## tarea2025050-print-pdf

## Source details

**Canonical URL:** [tarea2025050-print-pdf](https://www.imf.org/-/media/files/publications/tar/2025/english/tarea2025050-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/tar/2025/english/tarea2025050-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/tar/2025/english/tarea2025050-print-pdf.pdf.json)

---

### Preface — mission, context, and operational challenges
- Mission: MCM mission visited Paramaribo from April 2–5, 2024, at the request of the Centrale Bank van Suriname (CBvS) to assist authorities in liquidity forecasting; met Governor, Deputy Governor, Directors, staff, and commercial banks.
- Policy regime shift: Suriname moved from an exchange rate to a monetary targeting regime and initiated open market operations (OMOs) in July 2021; CBvS started market operations using mainly term deposits in July 2021.
- Recent macro developments:
  - Suriname dollar devalued by 182.5 percent from September 2020 to June 2021.
  - Inflation rates above 60 percent driven by high exchange rate pass-through.
  - Widespread financial dollarization.
- Prior CBvS actions:
  - Added Central Bank Certificates (CBCs) in June 2022 to absorb excess liquidity.
  - Increased reserve requirement ratio from 39 to 44 percent in April 2023 and established a 20 percent annual credit growth cap for banks.
  - March 2024 underperformance on the reserve money target by SRD 537.7 million.
- Operational challenges:
  - Persistently high excess reserves and repeated target shortfalls.
  - Low bank engagement in auctions and high bid rates relative to bid quantities.
  - Absence of a well-functioning money market, FX market challenges, supervision requirements, and dual reserve-holding systems complicating measurement of excess and precautionary reserves.

### Executive Summary — IMF TA, framework, and key operational advice
- IMF technical assistance:
  - Deployed MCMCO statistical forecasting framework comprising 12 models across three categories: exponential smoothing (ETS), ARIMA (including seasonal and with exogenous regressors), TBATS, and volatility models (GARCH family).
  - Framework includes out-of-sample performance testing with four criteria assessing accuracy, bias, and confidence intervals; supplies code packages and simulations for staff training.
- Liquidity forecasting and OMOs calibration:
  - Developed liquidity forecasting table to track system available reserves and demand for excess reserves, guiding weekly OMO calibration.
  - Liquidity supply captured by autonomous factors and maturity of previous OMOs; demand determined by excess reserves and reserve requirements. Liquidity gap (supply minus demand) guides OMO allotments.
- Data and implementation:
  - CBvS should collect daily data and implement full granular model; consider integrating IMF framework into current operations.
- Bank engagement & market development:
  - Survey and workshop with banks: concerns about CBCs policy, high reserve requirements, high OMO interest rates driven by banks' bids.
  - CBvS developing a fully automated platform for standing facilities; pursue digitalization (electronic signatures) and regular surveys to capture banks’ expectations.
- Strategic reviews:
  - Review CBC strategy (eligibility, tenors, auction frequency, communication), simplify dual reserve systems, modernize infrastructure with government to foster interbank market development.

### Recommendations (Table 1 summary) — priorities and timing
- Liquidity Forecast
  - 1. Implement forecast with statistical models to produce demand of reserve forecast. (¶12) — High priority — Short-term
  - 2. Complement the CBvS’ forecast with statistical models to produce forecast for autonomous factors. (¶21) — High priority — Short-term
  - 3. Collect daily level data for autonomous factors, specifically for liabilities to government and net foreign assets. (¶24) — High priority — Short-term
  - 4. Consider reconciliating the IMF liquidity forecasting framework into current forecast operations. (¶27) — Medium priority — Medium-term
- Participation in Money Market Operations
  - 5. Redirect eligibility criteria to entities subject to the CBvS’ reserve requirements. (¶42) — High priority — Short-term
  - 6. Reduce tenors of CBCs for structural operations to a maximum of 180 days and limit term deposit tenors to maximum three (e.g., 1 week, 30 days, 90 days), ideally indexing to inflation. (¶42) — High priority — Short-term
  - 7. Consider ad-hoc fine-tuning operations for large and unpredicted liquidity fluctuations. (¶44) — High priority — Medium-term
  - 8. Strictly follow planned schedule of auctions as previously communicated. (¶46) — High priority — Short-term
  - 9. Conduct regular surveys (preferably monthly) to understand banks’ expectations on key macro variables. (¶47) — Medium priority — Medium-term
  - 10. Collaborate with government to modernize infrastructure to develop interbank market and improve auction participation. (¶52) — High priority — Short-term
  - 11. Simplify the two separate systems requiring holding reserves in the system. (¶60) — High priority — Short-term

### Implementation of the IMF statistical forecasting framework — features and operations
- Framework features:
  - 12 forecasting models across Naïve, ETS (including with regressors), ARIMA/SARIMA (with regressors), TBATS, and volatility (GARCH family).
  - Out-of-sample testing using RMSE, MAE, ME, and MIS for 95 percent prediction intervals; provides model selection, forecasting, and reconciliation methods.
  - Supports forecasting autonomous factors (CiC, liabilities to government, net foreign assets) and reserve demand; provides forecast uncertainty via confidence intervals.
- Operational recommendations:
  - Customize framework to weekly maintenance period starting Wednesday with five-day horizons.
  - Update forecasts weekly; evaluate model selection periodically (monthly or quarterly).
  - Collect daily stock-level data for autonomous factors, especially liabilities to government and net foreign assets.
  - Combine statistical forecasts with qualitative inputs (bank surveys, demand curves, precautionary buffers).

### Estimating demand for reserves — empirical patterns and model choices
- Commercial bank reserves balance
  - Patterns: upward trend since July 2021 with weekly and quarterly seasonality; drop Jan–Feb 2023; spike end-January 2024 attributed to a 45.45 percent increase in required reserve for two days.
  - Modeling: ARIMA with regression selected as best model across horizons; models tested include Naïve, ETS, ARIMA, TBATS; volatility modeled via GARCH family.
  - Performance metrics: RMSE, MAE, ME, MIS (95 percent).
- Excess reserves
  - Definition: bank reserve balance on account minus reserve requirement; recorded as daily working balance; CBvS adds a five percent SNPES buffer for effective excess reserves.
  - Patterns: large volatility over three years with upward trend; weekly and quarterly seasonality.
  - Modeling: ARIMA with regression selected as best model (lowest RMSE and MAE); ETS with regression less biased for short horizons (t+1, t+2); ETS best for t+4.
- Precautionary liquidity buffers (system-wide thresholds, July 2021–March 2023)
  - 95th percentile (entire banking system) for excess reserves: 2953 million SRD.
  - 5th percentile (entire banking system) for excess reserves: 969 million SRD.
  - Effective excess reserves thresholds (subtracting five percent for SNEPS):
    - 95th percentile: 2124 million SRD.
    - 5th percentile: 408 million SRD.
- Weekly average excess reserve by bank (5th and 95th percentiles)
  - Bank 1: 5th Percentile 306 ; 95th Percentile 1278
  - Bank 2: 5th Percentile 54 ; 95th Percentile 386
  - Bank 3: 5th Percentile 130 ; 95th Percentile 932
  - Bank 4: 5th Percentile 28 ; 95th Percentile 186
  - Bank 5: 5th Percentile 43 ; 95th Percentile 205
  - Bank 6: 5th Percentile 67 ; 95th Percentile 293
  - Bank 7: 5th Percentile 12 ; 95th Percentile 64
  - Bank 8: 5th Percentile 7 ; 95th Percentile 65
  - Bank 9: 5th Percentile 4 ; 95th Percentile 69
- Weekly average effective excess reserve by bank (5th and 95th percentiles)
  - Bank 1: 5th Percentile 117 ; 95th Percentile 1001
  - Bank 2: 5th Percentile -74 ; 95th Percentile 218
  - Bank 3: 5th Percentile 43 ; 95th Percentile 844
  - Bank 4: 5th Percentile -44 ; 95th Percentile 97
  - Bank 5: 5th Percentile 25 ; 95th Percentile 182
  - Bank 6: 5th Percentile -6 ; 95th Percentile 221
  - Bank 7: 5th Percentile 0 ; 95th Percentile 52
  - Bank 8: 5th Percentile -20 ; 95th Percentile 39
  - Bank 9: 5th Percentile -3 ; 95th Percentile 51
- Recommendation: incorporate statistical demand-for-reserve forecasts and combine with qualitative assessments (surveys, demand curves, precautionary buffers).

