## Executive Summary (tarea2025035-print-pdf)

## Source details

**Canonical URL:** [Executive Summary (tarea2025035-print-pdf)](https://www.imf.org/-/media/files/publications/tar/2025/english/tarea2025035-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/tar/2025/english/tarea2025035-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/tar/2025/english/tarea2025035-print-pdf.pdf.json)

---

### Mission focus and activities
- Focus: enhance the CBCS Financial Stability Report (FSR).
- Reviewed FSR (2023) composition, financial stability analyses, assessments, and analytical toolkit.
- Helped Financial Stability Division estimate sectoral credit risk models to enhance the forward-looking element of the FSR.
- Explained Bayesian Model Averaging (BMA) as an alternative to address model uncertainty.
- Provided R script with initial estimates for eight sectoral credit risk models covering:
  - mortgages to households, personal loans, corporate mortgages, and other corporate loans for Curaçao and Sint Maarten.
- Purpose of models: project impact of macroeconomic scenarios on banks’ balance sheets and produce NPL ratio projections using CBCS macroeconomic forecasting framework.
- Discussed: financial stability indicators, credit risk, stress testing, insurance & pension economic balance sheets and corresponding risks, climate risk basics, interconnectedness and contagion risk.

### Analytical toolkit and data management
- Initial sectoral credit risk models should be further improved and integrated into the FSR.
- Recommended data actions:
  - Explore housing price statistics to monitor real estate imbalances.
  - Collect data on debt service to income to monitor household and corporate debt burdens.
  - Organize all data in one data warehouse and fully utilize for analytics; set up a mirroring database for analytical processing (e.g., to support R) and process data directly at the server.
  - Explore setting up a centralized credit register at the bank using data already reported by banks to private credit registers.
  - In the interim, cover data gaps with surveys.
  - Clearly communicate cut-off dates for different data sources and explore all existing data sources not yet employed (e.g., insolvency data).

### Key findings on FSR content and structure
- CBCS has published the FSR since 2022.
- FSR strengths:
  - Clear and structured messages with a macro-financial narrative on key risks and real sector developments.
  - Covers macro-financial developments, banks, insurers, pension funds, interconnectedness, outlook, and policies; includes charts on financial system composition.
  - Foreword with key messages from the President enhances external profile.
- Areas for enhancement:
  - Streamline text to avoid repetition and focus on key risk drivers.
  - Substantially reduce nominal values in text—use relative indicators or move to statistical annex.
  - Add mission statements/themes for paragraphs to assist navigation.
  - Reduce number of pages of basic text without annexes.
  - State objective/financial stability definition at report start rather than in policy chapter; clarify supervisory responsibility.
  - Communicate financial stability indicators with key aspects of existing regulatory framework for banks, insurers, pension funds; note liabilities might not be fully market-consistent and some indicators may not capture underlying risks.
  - Improve non-bank sector analysis considering upcoming IFRS17 reporting for insurance.
  - Consistently cover emerging risks (cyber and climate) with brief updates and references to prior discussions.
  - Enhance internal and external communication (cross-departmental discussions, mailing lists, press conference, interviews/YouTube, communication strategy).

### Financial system composition and vulnerabilities (end-2022)
- Total assets of the financial sector: 319% of GDP in 2022.
- Composition of total financial assets (end-2022):
  - Banks: 156% of GDP, 48.9% of TFA
  - Pension funds: 105% of GDP, 41.4% of TFA
  - Life insurers: 58% of GDP, 7.9% of TFA
  - Non-life insurers: 1.8% of TFA
- Financial system compiled by 49 local financial institutions:
  - eight banks, ten life insurance companies, nineteen non-life insurance companies, and twelve pension funds.
- Banking sector:
  - Banking assets = 156 % of union GDP in 2022.
  - Top three banks = half of total banking assets.
  - Two banks have related party connections with non-bank financial companies.
  - Three banks are foreign-owned (two by a Canadian bank and one by a bank from Trinidad and Tobago), and one is minority foreign-owned.
  - Offshore banks (much bigger in total size) are disconnected from local deposit and loan markets.
- Pension and insurance sector sizes:
  - Pension sector: 132% of union GDP in 2022.
  - Insurance sector: 31% of union GDP in 2022.

### Macroeconomic and fiscal background and systemic risks
- Curaçao:
  - Real GDP grew by 7.9% in 2022.
  - Growth projected to moderate to 3.5 % in 2023 and 3% in 2024.
  - Inflation projected to decline to 3.3% in 2023.
  - External current account deficit in 2023 projected at 24.5 percent of GDP.
  - Overall fiscal balance expected to improve to a surplus of 1.1% of GDP in 2023 from 0.4% of GDP in 2022.
  - Government debt projected to subside to 72% of GDP in 2023 from 77% of GDP in 2022.
  - Fiscal outlook critically depends on the resolution strategy for Ennia.
- Sint Maarten:
  - Growth projected to decelerate to 4% in 2023 from 10% in 2022.
  - Inflation projected to subside to 3% in 2023 from 3.8% in 2022.
  - Current account expected to improve, leading to a small surplus.
  - Overall deficit expected to narrow to 0.3% of GDP from 1% of GDP last year.
  - Public debt envisaged to decline to 48% of GDP from 50% of GDP in 2022.
  - Sint Maarten is refinancing Covid-related loans at the Dutch capital market yields (currently at around 3.4%).
- Systemic vulnerabilities and risks:
  - Global slowdown is key financial stability risk given high dependence on tourism.
  - High inflation reduces equity and bond prices, harming investment portfolios, especially pension funds; pension fund withdrawals from bank deposits observed.
  - Banks are relatively risk-averse with high credit standards; increased credit risk arises via tourism exposure.
  - Bank profitability improved in 2021 and continued in 2022, with net income after taxes increasing by 47.5%.
  - Credit risk concern: consistently high volume of NPLs, though total NPLs decreased at end-2022.
  - Low profitability and de-risking led some foreign banks to withdraw; trend has stabilized.

