.0.
  - AAL and SD(AAL) are NDs average annual losses and standard deviation.
  - Attachment point and maximum coverage set according to estimated damages along expected loss function for 20- and 100-year estimated damages respectively.
  - CCRIF payouts triggered per ND algorithm and determined in proportion to simulated losses.
  - Due to imperfect correlation:
    - CCRIF payouts discounted by factor 0.5 for 1/20 year losses (payout = ½ of losses on average for smaller NDs).
    - Correlation assumed to increase linearly to convergence to 1 for 1/100 year loss (payout proportional to loss covered).
  - Correlation and multiplier assumptions: correlation 0.5 aligns with CCRIF estimates of insurance multipliers near 1 for relatively smaller/frequent NDs (1/20 year), and around 2 for large NDs (1/100 year).
- State-contingent debt (CAT bonds):
  - Governments issue CAT bonds for debt service relief in case of a ND.
  - Proceeds invested in safe asset with returns equivalent to Libor and held in SPV.
  - Governments hold a call option on SPV principal; if ND occurs, governments withdraw funds to pay claims and interest/principal forgiven; otherwise investors receive principal and interest.
  - Assumed CAT bonds: 3-year maturity; interest = Libor plus 500 basis points.
  - Amount of CAT bond issuance determined by residual coverage needed to top up SF and maximum CCRIF coverage.
  - If maximum CCRIF coverage insufficient to reach 99 percent overall coverage, simulations assume issuance of US$ 10 to 100 million in CAT bonds as needed to reach 99 percent target.

### V. Results — key quantitative findings and sensitivities
- Layer 1 (Self-insurance with SFs):
  - Off-sample Dominica example: 15-year draw with three NDs in years t+5, t+9, t+12; budget savings in non-ND years at 1.5 percent of previous-year’s GDP; starting stock 10 percent of GDP.
  - Simulated ND sizes vary; example shows one ND with budget financing near 5 percent of GDP and two NDs with financing flows over 10 percent of GDP.
  - Possibility of SF depletion exists in extreme NDs; depletion probability is a calibration choice reflecting risk tolerance.
- Across ECCU plausible calibrations:
  - SF stocks needed in the range of 6-12 percent of GDP.
  - Annual savings required in the range of 0.5-1.9 percent of GDP to achieve financial sustainability with a low probability of depletion.
  - Sensitivity: setting ND probability one notch higher increases annual budget saving needs (examples: from around 0.2 percent of GDP for Grenada; St. Lucia; St. Vincent and the Grenadines to around 0.5 percent of GDP for Antigua and Barbuda; Dominica; St. Kitts and Nevis).
  - Lowering the proposed SF stock by 2 percent of GDP increases probability of depletion by about 10 percent or more.
- Layers 2 and 3 to reach 99 percent overall coverage:
  - Structure: 95 percent of expected cost covered by SFs (layer 1), then maximum CCRIF (layer 2), then CAT bonds (layer 3).
  - Simulation results indicate ECCU countries would require total coverage equivalent to 13-31 percent of GDP under illustrative assumptions.
  - Under assumptions:
    - SFs cover initial 6-12 percent of GDP.
    - Maximum CCRIF access coverage estimated in range 2-17 percent of GDP.
    - Residual shortfall covered by CAT bonds in range 1.7-4.6 percent of GDP.
  - Thresholds and coverage levels illustrative and adjustable by countries’ risk-aversion and institutional capacity.