### Forecasting autonomous factors and net liquidity — model selection and reconciliation results
- Currency in Circulation (CiC)
  - Patterns: upward trend; weekly, quarterly, yearly seasonality; increases at end of year and decreases in first quarter.
  - Model selection: ARIMA with regression best for one-, two-, and four-week forecasts; ETS with regression best for cash in vault; reconciliation (STR) improves RMSE at all horizons.
  - Model selections (Table 4): Currency Outside Banks: ARIMA with Regression; Cash in Vault: ETS with Regression; Currency in Circulation: ARIMA with Regression.
- Liabilities to Government
  - Patterns: monthly and quarterly seasonality; lower in Q2 and Q3 vs Q1 and Q4; weekday distribution stable.
  - Model selection: TBATS ranked best for forecast accuracy; seasonal ARIMA closer to unbiased position for forecast bias.
  - Data recommendation: collect daily stock-level liabilities to government (current model uses net changes with initial level start-2019).
- Net Liquidity (NL)
  - Model selection: TBATS best at t+1 week (lowest RMSE and MAE); ARIMA with regression better at t+2 and t+4 weeks.
  - Reconciliation methods tested: Bottom Up, MinT, STR, OLS.
    - For one-week horizon: STR recommended (highest forecast accuracy for one week).
    - For two- and four-week horizons: Bottom Up provides lowest RMSE and MAE.
    - In bias (ME): Bottom Up has the lowest ME across horizons.
  - Predictive accuracy and bias for reconciled forecasts (Table 5)
    - STR:
      - RMSE: Week1 227.38 ; Week2 284.67 ; Week4 384.07
      - MAE: Week1 152.92 ; Week2 204.24 ; Week4 284.29
      - ME: Week1 4.44 ; Week2 16.65 ; Week4 37.34
    - OLS:
      - RMSE: Week1 227.70 ; Week2 286.35 ; Week4 388.31
      - MAE: Week1 153.31 ; Week2 205.77 ; Week4 288.19
      - ME: Week1 5.64 ; Week2 18.91 ; Week4 42.31
    - Bottom Up:
      - RMSE: Week1 228.71 ; Week2 283.05 ; Week4 376.95
      - MAE: Week1 153.44 ; Week2 202.22 ; Week4 277.23
      - ME: Week1 0.86 ; Week2 9.89 ; Week4 22.42
    - Base (Unreconciled):
      - RMSE: Week1 229.48 ; Week2 291.35 ; Week4 399.39
      - MAE: Week1 155.02 ; Week2 210.05 ; Week4 297.97
      - ME: Week1 8.03 ; Week2 23.42 ; Week4 52.26
    - MinT:
      - RMSE: Week1 245.19 ; Week2 299.03 ; Week4 390.35
      - MAE: Week1 169.17 ; Week2 215.38 ; Week4 290.62
      - ME: Week1 -5.78 ; Week2 -3.18 ; Week4 -0.12

### Liquidity table, calibration, and caveats for implementation
- Liquidity table (Table 6) use:
  - Summarize forecasts for flows of liquidity supply (autonomous factors + maturity of previous OMOs) and demand (excess reserves + reserve requirements) to compute liquidity gap and suggested OMO operations.
  - Forecast autonomous factors daily for five-day horizon (skip weekends) aligned to weekly maintenance period.
  - CBvS’ four-week averaging used for reserve requirement input for five-day horizon.
- Calibration caveats:
  - IMF framework uses pure statistical models; CBvS must align outputs with reserve money targets and institutional information and apply expert judgment in some scenarios.
  - Time series already incorporate maturing securities; avoid misleading estimates when significant securities mature by calibrating projections with additional maturing volumes and setting thresholds based on historical distributions.
  - Use precautionary liquidity buffers as thresholds to monitor banks’ demand-side behavior.

### Money market operations, CBCs, and auction mechanics — institutional context and observed impacts
- Instruments and cadence:
  - Two main instruments: term deposits (TDs) and Central Bank Certificates (CBCs).
  - Weekly multiple-price TD auctions with tenors one week, one month, three months; quarterly CBC auctions with tenors six and nine months.
- CBCs:
  - Introduced June 2022, unconventional and temporary; available to wholesale and retail segments (non-bank financial institutions, legal entities, households).
  - Concerns: eligibility of non-reserve holders undermines CBvS’ ability to meet reserve money target; retail participation large (household applications increased >700 percent to 1298 in November 2023).
  - Participation by volume in CBCs: Households 1.6 billion SRD (38.5 percent of SRD 4.3 bn total); Companies 25.5 percent; Banks 14 percent.
- Auction participation and market fragmentation:
  - Low participation, high bid rates disproportionate to quantities; smaller banks often bid higher rates for low volumes; larger banks bid larger amounts at lower rates.
  - Interest from OMOs contributed approximately 70 percent in net income of banks in 2023.
  - Participation trends: in 2021 four most active banks participated in 77 percent of auctions; in 2024 four most active banks engaged in 46 percent of auctions; more than half of participations typically by same banks representing ≈55 percent of total banking sector assets.
- Market transmission and rates:
  - Deposit rates rose from 7.1 percent in August 2022 to 11.6 percent in January 2024.
  - Lending rates rose from 14.8 percent to 15.1 percent over same period, largely driven by increase in reserve requirements from 39 percent to 44 percent in April 2023.
- Suggested CBC policy adjustments:
  - Reassess eligibility criteria, tenors, and auction frequency.
  - Redirect eligibility to entities subject to reserve requirements and supervised by CBvS.
  - Reduce CBC tenors to maximum 180 days; limit TD tenors to three (e.g., 1 week, 30 days, 90 days); ideally index to inflation.
  - Phase out CBC issuances carefully to avoid liquidity disruptions.

### Banks’ operational suggestions and CBvS modernization
- Banks’ suggestions to develop interbank market and lower OMO rates:
  - Improve infrastructure and conditions; establish automated borrowing and placement facility on single platform; allow electronic approvals.
  - Introduce instruments: repos, Treasury Bills, other fixed-income instruments; reduce SRD interest rates; develop modern stock exchange.
  - CBvS and Ministry of Finance to actively buy/sell instruments to bolster local market.
- CBvS modernization steps:
  - Developing fully automated platform for standing facilities; digitalization to reduce auction errors and processing delays.
  - Testing an electronic FX trading platform with banks and cambios; gradual access to exporters and cambios.
- FX market challenges and constraints:
  - February 2023 NOP regulation: limit 10 percent of Tier 1 capital for each foreign currency; excess must be sold to CBvS.
  - CBvS constrained under Extended Fund Facility to purchase FX from banks and cambios exclusively via auctions.
  - Cash FX transportation limited; CBvS cautious in increasing frequency due to liquidity and exchange rate pressures.

### Supervision requirements and liquidity management
- Banks' concerns:
  - Supervision requirements (HQLA buffers, CFPs) perceived as tying up funds that could be used for lending or investments; CFPs resource-intensive.
  - TA report (December 2020) recommended phasing out five percent liquidity buffer.
- CBvS response and actions:
  - CBvS views regulation as not restrictive; implementing strategic adjustments acknowledging prior overestimation of realizable asset values in stress tests.
  - Establishing enhanced AML/CFT oversight; implementing regulations on credit classification and provisioning; conducted AQRs for 2022 and 2023; involved in recapitalization and NPL directives.
- Dual reserve processes complexity:
  - Two systems complicate precise assessment:
    - 44 percent reserve requirement within 14-day maintenance period.
    - Five percent non-binding liquidity buffer (SNEPs) based on different reserve base and period.
  - Recommendation: review and simplify dual systems to improve forecasting accuracy and reduce incentives for precautionary excess liquidity.

