## 1uaeea2022003 ## Source details **Canonical URL:** [1uaeea2022003](https://www.imf.org/-/media/files/publications/cr/2022/english/1uaeea2022003.pdf) ## Other formats - [Markdown version](/-/media/files/publications/cr/2022/english/1uaeea2022003.pdf.md) - [Structured JSON version](/-/media/files/publications/cr/2022/english/1uaeea2022003.pdf.json) --- ### Executive summary — New Dirham Monetary Framework, liquidity stance, and key metrics - Framework design: - Floor interest rate corridor system implemented to steer the O/N interbank rate towards the policy rate anchored on the US Federal Reserve’s IORB since July 29, 2021. - OMOs used to drain structural liquidity surplus and support money market development. - Current liquidity assessment: - IMF estimation: operate a structural liquidity surplus of around AED 50 bn. - Interbank market turnover has shrunk over the last two years by around two thirds. - Weighted average yield on M-Bills across all tenors is 0.16 percent and duration of 47 days. - Islamic CDs duration is 18 days and weighted average YTM is 0.13 percent. - Key operational challenge: - Strengthen capacity to identify and absorb volatility that does not represent a fundamental change in the exchange rate equilibrium through: - Forecasting autonomous factors (AF). - Assessing precautionary demand for banks reserves. - Better calibration of monetary policy instruments. ### Operational design and instruments (four pillars) - Four pillars: standing facilities, OMOs, liquidity insurance, and reserve requirements. - Standing facilities and parameters: - Corridor width: initially 100 bps, reduced to 50 bps on January 11, 2021. - Overnight Deposit Facility (ODF): introduced July 12, 2020; ODF rate is the base rate. - Overnight Lending Facility (OLF)/Collateralized Murabaha: ceiling facility; consolidated with REPO into single OLF; eligible collateral: M-Bills/Islamic CDs, T-Bills, USD, UAE and foreign securities rated minimum AA. - Intraday Liquidity Facility (ILF): collateralized intraday liquidity at zero cost; collateral: M-Bills/Islamic CDs or T-Bills. - M-Bills and OMOs: - M-Bills maturity: <= 12 months; Frequency: Fortnightly; Settlement: T+2. - Regular OMO Repo: maturity <= 7 days; Frequency: Weekly; Settlement: T+0. - Fine-tuning OMO instruments (FX swaps): maturity <= 14 days; Frequency: Weekly; Settlement: T+0. ### FX swap facility and collateral treatment - Current issues: - One-way FX swap facility priced lower than OLF; tenors pricing requires CBUAE to set term differentials. - Three banks capture 80 percent of facility’s volume. - Implied rate in FX swap facility around 22 bps on average during 2021, lower than OLF. - Recommendations: - Discontinue FX Swap Facility once OLF collateral is broadened to include US dollars. - Interim: announce market at least three months before discontinuation and after OLF accepts dollars as collateral. - Consider paying interest on USD collateral taking into account overnight reserves returns, operational cost, and ensure rate lower than alternatives to avoid use as USD investment vehicle. ### Structural OMOs, M-Bills issuance and primary dealer (PD) arrangements - Improvements proposed: - Announce volumes of M-Bills ahead according to AF projections and allocate fully, retaining minimal discretion to reject out-of-market bids via internal rule. - Evaluate pre-registering and obtaining ISIN for M-Bills one quarter ahead to reduce the eight-day window between order-book and auction. - Avoid non-competitive, bilateral operations where possible; consider relaxing the 45 percent cap of holdings by a single PD per issuance line conditionally. - Align M-Bills calendar with the reserve maintenance period (RMP) by allowing settlement on initial day of the RMP. - Consider increasing issuance of three-month M-Bills and allocate higher amounts in long-term M-Bills and re-tranche to promote secondary market transactions. - Primary dealer program: - Maintain PD arrangement; tighten underwriting obligation and conduct performance assessment after inaugural year using secondary-primary quote differences and bid-ask spreads. ### Fine-tuning OMOs — timing, modalities, and auction design - Purpose and timing: - Manage unexpected liquidity fluctuations during the RMP; correct forecast deviations and overbidding/underbidding behaviors. - Conduct at the end and/or on any day of the RMP; settle all fine-tuning OMOs for value T+0. - Instruments and bidding: - Use FX swaps and AED Repo; if monetary coverage rule not constraining, conduct both auctions simultaneously and in same direction. - DTIs bid in forward points for FX Swaps; bid in interest rates for Repos. - Spot rate for FX Swaps should equal CBUAE spot facility rate considering the long leg. - For Repo transactions, all rates allotted should be in the corridor. - Auction design: - Prefer multiple price auctions in practice for robustness; single price auctions have theoretical benefits but higher collusion risk. - Maintain transparency: ex-ante and ex-post publication of auction rules, access criteria, refusal procedures, size, timing, composition, and detailed post-results. ### Reserve requirements (RR) — level, averaging, remuneration, and currency matching - Current settings and changes: - RR ratios: demand deposits 7 percent; time deposits 1 percent. - Foreign currency deposits account for around 30 percent of total deposits. - Penalty for shortfall: CBUAE base rate plus 400 bp. - Reserve maintenance period extended from 7 to 14 days (effective October 28, 2020); full reserve averaging allowed. - Fulfillment currency changed to dirham for both local and foreign currency required reserves. - Policy recommendations: - Optimal RR should be in range of AED 55 billion corresponding approximately to the current 7 percent ratio. - Calibrate required reserves to act as buffer against autonomous-factor forecasting errors; use averaging to reduce interbank rate volatility. - Consider gradually reinstating fulfillment of RR on FX deposits in USD to avoid distortions and reinforce prudential objective. - Consider remunerating required reserves after careful assessment of balance sheet implications; if remunerating at base rate not feasible consider lower remuneration. - In medium term, extend RMP up to 28 days as appropriate. ### Liquidity forecasting framework — design, staffing, data, and software - Forecasting framework features: - Forecast each autonomous factor separately: Currency in Circulation (CiC), State Account Balance (SAB), Net Foreign Assets (NFA). - Tests different families: ARIMA, ETS, TBATS; for NFA use GARCH/eGARCH for volatility. - Integrated model selection based on information criteria and out-of-sample performance; model dynamic selection and model averaging. - Probabilistic forecasts with automatic confidence interval estimation to calibrate operations. - Reconciliation approaches: bottom-up, forecast Total directly, or forecast all and reconcile; OLS reconciliation and MinT recommended. - Operational recommendations: - Implement IMF-designed R-based software and publish whole liquidity forecast and associated methodology. - Staff a forecasting unit within Monetary Management Department; at minimum two people with statistics, forecasting and programming background; expand to include econometricians and modelers. - Upgrade data management: daily granular CBUAE balance sheet, payment system data, and MoU with Federal MoF and local Treasuries to share Treasury future cash flows; implement TSA at CBUAE. - Use probabilistic forecasts and calibrate statutory reserves on conditional forecasting errors at a chosen confidence level (e.g., 95 percent). ### Forecasting empirical findings and model guidance - Empirical model performance: - ARIMA with regression leads to best forecasting performance for CiC, SAB and the Total. - For NFA, eGARCH recommended empirically; GARCH family yield similar results. - Model averaging improves point forecasts but not interval forecasts. - OLS reconciliation is robust and outperforms bottom-up, especially for intervals. - Forecast horizons and metrics: - Forecast horizons: one-day ahead up to 14-day ahead during RMP; up to 10 forecasts per two-week period. - Point forecast metrics: RMSE and MAE; prediction interval metric: Winkler score. - Interval forecast calibration: - Hit rates for 80 percent prediction intervals show under-coverage at short horizons; numbers near 0.2 indicate better calibration. - TBATS tends to under-cover; ARIMA with regression performs well for interval forecasts at longer horizons. ### Optimal liquidity surplus, estimation, and sterilization mix - Conceptual objective: - Optimal liquidity surplus E_t^* defined as minimal excess reserves for which P(|i_t − ī|)=0 given policy priority to anchor interbank rate. - Empirical estimation: - Unconditional parametric and non-parametric fits estimate optimal surplus between AED 50–60 bn. - Conditional cross-validated IV-2SLS instrumental approach (using level and volatility of AF as instruments) yields optimal liquidity surplus around AED 50 bn. - R2 of conditional CV IV-2SLS fits between 65 and 78 percent; first-stage F-statistic near 10. - IMF chooses a tolerance zone length of 25 bps (roughly lowest 20th percentile of empirical DONIA spread) to define anchored interbank rate. - Sterilization and calibration: - Sterilization instruments: required reserves with averaging, long-term securities B_t^{LT} (up to one year), short-term securities B_t^{ST}. - Optimal securities issuance: B_t^* = I(AN_t) − E_t^* − R^* where R^* calibrated to conditional forecasting errors. - Split structural vs short-tenor issuance via time-series filtering (HP filter for trend with high smoothing). - Recommendation: gradually adjust toward AED 50 bn surplus with close market monitoring; ensure minimum benchmark size for marketable securities. ### Money markets development, benchmarks, and market infrastructure - Benchmark transition: - Transition from EIBOR to DONIA for overnight market recommended; CBUAE to publish DONIA daily. - Integrate bank borrowing from non-bank wholesale counterparties into DONIA; CBUAE can collect t−1 data for calculation at t. - Short-run term rates: backward-looking compounded O/N rates; long-run: develop forward-looking term rates as derivatives markets evolve. - Market assessments: - FX Swap market: most developed; 2019 BIS survey daily turnover 30,769 million dollars; 84 percent volume maturity < 14 days. - Repo market: infancy stage; half of banks lack MRA. - M-Bills secondary market: started to develop but limited liquidity; average secondary market amount 4.7 million; overnight volume stands at 10 billion AED average monthly. - Market development actions: - Set market conventions for FX SWAP pricing and track implicit AED rate from FX SWAP points; consider FX global code of conduct. - Support repo market development: enforce MRAs, consider central counterparty, align issuance across tenors to supply collateral. - Publish more market information and enhance transparency (auction results, interbank repo rates, FX swap rates, volumes, collateral used, active participants). ### Monetary policy