### Representative numeric highlights and calibration examples
- 2000 simulations.
- ND identification: highest dd percent tail of pdf.
- Example calibration (Dominica):
  - P[d] = 0.2.
  - Initial SF stock = 10 percent of GDP.
  - Probability of depletion within next ten years = 0.08.
  - Target simulation probability of depletion = 5 percent.
  - Budget saving flows in non-ND years = 1.5 percent of previous-year’s GDP.
  - Rate of return Gt = 2 percent.
- Insurance parameters:
  - CCRIF insurance multipliers range = 1.5-2.0.
  - CCRIF payout discount factor for 1/20 year losses = 0.5.
  - CAT bonds maturity = 3-year; interest = Libor plus 500 basis points.
  - CAT bond issuance size if needed = US$ 10 to 100 million.
- Coverage targets:
  - Overall ND fiscal costs coverage target = 99 percent.
  - SFs coverage target of expected damages = 95 percent (layer 1).
  - Simulated total coverage needed for ECCU countries = 13-31 percent of GDP.
  - SFs’ illustrative coverage = 6-12 percent of GDP.
  - CCRIF illustrative coverage = 2-17 percent of GDP.
  - CAT bonds illustrative coverage = 1.7-4.6 percent of GDP.
- Text-table excerpt (Fund metrics, in percent of GDP unless indicated):
  - Fund size: ATG 12.00, DMA 10.00, GRD 6.00, KNA 10.00, LCA 8.00, VCT 8.00
  - Annual budget saving if no ND: ATG 1.90, DMA 1.50, GRD 0.35, KNA 1.90, LCA 0.90, VCT 0.95
  - Probability of fund depletion, units: ATG 0.05, DMA 0.08, GRD 0.04, KNA 0.07, LCA 0.05, VCT 0.03
  - Annual expected use of the fund if ND: ATG 8.7, DMA 5.9, GRD 3.0, KNA 4.8, LCA 3.0, VCT 2.6
  - Average fiscal cost of simulated ND 1/: ATG 57.9, DMA 29.3, GRD 29.7, KNA 19.3, LCA 15.0, VCT 12.9
    - Note: 1/ Includes estimated average decline in revenues and increase in primary expenditures in the simulations, net of expected increase in donor grants.

### Insurance layering costs, donor subsidy proposal, and policy implications
- Insurance layering with 99 percent coverage implies annual fiscal costs in the range of 0.5-1.9 percent of GDP.
- Higher costs noted for Dominica, Antigua and Barbuda and St. Kitts and Nevis, reflecting higher estimated Average Annual Losses.
- Parametric insurance pricing multipliers of 1.5-2.0 imply insurance costs worsen fiscal sustainability in expected terms while reducing dispersion of debt outcomes.
- Role of SFs and layering:
  - Increasing share of coverage with SFs could reduce costs, but maintaining SFs is challenging given developmental needs and political pressures.
  - Self-insurance with an SF is argued as needed given high cost of parametric instruments; parametric instruments can top-up SFs for tail risks.
- Donor subsidy proposal:
  - Donors could subsidize insurance premia to remove incentives for underinsurance and make insurance fiscally neutral from governments’ perspective (insurance multipliers effectively equal 1).
  - Under simulation assumptions, making insurance fiscally-neutral would require grants in the range of 0.2-1.1 percent of GDP per year — equivalent to about US$ 40 million per year for all ECCU countries.
  - Subsidy design would be incentive-compatible: equalize governments’ insurance costs with expected average annual losses.
- Caveats and lower-bound nature of estimates:
  - Estimates are lower bounds because fiscal space for reprioritization may be lower than assumed; methodology omits long-term increases in disaster intensity/frequency from climate change; permanent/structural output effects not incorporated.
  - Resilience investments in infrastructure would reduce damage extent and fiscal cost, lowering coverage need and insurance cost.
  - Need to periodically reassess and recalibrate insurance needs.

### Institutional design recommendations for SFs and parametric use
- Strong institutional setup critical to contain political pressures and opportunistic appropriations.
- Recommended features:
  - Unambiguous budget contribution and disbursement rules with triggers based on verifiable criteria.
  - Clearly-stated objective.
  - Strict information disclosure requirements for transparency.
- CCRIF and market catastrophe insurance can complement self-insurance for medium and large disasters.
- Parametric triggers imperfectly correlated with damages for low and medium intensity NDs; more likely to trigger for large and extreme NDs and top-up SFs coverage.
- High cost of parametric insurance makes economic/welfare justification questionable; however, social risk aversion could justify purchase to reduce debt-outcome dispersion despite worsening expected debt sustainability.

*Source: wpiea2020266-print-pdf — IMF staff simulations and text as provided.*

### REFERENCES _____________________________________________________________________________________________21

### wpiea2020266-print-pdf - REFERENCES _____________________________________________________________________________________________21