### Methodological highlights (Annex II) — models, seasonality, and volatility
- Time series models included: Naïve (random walk), Seasonal Naïve, ETS (level, trend, seasonality with smoothing parameter α), ARIMA/SARIMA (orders p,d,q and seasonal P,D,Q[m] with stepwise selection), TBATS (handles multiple seasonality with trigonometric terms, Box-Cox, ARMA errors), volatility models (GARCH, eGARCH, GJR-GARCH).
- Multiple seasonal cycles:
  - Daily series can exhibit day-in-week, day-in-month, and day-in-year cycles; trigonometric indicators (cosines and sines) used to encode seasonality parsimoniously.
  - Lasso regression recommended to filter redundant seasonal indicators for parsimony.
- Forecast reconciliation:
  - Approaches: Bottom-up, aggregate-level, hybrid; reconciliation methods include OLS, MinT, STR to ensure forecasts add up and exploit error correlations.
- Volatility modeling:
  - GARCH family specifications and eGARCH/GJR-GARCH asymmetry formulations highlighted for high-volatility series (e.g., Net Foreign Assets).

*Italic: Source — IMF Technical Assistance Report — Preface, Executive Summary, Recommendations, and relevant annexes as provided in the content unit.*

### Preface ................................................................................................................

### Preface

### Mission and context
- A Monetary and Capital Markets (MCM) Department mission visited Paramaribo from April 2–5, 2024, at the request of the Centrale Bank van Suriname (CBvS) to assist the authorities in liquidity forecasting.
- The mission met with the Governor, the Deputy Governor, Directors, staff members, and the commercial banks.

### Key operational change motivating the mission
- Suriname shifted from an exchange rate to a monetary targeting regime and initiated open market operations (OMOs) in July 2021.
- The CBvS started conducting market operations using mainly term deposits in July 2021.

### Recent macroeconomic developments cited
- The Suriname dollar devalued by 182.5 percent from September 2020 to June 2021.
- Resulting macroeconomic problems included inflation rates above 60 percent driven by high exchange rate pass-through.
- Widespread financial dollarization emerged as households and economic entities hedged against currency depreciation.

### Central bank responses prior to and during mission
- To absorb excess liquidity the CBvS added Central Bank Certificates (CBCs) in June 2022.
- In April 2023, the CBvS increased the reserve requirement ratio from 39 to 44 percent and established a 20 percent annual credit growth cap for banks.
- Despite these measures, in March 2024 the CBvS underperformed on the reserve money target by SRD 537.7 million.

### Operational challenges identified
- Persistently high excess reserves and repeated target shortfalls.
- Low engagement by banks in auctions and submission of disproportionately high rates relative to bid quantities.
- Absence of a well-functioning money market, challenges in the FX market, supervision requirements, and dual reserve-holding systems complicating accurate measurement of excess and precautionary reserves.

---

### Executive Summary

### IMF technical assistance provided
- The mission deployed an IMF MCMCO statistical forecasting framework to refine CBvS’ liquidity forecasting.
- The IMF framework comprises 12 forecasting models across three categories: exponential smoothing (simple, with exogenous regressors, and seasonal), Auto Regressive Integrated Moving Average (ARIMA, simple, with exogenous regressors, and seasonal), TBATS (Trigonometric Seasonality, Box-Cox Transformation, ARMA Errors, Trend and Seasonal), and volatility models.
- The framework includes an out-of-sample performance testing system grounded in four criteria that evaluate accuracy, bias, and confidence intervals, enabling model selection, forecast generation, and forecast reconciliation.
- The mission provided simulations for staff training and code packages for forecasting autonomous factors and reserve demand.

### Liquidity forecasting and operational calibration
- A liquidity forecasting table was developed during the mission to track predictions on system available reserves and demand for excess reserves, facilitating calibration of weekly OMOs.
- Liquidity supply is captured by autonomous factors and the maturity of previous OMOs; demand for reserves is determined by excess reserves and reserve requirements.
- The liquidity gap (liquidity supply minus demand for reserves) guides allotment decisions for the volume of OMOs.

### Data and implementation recommendations
- The CBvS should collect daily data and complete implementation of the full granular version of the model as soon as possible.
- Consider integrating the IMF liquidity forecasting framework into current forecasting operations.

### Bank engagement and market development
- The mission conducted a survey of commercial banks and a joint workshop with CBvS and banks to discuss challenges and suggestions.
- Banks expressed concerns about:
  - The policy on CBCs and their market impact.
  - High reserve requirements.
  - High OMO interest rates influenced by banks' own high bid rates.
- The CBvS is developing a fully automated platform for standing facilities and should pursue digitalization (including electronic signatures for auction documents) and regular surveys to capture banks’ expectations.

### Strategic reviews advised
- The CBvS should review its CBC strategy, including eligible criteria, tenors, auction frequency, and communication strategy, to assess repercussions of non-bank participation, risks from longer tenors and high interest rates, and impacts on investment, spending, liquidity buffers, and macroeconomic objectives.
- Simplify the two separate systems requiring reserves to be held in the system to provide an accurate understanding of actual excess and precautionary reserves and facilitate calculations for both CBvS and banks.
- Work closely with the government to modernize infrastructure to foster interbank market development and improve bank participation in auctions.

---

### Recommendations (Table 1 summary)
- Near term: < 12 months; Medium term: 12 to 24 months.

- Liquidity Forecast
  - 1. Implement forecast with statistical models to produce demand of reserve forecast. (¶12) — High priority — Short-term
  - 2. Complement the CBvS’ forecast with statistical models to produce forecast for autonomous factors. (¶21) — High priority — Short-term
  - 3. Collect daily level data for autonomous factors, specifically for liabilities to government and net foreign assets. (¶24) — High priority — Short-term
  - 4. Consider reconciliating the IMF liquidity forecasting framework into current forecast operations. (¶27) — Medium priority — Medium-term

- Participation in Money Market Operations
  - 5. The eligibility criteria to participate in OMOs should be redirected only to entities subject to the CBvS’ reserve requirements. (¶42) — High priority — Short-term
  - 6. The tenors of the CBCs for structural operations should be reduced to a maximum of 180 days and operations with term deposits should be limited to maximum three different tenors, for instance with 1 week, 30 days and 90 days, ideally indexing them to inflation to ensure its real value. (¶42) — High priority — Short-term
  - 7. Ad-hoc fine-tuning operations should be considered for large and unpredicted fluctuations in liquidity. (¶44) — High priority — Medium-term
  - 8. The CBvS should strictly follow the planned schedule auctions as previously communicated. (¶46) — High priority — Short-term
  - 9. The CBvS should conduct a survey on a regular basis (preferably monthly) to understand banks’ expectations regarding the key macro variables related to CBvS’ objective. (¶47) — Medium priority — Medium-term
  - 10. The CBvS, in collaboration with the government, should make further efforts to modernize its infrastructure to improve the development of the interbank market, fortify its understanding of the banking sector’s dynamics and improve banks’ participation in auctions. (¶52) — High priority — Short-term
  - 11. The existence of the two separate systems requiring holding reserves in the system should be simplified. (¶60) — High priority — Short-term

---

### I. Introduction — Key points
- Suriname initiated open market operations in July 2021 after shifting to a monetary targeting regime; M0 is set as the operating target and term deposits (TD) and central bank deposit certificates (CDs) are used as monetary policy instruments. (¶1)
- Accurate liquidity forecasting is essential for the effectiveness of the monetary targeting regime; the mission provided advanced statistical models to enhance CBvS forecasts. (¶2)
- The CBvS has faced persistent excess reserves and underperformance in achieving monetary targets since March 2024 amid low bank participation in OMOs; high inflation, exchange rate fluctuations, and expansionary fiscal policies increased base money and credit, especially after June 2022. (¶3)

*Source: IMF Technical Assistance Report — Preface, Executive Summary, Recommendations, and Introduction.*

### 4.    The mission implemented the MCMCO-developed statistical forecasting framework to

### The mission implemented the MCMCO-developed statistical forecasting framework to enhance the CBvS' liquidity forecasting

### Implementation overview
- The IMF framework consists of 12 forecasting models and includes a system for testing out-of-sample performance based on accuracy, bias, and confidence intervals.
- Functions provided:
  - Supports model selection, forecast creation, and reconciliation to improve accuracy.
  - Supplies code packages for forecasting autonomous factors and reserve demand.
  - Includes simulations for staff training.
- Complementary activities:
  - Survey among commercial banks to understand challenges in managing liquidity and low participation in market operations.
  - Workshop organized in collaboration with the CBvS.
  - Report quantitatively substantiates survey findings using available data.