transmission and data needs - Data collection: - Collect average monthly rates on various categories of deposits and loans bank-by-bank and compile weighted averages for the system. - Use long time series to estimate interest rate pass-through from Base Rate to DONIA and to short-term bank rates. - Pass-through estimation: - Use markup pricing models and error-correction/ARDL representations to estimate short-run and long-run pass-through coefficients (iR = α + β · iM; β < 1 typically). ### Key operational and policy recommendations (selection and sequencing) - Immediate (NT) actions: - Allocate M-Bills fully in line with calibration exercise. - Evaluate feasibility of pre-registration and ISIN one quarter ahead. - Discontinue bilaterally negotiated operations/non-competitive auctions; consider relaxing 45 percent PD holding cap and tighten underwriting obligations. - Consider increasing issuance of three-month M-Bills. - Align M-Bills calendar with RMP to allow settlement on initial day. - Discontinue FX Swap facility and accept US dollars as collateral for OLF; consider remuneration of USD collateral. - Conduct fine-tuning OMOs (FX swaps and AED Repo) at end and/or on any day of RMP; DTIs bid in forward points for FX Swaps and in interest rates for Repos. - Settle fine-tuning OMOs for value T+0. - Implement IMF forecasting framework; staff modelers and publish forecasts and methodology. - Operate a liquidity surplus around AED 50 bn and adjust gradually. - Medium-term (MT) actions: - Reinstate RR fulfillment on FX deposits in USD gradually. - Extend reserve maintenance period up to 28 days. - Consider remuneration of required reserves after balance-sheet assessment. - Develop risk-free AED yield curve jointly with Federal MoF. - Follow-up: - Conduct a follow-up IMF TA mission to assess and project monetary policy costs and implications for CBUAE income statement. - Strengthen market surveillance, data collection, and quantitative research capacity. ### Technical appendix highlights — statistical models and software - Forecast models and choices: - ARIMA (including SARIMA and ARIMA with regression), ETS, TBATS; for NFA volatility: GARCH, eGARCH, GJR-GARCH. - Model selection via AICc; auto.arima implementation referenced. - Structural breaks and holiday effects modeled via dummies and smooth DaysToHoliday constructs. - Forecast reconciliation: OLS and MinT; OLS found most robust for intervals. - Software and implementation: - Forecasting framework programmed in R; RStudio recommended; R version 3.6.3 recommended. - Four R scripts: step_000_dataclean.R, step_001_features.R, step_002_model.R, step_003_summary.R. - Use of Python and R interoperable tools recommended; version control (git/Github) advised. - Hardware: high-performance server infrastructure recommended; cloud computing options noted. - Documentation and support: - IMF supplied code, config file and documentation; IMF experts available for extra support and training. *Source: IMF | United Arab Emirates Liquidity Management and Forecasting (content unit 1uaeea2022003).* ### Preface................................................................................................................. ### Preface ### Mission purpose and scope - At the request of the Central Bank of the United Arab Emirates (CBUAE), a virtual Monetary and Capital Markets (MCM) Department mission was conducted from June 28 to July 26, 2021, to assist the authorities in strengthening the liquidity management and forecasting frameworks. ### Mission participants and contacts - The mission met with: - Wassim Slama (Director of Monetary Policy) - Khalifa Alfaheem (Director of Monetary Operations) - Michael Creed (Assistant Director—Monetary Management Framework) - Other relevant staff from the Monetary Management Department - The mission also held meetings with several foreign commercial banks and a state-owned bank. ### Acknowledgements - The mission wishes to thank the authorities, in particular Wassim Slama, for their cooperation and productive discussions. *Source: Preface, IMF | United Arab Emirates Liquidity Management and Forecasting* ### EXECUTIVE SUMMARY ### EXECUTIVE SUMMARY ### New Dirham Monetary Framework — overview - The new Dirham Monetary Framework is largely consistent with an operational framework under a fixed exchange rate regime and capital mobility. - A floor interest rate corridor system was implemented to steer the O/N interbank rate towards the policy rate set at a level consistent with the currency peg. The Base Rate has been anchored on the US Federal Reserve’s Interest on Reserves Balance (IORB) since July 29, 2021. - Open market operations (OMOs) are used to drain the structural liquidity surplus and support money market development. - Key operational challenge: strengthen capacity to identify and absorb volatility that does not represent a fundamental change in the exchange rate equilibrium. This requires: - Anticipation of changes in the supply of banks reserves through forecasting of the autonomous factors (AF). - Assessment of the precautionary demand for banks reserves. - Better calibration of monetary policy instruments to offset interest-rate impacts from non-fundamental liquidity shocks. ### Foreign exchange (FX) swap facility and collateral - Recommendation: discontinue the FX swap facility once eligible collateral for the OLF is broadened to include US dollars (USD). - Rationale and issues: - The one-way swap facility is currently priced at a lower level than the OLF. - Current pricing of tenors requires the CBUAE to set term interest rate differentials rather than letting the market price them through price-discovery auction systems. - Consider paying interest on US dollars collateral, taking into account returns from overnight reserves management, operational cost, and alternative rates to prevent usage as an investment vehicle. ### Structural OMOs and M-Bills - Improvements suggested for structural OMOs and M-Bills issuance: - Primary dealer (PD) arrangements for M-Bills issuances aid secondary market development but require tightening of the underwriting obligation. - Volumes of M-Bills should be calibrated according to projections of the autonomous factors and allocated fully. - Feasibility of pre-registering and getting ISIN for M-Bills one quarter ahead should be evaluated to reduce the eight-day window between collection of the order book and auction. - Non-competitive, bilateral operations should be avoided to the extent possible; this may require relaxing the current 45 percent cap of holdings by a single primary dealer (PD) per issuance line. - Align the calendar of M-Bills with the reserve maintenance period (RMP) by allowing settlement of M-Bills on the initial day of the RMP to enhance banks’ liquidity management. - Consider increasing issuance of the three-month M-Bills and allocate higher amounts in long-term M-Bills, and re-tranche them to promote secondary market transactions. ### Fine-tuning OMOs - The CBUAE should implement fine-tuning OMOs to manage unexpected liquidity fluctuations during the RMP. Key features: - Fine-tuning OMOs can correct deviations in liquidity forecasts and overbidding/underbidding behaviors not allocable in the interbank market. - Fine-tuning OMOs should be carried out at the end and/or on any day of the RMP as deemed necessary. - The CBUAE may use FX Swaps and Repurchase agreements (Repo) on AED. If the monetary coverage rule is not constraining, both auctions should be conducted simultaneously in the same direction through a fixed-volume variable-rate allotment system. - For FX Swap auctions, Deposit Takers Institutions (DTI) bid in terms of forward points for the term of the swap; for Repo auctions, bids are in terms of interest rates. - The spot rate for FX Swap instruments should be the same as the rate offered in the CBUAE spot facility considering the long leg of the swap. - For Repo transactions, all rates allotted should be in the corridor. - CBUAE should keep minimal discretion to reject bids that do not represent market conditions. - Settle all fine-tuning OMOs instruments for value T+0. ### Reserve requirements (RR) - Reserve requirements are primarily a monetary policy and liquidity management tool; their nonremuneration acts as an indirect tax on financial intermediation. - Reserve averaging reduces interbank rate volatility and frequency of OMOs. - Given the current largest confidence interval of the autonomous factors’ forecasts, optimal reserves requirements should be in the range of AED 55 billion, which corresponds approximately to the current seven percent ratio. - Extension of the reserve maintenance period to two weeks simplifies banks’ liquidity management and could be extended in the medium term up to 28 days. - It would be prudent to reinstate gradually the fulfillment of the RR on FX deposits in USD to avoid distortions on financial intermediation and reinforce the prudential objective. - The CBUAE should consider remunerating required reserves after careful assessment of balance sheet implications. ### Liquidity forecasting framework - The CBUAE should implement the designed state-of-the-art, full-fledged forecasting framework for each of the three main autonomous factors and a reconciliation strategy to forecast their sum with an estimation of the confidence interval. - Features of the IMF-designed framework: - Tests different families of forecasting models with integrated selection of the best performer based on information criteria and out-of-sample performance. - Allows automatic reparameterization of models to account for structural breaks and new data. - Uses a probabilistic model to forecast net liquidity injection stemming from autonomous factors, including net foreign assets. - Implements model dynamic selection and model averaging to improve accuracy and reduce modeling risk. - Automatically estimates probabilistic confidence intervals of projections at a given tolerance level to calibrate operations. - Operational recommendations: - Forecast all autonomous factors separately and generate probabilistic confidence intervals for net liquidity injection. - Staff an analytical section of modelers and econometricians to estimate and fine-tune forecasting models and analyze market developments. - Implement robust analytical and data management frameworks with comprehensive daily data analyzed through statistical software. - Enhance forecasts by incorporating expert judgement and granular CBUAE knowledge. - Publish the whole liquidity forecast and associated methodology to help anchor market expectations. - Operationalize a MoU with the Federal MoF and other Departments of Finance (DoFs) to share Treasuries’ future cash-flow information. - Federal and local Treasuries should operate a Treasury Single Account; remuneration and fees terms of Treasury accounts at the CBUAE should be set. ### Liquidity level, sterilization, and calibration - Current situation: the CBUAE has been operating an excessive liquidity surplus, which harms interbank market activity and the establishment of a sovereign yield curve. - Interbank market turnover has shrunk over the last two