### I. Introduction — scale and fiscal implications of natural disasters in the ECCU
- Natural disasters (ND) cause human loss, infrastructure destruction, and fiscal costs in the Eastern Caribbean Currency Union (ECCU).
- Large economic disasters in advanced economies imply welfare costs equivalent to about 20 percent of annual GDP (Barro (2006, 2009)).
- Cantelmo, Melina and Papageorgiou (2019) estimate the welfare loss from natural disasters is equivalent to a permanent decline in consumption of 1.6 percent of GDP.
- For ECCU small states, which face larger and more frequent disasters, effects are expected to be greater.
- Private-sector ND insurance is generally insufficient; governments act as the de-facto ultimate insurer covering public infrastructure damage and a significant share of private losses and social support.
- CCRIF provides quick parametric disbursements within two weeks but:
  - payouts are insufficient relative to needs in extreme events;
  - CCRIF payout is limited to up to about $100 million per event, insufficient to cover fiscal costs;
  - expected payouts are about half of insurance premia.
- Insurance multipliers (ratio of insurance premia to expected payout) for market instruments (including CAT bonds) are in the range of about 1.5-2.0; CCRIF insurance multipliers are around 2, and higher for extreme events.
- Parametric triggers are imperfectly correlated with actual ND damages; imperfect correlation is less likely at less frequent but more intense disasters.
- High cost and imperfect correlation imply these instruments worsen debt sustainability outlook in expected terms but help reduce debt volatility.

### II. Rationale for a Saving Fund (SF) as the first insurance layer
- A SF can provide public self-insurance for immediate expenditure needs, rehabilitation and reconstruction, strengthening fiscal sustainability.
- Practical challenges to relying on post-disaster financing:
  - Access to rapid financing is typically not immediate for small countries.
  - Official loan and grant disbursement processes are lengthy.
  - Domestic financing can be limited if NDs affect financial institutions’ asset quality and deposits.
- SFs with recurrent budget savings internalize expected periodic ND costs, reducing the need for debt issuance and supporting fiscal sustainability.
- Existing regional funds exist but none specifically target ND fiscal costs; regional experience shows risks from political influence and capture, highlighting the need for strong institutional design and oversight.
- The paper assumes start-up cost of SFs is financed with Citizenship by Investment (CBI) program revenues in all ECCU countries except St. Vincent and the Grenadines.
  - For St. Vincent and the Grenadines (no CBI), it is assumed the SF is built gradually with additional annual saving of 1 percent of GDP.
  - Alternatively, starting a SF for ND may require debt issuance.
- CBI revenues are significant but uncertain and volatile, posing macroeconomic management risks; saving CBI flows for SFs can reduce debt issuance after NDs and reduce use of an unreliable revenue source for current spending.

### III. Methodology — empirical model and Monte-Carlo simulation framework
- Empirical approach:
  - An unrestricted Vector Auto-regression Model (VAR) is estimated for each country, including fiscal determinants of public debt dynamics.
  - Endogenous variables include the cyclical components of GDP; government revenues excluding grants; grants; current primary expenditures; and capital expenditures — each expressed as a share of its trend.
  - Cyclical components of GDP are estimated using the Hodrick-Prescott filter on 1990-2017 annual data. All variables expressed in real terms using the GDP deflator.
  - Exogenous control variables (Xj) include the U.S. real effective exchange; the oil price; the cyclical component of U.S. output; and a dummy for the September 11, 2001 terrorist attack in the United States.
  - Estimated residuals εt ~ N(0, Ω) are orthogonal to control variables; εt includes ND and “small” shocks. The variance-covariance matrix Ω captures contemporaneous disturbance properties.
- Monte-Carlo experiment:
  - Random vectors ε̂t+1,...,ε̂T are generated with ετ = W vτ where vτ ~ N(0,1) and W is the Choleski factorization of Ω.
  - The VAR is used to generate 2000 forecasts Yt for each country using the randomly generated shocks.
  - Simulations produce joint dynamic responses for the five endogenous variables and are used to compute probability density functions for each variable for the period 2017-2030.
- Debt dynamics and SF interactions:
  - Public debt is computed via the debt accumulation identity:
    - Dt+l+1 = (1 + it+l+1) Dt+l − PBt+l + (IFS S t+l − OFS S t+l); Dt+l > 0; l = t+1,...,T
    - where PBt is the primary balance from simulated revenue and primary expenditure series; it is the implicit interest rate (interest expenditures in year t divided by public debt stock in t-1); IFS S − OFS S are below-the-line inflows and outflows from the SF for NDs.
  - Different sequences of ND occurrence across simulations produce different debt paths; SF flows replace debt issuance when used vis-à-vis the budget.
  - Debt stock projections are not affected in expected terms, provided savings into the fund are utilized in the long-term and across simulations in a given period.
- Converting deviations to percent of GDP:
  - Projections are obtained as deviations from trend and calculated as percent of GDP.
  - Deterministic trends are projected assuming all trends grow at the same constant rate g starting from the end point of the estimated trend in the sample period.
  - Endogenous variables in real-term levels are expressed as percent of GDP by dividing by the GDP projection in each vector simulation.
  - Starting points of trend projections are set at constant prices of the last year of the sample to ensure consistency with data.