### Recommendations, priority and timeframe
- Definitions: Near term: < 12 months; Medium term: 12 to 24 months.

- Content of report and analytical toolkit
  1. Streamline FSR text to avoid repetition and focus on key risk drivers.
     - Priority: High
     - Timeframe: Near-term
  2. Strengthen FSR and toolkit to be more forward-looking.
     - Priority: High
     - Timeframe: Medium-term
     - a. Improve credit risk and market risk analysis.
       - Priority: High
       - Timeframe: Medium-term
     - b. Further improve initial sectoral credit risk model estimates.
       - Priority: High
       - Timeframe: Medium-term
  3. Communicate financial stability indicators together with key aspects of the regulatory framework.
     - Priority: High
     - Timeframe: Near-term
  4. Consistently cover emerging risks (cyber and climate) in the FSR.
     - Priority: High
     - Timeframe: Near-term
  5. State definition of financial stability at beginning of report and clarify supervisory responsibility.
     - Priority: High
     - Timeframe: Near-term

- Organizational and communication
  6. Enhance structural cross-departmental discussions before and during drafting.
     - Priority: High
     - Timeframe: Near-term
  7. Enhance external communication and active promotion of the FSR.
     - Priority: High
     - Timeframe: Near-term

- Data and management
  8. Organize all data in one data warehouse and fully utilize for analytical purposes.
     - Priority: High
     - Timeframe: Medium-term
  9. Explore setting up a centralized credit register.
     - Priority: Medium
     - Timeframe: Medium-term
  10. Clearly communicate cut-off dates for different data sources in the FSR.
      - Priority: Medium
      - Timeframe: Medium-term
  11. Explore existing data sources not yet employed (e.g., insolvency data) and initiate surveys to cover data gaps.
      - Priority: High
      - Timeframe: Near-term

*IMF | CARTAC Curaçao and Sint Maarten — Executive Summary (tarea2025035-print-pdf)*

### Key financial-sector metrics and context
- ROA was 2%.
- ROE 13.5%.
- NIM 4.8%.
- NPL ratio 9% (7.2% for business loans, 6.5% for consumer loans and 5.9% for mortgage loans).
- Life insurers: profitability and solvency remain challenging.
- Non-life insurers: expected to remain stable; facing increasing claims due to economic pickup and claims inflation; premiums steadily increasing.

### Organizational and production arrangements for the FSR
- FSR is the CBCS flagship product for financial stability; published annually since 2022.
- Primary responsibility: Financial Stability Division, in particular the Macroprudential Unit.
- Macroeconomic data and projections provided by Economic and Statistics Division of the Economic and Research Department; potential to provide macroeconomic scenarios when moving from sensitivity analysis to macro stress tests with macro-financial consistent scenarios.
- Production plan: prepared by head of Macroprudential Unit; publication day fixed between the end of April and the beginning of May; drafting schedule driven by data availability.
- Recommendation: enhance structural cross-departmental discussions with more extensive involvement of supervision.

### Communication and publication improvements
- Promote the FSR as the key communication tool for financial stability.
- Suggested actions: press conference follow-ups, interviews with key media, YouTube videos with management and experts.
- Provide all chart/table data together with the report on the website.

### Structure, content, and presentation recommendations
- Avoid repetition; focus on central risk story.
- State objective/definition of financial stability at the start and clarify supervisory responsibility.
- Replace nominal values in text with relative indicators or move them to a statistical annex.
- Add mission statements/themes per paragraph and state data cut-off dates.
- Accompany all ratios and indicators with definitions (example: liquidity asset ratio for commercial banks or transferable deposits).
- Add description of relevant regulatory framework elements for different financial segments, including market-sensitivity and risk-based capital approach.

### Emerging risks: climate and cyber
- Climate risk:
  - Caribbean vulnerability to increased frequency/intensity of natural catastrophes (hurricanes) is an existential threat and relevant for Curaçao and Sint Maarten.
  - Actions cited: regional climate workshop (February 2024) and peer-to-peer visit to Central Bank of Barbados (April 2024).
  - Insurance-related climate analysis: calculate climate protection gap (share of non-insured economic losses in total losses after climate-related catastrophe events) for domestic economy.
- Cyber risk: requires more attention given increasing sophistication of cyber-attacks.

### Data and analytical toolkit enhancements (detailed)
- Make FSR and toolkit more forward-looking: improve credit risk and market risk analysis capturing trends in average PD or LGD for segments/sectors.
- Collect granular banking data for credit risk models by loan segments (mortgages, consumer loans, industry-specific).
- Collect PD and LGD data and average LTV ratios by financial institution to compute cross-sector averages and monitor bracket trends.
- For market risk: collect duration for both assets and liabilities; note typical negative duration mismatch for pension funds and life insurers and positive duration mismatch for banks.
- Establish a comprehensive housing price index to monitor price trends, affordability, and regional disparities.
- Collect debt service to income data.
- Insurance analysis improvements:
  - Clarify valuation framework and discounting of liabilities.
  - Provide information on prevailing reinsurance types (proportional or non-proportional).
  - Consider reducing six-month lag in insurance data reporting.
- Pension sector improvements:
  - Provide information on type of pension plans (defined benefit (DB) vs. defined contribution (DC)).
  - Clarify discounting of liabilities and discuss impact of yield changes.
- Interconnectedness analysis: add exposures via bonds and equities (important for pension funds and insurance companies even if banks have limited trading books).

### Stress testing methodology and specific tests
- Current bank stress testing: simple static methodology; annual exercise including:
  - Capital adjustment stress test.
  - Large exposure stress test.
  - Deposit run-off stress test.
  - Large deposit stress test.
- Scenarios and assumptions:
  - Moderate scenario: 10% of the performing loans become nonperforming.
  - Severe scenario: 30% of the performing loans become nonperforming.
  - Additional provisioning assumption: 50%.
  - Capital adequacy target: tests assess whether CAR exceeds supervisory requirement of 10.5% after shocks.
  - Large exposure test: moderate assumes three large loans become nonperforming; severe assumes five large loans become nonperforming.
  - Deposit run-off test: moderate run-off of 20% of deposits; severe run-off of 40% of deposits; test assesses whether liquid assets to required liquidity after run-off > 100 percent.
  - Large deposit stress test: withdrawal of three large deposits under moderate scenario and five large deposits under severe scenario; assesses whether resulting liquid assets to total assets remain above supervisory regulation of 20%.
- Methodology limitations:
  - Static approach does not link balance sheet development with macro-financial scenarios (sensitivity analysis rather than stress testing).
  - Recommendation: move to dynamic balance sheet approach and multi-period, multi-factor solvency stress testing tool (foreseen CARTAC TA mission).