years by around two thirds. - Excessively liquid banks deposit excess liquidity at the CBUAE rather than trading among each other. - Insufficient liquidity absorption limits CBUAE’s issuance of securities and weakens sovereign yield curve estimation. - IMF estimation: CBUAE should operate a structural liquidity surplus of around AED 50 bn. - This estimate derives from a comprehensive econometric model with conditional analysis based on non-linear, instrumented and cross-validated estimator; robustness analyses confirm the result. - The CBUAE should reach this level by progressive sterilization and monitor market reactions. - CBUAE should calibrate the operational mix between short-term securities issuance, structural operations, and optimal statutory reserves. - The IMF team designed an integrated approach articulating monetary objectives, market development and liquidity forecasting to determine optimal combinations of sterilization instruments. - The approach offers guidance to policymakers and provides robustness and consistency over time. ### Money markets development and benchmarks - Development of money markets is key to enhance monetary policy transmission. - Transition recommended from the Emirates Interbank Offered Rate (EIBOR) to the Dirham Overnight Index Average (DONIA) for the overnight market. - DONIA calculation should integrate bank borrowing from a wide range of non-bank wholesale counterparties. - As derivative markets linked to the new benchmark develop, forward-looking term rates could be constructed. - M-Bills secondary market has started to develop but remains limited in liquidity and depth. - PDs program can be adapted to encourage greater secondary market activity; consider allocating higher amounts in long-term M-Bills and re-tranching to increase outstanding amounts. - FX Swap market is most developed; would benefit from common agreement on market conventions for pricing FX SWAP and tracking implicit AED rate from FX SWAP points, and adoption of the FX market global code of conduct. - Onshore repurchase agreements (Repo) market is at an infancy stage. - The development of a risk-free yield curve in AED jointly between the CBUAE and the Ministry of Finance should be pursued as a building block for secondary markets. - More disclosures are needed to enhance price discovery in secondary markets. ### Monetary policy transmission and data needs - In the medium term, the CBUAE should estimate interest rate pass-through to assess monetary transmission channels. - Collect average monthly rates applied on various categories of deposits and loans on a bank-by-bank basis. - Compile weighted average monthly rates for each category across the banking system. - Assess pass-through from the Base Rate to DONIA and then to short-term bank rates. ### Key operational and policy recommendations (selection) - Allocate the M-Bills fully in line with the calibration exercise. (CBUAE — NT) - Evaluate feasibility of pre-registration and ISIN for M-Bills auctioned one quarter ahead. (CBUAE — NT) - Discontinue bilaterally negotiated operations or non-competitive auctions; consider relaxing the 45 percent PD holding cap and tighten underwriting obligations. (CBUAE — NT) - Consider increasing issuance of three-month M-Bills. (CBUAE — NT) - Align M-Bills calendar with the RMP to allow settlement on initial day. (CBUAE — NT) - Discontinue FX Swap facility and accept US dollars as collateral for OLF; consider remuneration of US collateral. (CBUAE — NT) - Conduct fine-tuning OMOs (FX swaps and AED Repo) at end and/or on any day of RMP; auctions to be transparent and consistent; DTIs bid in forward points for FX Swaps and in interest rates for Repos. (CBUAE — NT) - Settle fine-tuning OMOs for value T+0. (CBUAE — NT) - Calibrate required reserves to act as a buffer against autonomous factors forecasting errors — unchanged ratio of 7 percent (AED 55 billion). (CBUAE — NT) - Reinstate RR fulfillment on FX deposits in USD gradually. (CBUAE — MT) - Extend reserve maintenance period up to 28 days. (CBUAE — MT) - Remunerate required reserves after balance sheet assessment. (CBUAE — MT) - Implement IMF forecasting framework for autonomous factors and net liquidity injection; staff modelers/econometricians and publish forecasts and methodology. (CBUAE — NT) - Operate a liquidity surplus around AED 50 bn, adjusting gradually with close market monitoring. (CBUAE — NT) - Transition from EIBOR to DONIA for overnight market; adopt FX market global code of conduct; develop risk-free AED yield curve jointly with Federal MoF. (CBUAE — ST/MT) *Source: EXECUTIVE SUMMARY, IMF technical assistance report — United Arab Emirates: Liquidity Management and Forecasting.* ### 6. The main objective of the new framework is to steer the overnight (O/N) interbank ### 6. The main objective of the new framework is to steer the overnight (O/N) interbank rate towards the policy rate (base rate) set at a level consistent with the currency peg. ### Objectives and operational design - Primary objective: - Steer the overnight (O/N) interbank rate towards the policy rate (base rate) set at a level consistent with the currency peg. - Operational approach: - Implemented a floor interest rate corridor system with excess reserves placed at the Overnight Deposit Facility (ODF), remunerated at the base rase. - Choice of the floor system driven by its capacity to provide a stable anchor to interbank rates without requiring accurate liquidity forecasting of the autonomous factors, given the high volatility of the Net Foreign Assets. - Since July 12, 2020, the CBUAE has set the level of the base rate in line with the US Federal Reserve’s Interest on Excess Reserves (IOER) rate. This alignment is based on the Uncovered Interest Rate Parity (UIP). - Currently, the Base Rate is higher that the Fed Fund rates, implying the existence of a small risk premium to balance FX flows in support of the peg. - Secondary objectives: - Promote market development and increase incentives for banks to manage liquidity more actively. - Complement the floor system with a regular program of central bank M-Bill issuances to support the primary and secondary money market development. - Operational framework requires active allocation of reserves across instruments and estimation of precautionary unremunerated reserves. - Subsidiary objective: - Financial stability: design of a Contingent Liquidity Insurance Facility (CLIF) to provide liquidity for potentially a long maturity during stress (idiosyncratic or market wide). - CLIF collateralized with a broader range of collateral including foreign sovereign, corporate and banks bonds. - Liquidity provided by uniform priced auctions or bilaterally to all DTI depending on stress level. - CLIF scheduled to be activated by end of 2021. - COVID-19 response: - Targeted Economic Support Scheme (TESS) offering payment holidays to bank customers. - Allocated AED 257 billion volume set in line with the monetary base coverage rule. - TESS take-up about 37 percent; deferrals extended to December 31, 2021, and working capital loan deferrals to June 30, 2022. ### Features of the operational framework (four pillars) - Four pillars: standing facilities, OMOs, liquidity insurance, and reserve requirements. - Banks’ usage: - Most banks indicated use of CBUAE instruments as part of liquidity management and regular participation in OMOs. ### Standing facilities (corridor and facilities) - Corridor: - Width initially set at 100 bps and reduced to 50 bps on January 11, 2021 (date of the launching of M-Bills). - Overnight Deposit Facility (ODF): - Introduced on July 12, 2020; sets the floor rate for overnight money market rates and signals monetary policy stance. - Rate applied to the ODF is the main policy rate (base rate), initially anchored on the IOER and aligned to the IORB since July 29, 2021. - Facility well accepted and most banks used daily. - Overnight Lending Facility (OLF)/Collateralized Murabaha Facility: - Sets the ceiling for short-term money market rates. - OLF and REPO consolidated into a single overnight lending facility. - Given structural excess liquidity, few banks use the OLF. - Eligible collateral: Monetary Bills (M-Bills)/Islamic CDs issued by CBUAE or T-Bills issued by UAE Federal or Local governments, USD currency, and UAE and Foreign Securities rated with a minimum AA rating. - Intraday Liquidity Facility (ILF): - Provides collateralized intraday liquidity at zero cost to manage payment disruptions or gridlocks. - ILF collateral: M-Bills/Islamic CDs issued by CBUAE or T-Bills issued by federal or local governments. ### Open Market Operations (OMOs) and M-Bills program - Structural OMOs: - Drain structural surplus liquidity and smooth short-term imbalances that may influence interest rates. - M-Bills program: - New tradable securities issuance program replacing Certificates of Deposits. - Issuance calendar announced with dates and tenors allocated through auctions for primary dealers (PDs). - CBUAE continues Islamic CDs program until Islamic M-Bill launch. - Weighted average yield on M-Bills across all tenors is 0.16 percent and a duration of 47 days. - Islamic CDs duration is only 18 days and the weighted average YTM is 0.13 percent. - CBUAE performs market intelligence with dealers eight days before auction to estimate demand and currently allots a percentage below 100 percent to promote competition. - CBUAE may conduct buy-backs or repurchase operations for liquidity provision. - Market signals: - Flattening of the M-Bill curve indicates excess liquidity beyond optimal surplus; implies banks are less risk averse in search of yield. - Decomposition of the M-Bill primary yield curve indicates term, liquidity, and other premiums; flattening implies reduced risk aversion. - Fine-tuning OMOs: - Intended to manage unexpected liquidity fluctuations and reduce volatility of domestic money market interest rates. - Not yet conducted due to lack of accurate liquidity forecasts. - When operational, will use (i) repurchase transactions at different tenors; and (ii) matched transactions of securities and FX transactions including FX swaps. - Purpose includes correcting deviations during the RMP and adjusting for banks’ overbidding or underbidding behaviors. ### FX Swap Facility - History and purpose: - Introduced during 2008–2009 crisis to provide liquidity from seven days up to 12 months. - Used during stress periods; recently three banks capture 80 percent of facility’s volume. - Perceived useful as a cap for the FX Swap market in stress times (e.g., COVID-19 turmoil in 2020). - Implied rates: - FX swap market implied rates are used as reference to quote other instruments such as M-Bills. - Implied rates calculated assuming 3.6730 in the Spot and Libor as a USD rate. - On average during 2021, the implied rate in the FX swap facility was around 22 bps, lower than the OLF rate which could disincentivize OLF use. - Next phase: - FX swap facility to be absorbed by the OLF, allowing banks to get liquidity using US dollars as collateral (synthetic overnight FX swap, settled at T+0). ### Reserve requirements (RR) - Applicability: - Imposed on