### Estimating Demand for Reserves — Commercial Bank Reserves Balance
- Key empirical patterns:
  - Upward trend in commercial bank reserves balance starting from July 2021 with weekly and quarterly seasonality; a drop observed from January to February 2023.
  - A spike at end-January 2024 attributed to a 45.45 percent increase in the required reserve at the beginning of February 2024 for two days.
  - Weekly seasonal pattern captured via day-of-the-week binary indicator variables (four indicators). One transitory structural break included from January 31 until February 3, 2024. Ten trigonometric indicators used for longer-term seasonality.
- Model findings:
  - ARIMA with regression emerges as the best model to predict the commercial bank balance across time horizons.
  - The framework encompasses time series models: Naïve, Exponential Smoothing (ETS), ARIMA, and TBATS; volatility models use the GARCH family.
  - Performance metrics used: Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Error (ME), and Mean Interval Score (MIS) for 95 percent prediction intervals.
  - ARIMA with regressors: best out-of-sample forecast performance (accuracy and improvement over Naïve). Seasonal ARIMA shows better forecast bias performance (ME closer to zero).
  - Forecast presentation: ARIMA with regression yields narrower 80 percent and 95 percent prediction intervals than the equally weighted average of all models.

### Estimating Demand for Reserves — Excess Reserves
- Definitions and patterns:
  - Excess reserves = bank reserve balance on account minus reserve requirement in the CBvS liquidity monitoring framework; recorded as daily working balance.
  - CBvS adds a five percent SNPES buffer for effective excess reserves.
  - Excess reserves: large volatility over past three years but an overall upward trend; weekly and quarterly seasonality observed.
- Model selection and performance:
  - Same selection process as for bank reserves: both time series and variance models tested; time series models outperform variance models.
  - ARIMA with regression selected as best model with the lowest RMSE and MAE across forecast horizons.
  - In terms of forecast bias: ETS with regression most unbiased for t+1 and t+2 weeks; ETS best for t+4 weeks.

### Precautionary Liquidity Buffers
- Purpose and methodology:
  - Liquidity buffers are part of excess reserves held for precautionary purposes; methodology introduced in prior MCM TA mission.
  - Estimation based on historical weekly average distributions of excess reserves aggregated and by bank.
  - Recommended periodic review of buffers and establishment of excess reserve thresholds at the 5th and 95th percentiles.
- System-wide thresholds (July 2021–March 2023):
  - 95th percentile for entire banking system's excess reserves: 2953 million SRD.
  - 5th percentile for entire banking system's excess reserves: 969 million SRD.
  - Effective excess reserves thresholds (subtracting five percent for SNEPS):
    - 95th percentile: 2124 million SRD.
    - 5th percentile: 408 million SRD.
- Weekly average excess reserve by bank (5th and 95th percentiles) — Table 2:
  - Bank 1: 5th Percentile 306 ; 95th Percentile 1278
  - Bank 2: 5th Percentile 54 ; 95th Percentile 386
  - Bank 3: 5th Percentile 130 ; 95th Percentile 932
  - Bank 4: 5th Percentile 28 ; 95th Percentile 186
  - Bank 5: 5th Percentile 43 ; 95th Percentile 205
  - Bank 6: 5th Percentile 67 ; 95th Percentile 293
  - Bank 7: 5th Percentile 12 ; 95th Percentile 64
  - Bank 8: 5th Percentile 7 ; 95th Percentile 65
  - Bank 9: 5th Percentile 4 ; 95th Percentile 69
- Weekly average effective excess reserve by bank — Table 3:
  - Bank 1: 5th Percentile 117 ; 95th Percentile 1001
  - Bank 2: 5th Percentile -74 ; 95th Percentile 218
  - Bank 3: 5th Percentile 43 ; 95th Percentile 844
  - Bank 4: 5th Percentile -44 ; 95th Percentile 97
  - Bank 5: 5th Percentile 25 ; 95th Percentile 182
  - Bank 6: 5th Percentile -6 ; 95th Percentile 221
  - Bank 7: 5th Percentile 0 ; 95th Percentile 52
  - Bank 8: 5th Percentile -20 ; 95th Percentile 39
  - Bank 9: 5th Percentile -3 ; 95th Percentile 51

### Recommendation on reserves demand forecasting
- The mission recommends incorporating statistical models for demand-for-reserve forecasts to extract information from historical time series.
- CBvS should combine statistical forecasts with qualitative evaluations (bank surveys, demand curves, precautionary buffers) to better understand liquidity from the demand side.

### Forecasting Autonomous Factors and Net Liquidity — Overview
- Framework objectives:
  - Introduce statistical models and reconciliation techniques to forecast autonomous factors and net liquidity.
  - Forecasts include currency in circulation (CiC) and central bank liabilities to the government.
  - Use statistical reconciliation to align sum of autonomous-factor forecasts with aggregated net liquidity forecast.
  - Provide forecast uncertainty via confidence intervals.
- Benchmarking with CBvS practice:
  - CBvS currently uses ARIMA for daily CiC and models monthly CiC as a function of lagged inflation and seasonal factors; daily forecasts for government transactions updated weekly.
  - CBvS forecasts can incorporate qualitative/internal information on large government transactions; combining statistical forecasts with policy judgment is recommended.
  - Note: liabilities to government in the IMF framework computed by tracking net changes from an initial level; differences between level and net-change data may introduce additional uncertainty.

### Currency in Circulation (CiC)
- Patterns:
  - Total CiC shows an upward trend. Cash outside the bank and cash in vault show weekly, quarterly, and yearly seasonality; total CiC shows weekly and quarterly seasonality.
  - CiC tends to increase at end of each year and decrease during the first quarter of the next year.
- Model selection and reconciliation:
  - ARIMA with Regression selected for CiC for one-week, two-week, and four-week forecasts.
  - Seven models tested including Naïve and Naïve seasonal. ARIMA with Regression ranks top for accuracy.
  - Regressors include day-of-the-week binary indicators (four indicators) and four trigonometric indicators for longer-term seasonality.
  - Mean error (ME): ARIMA and seasonal ARIMA closest to zero for t+1 and t+2; ARIMA with regression lowest for t+4.
  - Model reconciliation improves CiC forecast quality: reconciled forecasts (STR) show lower RMSE at one-, two-, and four-week horizons.
  - ARIMA with regression reconciled by STR appears best-performing for CiC in terms of RMSE at all horizons.
- Model selections summarized (Table 4):
  - Currency Outside Banks: ARIMA with Regression
  - Cash in Vault: ETS with Regression
  - Currency in Circulation: ARIMA with Regression

### Liabilities to Government
- Patterns:
  - Monthly and quarterly seasonality observed; liabilities lower in Q2 and Q3 relative to Q1 and Q4.
  - Weekday distribution relatively stable.
- Model findings:
  - TBATS ranked best for forecast accuracy for liabilities to government (allows multiple seasonality with trigonometric terms, exponential smoothing, ARMA errors, Box-Cox transformation).
  - Seasonal ARIMA shows closer distance to unbiased position for forecast bias.
- Data recommendation:
  - Collect daily stock-level data of liabilities to government in future operations (current model uses net changes with initial level at start-2019).