*Source: wpiea2020266-print-pdf - REFERENCES _____________________________________________________________________________________________21*

### 17.      The third step is to identify the occurrence of natural disasters in each

### 17.      The third step is to identify the occurrence of natural disasters in each

### Identification of Natural Disasters (NDs) in simulations
- Simulations include an algorithm that identifies as a ND the largest 푑푑 percent fiscal deteriorations across draws.
- Fiscal deteriorations are computed as the sum of year-on-year changes of:
  - (i) non-grant revenue 푁푁푁푁
푡푡 (with a negative sign);
  - (ii) grant revenues 푁푁퐺퐺
푡푡;
  - (iii) current primary expenditure 푁푁
푡푡;
  - (iv) capital expenditure 퐾퐾
푡푡.
- The fiscal aggregate random variable used to identify simulated NDs is written as:
  - 푧푧
푡푡 = △푁푁퐺퐺
푡푡 − △푁푁푁푁
푡푡 + △푁푁퐺퐺퐺퐺
푡푡 + △푁푁퐾퐾 퐺퐺
푡푡
- Simulated NDs are the largest random draw realizations of 풛풛
풕풕 across the 2000 simulations: the algorithm computes the distribution of the sum in every simulation year t+1, l=t+1,...T, and labels as a ND all realizations that fall in the highest 푑푑 percent tail of the pdf of this sum.
- The ND identification rests on the assumption that all other major sources of large shocks have been controlled for in the VARs; remaining negative large shocks are treated as NDs.

### A. Saving Funds for Self-insurance (SFs): design and mechanics
- In years with no ND (as identified by the algorithm) the budget contributes savings to the SF.
  - Budget contributions are modeled as fixed parameter 휃휃 as a percent of previous-year GDP, calibrated to achieve financial sustainability with a sufficiently low probability of depletion.
  - If simulations result in a fiscal deficit, issuing public debt may be needed to finance the required contribution to the Fund.
- When a ND occurs, SF → budget financing equals the sum of four components:
  - + Gap of non-grant revenues below trend (decline in tax and non-tax revenues).
  - - Gap of grant revenues above trend (donor support reducing need for SF financing).
  - + Gap of current primary expenditure above trend (higher social support and rehabilitation); minus a fixed reprioritization term denoted 휌휌
퐺퐺
 푁푁
... with 휌휌
퐺퐺 ∈(−1,1).
  - + Gap of capital expenditure above trend (reconstruction public investment); minus a fixed reprioritization term denoted 휌휌
퐾퐾
... with 휌휌
퐾퐾 ∈(−1,1).
- Denote the random variable from the sum of the four components as 휎휎
푡푡, where the explicit formula in the source defines the component terms and reprioritization subtractions.
- Contributions to the budget continue until each indicator returns below the pre-ND year value; the SF finances the “hump”.
- Re-prioritization is key: reconstruction expenditures are often obtained by reallocations and postponed or cancelled pre-ND allocations (illustrated by Dominica after Tropical storm Erika, August 2015).

### SF stock evolution and operational assumptions
- SF stock evolution per simulation:
  - 푆푆푆푆
푡푡+1 = 푆푆푆푆
푡푡 (1 + 퐺퐺
푡푡) + 퐼퐼
푡푡+1
푆푆푆푆 − 푂푂
푡푡+1
푆푆푆푆
  - Equivalently: 푆푆푆푆
푡푡 (1 + 퐺퐺
푡푡) + (1 − 푁푁퐷퐷
푡푡) 휃휃 푁푁퐷퐷푃푃
푡푡−1 − 푁푁퐷퐷
푡푡 휎휎
푡푡.
- Definitions and calibration links:
  - 휃휃 푁푁퐷퐷푃푃
푡푡+푙푙−1 = 0 in years when algorithm identifies a ND (푁푁퐷퐷
푡푡 =1).
  - Probability P[푑푑] = P[휎휎
푡푡 ≥ 휎휎�] = 푝푝, where 휎휎� is set in calibration to match historical ND frequency (example: if ND every 5 years on average, set 푝푝=0.2).
  - 퐺퐺
푡푡 is the rate of return of assets in the SF.
  - If SF is depleted (푆푆푆푆
푡푡 = 0), remaining financing is covered with public debt issuance.
- Initial stock (start-up cost) 푆푆푆푆
0 > 0 affects probability of depletion; start-up cost assumed funded with existing CBI assets and not added to debt stock.