all deposit-taking licensed financial institutions. - Deposit base includes government deposits. - Ratios: - RR ratio lower for time deposits (1 percent) than for demand deposits (7 percent). - Same RR ratio applies to foreign currency deposits to prevent distortions; foreign currency deposits account for around 30 percent of total deposits. - Penalty: - Penalty for non-compliance is CBUAE’s base rate plus 400 bp on any shortfall in required reserve—well above the lending facility. - Changes in 2020: - Lower RR ratio: demand deposit RR lowered from 14 percent to 7 percent on April 5, 2020; time deposits remained 1 percent. - Introduction of the ODF: since July 12, 2020, banks can hold excess reserves in the ODF remunerated at the IOER. - Extension of the reserve maintenance period: from 7 to 14 days starting on a Wednesday and ending on a Tuesday (effective October 28, 2020). Full reserve averaging allowed. - Fulfillment currency: fulfillment of the FX reserve requirement changed to dirham for both local and foreign currency required reserves. - Remuneration and mechanics: - Required Reserves are unremunerated; only excess reserves held overnight in the ODF are remunerated at the IOER. - Banks manage reserves between Reserve Account (counts towards RR) and ODF (remunerated); transfers to ODF by 8:00 pm daily. - Bank behavior: - Banks generally adopt conservative reserve management and hold ample reserves; overall reserves far exceed requirements and remain stable over maintenance period. - No front or back-loading observed in extended RMP; many banks do not shift funds into the ODF on Saturdays, over-fulling the RR and allowing more transfers into the ODF on weekdays. - Distribution of reserves skewed: a few large banks (e.g., First Abu Dhabi Bank) hold outsized shares; smaller banks often maintain large reserve-account balances foregoing ODF remuneration. - Reserve positions vary across banks; some face tighter liquidity despite overall surplus. ### Recommendations (selected) - OLF and FX Swap Facility: - Support decision to discontinue the FX Swap Facility; one-way SWAP priced lower than OLF can create distortions. - Interim period: SWAP facility to remain as liquidity-provision tool; CBUAE should announce market at least three months before discontinuation and once OLF accepts dollars as collateral. - Broadening OLF collateral to include US dollars recommended; operational considerations: - CBAUE may pay an interest for USD collateral considering: (i) rates at which CBAUE may invest these amounts overnight minus adequate basis points to compensate CBAUE for related operational cost; (ii) total rate should be lower than other alternatives for DTIs to prevent use as a USD investment vehicle. - Structural OMOs and M-Bills auctions: - Improve M-Bills auctions by: - Announcing volume of M-Bills to be auctioned ahead according to projections of autonomous factors and allocate it fully, retaining minimal discretion to reject out-of-market bids via an internal rule. - Suggest rejection thresholds defined by last price or mid price before the auction plus or minus the bid-ask average plus two standard deviation divided by 2. - Evaluate feasibility of pre-registering and obtaining ISIN for M-Bills one quarter ahead to reduce the eight-day window between order book collection and auction. - Avoid non-competitive, bilateral operations to limit moral hazard and preserve price-setting in primary auctions. - Consider relaxation of the current 45 percent cap of holdings by a single PD per issuance line to better reflect skewed liquidity holdings; condition such relaxation on inventories or past market-making activities. - Align calendar of M-Bills/ICDs with RMP by allowing settlement of M-Bills/ICDs on the initial day of the RMP. - Continue reoffering of auctioned M-Bills to increase liquidity, particularly for longer tenors. - Consider increasing issuance of three-month tenor M-Bills as an important money market benchmark that falls within HQLA classification and Basel III prudential liquidity requirements. *IMF | United Arab Emirates Liquidity Management and Forecasting (chapter excerpt).* ### 26. The mission recommends maintaining the primary dealer arrangement subject to a ### 26. The mission recommends maintaining the primary dealer arrangement subject to a ### Primary dealer arrangement — recommendation and safeguards - Maintain the primary dealer (PD) program to develop the secondary market through the obligation of making a two-way quotation. - Tighten the underwriting obligation to address potential risks from an increased two-tier banking system; make the underwriting obligation further explicit, for example by specifying a certain percentage for secondary market operations. - Observation: Currently, PDs do not seem to add any markup on their clients’ bids in the primary market. - Performance assessment: conduct a planned assessment of the performance of each PD after the inaugural year. - Suggested assessment criteria: cover the difference between the secondary and primary quotes as well as the bid-ask spread of secondary market operations. ### Fine-Tuning Open Market Operations (OMOs) — timing and conduct - Fine-tuning OMOs recommended at the end of the RMP and/or on any day of the RMP, whenever deemed necessary. - Fine-tuning operations should take place at the opening of the market. - Announcement timing: could be done in the afternoon of the previous day to allow DTIs to estimate needs properly. - Once banks gain operational experience and improve internal liquidity forecasting, fine-tuning operations could be announced a couple of hours before execution. - Transactions should settle on the same day to have immediate traction on interbank rates. - Cautionary advice: avoid too-often fine-tuning operations in the early stages of the new liquidity forecasting and management framework, as multiple attempts to correct liquidity surplus within a specific RMP can be counter-productive. ### Instruments and modality for fine-tuning operations - The CBUAE may use FX swaps and Repurchase agreements (Repo) on AED to conduct fine-tuning operations. - Key considerations: - The monetary base coverage rule by foreign reserves should be the primary guide for choosing between FX swap and repo fine-tuning operations: repo operations expand the CBUAE balance sheet while the FX swap is recorded off balance sheet. - If the coverage rule is not constraining, conduct FX swap and repo market auctions simultaneously with tenors up to 14 days for the FX swap and repo transactions. Both auctions should be in the same direction (absorption or injection). - After deciding amount and tenors, distribute allocation according to received bids. At the time of the mission, repo market operations were very scarce; the CBUAE may encourage repo usage to develop the market. - Auction mechanism: should be transparent and consistent to maximize participation, ensure efficient allocation and promote price discovery. - Multiple price auctions generally work better in practice and are more robust to collusion and misinterpretation risks in developing or less competitive markets. - Single price auctions theoretically reduce the “winner’s curse” and increase incentive to bid competitively but may increase incentives to collude. - Transparency: continue a framework providing significant transparency for ex-ante and ex-post publication of auction rules, access criteria, procedures for refusal of bids, size, timing and composition of operations, and detailed post-results to aid price discovery in underdeveloped markets. - Bidding formats: DTIs bid in terms of forward points for FX swaps and in terms of interest rates for repo auctions. - The spot rate for FX swaps should be the same as the rate offered in the CBUAE spot facility considering the long leg of the swap. - For Repo transactions, all rates allotted should be in the corridor. - CBUAE should retain discretionary right to reject bids that do not represent market conditions. - Main terms summarized (as provided in source): - Structural OMO: M-Bill Issuances - Counterparty: Primary dealers - Collateral: N/A - Pricing: Auction determined; Discounted price - Maturity: <= 12 months - Frequency of Operation: Fortnightly - Settlement: T+2 - Regular OMO: Repo - Counterparty: DTIs - Collateral: A and C classes USD and AED - Pricing: Auction determined; Rate quotes - Maturity: <= 7 days - Frequency of Operation: Weekly - Settlement: T+0 - Fine-tuning OMO Instruments: FX swaps - Counterparty: DTIs - Collateral: USD and AED - Pricing: Auctions determined; Swap point quotes - Maturity: <= 14 days - Frequency of Operation: Weekly - Settlement: T+0 ### Reserve requirements (RR) — level, remuneration, and currency matching - General principles: - RR ratio should be no higher than necessary to meet stated objectives. - RR remain an important tool in the UAE to support the fixed exchange rate arrangement and manage liquidity. - While prudential objectives are largely achieved through macroprudential requirements, RR—especially on FX deposits—may be beneficial as a second line of defense in large FX liquidity shocks. - Level and role: - Optimal RR ratio depends on desired overall liquidity and distribution across the toolkit. - Since 2020, a high level of excess liquidity has decreased interbank market activity, suggesting further sterilization may be warranted. - A higher RR ratio combined with reserve averaging provides a flexible tool to reduce system liquidity. - Example: re-instating the pre-crisis RR ratio of 14 percent could absorb a greater share of excess liquidity, especially if operational constraints make increasing M-Bill auction volumes difficult in a small market. - Cost and remuneration: - More unremunerated RR are effectively a higher tax on financial intermediation. - A higher RR increases the burden on banks who must hold larger amounts in unremunerated reserves. - To avoid distortions, required reserves should generally be remunerated at the opportunity cost—here the base rate of the ODF. - A high level of unremunerated RRs, especially if the IOER increases, could lead to circumvention of RR via funding shifts and risk a slowdown in credit growth as banks pass on costs. - Recommendation: CBUAE should consider remunerating required reserves after careful assessment of balance sheet implications. - Central bank operations need to be sustainable; remunerating reserves adds cost. - If remunerating at the base rate is not feasible, remuneration at a lower rate could help alleviate distortions. - Matching fulfillment currency to liabilities: - Fulfilling RR only in local currency can create distortions and undermine prudential objectives. - If RR on FX liabilities is fulfilled in foreign currency, the requirement is less distortionary for financial intermediation and ensures reserves function as an additional source of HQLA. - CBUAE has limited ability to act as lender of last resort in foreign currency; moving fulfillment to local currency only has eliminated RR as a second line of defense to FX liquidity shocks. - Recommendation: gradually reinstate fulfillment of the RR on FX deposits