### Net Liquidity
- Model selection and reconciliation:
  - TBATS selected as best model for net liquidity at t+1 week (lowest RMSE and MAE). ARIMA with Regression performs better at t+2 and t+4 weeks.
  - Reconciliation methods tested: bottom up, MinT, STR, and OLS.
  - For one-week horizon: STR recommended (highest forecast accuracy for one week).
  - For two- and four-week horizons: Bottom Up method provides lowest RMSE and MAE.
  - In terms of bias (ME), Bottom Up has the lowest ME across all horizons.
- Predictive accuracy and bias for reconciled forecasts (Table 5):
  - STR:
    - RMSE: Week1 227.38 ; Week2 284.67 ; Week4 384.07
    - MAE: Week1 152.92 ; Week2 204.24 ; Week4 284.29
    - ME: Week1 4.44 ; Week2 16.65 ; Week4 37.34
  - OLS:
    - RMSE: Week1 227.70 ; Week2 286.35 ; Week4 388.31
    - MAE: Week1 153.31 ; Week2 205.77 ; Week4 288.19
    - ME: Week1 5.64 ; Week2 18.91 ; Week4 42.31
  - Bottom Up:
    - RMSE: Week1 228.71 ; Week2 283.05 ; Week4 376.95
    - MAE: Week1 153.44 ; Week2 202.22 ; Week4 277.23
    - ME: Week1 0.86 ; Week2 9.89 ; Week4 22.42
  - Base (Unreconciled):
    - RMSE: Week1 229.48 ; Week2 291.35 ; Week4 399.39
    - MAE: Week1 155.02 ; Week2 210.05 ; Week4 297.97
    - ME: Week1 8.03 ; Week2 23.42 ; Week4 52.26
  - MinT:
    - RMSE: Week1 245.19 ; Week2 299.03 ; Week4 390.35
    - MAE: Week1 169.17 ; Week2 215.38 ; Week4 290.62
    - ME: Week1 -5.78 ; Week2 -3.18 ; Week4 -0.12

*IMF staff calculation, Centrale Bank van Suriname.*

### 22.  The mission recommends the CBvS to incorporate the IMF liquidity forecasting framework

### 22.  The mission recommends the CBvS to incorporate the IMF liquidity forecasting framework

### IMF liquidity forecasting framework: operational recommendations
- Incorporate the IMF liquidity forecasting framework into the CBvS current liquidity forecasting framework.
- Customize the framework to align with the weekly maintenance period starting every Wednesday with five-day horizons until the next maintenance day on Wednesday.
- Update the forecast generated from the model every week.
- Periodically evaluate model selection (i.e., run model validations every month or every quarter, etc.) and change model selection if necessary to incorporate updated inputs from the latest data feed.
- Collect daily level data for autonomous factors to monitor trend and seasonal variations of time series more accurately.
  - CBvS currently uses net government transactions as input; forecast would benefit from transmitting transaction data to daily stock data of liabilities to the government.
  - Consider incorporating net foreign asset into the current forecasting framework.

### Liquidity table and calibration
- Use a liquidity forecasting table (Table 6) to summarize information and record forecasts on:
  - flows of liquidity supply;
  - flows of liquidity demand;
  - conclusion of forecasted liquidity gap leading to suggested operations.
- Table features and inputs:
  - First three items: autonomous factors and maturity of previous open market operations provide liquidity supply flows.
  - Forecasts of autonomous factors are generated from the IMF framework on a daily basis for a horizon of five days ahead, skipping weekends and aligning with the current weekly liquidity maintenance period.
  - Forecast horizon can be extended to a longer horizon if needed.
  - Upcoming maturity amount of previous OMO is based on CBvS records.
  - From the fourth item onward: projections of demand for excess reserves.
  - The table uses CBvS’ forecast averaging for four weeks as input, given reserve requirements remain the same for a five-day ahead horizon and forecast error of reserve requirement in the longer term is relatively small.
  - Combining reserve requirements and excess reserves generates total forecasted liquidity demand.
- From supply and demand information, CBvS can calculate the liquidity gap: difference between available reserves and reserve demand from banks.
- Training and practice:
  - IMF provided training to technical staff; CBvS can maintain and update the IMF liquidity framework.
  - Mission conducted a simulation with CBvS staff to practice using the liquidity forecasting table and updating it with the most recent data.

### Calibration and implementation caveats
- Reconcile/ integrate the IMF liquidity forecasting tool into current forecast operations as a next step.
- IMF framework uses pure statistical models to estimate the liquidity gap to guide issuance amount of term deposits.
- CBvS needs guidance for full implementation, including:
  - aligning with reserve money targets;
  - consolidating information from institutional arrangements and sources;
  - navigating scenarios requiring expert judgment.
- Time series used for projections already incorporate maturing securities, accounting for additional liquidity.
- To avoid misleading estimations when significant volumes of securities mature within the projection period:
  - Calibrate projection results by considering additional volume from maturing securities in auctions.
  - Adjust by setting a specific threshold based on historical distributions of maturing securities.
  - Use estimations of precautionary liquidity buffers (from the previous section) as a threshold to monitor banks’ behavior from the demand side when considering liquidity forecasting.

### Money market operations: institutional context and instruments
- Transition and operations:
  - Following transition from an exchange rate regime to a monetary targeting regime, Suriname commenced open market operations in July 2021.
  - Between September 2020 and June 2021, the Suriname dollar experienced a 182.5 percent devaluation during the transition to a freely floating exchange rate system.
  - Resulting macroeconomic challenges included inflation rates that exceeded 60 percent in the following months and extensive financial dollarization.
  - In June 2021, Suriname started its first market operations aiming at achieving its M0 target.
- CBvS instruments and auction cadence:
  - Two main instruments: term deposits (TDs) and Central Bank Certificates (CBCs).
  - Weekly multiple-price auctions for term deposits with tenures of one week, one month, and three months.
    - Quantity is announced; counterparties bid desired volume and rates.
    - Cut-off rate determined by the highest acceptable bid until total quantity is reached.
  - Quarterly auctions for CBCs with tenures of six and nine months.
- CBCs:
  - Introduced in June 2022; viewed as unconventional, temporary, not part of standard monetary policy toolkit.
  - Used to absorb structural excess liquidity and stimulate development of money and capital markets, including interest rate transmission.
  - Available to wholesale and retail segments: non-bank financial institutions, legal entities, households.