### Macroeconomic scenario design and demand-shock scenario
- Economic Analysis and Research Department produces quarterly macro outlook (March, June, September, and December) and a 5-year forecast of key macro variables.
- Model usage: generate adverse scenarios for financial stability assessment and project sector NPL ratios.
- Post-mission demand shock scenario assumptions constructed by CBCS:
  - The real GDP of the Netherlands and the US will contract by 5 percent in 2024 and 2 percent in 2025.
  - Inflation in both the Netherlands and the US will decline by 2% in 2024 and 1% in 2025.
  - Main trading partners’ economic deterioration adversely affects domestic tourism sectors, GDP growth, and employment; inflation subdues.

### Econometric and modelling recommendations
- Strengthen toolkit with econometric models to project key financial stability variables (expected NPLs, net interest rate income).
- Transition from current stress testing framework to multi-period, multi-factor stress tests using consistent macroeconomic scenarios.
- Bayesian Model Averaging (BMA) presented as alternative to address model uncertainty:
  - BMA flexibility helps handle uncertainty in predictions and parameter estimates due to model specification, variable selection, and dynamic relationships.
  - Caveat: BMA helps mitigate issues with short time series but does not inherently overcome small sample size limitations.

### R script and sectoral credit risk models (initial and revised)
- Mission provided R script with initial estimates for eight sectoral credit risk models covering mortgages to households, personal loans, corporate mortgages, and other corporate loans for Curaçao and Sint Maarten.
- NPL ratio used as proxy for credit risk; potential explanatory variables: real GDP growth, unemployment, inflation, and number of tourist arrivals (yoy change). Time series and sectoral NPL ratios available quarterly.
- Initial empirical findings:
  - Household mortgages (Curaçao and Sint Maarten): NPL ratio rises with lower GDP growth, higher unemployment, and lower tourist arrivals.
  - Household loans without mortgages: similar pattern; tourist arrivals impact non-significant.
  - Commercial mortgages and other loans: unemployment and tourist arrivals effects similar; inflation effect negative (possible tourism revenues effect).
  - Sint Maarten results less convincing than Curaçao due to small, volatile tourism-dependent economy affected by COVID-19; NPL ratios did not increase significantly due to extensive fiscal measures.
- Recommendation: CBCS become familiar with R script and improve initial estimates; consider additional variables as available.

### Model revisions and methodological changes (post-mission)
- CBCS revised sectoral credit risk model and simplified to two models per country (household and corporate).
- Explanatory variable "number of tourist arrivals" replaced with "exports of goods and services".
- Dependent variable (NPL ratio) transformed using logit transformation.
- Interest rate omitted because not part of current CBCS macroeconomic projection model and limited transmission to domestic interest rates given absence of domestic sovereign market and current fixed-interest mortgage practices.
- Models should be further tested and included (with adjustments if needed) in multi-factor, multi-period stress testing framework to be developed with CARTAC.
- CBCS models updated after the mission; results provided in Annex II.