in USD starting with implementation on a specified fraction of FX liabilities to restore prudential role of FX RR and limit distortions. - Reserve averaging and maintenance period: - Extension of the reserves maintenance period to two weeks simplifies liquidity management for banks by giving more flexibility to smooth liquidity shocks. - Reserve averaging effectively flattens the reserve demand curve and cushions interbank rate against high-frequency imbalances. - Empirical evidence on the reduction in interbank rate volatility is limited given the short sample and parallel framework changes; however, since the introduction of the ODF and the extended RMP, interbank rate volatility has fallen sharply. - In the medium-term, CBUAE could consider extending the RMP up to 28 days. - Cross-country experience suggests reserve maintenance periods of four to five weeks are most effective to reap full benefits of reserve averaging. - A too-long RMP could undermine the prudential function of RR, especially for FX RR. - Distributional considerations: - High concentration of reserves in a few banks can lead to spikes in the interbank rate if some banks hoard reserves. - High concentration may require a higher level of excess reserves to anchor the interbank rate to the policy rate. - Recommendation: monitor distribution of excess reserves across banks to inform liquidity management in addition to aggregate variables. - The distribution of reserves in the UAE is highly uneven and likely impacts interbank rate volatility as liquidity surplus falls. - Regular monitoring helps identify bottlenecks and complements aggregate analysis to inform liquidity operations. ### Liquidity forecasting framework — organization, staffing, and data - Organizational recommendation: - Assign forecasting responsibility to a section within the CBUAE Monetary Management Department to ensure close link between forecasting and management and establish clear accountability. - The forecasting section/unit acts as the “center” where demand and supply components of bank reserves are aggregated; information sources include federal Ministry of Finance, CBUAE foreign exchange operations and accounting departments. - Forecasting section responsibilities: - (i) communicating with different information sources and ensuring timely data receipt; - (ii) aggregating the analytical central bank’s balance sheet to obtain autonomous factors; - (iii) forecasting individual liquidity components; - (iv) supervising consistency of forecasted components; - (v) preparing an overall liquidity projection to be updated on a daily basis; - (vi) assessing forecasting errors. - Staffing and capacity: - CBUAE is severely understaffed and not able to forecast liquidity using modern methods and manage liquidity appropriately. - At a minimum, two people with a solid background in statistics, forecasting and programming would be necessary to produce regular liquidity forecasts for market operations. - The IMF team provided a full-fledged liquidity forecasting framework and software infrastructure but in-house econometrics and statistical expertise is crucial to fine-tune and adjust models to new market conditions and structural breaks. - Data management and inter-agency arrangements: - CBUAE needs to upgrade its data management system to ensure timely, high-frequency access to databases; liquidity forecasting requires daily CBUAE balance sheet data and detailed monetary operations stored in granular time-series databases. - The current draft Memorandum of Understanding (MoU) between CBUAE and the DoF should be extended to the Federal and other local Treasuries from different Emirates; MoUs should cover regular exchange of granular information about future Treasury cash flows (including public institutions) and be implemented within one year at latest. - The Federal Treasury should operate a Treasury Single Account (TSA) at the CBUAE to optimize cash management and enhance CBUAE liquidity management. - Currently the Federal Treasury manages liquidity largely through commercial banks, complicating CBUAE liquidity management. - A TSA would streamline and strengthen cash-flow management and improve forecasts of liquidity. - CBUAE should stand ready to remunerate outstanding government balances at the ODF rate; advances for non-cash management purposes should be granted at the OLF rate. - Applying a TSA to other local treasuries is also advisable. ### Forecasting framework design and tools - The IMF team designed a state-of-the-art forecasting framework for the three main autonomous factors: currency in circulation (CiC), State Account Balance (SAB), and net foreign assets (NFA), with a reconciliation strategy and confidence interval estimation. - Methodology: - Framework tests different families of forecasting models with integrated selection of best performer based on information criteria and out-of-sample performance. - Allows automatic reparametrization to account for structural breaks and new data developments. - Software and implementation: - Forecasting framework programmed entirely under R. - Infrastructure is user-friendly, automatized, generates charts and tables on the fly, code is fully documented. - IMF experts can provide extra support and training to authorities to ensure operational readiness. - Data characteristics (example: CiC): - CiC series characterized by trends, strong weekly and monthly patterns, and spikes around major religious and public holidays. - Despite increasing uptake of cashless transactions, CiC exhibits an upward trend (possibly due to increased tourism). - Weekly and monthly patterns attributable to pay days (many employees paid on the 26th of the month). - Structural break in early 2020 at onset of COVID crisis—may relate to precautionary cash holdings or printing of new banknotes. *Italic source attribution: IMF | United Arab Emirates Liquidity Management and Forecasting (excerpts from content unit 1uaeea2022003).* ### 49. Net foreign assets exhibit strong volatility without clear pattern for the mean. The ### 49. Net foreign assets exhibit strong volatility without clear pattern for the mean. The ### Data patterns and notable observations - Net foreign assets (NFA) exhibit strong volatility and the mean is difficult to predict; conditional heteroskedasticity is usually present. - There is a clear outlier at the beginning of 2016 that can be excluded or handled with a dummy variable. - State account balance (SAB) data exhibit weekly and monthly seasonality, particularly in the latter half of the sample. - Outliers in early 2020 (onset of COVID-19) occur in SAB and can be controlled using dummy variables. - Currency in circulation (CiC) experienced a dramatic level shift at the onset of COVID-19 (structural break). - Cash in circulation tends to rise in the lead up to holidays (example: Eid-al-Fitr). ### Modeling approaches tested - Three main families tested to forecast the conditional mean of autonomous factors: ARIMA models, ETS models, and TBATS models. - Standard ARIMA fitted alongside seasonal ARIMA and ARIMA with regression. - ETS family models considered both with and without seasonality. - For NFA, models for conditional heteroskedasticity (GARCH/eGARCH) are used. - Model order for ARIMA selected using the Hyndman and Khandakar algorithm (Hyndman and Khandakar (2008)). - Best models chosen according to the Akaike Information criterion (AIC). ### Structural breaks and holiday effects - Structural breaks controlled by introducing a dummy equal to 0 before the break and 1 after the break. - Holiday effect modeling: - Dummy variables are possible but a smooth approach was considered. - Construct DaysToHoliday (DDDH_t) taking 0 on holiday, 1 the day after, −1 the day before, etc. - Eid-al-Adha effect defined as EID_t = max(0, (1 − DDDH_t^2)/7) which creates a smooth hump for seven days around the holiday. - Similar variables suggested for Eid-al-Fitr, UAE National Day, and New Years’ Day. ### Covariates and capital flows - Capital flows data (incoming flows, outgoing flows, incoming number of transactions, outgoing number of transactions) were considered as covariates for NFA. - These covariates did not improve forecasting performance for NFA; the mean remains difficult to predict. ### Covariate selection and best regressors (summary table) - Regressors selected by fitting all possible regression specifications. - Table of regressors used for each series (Yes/No): - Day of week (Dummies): CiC Yes, SAB Yes, Total Yes - Day of Month (Fourier): CiC No, SAB Yes, Total Yes - Day of Quarter (Fourier): CiC No, SAB No, Total No - Day of Religious Year (Fourier): CiC No, SAB No, Total No - Day of Civic Year (Fourier): CiC No, SAB No, Total Yes - Holiday Effects: CiC Yes, SAB No, Total Yes - COVID-19 Dummy: CiC Yes, SAB Yes, Total Yes - For NFA the random walk model was adequate for the mean; variance is more forecastable. - For CiC and SAB the best specification was a seasonal ARIMA with m= 7. - For the Total, standard ARIMA led to the lowest AIC. ### Forecast combination and model averaging - Model averaging used to ensure robustness against misspecification. - Due to limited test sample, simple averages of the best model from each model family are used. - Dynamic model selection: for each evaluation period the best model within each family is selected according to AIC; combination occurs across families via simple averages. ### Forecast evaluation design - Reserves maintenance period (RMP) is a two-week period beginning on Wednesday and finishing the following Tuesday. - Target: value of autonomous factors on the final day of the RMP. - Forecast horizons: one-day ahead up to 14-day ahead at the outset of the RMP; up to 10 (two working weekdays worth) forecasts generated per two-week period. - Point forecast metrics: root mean square error (RMSE) and mean absolute error (MAE). - Prediction interval metric: Winkler score for an 훼% prediction interval (l,u) and realised value y: - W = (u − l) + (2/훼)(l − y) if y < l - W = (u − l) if l ≤ y ≤ u - W = (u − l) + (2/훼)(y − u) if y > u ### Forecast reconciliation methods - Forecasts produced for: - Currency in Circulation (CiC) - Net Foreign Assets (NFA) - State Account Balance (SAB) - Net liquidity injection = Net Foreign Assets − Currency in Circulation − State Account Balance. - Reconciliation approaches: - Bottom-up (sum individual forecasts). - Forecast the Total directly. - Produce forecasts for all autonomous factors and the Total, then reconcile to add up. - Summing matrix S: - S = [[1 1 1],[1 0 0],[0 1 0],[0 0 1]] (as presented) - OLS reconciliation: - y_reconciled = S (S′ S)^{−1} S′ y_unreconciled - MinT reconciliation (exploits forecast-error covariance Σ): - y_reconciled = S (S′ Σ^{−1} S)^{−1} S′ Σ^{−1} y_unreconciled - For reconciliation, ARIMA with regression used for Total, CiC and SAB; eGARCH used for NFA. ### Point forecast performance summary - Relative RMSEs for the Total (rows correspond to horizons, columns to methods: ARIMA, ETS, ETS (No Seas), Reg. ARIMA, Seas. ARIMA, TBATS, Mod. Av.): - Row 1: 1.035 0.966 0.98 1.013 1.029 1.047 1 - Row 2: 1.018 1.025 0.977 1.02 1.04 1.048 