### Challenges: excess reserves, auction participation, and market effects
- Causes and policy responses:
  - Persistent high excess reserves driven by persistent high inflation, exchange rate volatility, and expansionary fiscal policies leading to increases in base money and surge in credit (in particular after June 2022).
  - Average bid rates increased substantially until the end of the year while CBvS was not successful in mopping up excess liquidity.
  - CBvS introduced CBCs in June 2022 and increased reserve requirement ratio from 39 to 44 percent in April 2023.
  - CBvS imposed a 20 percent limit on annual credit growth for the banking sector in April 2023.
  - As of March 2024, CBvS realized an underperformance of SRD 537.7 million.
- Survey of commercial banks:
  - Survey responses from seven out of nine banks, covering approximately 92 percent of total banking sector assets.
  - Questionnaire contained 27 questions about liquidity management and market environment.
  - Banking sector:
    - Covers approximately 76.5 percent of the financial assets of the country, which is equivalent of 91 percent of GDP.
    - Nine active banks; four classified as systemic.
    - Ownership: one bank under foreign ownership; six majority private domestic banks; two fully state-owned banks.
  - Main concerns cited by banks:
    - High OMO interest rates;
    - CBCs policy and its market impact;
    - High reserve requirements;
    - Absence of a well-functioning money market;
    - Challenges in FX market;
    - Supervision requirements.
- Auction participation and bidding behavior:
  - Low participation and high rates, often disproportional to quantities bid; high discrepancy across rates for TDs and CBCs.
  - CBvS observation: smaller banks often submit highest bids for low volumes; larger banks demand larger amounts at relatively lower rates (particularly for TDs).
  - Example: reserve demand curve from December 7, 2022, one-week tenor — largest bid came from largest bank while marginal amounts demanded by smaller banks at higher rates.
  - Interest from OMOs contributed approximately 70 percent in net income of the banks in 2023.
  - Scatter-plot analysis suggests a weak or complex relationship between bid rates and Return on Assets (ROA); significant uncertainty remains.
  - Participation frequency trends:
    - Rise in participation frequency since July 2021 auctions.
    - More than half of participations typically taken by same banks, representing approximately 55 percent of total banking sector assets.
    - In 2021, four most active banks participated in 77 percent of auctions; three banks did not participate at all.
    - In 2024, four most active banks engaged in 46 percent of auctions; all banks took part in some auctions.
    - Heatmaps suggest banks with lower asset quality and relatively worse liquidity positions were more frequent participants in 2022 and 2023.
    - CBvS could conduct further analysis with more granular, higher-frequency data to observe indicator evolution and auction participation given seasonal behavior of some indicators.
- Market fragmentation and transmission:
  - In a fragmented banking sector with limited monetary policy transmission, some banks view elevated OMO rates as obstacle to liquidity management.
  - Deposit rates increased on average from 7.1 percent in August 2022 to 11.6 percent in January 2024.
  - Lending rates increased marginally from 14.8 percent to 15.1 percent in the same period, largely driven by increase in reserve requirements from 39 percent to 44 percent in April 2023.
  - Fragmentation implications:
    - Some banks face heightened need for funds and tighter margins if unable to compete for deposits.
    - Banks with more abundant funds can maintain margins and are less impacted by OMO rates.
- CBCs and retail participation effects:
  - With non-bank financial institutions and households eligible for CBCs, banks concerned about potential fund withdrawals to invest in CBCs, intensifying fund scarcity and stability risks.
  - Applications from households increased more than 700 percent in one year, reaching 1298 in November 2023.
  - Household participation by volume in CBCs: 1.6 billion, representing 38.5 percent of total volume (SRD 4.3 bn).
  - Companies: 25.5 percent of total volume.
  - Banks: 14 percent of total volume.

*Italic: Source — IMF staff technical assistance mission text as provided in the content unit.*

### 41.  Banks’ suggestions to lower the OMO rates as well as the volumes and frequency of CBCs

### 41.  Banks’ suggestions to lower the OMO rates as well as the volumes and frequency of CBCs

### OMO rates and auction mechanics
- In the current RTM framework, auction rates are set by banks; the central bank focuses on narrow money as its operational target.
- The cut-off rate equals the highest acceptable bid until the total quantity of the auction (OMO volume) is reached.
- Since March 2024, the CBvS has realized an underperformance with respect to the target; as a result the cut-off bid has equaled the highest bid submitted, in some cases as high as 70 percent.
- Workshop clarification: the high level of OMO rates is not a direct policy decision by the central bank; limitations of monetary policy transmission were highlighted.

### Strategy and design of CBCs (certificates)
Findings and concerns:
- CBvS acknowledged need to review the policy regarding issuance of CBCs and will review without disclosing further details.
- Allowing participants that do not have reserves in the CBvS to participate in OMOs could impair CBvS’ ability to meet its reserve money target and adversely affect banks with liquidity shortages.
Recommendations:
- Reassess eligibility criteria, tenors of the certificates, and range of available tenors.
- Redirect eligibility criteria to only entities that are subject to reserve requirements and supervised by CBvS.
- Reduce tenors of structural operations using CBCs to maximum 180 days.
- Limit operations using term deposits to maximum three different tenors, for instance: one week, 30 days and 90 days.
- Ideally index the securities to inflation to ensure real value.
- Limit maturity options to short, medium, and longer terms to set benchmarks for trading and enhance transmission.

### Potential macroeconomic and liquidity impacts of CBCs
- High-return CBCs for extended periods can:
  - Impair investments and spending by drawing capital away from private and public investment.
  - If redeemed when inflation aligns with target, large redemption volumes could disrupt price stability efforts.
  - Cause banks with long-term CBCs to hold larger precautionary buffers, exacerbating excess liquidity.
- Abruptly halting CBC issuance could risk promoting excessive liquidity growth if CBvS cannot absorb liquidity with TD instruments.
- Recommendation: phase out issuances carefully and evaluate pace of phase-out.

### Fine-tuning operations for liquidity fluctuations
- When main need is to address large and unpredicted movements in liquidity, CBvS should consider implementing ad hoc fine-tuning operations that clearly articulate this objective.
- Such operations would improve CBvS’s capacity to meet the monthly target regarding average excess reserves and respond to unexpected situations.

### Communication, credibility, and market expectations
Findings:
- A bank suggested communication on CBCs should be limited to a short period and not for the entire year to avoid driving up interest rates.
- In 2024, CBvS decided to have two issuances in the first quarter despite earlier communication of one issuance per quarter.
Recommendations:
- Maintain actions aligned with communication strategy to preserve credibility.
- Design a well-defined strategy and be transparent in communication; inconsistent forward guidance undermines credibility and increases market uncertainty.
- CBvS requested TA focused on communications to take place still in 2024.
- CBvS should survey banks’ expectations on key variables (inflation, GDP growth, base money, reserve money) initially monthly, with possible transition to quarterly.

### Absence and underdevelopment of the money market / interbank market
Key statistics and recent activity:
- Interbank lending: approximately 0.2 percent of total balance in 2019, declining to 0.1 percent in 2023.
- Standstill from November 2022 to June 2023 of no transactions; two banks resumed activities in July 2024, trading a monthly amount of 250 million, with interest rates initially at 60 percent, later decreasing to 30 percent by January 2024.
- Composition of interbank claims:
  - FX claims: 94.7 percent of total claims.
  - SRD: 1.8 percent.
  - Shares: 3.5 percent.
- Within FX claims:
  - Long-term claims (> one year): 76 percent.
  - Short-term claims (up to one year): two percent.
  - Current account deposits: 12 percent.
  - Short-term deposits: eight percent.
  - Time deposits with maturity > one year: two percent.

Banks’ diagnosis of underdevelopment:
- Causes cited: funding scarcity, market imbalances, limited data, significant financial risks, scarcity of SRD funding, absence of active secondary market for overnight and short-term placements, reliance on OMO rates as reference making interbank lending prohibitively expensive, insufficient public data, country and counterparty risks, illiquid stock market, highly volatile interest rates.

Banks’ suggestions to develop interbank market:
- Improve infrastructure and general conditions.
- Establish an automated borrowing and placement facility on a single platform to eliminate paper approvals (CBvS currently legally prohibited from accepting electronic signatures).
- Introduce instruments such as repurchase agreements (repos), Treasury Bills (T-Bills), and other fixed-income instruments.
- Reduce SRD interest rates, develop economic conditions to boost confidence, introduce a modern stock exchange market.
- CBvS and Ministry of Finance to actively buy and sell financial instruments to bolster local market.

CBvS modernization steps:
- Developing a fully automated platform for standing facilities.
- Standardized digital processes would reduce auction errors (incorrect order numbers, missing stamps, late submissions).
- Important modernization element: substitute paper approvals with efficient and secure digital systems to boost market confidence and transaction speed.

### FX market challenges and CBvS role
Regulatory change:
- February 2023: Supervision Department updated regulations on banks' net open positions (NOP), setting a limit of 10 percent of Tier 1 capital for each foreign currency; banks exceeding limit required to sell excess to CBvS.