### Calibration of simulation parameters
- Parameters calibrated consistent with macroeconomic frameworks for each ECCU country in the WEO database.
- Macroeconomic/fiscal calibration set includes potential GDP growth rate; implicit interest rate on public debt; fiscal consolidation targets in percent of GDP.
- Parameters affecting the SF calibrated to cover full fiscal deteriorations after NDs in 95 percent of simulations (i.e., SFs depleted in largest 5 percent of simulated ND shocks).
- Illustrative Dominica calibration examples:
  - ND probability set at P[d] = 0.2 (ND every 5 years on average).
  - Initial size of Fund stock set at 10 percent of GDP to obtain a probability of depletion within the next ten years of 0.08.
  - In simulations, target probability of depletion is 5 percent.
  - Budget saving flows into SF in years without ND set at 1.5 percent of previous-year’s GDP to obtain sustainable SF stock in expected terms.
- Rate of return on SF 퐺퐺
푡푡 set at 2 percent.
- Parameters 휌휌
퐺퐺 and 휌휌
퐾퐾 set to capture realistic expenditure reprioritization (base levels defined as spending in years without ND-related spending).

### B. Parametric Insurance Instruments (Layers 2 and 3)
- CCRIF:
  - Premium 푃푃
퐶퐶퐶퐶퐶퐶퐶퐶  푆푆 = 푎푎 퐴퐴퐴퐴퐴퐴 + 푏푏 푆푆퐷퐷(퐴퐴퐴퐴퐴퐴)
  - 푎푎 and 푏푏 calibrated to match observed premia; insurance multipliers in the range 1.5-2.0.
  - 퐴퐴퐴퐴퐴퐴 and 푆푆퐷퐷(퐴퐴퐴퐴퐴퐴) are NDs average annual losses and standard deviation.
  - Attachment point and maximum coverage set according to estimated damages along expected loss function for 20- and 100-year estimated damages respectively.
  - CCRIF payouts triggered per ND algorithm and determined in proportion to simulated losses.
  - Due to imperfect correlation between parametric triggers and actual damages:
    - CCRIF payouts discounted by factor 0.5 for 1/20 year losses (payout = ½ of losses on average for smaller NDs).
    - Correlation assumed to increase linearly to convergence to 1 for 1/100 year loss (payout proportional to loss covered).
  - Correlation and multiplier assumptions: correlation 0.5 aligns with CCRIF estimates of insurance multipliers near 1 for relatively smaller/frequent NDs (1/20 year), and around 2 for large NDs (1/100 year).
- State-contingent debt (CAT bonds):
  - Governments issue CAT bonds for debt service relief in case of a ND.
  - Proceeds invested in safe asset with returns equivalent to Libor and held in SPV.
  - Governments hold a call option on SPV principal; if ND occurs, governments withdraw funds to pay claims and interest/principal forgiven; otherwise investors receive principal and interest.
  - Assumed CAT bonds have 3-year maturity and interest rate equivalent to Libor plus 500 basis points.
  - Amount of CAT bond issuance determined by residual coverage needed to top up SF and maximum CCRIF coverage.
  - If maximum CCRIF coverage insufficient to reach 99 percent overall coverage of NDs’ fiscal costs, simulations assume issuance of US$ 10 to 100 million in CAT bonds as needed to reach 99 percent target.