### Key model results (selected, Annex II)
- Household loans in Curaçao (ranked by BIC)
  - model1: c = -0.389; gdp = -0.523; lag(HH,1) = 0.848; inf = NA; unp = NA; exp = NA; df = 4.000; logLik = 29.702; BIC = -46.862; delta = 0.000; weight = 0.565
  - model7: c = -1.979; gdp = NA; lag(HH,1) = 0.400; inf = NA; unp = 3.372; exp = NA; df = 4.000; logLik = 29.209; BIC = -45.875; delta = 0.987; weight = 0.345
  - model2: c = -1.894; gdp = NA; lag(HH,1) = 0.424; inf = -0.174; unp = 3.235; exp = NA; df = 5.000; logLik = 29.226; BIC = -42.774; delta = 4.088; weight = 0.073
  - model5: c = -0.004; gdp = NA; lag(HH,1) = 0.972; inf = -1.825; unp = NA; exp = NA; df = 4.000; logLik = 25.858; BIC = -39.174; delta = 7.689; weight = 0.012
  - model6: c = -0.001; gdp = NA; lag(HH,1) = 1.002; inf = NA; unp = NA; exp = -0.043; df = 4.000; logLik = 24.340; BIC = -36.138; delta = 10.725; weight = 0.003
  - model4: c = 0.013; gdp = NA; lag(HH,1) = 0.981; inf = -1.725; unp = NA; exp = -0.009; df = 5.000; logLik = 25.882; BIC = -36.087; delta = 10.775; weight = 0.003
  - Note: GDP = year-on-year change of the gross domestic product, HH = logit transformation of total household NPLs, inf = inflation, unp = unemployment rate, Exp = year-on-year change of the exports of goods and services. All variables are stationary.
- Household loans in Sint Maarten (ranked by BIC)
  - model1: c = -0.568; gdp = -0.377; lag(HH,1) = 0.748; inf = NA; unp = NA; exp = NA; df = 4.000; logLik = 25.723; BIC = -38.905; delta = 0.000; weight = 0.626
  - model7: c = -2.239; gdp = NA; lag(HH,1) = 0.151; inf = NA; unp = 3.734; exp = NA; df = 4.000; logLik = 24.458; BIC = -36.374; delta = 2.531; weight = 0.177
  - model2: c = -2.117; gdp = NA; lag(HH,1) = 0.195; inf = -2.492; unp = 3.946; exp = NA; df = 5.000; logLik = 25.995; BIC = -36.312; delta = 2.593; weight = 0.171
  - model6: c = -0.151; gdp = NA; lag(HH,1) = 0.931; inf = NA; unp = NA; exp = -0.024; df = 4.000; logLik = 21.710; BIC = -30.879; delta = 8.026; weight = 0.011
  - model5: c = -0.141; gdp = NA; lag(HH,1) = 0.923; inf = -2.070; unp = NA; exp = NA; df = 4.000; logLik = 21.699; BIC = -30.856; delta = 8.049; weight = 0.011
  - model4: c = -0.046; gdp = NA; lag(HH,1) = 0.967; inf = -1.591; unp = NA; exp = -0.019; df = 5.000; logLik = 22.143; BIC = -28.609; delta = 10.296; weight = 0.004
  - Note: definitions as above; all variables are stationary.
- Corporate loans in Curaçao (ranked by BIC)
  - model5: c = -0.205; gdp = NA; lag(CT,1) = 0.864; inf = -2.485; unp = NA; exp = NA; df = 4.000; logLik = 21.889; BIC = -31.236; delta = 0.000; weight = 0.340
  - model2: c = -0.807; gdp = NA; lag(CT,1) = 0.695; inf = -2.312; unp = 1.542; exp = NA; df = 5.000; logLik = 23.308; BIC = -30.938; delta = 0.298; weight = 0.293
  - model7: c = -0.798; gdp = NA; lag(CT,1) = 0.761; inf = NA; unp = 1.748; exp = NA; df = 4.000; logLik = 20.890; BIC = -29.238; delta = 1.999; weight = 0.125
  - model6: c = 0.013; gdp = NA; lag(CT,1) = 1.014; inf = NA; unp = NA; exp = -0.074; df = 4.000; logLik = 20.807; BIC = -29.073; delta = 2.164; weight = 0.115
  - model4: c = -0.149; gdp = NA; lag(CT,1) = 0.897; inf = -2.098; unp = NA; exp = -0.025; df = 5.000; logLik = 22.005; BIC = -28.332; delta = 2.904; weight = 0.080
  - model1: c = -0.221; gdp = -0.212; lag(CT,1) = 0.908; inf = NA; unp = NA; exp = NA; df = 4.000; logLik = 19.902; BIC = -27.262; delta = 3.975; weight = 0.047
  - Note: CT = logit transformation total corporate NPLs; all variables are stationary.
- Corporate loans in Sint Maarten (ranked by BIC)
  - model7: c = -2.428; gdp = NA; lag(CT,1) = -0.037; inf = NA; unp = 4.638; exp = NA; df = 4.000; logLik = 13.786; BIC = -15.030; delta = 0.000; weight = 0.788
  - model2: c = -2.424; gdp = NA; lag(CT,1) = -0.036; inf = -0.058; unp = 4.640; exp = NA; df = 5.000; logLik = 13.786; BIC = -11.896; delta = 3.135; weight = 0.164
  - model1: c = -1.076; gdp = -0.401; lag(CT,1) = 0.433; inf = NA; unp = NA; exp = NA; df = 4.000; logLik = 10.641; BIC = -8.741; delta = 6.290; weight = 0.034
  - model6: c = -1.038; gdp = NA; lag(CT,1) = 0.453; inf = NA; unp = NA; exp = 0.015; df = 4.000; logLik = 9.003; BIC = -5.465; delta = 9.566; weight = 0.007
  - model5: c = -1.020; gdp = NA; lag(CT,1) = 0.464; inf = 0.500; unp = NA; exp = NA; df = 4.000; logLik = 8.906; BIC = -5.271; delta = 9.760; weight = 0.006
  - model4: c = -1.043; gdp = NA; lag(CT,1) = 0.452; inf = 0.068; unp = NA; exp = 0.015; df = 5.000; logLik = 9.004; BIC = -2.330; delta = 12.701; weight = 0.001
  - Note: definitions as above; all variables are stationary.

### Data and analytical recommendations linked to model use
- Further test models and include in multi-factor, multi-period stress testing framework to be developed with CARTAC.
- Organize all data in one data warehouse and fully utilize them for analytics; set up a statistical mirroring database to support analytical tools for server-side processing.
- Explore establishing a centralized credit register; many banks already report to private credit registers.
- Ensure credit register and data warehouse are linkable to calculate default rates and average LGDs for segments (mortgages, consumer loans).
- In absence of credit register, use targeted surveys to fill data gaps (average PDs/default rates, average LGDs, average lending rates, SME vs large corporate breakdown, type of collateral, average deposit rates).
- Surveys could also cover cyber risk data gaps (number of cyber incidents exceeding threshold; total losses; breakdown by incident type).

*IMF | CARTAC Curaçao and Sint Maarten*

### Executive Summary

### Executive Summary

### Mission focus and activities
- The mission focused on enhancing the CBCS (Central Bank of Curaçao and Sint Maarten) Financial Stability Report (FSR).
- Reviewed the latest available FSR (2023) and discussed composition, financial stability analyses, assessments, and analytical toolkit.
- Helped the Financial Stability Division estimate sectoral credit risk models to enhance the forward-looking element of the FSR.
- Explained the Bayesian Model Averaging (BMA) approach as an alternative methodology to address model uncertainty.
- Provided R script with an initial estimate for eight sectoral credit risk models:
  - mortgages to households, personal loans, corporate mortgages, and other corporate loans for both countries — Curaçao and Sint Maarten.
- The estimated credit risk models allow the CBCS to project the impact of different macroeconomic scenarios on banks’ balance sheets and make NPL ratio projections employing the macroeconomic forecasting framework available to the CBCS.
- Discussed financial stability indicators, credit risk, stress testing, insurance & pension economic balance sheets and corresponding financial stability risks, basic elements of climate risk, interconnectedness and contagion risk.

### Analytical toolkit and data management
- Initial sectoral credit risk models developed during the mission should be further improved and integrated into the FSR.
- The mission recommended exploring housing price statistics as essential information to monitor potential imbalances in the real estate market.
- Recommended collecting data on debt service to income to complement monitoring household and corporate debt burdens.
- Recommended organizing all data in one data warehouse and fully utilizing them for analytical purposes; set up a mirroring database for analytical processing (e.g., to support R) and process data directly at the server.
- Recommended exploring how to set up a centralized credit register at the bank using data already reported by banks to private credit registers.
- In the interim, existing data gaps could be covered by surveys.
- Cut-off dates for different data sources should be clearly communicated in the report and all existing data sources not employed so far should be explored (e.g., insolvency data).