Banks’ concerns:
- Non-compliance of open currency position, high influx of FX, increasing difficulty to sell FX and unrealized losses.
- Following high FX inflows, banks must purchase all FX exporters request to sell and find it difficult to sell this FX to other clients.
- Banks suggested CBvS should buy all FX above open currency limits from commercial banks.
- Mixed assessments of FX operations with CBvS: neutral, complicated, or very complicated.
- Suggested an automated FX platform to improve efficiency.

Program constraint:
- Under the Extended Fund Facility, CBvS is restricted to purchasing FX from banks and exchange offices (cambios) exclusively through auctions.

Cash FX transportation:
- After losing correspondent relationships, some banks hindered in transporting cash FX and urge CBvS for more frequent FX shipments.
- FX market sources cash euros mainly from visitors from French Guiana and the Netherlands via cambios; predominant household/business demand is cash and non-cash U.S. dollars.
- CBvS performs cash transportations on a quarterly basis under strict compliance; CBvS cautious about increasing frequency.

CBvS actions:
- Exercises caution in acquiring FX to avoid exacerbating surplus SRD liquidity and exerting upward pressure on exchange rate.
- Noted decrease in FX demand due to reduced purchasing power from high inflation and decreased demand from foreign-owned fuel companies.
- In process of completing development of an electronic FX trading platform; has tested platform with banks and cambios and will gradually allow other participants (exporters and cambios).

### “Tight” monetary stance, reserve requirements, and liquidity buffers
Findings and clarifications:
- Some banks characterize monetary policy as “tight,” pointing to high reserve requirements and strict/punitive averaging mechanism.
- Fund and CBvS clarified RTM mechanism: reserve requirements aim to curb inflation and stabilize the economy.
- Averaging system provides flexibility; commercial banks can use up to 20 percent of their required reserves as a buffer to absorb short-term liquidity shocks.
- Concepts of excess reserves and precautionary buffers were discussed; precautionary buffers appear important in Suriname’s banking system.

Complexity from multiple reserve processes:
- Two different requirements complicate assessment of total excess reserves:
  - 44 percent reserve requirement within a 14-day maintenance period.
  - Five percent non-binding liquidity buffer calculated based on a different reserve base and treated as an indicative precautionary balance.
- Periods used for the two calculations do not coincide; CBvS faced challenges aligning the calculations initially.

Implications and recommendation:
- Existence of two separate processes may heighten challenges in calculating precise liquidity buffers, potentially exacerbating banks’ tendency to hold excess liquidity.
- Data (Annex III referenced) show banks generally meet 44 percent requirement with considerable margin, but after discounting the five percent SNEPs banks often face scarcity in work accounts—an incentive to hold precautionary buffers.
- Recommendation: review and simplify the two separate systems to improve liquidity forecasting accuracy and reduce incentives for precautionary excess liquidity.

*IMF Technical Assistance Report — content unit: 41. Banks’ suggestions to lower the OMO rates as well as the volumes and frequency of CBCs*

### 61.  Some banks also view the supervision requirements as a key obstacle to manage liquidity.

### tarea2025050-print-pdf - 61.  Some banks also view the supervision requirements as a key obstacle to manage liquidity.

### Banks' concerns on supervision requirements and liquidity management
- Paragraph 61: Some banks view supervision requirements as a key obstacle to manage liquidity.
- Requirements and perceived effects:
  - Banks are required to maintain a buffer of freely available (unencumbered) high-quality liquid assets to protect against a range of stress scenarios, including those involving loss or impairment of unsecured and normally available funding sources.
  - Some banks believe these requirements tie up funds that could otherwise be used for profit-generating activities, like loans or investments.
  - Banks should ensure there are no legal, statutory, or operational restrictions on the use of these resources to meet the funding requirement.
  - Banks must have a formal Contingency Funding Plan (CFP) describing strategies for dealing with liquidity shortages in various emergency situations, including rules for managing different stress scenarios, a clear definition of responsibilities and procedures for implementing the plan.
  - Dome banks claim that CFPs can be resource-intensive and may divert attention from other business priorities.
- Related TA recommendation (footnote): The TA report “Monetary Policy Framework and Foreign Exchange Operation,” from December 2020, recommended that the five percent liquidity buffer should be phased out.

### Centrale Bank van Suriname (CBvS) stance and actions
- Paragraph 62: CBvS position and ongoing measures:
  - CBvS reiterated its view that the regulation is not restrictive and it is not an obstacle for efficient liquidity management by the banks.
  - CBvS is implementing strategic adjustments to its liquidity rules, acknowledging that the realizable value of assets has been overestimated in previous stress tests.
  - CBvS is establishing an enhanced oversight mechanism for Anti-Money Laundering and Counter-Terrorism Financing (AML/CFT).
  - CBvS is implementing regulations on credit classification and provisioning.
  - CBvS has conducted Asset Quality Reviews (AQRs) for the years 2022 and 2023.
  - CBvS has been involved in recapitalization efforts and issued directives on Non-Performing Loans (NPLs).
  - CBvS indicated it is formulating regulations to ensure fair competition in the financial sector (no further details provided).

### Conclusion and operational recommendations for CBvS
- Paragraphs 63–66: Summary of transition and recommended next steps:
  - CBvS shifted from an exchange rate to a monetary targeting regime and initiated open market operations (OMOs) in July 2021, with the transition aiming to tackle economic challenges but introducing devaluation of the Suriname dollar and subsequent financial dollarization.
  - The mission provided advanced statistical models for liquidity forecasting, including a forecasting framework comprising 12 models across three categories to enhance the CBvS’s ability to anticipate future liquidity needs and adjust OMOs.
  - A liquidity forecasting table was developed to facilitate calibration of weekly OMOs, and efforts were made to ensure staff dedication to learning the new methodology.
  - CBvS conducted a survey and workshop with commercial banks to address concerns regarding CBCs, reserve requirements, and OMO interest rates; highlighted need to review CBs strategy concerning CBCs and the dual systems of requiring reserves.
  - Recommended next steps:
    - Continue to refine liquidity management strategies and consider simplifying systems to improve forecasting accuracy.
    - Modernize infrastructure and develop a fully automated platform for standing facilities to foster development of the interbank market and enhance understanding of banking sector dynamics.
    - These steps aim to improve operational efficiencies, tailor strategies more effectively, and contribute to more informed policymaking.

### Liquidity forecasting framework — models and methodological highlights (Annex II)
- Overview:
  - The IMF’s Liquidity Forecasting Framework includes 12 models across three categories; time series models typically forecast currency in circulation and government accounts balance and focus on modelling trend and seasonality.
  - Four types of time series models included: the Naïve, Exponential Smoothing, ARIMA and TBATS.
- Naïve model (random walk):
  - Forecast equation: ŷt+ h = yt.
  - Seasonal Naïve: ŷt+ h = yt− s + h (where s is the seasonal period, corresponding to the number of days in the week (5 without the weekend)).
  - Useful as a simple benchmark; requires no parameter estimation.
- Exponential Smoothing (ETS):
  - Models time series as components: level, trend, and seasonality; framed in state-space form: ŷi = f(μ̂i, εi).
  - Example fully additive form:
    - ŷi = μ̂i + εi
    - μ̂i = leveli + seasonali + seasonal_noni
  - Level update (additive level example): leveli = leveli−1 + α⋅errori−1, where α is a smoothing parameter between 0 and 1.
  - Interpretation of α:
    - Example α = 0.2: last estimated level is updated by 20 percent of the last observation.
    - Low α → long-weighted moving average (resilient to noise); high α → reactive to recent observations.
    - Examples discussed: α = 0.1 and α = 0.5 illustrating different reactivity to noise.
  - Numerical optimization typically used to choose smoothing parameters by minimizing a loss function (e.g., Mean Squared Error, MSE).
  - Model selection can use information criteria such as Akaike Information Criterion (AIC), with simplified AIC expression noted: AIC = 2√MSE + 2k (where k is number of model parameters).
  - Regressors can be included by augmenting μt in the same fashion as conventional regression modelling.
- ARIMA / SARIMA:
  - General ARIMA form: (1 − φ(B))(1 − B)d yt = (1 + θ(B))εt.
  - Nomenclature ARIMA(p,d,q); seasonal extension SARIMA(p,d,q)(P,D,Q)[m].
  - Orders and estimation:
    - Stepwise algorithm (Hyndman and Khandakar 2008) used to select orders:
      i. Find d using the KPSS test.
      ii. Estimate four initial models and choose the best.
      iii. Expand candidate model set by considering models with s or q differing from the current best by 1.
      iv. Iterate until no improvement is made.
    - Selection criterion: Akaike Information Criterion corrected for small sample size (AICc).
  - SARIMA example: ARIMA(1,0,0)(0,1,1)[5] equivalent equation provided for seasonal patterns.
  - ARIMA models can incorporate covariates analogous to regression modifications.
- Seasonality and indicators:
  - Seasonality can be modelled using indicator variables (e.g., day-of-week dummies for Sunday, Monday, Tuesday, Wednesday, Thursday).
  - Structural breaks can be encoded using a continuous indicator D t = 1 if t occurs after the structural break, 0 otherwise.
- Notes on data and figures:
  - Annex I figures: distributions of weekly average excess reserve by bank and weekly average effective excess reserve by bank.
  - Figure source: IMF staff calculation, Centrale Bank van Suriname.
  - Figure notes: Observations starting from 07-06-2021 to 03-27-2024; alternate note lists observations starting from 2021-07-06 to 2024-03-27 and references CBvS definition of effective excess reserve based on the CBvS framework.