### V. Results — key quantitative findings and sensitivities
- Layer 1 (Self-insurance with SFs):
  - Off-sample example (Dominica): illustrative 15-year draw with three NDs in years t+5, t+9, t+12; budget savings in non-ND years at 1.5 percent of previous-year’s GDP; starting stock 10 percent of GDP; SF increases in no-ND years and declines when NDs occur.
  - Simulated ND sizes vary; example shows one ND with budget financing near 5 percent of GDP and two NDs with financing flows over 10 percent of GDP.
  - Possibility of SF depletion exists in extreme NDs; depletion probability is a calibration choice reflecting risk tolerance.
- Under plausible calibrations across ECCU:
  - SF stocks needed in the range of 6-12 percent of GDP.
  - Annual savings required in the range of 0.5-1.9 percent of GDP to achieve financial sustainability with a low probability of depletion.
  - Sensitivity: setting ND probability one notch higher increases annual budget saving needs (examples: from around 0.2 percent of GDP for Grenada; St. Lucia; St. Vincent and the Grenadines to around 0.5 percent of GDP for Antigua and Barbuda; Dominica; St. Kitts and Nevis).
  - Lowering the proposed SF stock by 2 percent of GDP increases probability of depletion by about 10 percent or more.
- Layers 2 and 3 (Parametric insurance and CAT bonds) to reach 99 percent overall coverage:
  - Structure: 95 percent of expected cost covered by SFs (layer 1), then maximum CCRIF (layer 2), then CAT bonds (layer 3) if needed.
  - Simulation results indicate ECCU countries would require total coverage equivalent to 13-31 percent of GDP under illustrative assumptions.
  - Under assumptions:
    - SFs cover initial 6-12 percent of GDP.
    - Maximum CCRIF access coverage estimated in range 2-17 percent of GDP.
    - Residual shortfall covered by CAT bonds in range 1.7-4.6 percent of GDP.
  - These thresholds and coverage levels are illustrative and can be modified according to countries’ risk-aversion preferences and institutional capacity to safeguard SF integrity.

### Representative numeric highlights and calibration examples (as reported)
- 2000 simulations.
- ND identification: highest 푑푑 percent tail of pdf.
- Example calibration (Dominica):
  - P[d] = 0.2 (ND every 5 years on average).
  - Initial SF stock = 10 percent of GDP.
  - Probability of depletion within next ten years = 0.08.
  - Target simulation probability of depletion = 5 percent.
  - Budget saving flows in non-ND years = 1.5 percent of previous-year’s GDP.
  - Rate of return 퐺퐺
푡푡 = 2 percent.
- Insurance parameters:
  - CCRIF insurance multipliers range = 1.5-2.0.
  - CCRIF payout discount factor for 1/20 year losses = 0.5.
  - CAT bonds maturity = 3-year; interest = Libor plus 500 basis points.
  - CAT bond issuance size if needed = US$ 10 to 100 million.
- Coverage targets:
  - Overall ND fiscal costs coverage target = 99 percent.
  - SFs coverage target of expected damages = 95 percent (layer 1).
  - Simulated total coverage needed for ECCU countries = 13-31 percent of GDP.
  - SFs’ illustrative coverage = 6-12 percent of GDP.
  - CCRIF illustrative coverage = 2-17 percent of GDP.
  - CAT bonds illustrative coverage = 1.7-4.6 percent of GDP.
- Text-table excerpt (Fund metrics, in percent of GDP unless indicated) reported for selected countries:
  - Fund size: ATG 12.00, DMA 10.00, GRD 6.00, KNA 10.00, LCA 8.00, VCT 8.00
  - Annual budget saving if no ND: ATG 1.90, DMA 1.50, GRD 0.35, KNA 1.90, LCA 0.90, VCT 0.95
  - Probability of fund depletion, units: ATG 0.05, DMA 0.08, GRD 0.04, KNA 0.07, LCA 0.05, VCT 0.03
  - Annual expected use of the fund if ND: ATG 8.7, DMA 5.9, GRD 3.0, KNA 4.8, LCA 3.0, VCT 2.6
  - Average fiscal cost of simulated ND 1/: ATG 57.9, DMA 29.3, GRD 29.7, KNA 19.3, LCA 15.0, VCT 12.9
  - Note: 1/ Includes estimated average decline in revenues and increase in primary expenditures in the simulations, net of expected increase in donor grants.

*Source: wpiea2020266-print-pdf — IMF staff simulations and text as provided.*

### 37.      The insurance layering with 99 percent coverage implies annual fiscal costs in

### wpiea2020266-print-pdf - 37.      The insurance layering with 99 percent coverage implies annual fiscal costs in

### Insurance layering costs and country heterogeneity
- The insurance layering with 99 percent coverage implies annual fiscal costs in the range of 0.5-1.9 percent of GDP.
- The simulations indicate higher costs for Dominica, Antigua and Barbuda and St. Kitts and Nevis, reflecting higher estimated Average Annual Losses of NDs in those countries.
- Cost composition reflects the relatively expensive nature of parametric insurance; SFs are more cost effective relative to the significant level of coverage targeted.