### Key findings on the FSR content and structure
- The CBCS has started publishing the FSR since 2022.
- The report provides clear and structured messages, containing a macro-financial narrative on key risks and covering developments in the real sector.
- The FSR captures macro-financial developments, banks, insurers, pension funds, interconnectedness, outlook, and policies, providing clear charts presenting the composition of the financial system.
- The Foreword with the key messages from the President boosts the external profile.
- Areas for enhancement:
  - Streamline text to avoid repetition and focus on key risk drivers.
  - Substantially reduce nominal values reported in text — replace by relative indicators or move to the statistical annex.
  - Add mission statements/themes for each paragraph to help navigation.
  - Reduce the number of pages of the basic text without annexes.
  - State the objective/financial stability definition at the beginning of the report rather than in the policy chapter, and add clarification on supervisory responsibility.
  - Communicate financial stability indicators together with key aspects of the existing regulatory framework for all segments (banks, insurance companies, pension funds), noting liabilities might not be fully market-consistent and some indicators may not capture underlying risks.
  - Improve analysis on non-bank sectors considering upcoming IFRS17 reporting for the insurance sector.
  - Consistently cover emerging risks (cyber and climate risks) throughout the report with brief discussions on latest developments and references to previous discussions so the report is self-sustained.
  - Enhance internal and external communication, including structural cross-departmental discussion(s) before and during drafting, internal distribution mailing list, press conference followed by interviews and/or YouTube videos, and elaborated communication strategy.

### Financial system composition and vulnerabilities (end-2022)
- Total assets of the financial sector corresponded to 319% of GDP in 2022.
- Composition of total financial assets (end-2022):
  - Banks: 156% of GDP, 48.9% of TFA
  - Pension funds: 105% of GDP, 41.4% of TFA
  - Life insurers: 58% of GDP, 7.9% of TFA
  - Non-life insurers: 1.8% of TFA
- The financial system is compiled by 49 local financial institutions:
  - eight banks, ten life insurance companies, nineteen non-life insurance companies, and twelve pension funds.
- Banking sector characteristics:
  - Relatively concentrated with banking assets making up 156 % of the union's GDP in 2022.
  - The top three banks correspond to half of the total banking assets.
  - Two banks have related party connections with non-bank financial companies.
  - Three banks are foreign-owned (two by a Canadian bank and one by a bank from Trinidad and Tobago), and one is minority foreign-owned.
  - Offshore banks (much bigger in total size) are completely disconnected from local deposit and loan markets.
- Pension and insurance sector sizes:
  - Pension sector: 132% of union GDP in 2022.
  - Insurance sector: 31% of union GDP in 2022.

### Macroeconomic and fiscal background and systemic risks
- Curaçao:
  - Real GDP grew by 7.9% in 2022.
  - Growth projected to moderate to 3.5 % in 2023 and 3% in 2024.
  - Inflation projected to decline to 3.3% in 2023.
  - External current account deficit in 2023 projected at 24.5 percent of GDP.
  - Overall fiscal balance expected to improve to a surplus of 1.1% of GDP in 2023 from 0.4% of GDP in 2022.
  - Government debt projected to subside to 72% of GDP in 2023 from 77% of GDP in 2022.
  - Fiscal outlook critically depends on the resolution strategy for Ennia, a large insurance group.
- Sint Maarten:
  - Growth projected to decelerate to 4% in 2023 from 10% in 2022.
  - Inflation projected to subside to 3% in 2023 from 3.8% in 2022.
  - Current account expected to improve, leading to a small surplus.
  - Overall deficit expected to narrow to 0.3% of GDP from 1% of GDP last year.
  - Public debt envisaged to decline to 48% of GDP from 50% of GDP in 2022.
  - Sint Maarten is refinancing Covid-related loans at the Dutch capital market yields (currently at around 3.4%).
- Systemic vulnerabilities and risks:
  - The global slowdown is the key financial stability risk given high dependence on tourism.
  - High inflation drives down equity and bond prices, negatively impacting investment portfolios, especially pension funds; pension fund withdrawals from bank deposits have been observed.
  - Banks are relatively risk-averse with high credit standards, but increased credit risk arises via tourism exposure.
  - Bank profitability improved in 2021 and continued in 2022, with net income after taxes increasing by 47.5%.
  - Credit risk remains a concern given consistently high volume of NPLs, though total NPLs decreased at end-2022.
  - Low profitability and de-risking have driven some foreign banks to withdraw from the local market; this trend has been stabilized.

### Recommendations, priority and timeframe
- Near term: < 12 months; Medium term: 12 to 24 months.

- The content of the report including the analytical toolkit
  1. The text of the FSR could be streamlined to avoid repetition and focus on key risk drivers.
     - Priority: High
     - Timeframe: Near-term
  2. The FSR and the corresponding toolkit should be strengthened to be more forward-looking.
     - Priority: High
     - Timeframe: Medium-term
     - a. Credit risk and market risk analysis should be improved.
       - Priority: High
       - Timeframe: Medium-term
     - b. The initial estimates of sectoral credit risk models developed during the mission should be further improved.
       - Priority: High
       - Timeframe: Medium-term
  3. Financial stability indicators need to be communicated together with the key aspects of the existing regulatory framework.
     - Priority: High
     - Timeframe: Near-term
  4. New emerging risks, such as cyber and climate risks, should be consistently covered in the FSR.
     - Priority: High
     - Timeframe: Near-term
  5. The definition of financial stability could be stated at the beginning of the report rather than in the policy chapter, and some clarification on supervisory responsibility could be added.
     - Priority: High
     - Timeframe: Near-term

- The organizational and operational setup of the report and its communication
  6. Enhance structural cross-departmental discussion(s) before starting the report drafting and during the drafting process.
     - Priority: High
     - Timeframe: Near-term
  7. The external communication of the FSR could be further enhanced by active promotion of the report.
     - Priority: High
     - Timeframe: Near-term