1 - Row 3: 1.047 0.984 0.989 1.182 1.027 1.125 1 - Row 4: 1.023 1.064 1.048 1.028 1.052 1.083 1 - Row 5: 1.081 1.116 1.069 0.925 1.088 1.06 1 - Row 6: 1.091 1.051 1.099 0.981 1.077 1.046 1 - Row 7: 1.168 1.23 1.22 0.821 0.984 1.043 1 - Row 8: 1.079 1.244 1.249 1.006 0.905 0.978 1 - Row 9: 1.047 1.241 1.189 0.928 1.192 0.811 1 - Row 10: 1.027 1.266 1.2 0.911 1.19 0.837 1 - Interpretation: - Model averaging is robust at all horizons. - ARIMA with regression and TBATS can outperform model average, particularly at longer horizons. - ETS sometimes outperforms model average at short horizons. - Nemenyi multiple comparisons indicate that superior performance is often not statistically significant. ### Interval forecast performance and calibration - Hit rates for 80 percent prediction interval by horizon (h=1 h=2 h=3 h=6 h=7 h=8 h=9 h=10 h=13 h=14) — table of multiple rows (values listed verbatim): - 0.08 0.09 0.25 0.09 0.19 0.21 0.21 0.21 0.17 0.17 - 0.02 0.13 0.19 0.13 0.17 0.17 0.21 0.17 0.19 0.15 - 0.02 0.08 0.11 0.04 0.06 0.11 0.15 0.09 0.08 0.06 - 0.06 0.09 0.21 0.06 0.11 0.15 0.19 0.17 0.15 0.13 - 0.08 0.15 0.26 0.13 0.15 0.17 0.19 0.21 0.15 0.19 - 0.08 0.09 0.32 0.08 0.21 0.19 0.26 0.25 0.15 0.21 - 0.08 0.17 0.26 0.26 0.23 0.26 0.34 0.42 0.4 0.47 - 0.19 0.3 0.28 0.13 0.23 0.21 0.21 0.26 0.26 0.23 - 0.21 0.3 0.25 0.15 0.21 0.21 0.21 0.29 0.26 0.23 - 0.19 0.28 0.26 0.13 0.25 0.25 0.28 0.37 0.23 0.26 - 0.21 0.32 0.25 0.11 0.23 0.23 0.23 0.32 0.26 0.23 - 0.13 0.21 0.19 0.21 0.23 0.21 0.19 0.19 0.21 0.17 - 0.08 0.15 0.08 0.08 0.11 0.08 0.08 0.06 0.02 0.02 - 0.08 0.11 0.08 0.06 0.09 0.08 0.06 0.04 0.02 0.02 - 0.09 0.17 0.13 0.08 0.09 0.06 0.06 0.06 0.04 0.04 - 0.13 0.21 0.15 0.17 0.15 0.15 0.13 0.11 0.09 0.09 - 0.13 0.19 0.19 0.19 0.25 0.19 0.21 0.23 0.21 0.21 - 0.11 0.19 0.23 0.15 0.19 0.25 0.26 0.26 0.17 0.19 - 0.13 0.21 0.21 0.15 0.15 0.15 0.19 0.25 0.13 0.15 - 0.11 0.19 0.13 0.15 0.13 0.19 0.19 0.23 0.13 0.13 - 0.09 0.17 0.21 0.15 0.21 0.15 0.17 0.23 0.17 0.15 - 0.15 0.25 0.34 0.25 0.25 0.28 0.26 0.28 0.15 0.19 - 0.15 0.21 0.23 0.25 0.25 0.21 0.25 0.26 0.25 0.23 - 0.17 0.26 0.34 0.3 0.4 0.38 0.43 0.34 0.49 0.42 - Interpretation: - Numbers close to 0.2 indicate better calibrated 80% intervals. - Prediction intervals tend to be conservative, particularly for very short-term forecasts. - TBATS tends to lead to under-coverage. - Similar conclusions apply for 95% intervals. ### Interval forecast performance by method - ARIMA with regression performs well for interval forecasts, particularly at longer horizons, for CiC, SAB and the Total. - For NFA, all GARCH specifications (e.g., eGARCH) lead to similar performance. - The model average does not perform as well as ARIMA with regression for interval forecasts, particularly at longer horizons and for the Total. - Nemenyi plot for 95% prediction interval for the Total shows ARIMA with regression’s superior performance relative to model average is statistically significant. ### Reconciliation performance - Reconciliation methods compared (Total only, brevity): - OLS reconciliation is the most robust and best performing; MinT yields similar results. - For point forecasts, differences across reconciliation methods are not statistically significant. - Bottom-up (summing individual forecasts) is significantly worse at longer horizons than other methods. - For prediction intervals, OLS reconciliation significantly dominates all methods at short horizons and all methods except MinT at longer horizons (shown for Winkler score for 80% intervals). ### Conclusions (summary of empirical findings) - ARIMA with regression leads to the best forecasting performance for SAB, CiC and the Total. - All GARCH specifications yield similar results for NFA; eGARCH recommended empirically. - Model averaging improves point forecasts but not interval forecasts in this analysis. - Forecast reconciliation using OLS can lead to better point forecasts and better forecast intervals than forecasting the Total alone or summing forecasts of autonomous factors. ### Recommendations (policy and operational) - The CBUAE should forecast all autonomous factors separately, as well as the net liquidity injection, using the IMF-designed forecasting framework. - Use ARIMA with regression for forecasting CiC and SAB. - Use an EGARCH model for forecasting NFA. - Employ model dynamic selection and averaging to improve accuracy and reduce modeling risk; the IMF software infrastructure allows on-the-fly fitting, selection and averaging. - Use the IMF forecast reconciliation model (OLS/MinT) to forecast net liquidity injection and estimate associated confidence intervals rather than naively summing forecasts. - Incorporate CBUAE domain expertise and granular knowledge to refine indicators for special events and policy events and improve modeling choices. *Source: IMF staff — United Arab Emirates Liquidity Management and Forecasting (excerpt).* ### 72. The CBUAE should publish the whole liquidity forecast and associated methodology ### 1uaeea2022003 - 72. The CBUAE should publish the whole liquidity forecast and associated methodology ### Recommendations to CBUAE and Treasury (paras. 72–75) - The CBUAE should publish the whole liquidity forecast and associated methodology to help anchoring market expectations and support liquidity management of market participants (para. 72). - The CBUAE has the most extensive capacity in the market to provide the best forecasts of the UAE autonomous factors and changes in the overall liquidity conditions; releasing these forecasts would strengthen liquidity management by market participants and help the market form consistent expectations about short-term liquidity changes (para. 72). - The CBUAE should staff a small team of econometricians to estimate and fine-tune the liquidity forecasting framework; in-house expertise in liquidity forecasting is crucial given the importance of liquidity forecasts and the increasing complexity of the UAE financial system (para. 73). - The CBUAE should sign a memorandum of understanding with the Treasury to share information about the Treasury future cash-flows; forecasts from the Treasury about its expected cash-flows are crucial to support the projection of the state account balance (para. 74). - The Treasury should operate a Treasury Single Account to consistently manage its cash flows and simplify the CBUAE’s liquidity management; a Treasury Single Account at the central bank is considered an international best practice for Treasury cash-flow management and central bank liquidity management (para. 75). ### Calibration of monetary operations — Conceptual framework (paras. 76–92) - In a floor system, the central bank should maintain enough liquidity surplus to anchor the interbank rate at the policy rate (overnight deposit rate). At a minimum, the central bank should maintain a liquidity surplus greater than the aggregate precautionary demand of reserves from commercial banks so banks can settle payments (para. 76). - To anchor the interbank rate at the policy rate, the central bank should supply enough reserves to stimulate deposits of commercial banks at the overnight deposit facility, stabilize the interbank rate at the floor, and avoid excessive volatility (para. 76). - Tradeoff: anchoring the interbank rate versus supporting money market development and building the sovereign yield curve. Excessive structural liquidity anchors the interbank rate but reduces interbank money market trading and turnover (para. 77). - Empirical observation: as the CBUAE has been operating with an increasing level of excess reserves, interbank market turnover has shrunk over the last two years by around two thirds of the historical average (para. 77). - Sterilizing liquidity via issuance of central bank securities is important in a country lacking Treasury instruments to build the sovereign yield curve (para. 77). - The optimal amount of liquidity surplus should be determined by the market dynamic of the interbank rate spread with the policy rate and the central bank’s objectives; IMF proposes a framework relying on (i) a model of demand and supply of central bank reserves and (ii) formalization of central bank monetary objectives (para. 78). - Demand for excess reserves modeled on Poole (1968), customized for UAE specificities; key predictions: - Quantity of excess reserves demanded varies inversely with the opportunity cost of holding reserves. - Demand of excess reserves increases with uncertainty about reserves balances (para. 79). - Behavioral specification over a maintenance period t = 1, ..., T: commercial banks demand central bank reserves E_t^d for precautionary purposes and arbitrage; model (equation (1)) specifies dependence on: - money market interbank spread with the policy rate (i_t − ī) where the policy rate equals the US IOER in the UAE context (footnote; para. 80). - stochastic exogenous factors ω_t (e.g., oil prices, seasonal effects) (para. 80). - fulfillment of statutory reserves R over the maintenance period, depending on averaging strategy m() (para. 80). - spread between US T-Bills and the policy rate (r_t − ī) capturing alternative investment opportunities (para. 80). - Reserve accounting and incompressible demand: - E_t^d ≥ E_t^i is the incompressible demand for reserves; reserve requirement is met if cumulative actual reserves on last day ≥ T*R̄ (para. 80). - Excess reserves on each day: E_t = R_t − R̄ (para. 80). - Payment flow uncertainty increases precautionary demand and raises interbank rates and volatility (para. 81). Any factor causing banks to hoard liquidity raises rates and volatility; factors pushing undesired excess reserves upward (e.g., rise in capital inflows AF together with insufficient central bank sterilization via securities B) push rates down (para. 82). - Calendar effects: interbank rate varies predictably over maintenance period; volatility higher at end of maintenance period; period-average requirements can lower spreads and volatility but effect attenuates toward end of period; in abundant liquidity systems these patterns may not appear (para. 83). - Survey confirmation: excess reserves are held primarily to meet unpredictable payment flows; penalty for shortfalls in reserve requirements is among top four reasons for holding excess reserves; reserves count toward fulfillment of liquidity requirements (para. 84). - Supply of reserves determined by autonomous factors and central bank policy position: net foreign assets, net position of government, other items net inject liquidity; currency in circulation absorbs liquidity; central bank adjusts structural liquidity via open market operations, standing facilities and ELA (para. 85). - Central bank balance-sheet accounting equality (equation (2)) and simplified autonomous factors AN_t definition lead to supply of reserves expressed as E_t^s = AN_t − R̄ − B_t (equation (3)) (para. 86). - Market clearing E_t^s = E_t^d implies equation (4) relating i_t − ī to AN_t, B_t, M̄R̄, E, r_t − ī, and ω_t via inverted functions (para. 87). - Structural functions f, g, h, m potentially non-linear and difficult to estimate; IMF uses reduced-form approach due to estimation challenges (para. 88). - IMF models interest rate spread dynamic as a highly non-linear function with three phases (para. 89): - Below incompressible demand: banks stop trading (cannot settle trades). - Between incompressible demand and optimal level: scarcity drives rates above policy rate; active interbank trading and volatile rates. - Above optimal level (liquidity trap): excessive reserves reduce incentives to trade, banks park liquidity at ODF, interbank market shrinks. - Empirical observation: when excess reserves are below AED 25 bn, the DONIA spread surges quickly and exponentially as reserves approach incompressible demand (para. 90). - Central bank objectives used to define optimal liquidity surplus E_t^* (para. 91): - Maximize monetary policy traction (anchor interbank rate at policy rate). - Support interbank money market development and build sovereign yield curve. - Anchoring the interbank rate has highest priority; optimal reserves are minimal excess reserves for which interest rate spread is, in absolute terms, minimal: E_t^* = { min E_t | P(|i_t − ī|)=0 } (equation (7); para. 91). - Optimal liquidity absorption B_t^* + M̄R̄^* equals E[P[AN_t]] − E_t^* (equation (8)); intuitively sterilize expected liquidity injected by autonomous factors while leaving enough liquidity to meet incompressible demand and support market trading (para. 92). ### Empirical estimation of the optimal liquidity surplus (paras. 