### Parametric insurance pricing, multipliers, and fiscal implications
- In the simulations, insurance costs are calibrated with multipliers of 1.5-2.0 (ratio of insurance premia to expected payouts).
- Multipliers above 1 imply that insurance costs worsen fiscal sustainability in expected terms, albeit they reduce dispersion of expected debt outcomes by reducing need to issue debt after extreme NDs.
- High cost of parametric insurance could be difficult to accommodate given fiscal sustainability challenges in the ECCU.

### Role of Saving Funds (SFs) and layering strategy
- Increasing the share of coverage with SFs could reduce costs, but maintaining SFs in practice may be challenging given competing developmental needs and political pressures for spending.
- Targeting a lower coverage only implies no internalization of these costs and need for additional debt issuance ex-post.
- The paper argues that self-insurance with a SF is needed given high cost of parametric insurance instruments.
- A layering framework including parametric instruments can be used to cover tail NDs risk for relatively low probability events of high intensity and damage.
- Saving funds and the parametric instruments would provide quick access to resources in the aftermath of NDs, important for immediate relief and rehabilitation.

### Quantified SF and savings requirements (simulation results)
- Countries would need SF stocks in the range of 6-12 percent of GDP and annual savings of 0.5-1.9 percent of GDP to achieve financial sustainability with a low probability of depletion.
- These results are based on Monte Carlo experiments using VAR model estimates for each country that account for the impact of natural disasters on output and government finances.

### Donor subsidy proposal to reduce underinsurance
- International donors could reduce incentives for underinsurance with grants subsidizing insurance costs, as part of a comprehensive ex-ante resilience strategy.
- Donor grants for insurance payments could be designed to remove existing incentives for underinsurance due to high cost, while supporting fiscal sustainability by entrenching internalization of NDs fiscal costs.
- Donors could provide an “insurance subsidy” grant to make insurance costs fiscally neutral, calibrated to equalize insurance costs paid by governments with expected average annual losses of disasters—effectively resulting in insurance multipliers of 1 from the country’s perspective.
- Under the simulation assumptions above, making insurance fiscally-neutral in expected terms would require grants in the range of 0.2-1.1 percent of GDP per year –equivalent to about US$ 40 million per year for all ECCU countries.
- This design would be incentive-compatible: it incentivizes appropriate levels of insurance purchase from governments’ perspective and ensures donors’ allocation aligns with fiduciary mandates.

### Caveats, lower-bound nature of estimates, and resilience investments
- The insurance need estimated in this paper are a lower bound because:
  - The fiscal space for re-prioritization may be lower than assumed in the simulations.
  - The methodology focuses on cyclical fluctuations, thereby omitting long term increase in insurance coverage need as a result of increase in disaster intensity and/or frequency with climate change.
  - Natural disasters can have permanent or structural effects on output, which are not incorporated in the estimations in this paper.
- Investment in infrastructure resilient to natural disasters, if sustained, would reduce the extent of damage and the fiscal cost of natural disasters, reducing the coverage need and cost of insurance.
- These reasons imply the need to periodically reassess and recalibrate the insurance need of each country.

### Institutional design recommendations for SFs and use of parametric instruments
- Strong institutional setup of SFs is critical to contain political pressures for spending or opportunistic appropriations.
- A strong institutional design should include:
  - Unambiguous budget contribution and disbursement rules, with triggers based on verifiable criteria.
  - A clearly-stated objective.
  - Strict information disclosure requirements to ensure transparency of operations.
- CCRIF and other market catastrophe insurance can complement government self-insurance for medium and large disasters.
- Parametric triggers are imperfectly correlated with damages for low and medium intensity NDs; for large and extreme NDs, parametric thresholds are more likely to trigger and could top-up SFs coverage.
- High cost of parametric insurance makes the economic and welfare justification questionable, although risk aversion in social welfare criteria could justify some (non-negative) purchase of actuarially-unfair parametric options to reduce dispersion of debt outcomes at the cost of worsening expected debt sustainability.

*Source: IMF staff calculations based on CCRIF and authorities' data.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020266-print-pdf.pdf_