- Data and their management
  8. The CBCS should organize all data in one data warehouse and fully utilize them for analytical purposes to support the financial stability analytical toolkit.
     - Priority: High
     - Timeframe: Medium-term
  9. The CBCS should explore how to set up a centralized credit register.
     - Priority: Medium
     - Timeframe: Medium-term
  10. Clearly communicate the cut-off dates for different data sources in the FSR.
     - Priority: Medium
     - Timeframe: Medium-term
  11. Explore all existing data sources not employed so far for the FSR, e.g., insolvency data. Initiate surveys with the industry to cover the existing data gaps.
     - Priority: High
     - Timeframe: Near-term

*IMF | CARTAC Curaçao and Sint Maarten — Executive Summary (tarea2025035-print-pdf)*

### 12. Profitability and solvency remain challenging for the life insurance sector in the union. On the contrary,

### 12. Profitability and solvency remain challenging for the life insurance sector in the union. On the contrary,

### Key financial-sector metrics and context
- ROA was 2%.
- ROE 13.5%.
- NIM 4.8%.
- NPL ratio 9% (7.2% for business loans, 6.5% for consumer loans and 5.9% for mortgage loans).
- Life insurers: profitability and solvency remain challenging.
- Non-life insurers: expected to remain stable; facing increasing claims due to the economy picking back up and claims inflation, but premiums are steadily increasing.

### Organizational and production arrangements for the FSR
- The FSR is the flagship product of the CBCS for financial stability; published annually since 2022.
- Primary responsibility: Financial Stability Division, in particular the Macroprudential Unit.
- Macroeconomic data and projections provided by the Economic and Statistics Division of the Economic and Research Department; potential to provide macroeconomic scenarios when moving from sensitivity analysis to macro stress tests with macro-financial consistent scenarios.
- Production plan: prepared by the head of the Macroprudential Unit; publication day fixed between the end of April and the beginning of May; detailed drafting plan driven by data availability.
- Internal interaction: recommendation to enhance structural cross-departmental discussions before and during drafting, with more extensive involvement of supervision.

### Communication and publication improvements
- Promote the FSR as the key communication tool for financial stability.
- Suggested additional communication actions: press conference follow-ups, interviews with key media, YouTube videos with management and experts.
- Provide all data displayed in charts or tables together with the report on the website.

### Structure, content, and presentation recommendations
- FSR strengths: clear, structured macro-financial narrative; covers real sector, banks, insurers, pension funds, interconnectedness, financial sector outlook, and policies; includes foreword with key messages from the president of the CBCS.
- Areas to streamline:
  - Avoid repetition; focus on key risk drivers following a central story.
  - State definition of objective/financial stability at the beginning.
  - Clarify supervisory responsibility.
  - Reduce nominal values in text—replace by relative indicators or move to a statistical annex.
  - Add mission statements/themes for paragraphs and state data cut-off dates.
  - Reduce number of pages of text without annexes.
- Add description of relevant elements of the regulatory framework for different financial segments, including market-sensitivity and risk-based capital approach, especially for non-bank financial institutions.
- Accompany all ratios and indicators with definitions (example: liquidity asset ratio for commercial banks or transferable deposits).

### Emerging risks: climate and cyber
- Climate risk: elaborate more on what CBCS is doing/planning; Caribbean vulnerability to increased frequency/intensity of natural catastrophes (hurricanes) is an existential threat and relevant for Curaçao and Sint Maarten.
- Actions cited: regional climate workshop (February 2024) and peer-to-peer visit to Central Bank of Barbados (April 2024) to learn about climate risk in financial stability context.
- Cyber risk: requires more attention given higher sophistication of cyber-attacks over time.
- Insurance-related climate analysis: calculate climate protection gap (share of non-insured economic losses in total losses after climate-related catastrophe events) for domestic economy.

### Data and analytical toolkit enhancements
- Make the FSR and toolkit more forward-looking: improve credit risk and market risk analysis, capturing trends in average PD or LGD for segments/sectors.
- Collect granular banking data to monitor credit risk models for loan segments (mortgages, consumer loans, industry-specific).
- Collect data on PD and LGD and on average LTV ratios by financial institutions to compute cross-sector averages and monitor bracket trends.
- For market risk: collect data on duration for both assets and liabilities; note typical negative duration mismatch for pension funds and life insurers and positive duration mismatch for banks.
- Explore setting up housing price statistics: establish a comprehensive housing price index to monitor price trends, affordability, and regional disparities.
- Collect data on debt service to income and use them to complement monitoring household and corporate debt burdens.
- Insurance analysis improvements:
  - Clarify valuation framework and discounting of liabilities.
  - Provide information on prevailing types of reinsurance (proportional or non-proportional).
  - Consider reducing the six-month lag in insurance data reporting.
- Pension sector improvements:
  - Provide information on type of pension plans (defined benefit (DB) vs. defined contribution (DC)).
  - Clarify discounting of liabilities and discuss impact of changes in yields.
- Interconnectedness analysis: add exposures via bonds and equities (important for pension funds and insurance companies even if banks have limited trading books).

### Stress testing methodology and specific tests
- Current bank stress testing: simple static methodology, annual exercise including:
  - Capital adjustment stress test.
  - Large exposure stress test.
  - Deposit run-off stress test.
  - Large deposit stress test.
- Scenarios and assumptions:
  - Moderate scenario: 10% of the performing loans become nonperforming.
  - Severe scenario: 30% of the performing loans become nonperforming.
  - Additional provisioning assumption: 50%.
  - Capital adequacy target: tests show whether CAR exceeds the supervisory requirement of 10.5% after shocks.
  - Large exposure test: moderate scenario assumes three large loans become nonperforming; severe assumes five large loans become nonperforming.
  - Deposit run-off test: moderate scenario run-off of 20% of deposits; severe scenario run-off of 40% of deposits; test shows whether the liquid assets to required liquidity after the run-off is above 100 percent.
  - Large deposit stress test: withdrawal of three large deposits under moderate scenario and five large deposits under severe scenario; shows whether resulting liquid assets to total assets will be adequate and above the supervisory regulation of 20%.
- Methodology limitations: static approach does not link balance sheet development with macro-financial scenarios (sensitivity analysis rather than stress testing); recommendation to move to dynamic balance sheet approach and multi-period, multi-factor solvency stress testing tool (foreseen CARTAC TA mission).