*IMF Technical Assistance Report | content as provided*

### 12.  Daily time series can exhibit multiple seasonal cycles that must be accounted for in the

### 12.  Daily time series can exhibit multiple seasonal cycles that must be accounted for in the modeling.

### Multiple seasonal cycles and challenges
- Daily series can exhibit multiple seasonal cycles: day in the week, day in the month, and day in the year, corresponding to different cyclicities in the data.
- Many models typically incorporate a single seasonal periodicity; multiple seasonalities substantially complicate forecasting.
- Three modelling elements for multiple seasonalities:
  - the length of the seasonal cycles,
  - their encoding,
  - the efficiency (aiming for parsimonious models).
- Counting days in each periodicity resolves the first element: example given, there are five days in the week (without weekends).
- Day-in-month seasonality is challenging because months have different numbers of days; quarterly seasonality is used as a workaround because a quarter contains a fixed number of weeks, and by extension days.

### Encoding multiple seasonal cycles (trigonometric indicator variables)
- Trigonometric encoding: given season length s periods, construct s/2 pairs of trigonometric variables, with i = 1,..., s/2:
  - d_i = a_s s s? [original text presents trigonometric forms as]
  - d_i = cos(2 i π t / s),
  - d_{i+s/2} = sin(2 i π t / s),
  - where t = 1,..., n (with n being the sample size).
- When s is odd, s/2 is rounded up to the closest integer.
- This encoding is mathematically equivalent to using s binary indicator variables (each encoding the level of a particular day in the season); in some cases one indicator corresponds to a constant, resulting in s−1 informative indicator variables.
- Advantage of trigonometric representation: can encode complex seasonal effects, such as leap years, which binary indicator encoding cannot capture.
- For binary encoding, the number of trigonometric pairs is calculated as the floor of the true s. Example: for day in the year for a five-day week year (260 days) we will have 260/2 cosines and 260/2 sines.

### Parsimony: filtering redundant seasonal indicators
- To obtain parsimony, redundant seasonal indicators are filtered.
- Procedure:
  - Remove the time series trend using a centered moving average (Ord, Fildes, and Kourentzes 2017).
  - Subtract the centered moving average (which models the trend) from the data to obtain residuals.
  - Model residuals with different trigonometric indicator variables as explanatory variables using a regression model.
  - Use lasso regression to eliminate less informative inputs and obtain a sparse representation.
- Purpose of lasso regression: find a compromise between model fit and complexity measured by number of parameters, to avoid overfitting.

### Trigonometric versus binary encoding under sparsity
- Using all indicators, binary and trigonometric variables (for integer s) provide the same output.
- When terms are eliminated:
  - Binary encoding omits all seasonal information for that period.
  - Trigonometric encoding provides a smoother approximation of the seasonal profile, offering an efficient sparse approximation of seasonal patterns.
- The figure in the source exemplifies the above (figure not reproduced here).

### TBATS model: handling complex seasonality and calendars
- TBATS incorporates multiple discussed features:
  - Seasonality and trend handled via exponential smoothing (using trigonometric terms for seasonality).
  - Box-Cox transformation applied.
  - ARIMA innovations incorporated.
- TBATS allows seasonality to change over time and can handle multiple calendars.

### Volatility models for high-volatility series
- Volatility models are appropriate for forecasting series with high volatility; normally applied to forecast Net Foreign Assets.
- Three classes of models are fitted; the most popular family is GARCH.
- GARCH variance specification:
  - σ_t^2 = ω + Σ_{i=1}^p β_i σ_{t−i}^2 + Σ_{i=1}^p α_i l_{t−i}^2
- eGARCH specification:
  - log σ_t^2 = ω + Σ_{i=1}^p β_i log σ_{t−i}^2 + Σ_{i=1}^q α_i g(ε_{t−i}),
  - where g(ε_t) = θ ε_t + λ (|ε_t| − M(|ε_t|)).
  - Advantage: asymmetry—sign and magnitude of innovations have different effects on variance.
- GJR-GARCH specification:
  - σ_t^2 = ω + δ σ_{t−1}^2 + α ε_{t−1}^2 + φ ε_{t−1}^2 A_{t−1},
  - where A_{t−1} = 0 if ε_{t−1} ≥ 0 and A_{t−1} = 1 if ε_{t−1} < 0.
  - Like eGARCH, allows asymmetric effects.

### Forecast reconciliation for autonomous factors and net liquidity (NL)
- Forecasts (including prediction intervals) for all autonomous factors and NL can be reconciled in a combined NL forecast.
- Net liquidity injection (net foreign assets, currency in circulation, and state account balance) is the main quantity of interest.
- Approaches:
  - Bottom-up: simply add up forecasts of autonomous factors.
  - Aggregate-level model: develop a forecasting model for the total of autonomous factors.
  - Hybrid: produce forecasts for each autonomous factor and the total to hedge against model misspecification at either level; downside—forecasts may not add up correctly.
- Let the vector of four point forecasts that do not add up be ŷ.
- Summing matrix M given by:
  - M = [1 1 1; 1 0 0; 0 1 0; 0 0 1] (matrix as presented).
- OLS reconciliation (guarantees add-up):
  - ỹ = M (M′ M)^{−1} M′ ŷ.
- MinT method (exploits correlation between forecast errors) (guarantees add-up):
  - ỹ = M (M′ Σ^{−1} M)^{−1} M′ Σ^{−1} ŷ,
  - where Σ is the covariance matrix of one-step-ahead forecasting errors.

### Annex III — Banks Behavioral Profile in the Maintenance Period
- Charts represent banks’ behavior with respect to liquidity management during the maintenance period.
- Left charts: how banks manage liquidity on average to meet reserve requirements, currently established at 44 percent.
  - Blue line: average of the reserves kept in the central bank for each maintenance day since June 2021 as a percentage of the reserve requirements.
  - Blue range: minimum and maximum (percentage of reserve requirements) maintained in each maintenance period.
- Right charts: liquidity situation of banks after the 5 percent SNEPs are discounted from their work account.
  - Yellow line: effective excess reserves as a percentage of the work account.
- Banks profiled in the annex: Bank 1, Bank 2, Bank 3, Bank 4, Bank 5, Bank 6, Bank 7, Bank 8, Bank 9 (charts not reproduced here).

*Source: IMF staff computations.*

---


_Source: https://www.imf.org/-/media/files/publications/tar/2025/english/tarea2025050-print-pdf.pdf_