93–100) - Functional relationship between excess reserves and interbank spread is highly non-linear with complex exponential decay; standard linear regressions are inadequate (para. 93). - Two estimation approaches: - Unconditional fit via cubic splines or parametrized functional form. - Conditional fit, instrumented and cross-validated (para. 93). - IMF preference for parametrized (even over-parametrized) models to guarantee parametric stability for operational purposes (para. 93). - Data mass point issue: large mass at bottom-left of scatter requires treatment; options include rescaled Tobit estimator or aggregating data. IMF prefers aggregation (para. 94). - Two aggregation methods implemented: - max-min dispersion of the policy rate spread for each level of reserves (discretized each half AED billion). - weighted average of the DONIA spread for same level of reserves, weighted by market turnover on that day (para. 94). - Operationalization of CBUAE objective: optimal level of excess reserves defined as minimum level for which interbank rate is anchored to policy rate: E_t^* = { min E_t | P[|i_t − ī|] = 0 } (para. 95). - Market tolerance zone: interbank rate cannot be exactly at policy rate due to market noise; IMF defines tolerance zone length by volatility containment for most observations (e.g., more than 90 percent). IMF chooses 25 bps, corresponding roughly to the lowest 20th percentile of the empirical cumulative distribution function of daily DONIA spread (para. 96). - Unconditional estimation methods used: - Non-parametric fit via cubic splines. - Parametrized fit via decreasing double exponential functional form f(x) = α + β * e^{−λ x} (para. 97). - Decreasing double exponential parametrization: location α, scale β, decay rate λ; parameters estimated by non-linear least squares with Levenberg-Marquardt algorithm (para. 98). - Estimated optimal liquidity surplus under unconditional non-parametric and parametric approaches: between AED 50–60 bn. The optimal surplus depends on aggregation method and fit algorithm; after the AED 50–60 bn threshold most observations exhibit an interbank spread within the tolerance band of 25 bps (para. 99). - Factors impacting the reaction function suggest a conditional fit is preferable: - Bubble plot shows large excess reserves associated with tiny interbank turnover and low volatility (low trading activity). - Observations with largest spreads occurred with large turnover and high volatility. - These elements indicate conditional fit is crucial to recover the true functional shape of the interbank spread reaction to excess reserves (para. 100). *IMF | United Arab Emirates Liquidity Management and Forecasting — excerpts from content unit 1uaeea2022003* ### 101. The price quantity relationship between the excess reserves and the DONIA spread ### 101. The price quantity relationship between the excess reserves and the DONIA spread ### Endogeneity and instrumentation - The price (interbank rate) and the quantity (excess reserves available on the market) are simultaneously determined, creating endogeneity/simultaneity that requires instrumentation. - Autonomous factors are used as exogenous instruments: the team uses both the level and the volatility of the sum of the autonomous factors to design an over-identified regression set-up. - The conditional fit is presented in a univariate representation using residuals of the DONIA spread on exogenous regressors (application of the Frisch-Waugh theorem). ### Model specification and control variables - The IMF team estimates a conditional fit model controlling for: - Seasonal factors: religious and secular holidays, regular calendar seasonality. - Market turnover: market microstructure. - Fed IOER: uncovered interest parity and international arbitrage. - Oil prices (level and volatility): precautionary demand and drivers of the autonomous factors. - VIX: international risk sentiment. - Dubai stock market realized volatility and DONIA realized volatility: domestic risk sentiment. - Parametric specification (as used in the IV-2SLS procedure): 퐷퐷퐷퐷퐷퐷푖푖퐸퐸 푆푆푆푆푟푟푒푒퐸퐸푑푑 푡푡 = 훼훼 + 훽훽 * e^{−흀흀 / 퐸퐸푥푥퐸퐸퐸퐸퐸퐸퐸퐸 푅푅퐸퐸퐸퐸퐸퐸푅푅푅푅퐸퐸퐸퐸(푡푡)} + 훾훾 X 푡푡 + 휖휖 푡푡 - Most coefficients estimated by standard IV-2SLS; the decay coefficient 흀흀 cannot be estimated directly and is selected via cross-validation. ### Estimation method (cross-validated IV-2SLS) - Procedure: 1. For a given decay rate 흀흀, estimate 훼훼, 훽훽, 훾훾 via IV-2SLS. 2. Repeat over a grid of 흀흀 and select the best 흀흀 by maximizing an information criterion or minimizing residual sum of squares. - Cross-validation enables estimating a complex non-linear functional form using a standard linear IV-2SLS estimator. - The IV-2SLS algorithm used includes cross-validation (Python scripts available on request). ### Results, fit, and key statistics - Cross-validated IV-2SLS regressions in-sample fit: - R2 between 65 and 78 percent for max-min dispersion and weighted average respectively. - Regressors statistically significant at the 5th or at the 10th percentile threshold. - First-stage F-statistic near 10, indicating relatively decent first-stage fit and adequate endogeneity control. - Optimal liquidity surplus: - Under both aggregation approaches (max-min and weighted average), the optimal liquidity surplus estimated via the conditional CV IV-2SLS fit is around AED 50 bn. - At AED 50 bn, the interbank rate is anchored to the policy rate with minimal interbank rate volatility. - Above AED 50 bn, further anchoring is not possible and excessive liquidity would be detrimental for market development and yield curve building. - Unconditional vs conditional fits: - The conditional fit yields an optimal liquidity surplus slightly lower than the unconditional fit, reflecting omitted-variable bias in the unconditional estimation. ### Operational sterilization mix: instruments and optimal calibration - Sterilization instruments available to CBUAE: - Required reserves 푅푅 with averaging (smoothing across maintenance period). - Long-term central bank securities 퐵퐵 푡푡 퐿퐿푇푇 (up to one year) to develop capital markets and benchmark the yield curve. - Short-term instruments 퐵퐵 푡푡 푆푆 푇푇 to reflect liquidity conditions with minimal duration risk. - Total securities: 퐵퐵 푡푡 퐿퐿푇푇 + 퐵퐵 푡푡 푆푆 푇푇 = 퐵퐵 푡푡 - Accounting and optimality relations: - Optimal securities issuance based on forecasts: (9) 퐵퐵 푡푡 ∗ = ℑ(퐴퐴푁푁 푡푡 ) − 퐸퐸 푡푡 ∗ − 푅푅 ∗ - Clearing condition around optimal surplus 퐸퐸 푡푡 ∗: (10) 퐵퐵 푡푡 ∗ + 푚푚�푅푅 ∗ � + 푅푅 ∗ = 피피[퐴퐴푁푁 푡푡 ] − 퐸퐸 푡푡 ∗ - Optimal required reserves tied to forecasting errors: (11) 푚푚�푅푅 ∗ � = 피피[ℑ(퐴퐴푁푁 푡푡 ) − 퐴퐴푁푁 푡푡 ] - Under reserves averaging m(), the optimal statutory reserves equal the forecasting errors of the autonomous factors 피피[ℑ(퐴퐴푁푁 푡푡 ) − 퐴퐴푁푁 푡푡 ]; averaging acts as a buffer to absorb forecasting noise. - Recommendation for calibrating reserves and securities: - Use probabilistic forecasts of autonomous factors; calibrate statutory reserves on conditional forecasting errors at a chosen confidence level (e.g., 95 percent) rather than average past forecasting errors. - Advantages: forward-looking (confidence interval adapts to structural breaks) and frequency-based (confidence threshold determines frequency of adjusting operations). - Once optimal required reserves 푅푅 ∗ are set, infer optimal securities: - 퐵퐵 푡푡 퐿퐿푇푇 ∗ + 퐵퐵 푡푡 푆푆 푇푇∗ = 피피(퐴퐴푁푁 푡푡 ) − 퐸퐸 푡푡 ∗ − 푅푅 ∗ - Split between structural and short-tenor issuances via time-series filtering: - Structural (trend) via HP filter with a high degree of smoothing (1600*20^24): (13) 퐵퐵 푡푡 퐿퐿푇푇 ∗ = 트렌드(피피(퐴퐴푁푁 푡푡 ) − 퐸퐸 푡푡 ∗ − 푅푅 ∗) - Short-tenor issuances are the residual: (14) 퐵퐵 푡푡 푆푆 푇푇∗ = 피피(퐴퐴푁푁 푡푡 ) − 퐸퐸 푡푡 ∗ − 푅푅 ∗ − 퐵퐵 푡푡 퐿퐿푇푇 ∗ - Ensure minimum benchmark size for each security; adjust trend if necessary to guarantee representative benchmark volumes. - Yield curve estimation: - Recommend monotone convex interpolator (Hagan and West (2010)) for UAE sovereign yield curve due to short sample and lack of long-end data. - Over-parametrization provides parameter stability; basis splines and Nelson-Siegel-Svensson underperform in UAE context. ### Recommendations (policy and operational) - Operate with a liquidity surplus around AED 50 bn (subject to adjustment if market conditions and structural factors change). - Adjust toward the optimal level gradually with close market monitoring to avoid disruptions and anchor the policy rate. - Calibrate required reserves to create a buffer to absorb autonomous-factor forecasting errors, leveraging averaging to limit market operations and reduce interbank rate volatility. - Calibrate structural issuances using a filtering approach to ensure stable and sufficiently large issuances of long-term securities (up to one year) to support market development. - Conduct a follow-up IMF TA mission to assess and project monetary policy costs of the framework, including different scenarios and implications for the CBUAE income statement. - Strengthen market surveillance: solid data collection infrastructure, quantitative researchers on the market surveillance team, and a regular working group with market participants. ### Money markets development — key observations and metrics - Introduction of M-Bills and PDs is beginning to develop a secondary market for M-Bills and an interbank repurchase market. - Unsecured interbank market activity has plummeted since COVID-19 and accelerated after Q1–Q2 2020 following the new monetary policy framework; the fall is most visible at short maturities. - Network density declined: average number of market participants on both sides reduced from 20 to 5. - Concentration and liquidity: - Volumes have become highly concentrated at one or two lenders, increasing risk of interbank market breakdown. - Counterparty limits range from 65 million USD to 836 million USD. - Overnight volume stands at 10 billion AED on average over a full month (survey evidence). - Drivers of lower volumes: - Massive liquidity injection by the CBUAE. - Introduction of the ODF facility in July 2020; banks prefer to park excess reserves at the ODF remunerated at 10 bps rather than increase counterparty risk for slightly higher yields. - Reduction in reserve requirements in April 2020 from 14 percent to 7 percent, providing banks more leeway to settle payment flows without frequent interbank borrowing. - Consequence: permanent regulatory and framework changes make it unlikely that market volumes will recover substantially; market participation and interbank network structure remain impaired. - Market concentration indicator referenced: Herfindhal Index for O/N market (figure provided in source). *Source: IMF staff.