### Macroeconomic scenario design and specific demand-shock scenario
- Economic Analysis and Research Department produces quarterly macro outlook (March, June, September, and December) and a 5-year forecast of key macro variables.
- Model can be used to generate adverse scenarios for financial stability assessment and to project sector NPL ratios.
- Post-mission demand shock scenario assumptions constructed by CBCS:
  - The real GDP of the Netherlands and the US will contract by 5 percent in 2024 and 2 percent in 2025.
  - Inflation in both the Netherlands and the US will decline by 2% in 2024 and 1% in 2025.
  - Main trading partners’ economic deterioration adversely affects domestic tourism sectors, GDP growth, and employment; inflation subdues.

### Econometric and modelling recommendations
- Strengthen toolkit with econometric models to project key financial stability variables (expected NPLs, net interest rate income).
- Move from current stress testing framework to multi-period, multi-factor stress tests using consistent macroeconomic scenarios.
- Bayesian Model Averaging (BMA) presented as an alternative to address model uncertainty; BMA flexibility helps handle uncertainty in predictions and parameter estimates due to model specification, variable selection, and dynamic relationships.
  - Caveat: BMA helps mitigate issues with short time series but does not inherently overcome small sample size limitations.

### R script and sectoral credit risk models
- Mission provided an R script with initial estimates for eight sectoral credit risk models covering:
  - Mortgages to households, personal loans, corporate mortgages, and other corporate loans for both Curaçao and Sint Maarten.
- NPL ratio used as proxy for credit risk; potential explanatory variables considered: real GDP growth, unemployment, inflation, and number of tourist arrivals (year-on-year change). Time series and sectoral NPL ratios available quarterly.
- Initial empirical findings:
  - Household mortgages (Curaçao and Sint Maarten): NPL ratio rises with lower GDP growth, higher unemployment, and lower tourist arrivals.
  - Household loans without mortgages: same pattern; impact of tourist arrivals was non-significant.
  - Commercial mortgages and other loans: unemployment and tourist arrivals effects similar; effect of inflation was negative (possibly because higher inflation may increase profits in tourism as costs might increase less than revenues).
  - Initial results for Sint Maarten were less convincing than for Curaçao due to small, highly volatile tourism-dependent economy heavily affected by COVID-19; NPL ratios did not increase significantly due to extensive fiscal measures.
- Recommendation: CBCS should become familiar with the provided R script and further work on improvements to the initial estimates; consider additional variables as they become available.

*Italic: IMF | CARTAC Curaçao and Sint Maarten | excerpt from chapter on financial stability analytical framework and FSR production*

### 32. The CBCS further revised the sectoral credit risk model and simplified it to two models per country.

### 32. The CBCS further revised the sectoral credit risk model and simplified it to two models per country.

### Model revisions and methodological changes
- The team replaced the explanatory variable capturing a number of tourist arrivals with exports of goods and services.
- The dependent variable — NPL ratio — was transformed using logit transformation.
- The CBCS reduced the number of sectors per country to household and corporate sectors only.
- The models should be further tested and included (with potential adjustments if needed) in the multi-factor, multi-period stress testing framework that will be developed with the support of CARTAC.

### Rationale for variable choices and omitted variables
- Exports of goods and services were chosen because they have better statistical properties and are correlated with tourist arrivals.
- Interest rate was not considered as it is not part of the current CBCS macroeconomic projection model and the transmission of global interest rates to domestic interest rates is rather limited given the none-existence of the domestic sovereign market and current banking practices keeping rather fixed-interest rate on mortgages.

### Implementation and next steps
- The CBCS models were updated after the mission and results are provided in Annex II.
- The models should be further tested and potentially adjusted before inclusion in the multi-factor, multi-period stress testing framework developed with CARTAC.
- The CBCS should become familiar with the provided R script for the sectoral credit risk models and further work on improvements to the initial estimates.

### Key model results (Annex II) — household loans in Curaçao (ranked by BIC)
- model1: c = -0.389; gdp = -0.523; lag(HH,1) = 0.848; inf = NA; unp = NA; exp = NA; df = 4.000; logLik = 29.702; BIC = -46.862; delta = 0.000; weight = 0.565
- model7: c = -1.979; gdp = NA; lag(HH,1) = 0.400; inf = NA; unp = 3.372; exp = NA; df = 4.000; logLik = 29.209; BIC = -45.875; delta = 0.987; weight = 0.345
- model2: c = -1.894; gdp = NA; lag(HH,1) = 0.424; inf = -0.174; unp = 3.235; exp = NA; df = 5.000; logLik = 29.226; BIC = -42.774; delta = 4.088; weight = 0.073
- model5: c = -0.004; gdp = NA; lag(HH,1) = 0.972; inf = -1.825; unp = NA; exp = NA; df = 4.000; logLik = 25.858; BIC = -39.174; delta = 7.689; weight = 0.012
- model6: c = -0.001; gdp = NA; lag(HH,1) = 1.002; inf = NA; unp = NA; exp = -0.043; df = 4.000; logLik = 24.340; BIC = -36.138; delta = 10.725; weight = 0.003
- model4: c = 0.013; gdp = NA; lag(HH,1) = 0.981; inf = -1.725; unp = NA; exp = -0.009; df = 5.000; logLik = 25.882; BIC = -36.087; delta = 10.775; weight = 0.003
- Note: GDP = year-on-year change of the gross domestic product, HH = logit transformation of total household NPLs, inf = inflation, unp = unemployment rate, Exp = year-on-year change of the exports of goods and services. All variables are stationary.

### Key model results (Annex II) — household loans in Sint Maarten (ranked by BIC)
- model1: c = -0.568; gdp = -0.377; lag(HH,1) = 0.748; inf = NA; unp = NA; exp = NA; df = 4.000; logLik = 25.723; BIC = -38.905; delta = 0.000; weight = 0.626
- model7: c = -2.239; gdp = NA; lag(HH,1) = 0.151; inf = NA; unp = 3.734; exp = NA; df = 4.000; logLik = 24.458; BIC = -36.374; delta = 2.531; weight = 0.177
- model2: c = -2.117; gdp = NA; lag(HH,1) = 0.195; inf = -2.492; unp = 3.946; exp = NA; df = 5.000; logLik = 25.995; BIC = -36.312; delta = 2.593; weight = 0.171
- model6: c = -0.151; gdp = NA; lag(HH,1) = 0.931; inf = NA; unp = NA; exp = -0.024; df = 4.000; logLik = 21.710; BIC = -30.879; delta = 8.026; weight = 0.011
- model5: c = -0.141; gdp = NA; lag(HH,1) = 0.923; inf = -2.070; unp = NA; exp = NA; df = 4.000; logLik = 21.699; BIC = -30.856; delta = 8.049; weight = 0.011
- model4: c = -0.046; gdp = NA; lag(HH,1) = 0.967; inf = -1.591; unp = NA; exp = -0.019; df = 5.000; logLik = 22.143; BIC = -28.609; delta = 10.296; weight = 0.004
- Note: GDP = year-on-year change of the gross domestic product, HH = logit transformation of total household NPLs, inf = inflation, unp = unemployment rate, Exp = year-on-year change of the exports of goods and services. All variables are stationary.

### Key model results (Annex II) — corporate loans in Curaçao (ranked by BIC)
- model5: c = -0.205; gdp = NA; lag(CT,1) = 0.864; inf = -2.485; unp = NA; exp = NA; df = 4.000; logLik = 21.889; BIC = -31.236; delta = 0.000; weight = 0.340
- model2: c = -0.807; gdp = NA; lag(CT,1) = 0.695; inf = -2.312; unp = 1.542; exp = NA; df = 5.000; logLik = 23.308; BIC = -30.938; delta = 0.298; weight = 0.293
- model7: c = -0.798; gdp = NA; lag(CT,1) = 0.761; inf = NA; unp = 1.748; exp = NA; df = 4.000; logLik = 20.890; BIC = -29.238; delta = 1.999; weight = 0.125
- model6: c = 0.013; gdp = NA; lag(CT,1) = 1.014; inf = NA; unp = NA; exp = -0.074; df = 4.000; logLik = 20.807; BIC = -29.073; delta = 2.164; weight = 0.115
- model4: c = -0.149; gdp = NA; lag(CT,1) = 0.897; inf = -2.098; unp = NA; exp = -0.025; df = 5.000; logLik = 22.005; BIC = -28.332; delta = 2.904; weight = 0.080
- model1: c = -0.221; gdp = -0.212; lag(CT,1) = 0.908; inf = NA; unp = NA; exp = NA; df = 4.000; logLik = 19.902; BIC = -27.262; delta = 3.975; weight = 0.047
- Note: GDP = year-on-year change of gross domestic product, CT = logit transformation total corporate NPLs, inf = inflation, unp = unemployment rate, Exp = year-on-year change of exports of goods and services. All variables are stationary.

### Key model results (Annex II) — corporate loans in Sint Maarten (ranked by BIC)
- model7: c = -2.428; gdp = NA; lag(CT,1) = -0.037; inf = NA; unp = 4.638; exp = NA; df = 4.000; logLik = 13.786; BIC = -15.030; delta = 0.000; weight = 0.788
- model2: c = -2.424; gdp = NA; lag(CT,1) = -0.036; inf = -0.058; unp = 4.640; exp = NA; df = 5.000; logLik = 13.786; BIC = -11.896; delta = 3.135; weight = 0.164
- model1: c = -1.076; gdp = -0.401; lag(CT,1) = 0.433; inf = NA; unp = NA; exp = NA; df = 4.000; logLik = 10.641; BIC = -8.741; delta = 6.290; weight = 0.034
- model6: c = -1.038; gdp = NA; lag(CT,1) = 0.453; inf = NA; unp = NA; exp = 0.015; df = 4.000; logLik = 9.003; BIC = -5.465; delta = 9.566; weight = 0.007
- model5: c = -1.020; gdp = NA; lag(CT,1) = 0.464; inf = 0.500; unp = NA; exp = NA; df = 4.000; logLik = 8.906; BIC = -5.271; delta = 9.760; weight = 0.006
- model4: c = -1.043; gdp = NA; lag(CT,1) = 0.452; inf = 0.068; unp = NA; exp = 0.015; df = 5.000; logLik = 9.004; BIC = -2.330; delta = 12.701; weight = 0.001
- Note: GDP = year-on-year change of gross domestic product, CT = logit transformation total corporate NPLs, inf = inflation, unp = unemployment rate, Exp = year-on-year change of exports of goods and services. All variables are stationary.

### Data and analytical recommendations linked to model use
- Models should be further tested and included in the multi-factor, multi-period stress testing framework to be developed with CARTAC.
- The CBCS should organize all data in one data warehouse and fully utilize them for analytical purposes to support the financial stability analytical toolkit.
- A mirroring database should be set up for analytical purposes; the statistical mirroring database needs to support analytical tools to process large data directly at the server.
- The CBCS should explore setting up a centralized credit register at the bank as a rich data source; most banks and some financial specialized institutions already report to a private credit register.
- The credit register and data warehouse should be linkable to calculate crucial credit risk parameters such as default rates and average LGDs for different segments (e.g., mortgages, consumer loans).
- In the absence of a credit register, targeted surveys could cover data gaps (e.g., average PDs/default rates, average LGDs, average lending rates, SME vs large corporates breakdown, type of collateral, average deposit rates).
- Surveys could also address cyber risk data gaps (e.g., number of cyber incidents exceeding a threshold; total losses; breakdown by incident type).

*IMF | CARTAC Curaçao and Sint Maarten*

---


_Source: https://www.imf.org/-/media/files/publications/tar/2025/english/tarea2025035-print-pdf.pdf_