* ### 125. Interbank money market rates have displayed a negative spread to the policy rate ### 125. Interbank money market rates have displayed a negative spread to the policy rate ### Interbank rates: behavior and drivers - Interbank money market rates displayed a negative spread to the policy rate for about 24 months from mid-2017 to mid-2019. - Possible drivers identified: - Domestic banks intermediating the CBUAE facilities for foreign banks and earning the small spread. - Fall in the number of transactions causing EIBOR calculations to rely more on transactions with non-banks (DF2), such as wholesale deposits with money market funds and investment funds. - Non-banks typically have no access to the ODF explaining why EIBOR, which partly integrates transactions with non-banks (DF2), exhibits a spread to the policy rate. - Differences in regulatory costs imply that the cost of bank borrowing from non-banks is somewhat lower than comparable interbank rates. ### Market conditions since 2019 - Since 2019 market rates have remained stable and have oscillated close to the policy rate in the context of ample excess reserves. - Comparison of benchmarks: - EIBOR: based on quotations of a selected number of banks; remarkably stable throughout the reserve maintenance period. - DONIA: volume-weighted average rate using actual overnight interbank transactions; more volatile toward the end of the maintenance period and a better signal of market conditions (reflects variations in volume and uncertainty). - Puzzles: - Significant narrowing of term premia in EIBOR since 2019 despite reduced volumes, higher uncertainty, higher roll-over risk, and tighter regulation. - EIBOR quotations beyond one week rely heavily on expert judgement and are highly dispersed at the higher end of the tenor distribution. ### Transitioning from EIBOR to DONIA - Key complication: decline in interbank activity. - IMF team short-run recommendation: - Focus reform on linking the new benchmark with actual transactions in the most liquid segment of the money market, chiefly the overnight market. - Leverage CBUAE access to detailed interbank transactions data through its payment system database. - Calibrate the liquidity surplus to boost market activity (see section IV reference in source). - Data and calculation recommendations: - Integrate bank borrowing from a wide range of non-bank wholesale counterparties (money market funds and investment funds) into DONIA. - Expectation: bulk of transactions volume would come from wholesale deposit transactions with non-banks. - CBUAE could collect these data from the entire banking system at time t-1 for calculation at time t. - Locating calculation at the CBUAE advantages: - Rate based on transactions covering the entire banking system, not just selected banks. - CBUAE can carry out quality checks and cleaning of raw data before calculation. - Staffing implication: monetary policy directorate should include economists with expertise in econometric methods and experience analyzing large datasets. - Footnote example cited: wholesale deposit transactions with non-banks represent 70 percent of the volume of transactions used for the calculation of the SONIA rate by the Bank of England. ### Term rates: short-run and long-run approaches - Short-run recommendation: - Encourage markets to use backward-looking term rates (e.g., one-month interest payment obligations calculated as the compounded O/N reference rate over the same one-month period). - Advantages: simple calculation; can be calculated even if there are no transactions. - Weakness: does not integrate expectations of future interest rates and market conditions. - Long-run recommendation: - As derivative markets linked to the new benchmark develop, construct forward-looking term rates. - Forward-rates provide certainty to market participants because they are available at the beginning of the period to which they apply. - Operational recommendation: CBUAE should publish DONIA daily. - Implementation recommendation: - Delegate calculation of term rates to the market under the oversight and following the guidelines of an expert group set up by the CBUAE; market participants have access to pricing data and the expertise/resources for calculations. ### M-Bills secondary market: assessment and recommendations - Assessment: - M-Bills secondary market has started to develop but still faces problems of liquidity and deepness. - A deep and liquid M-Bills secondary market is a cornerstone for Repo market development since M-Bills are scarce marketable securities usable as underlying collateral. - Current risk-free curve is limited to M-Bills with maturity up to 336 days. - DoF plans to issue a bond in AED by the end of the year. - Observations: - PDs sometimes bid by themselves and on behalf of clients in primary auctions; allotted bids are generally close to market prices. - PDs have been trading in the secondary market close to auction rates and sometimes at lower rates. - Recommendations: - Evaluate increasing issuance in long tenors to provide more securities and improve liquidity, taking into account market demand and potential price impact. - Adapt the PDs program to encourage greater secondary market activity: - Relax cap on maximum holdings for a single PD in primary auctions (current cap pushes banks to buy-and-hold). - Publish ranking of PDs and fine-tune program parameters (bid-ask spread, daily amount PDs should quote). - Note: average amount in the secondary market is 4.7 million. - Consider different minimum quote size depending on M-Bill maturity (current minimum size is 10 million). - Allocate higher amounts across tenors and consider re-tranching long-term M-Bills to increase outstanding amounts and promote transactions. ### FX Swap market: assessment and recommendations - Assessment: - USD-AED FX Swap market is the most developed money market in the UAE. - 2019 BIS triennial survey: daily turnover of 30,769 million dollars; 84 percent of the volume has a maturity lower than 14 days. - Swap volume is five times the daily spot volume reported in the survey. - FX Swap market is the reference for other markets that use implied swap rates to quote rates such as EIBOR or M-Bills. - Recommendations: - CBUAE needs to set market conventions for pricing of FX SWAP and track the implicit AED rate derived from FX SWAP points. - Market convention: trade swaps in terms of forward points considering the long leg operation to determine relevant price bid or ask. - Forward points determined by interest rate differential between AED and USD rates in the same tenor. - To derive implied AED rate from forward points, one must assume the USD rate; according to DTIs, LIBOR can be used as a proxy for USD funding cost to get the implied AED. - Consider adoption of the FX global code of conduct as good practice. - DTIs are already using ISDA for long-term transactions (could be up to ten years). - Numerical/formula references provided in source for converting forward points to implied AED rates (as presented in original text). ### REPO market: assessment and recommendations - Assessment: - Onshore REPO market is at its infancy stage. - Half of the banks have not an MRA (Master Repurchase Agreement) in place; some banks suggest a central counterparty for repurchase operations could help development. - Recommendations: - Continue supporting REPO market development to reduce counterparty risk perceptions among banks. - Enforce use of Master Repurchase Agreement (MRA). - Enhance liquidity in secondary market through allocation across tenors and issuance policies aligned with liquidity projects and market demand. ### Market disclosure, communication, and monetary policy transmission - Disclosure and communication recommendations: - Improve communication strategy and create a website dedicated to CBUAE operations and market information. - Clarify which operations are regular versus exceptional. - Publish more market information: CBUAE auctions results, interbank repo rates, FX swap rates, volumes, collateral used, and active market participants. - For weighted average rates per tenor, CBUAE should compile all transactions and compute weighted averages using transaction volumes for tenors such as one day, seven days, etc. - Monetary policy transmission and data collection: - Market participants still rely on EIBOR as a reference; as EIBOR is discontinued, expect DONIA to become the new benchmark. - Some banks use internal FTP models including base rate, RR, and retail/wholesale funding costs. - CBUAE should collect average monthly rates applied on various categories of deposits and loans on a bank-by-bank basis and compile weighted average monthly rates for the banking system. - With long time series, CBUAE should estimate interest rate pass-through in UAE using markup pricing models summarized in the source: - Long-run relationship: iR = α + β · iM, where β is LR pass-through (β < 1 in practice). - Error-correction and ARDL representations provided in source to estimate short-run and long-run pass-through coefficients and persistence. - The mark-up pricing relationship implies β should be positive. *IMF | United Arab Emirates Liquidity Management and Forecasting (chapter excerpt).* ### APPENDIX II. UPDATE ON STATUS OF PREVIOUS TA MISSION’S RECOMMENDATIONS ### APPENDIX II. UPDATE ON STATUS OF PREVIOUS TA MISSION’S RECOMMENDATIONS ### Developing Monetary Policy Operational Framework — Recommendations and Implementation Status - Agree that the primary objective of the monetary policy implementation framework is to maintain short term interest rates in line with those consistent with maintaining the exchange rate peg, and that secondary objectives are to preserve financial stability and foster development of local markets. - Status: Fully implemented. - Accelerate the development of liquidity forecasting by: - Complete the identification of key data sources. - Develop a liquidity reconciliation and forecasting template. - Staff the forecasting and reconciliation function and merge with the market operations function. - Develop an information sharing protocol with the Ministry of Finance (MOF) and DoF Abu Dhabi to obtain forward schedules of FX transactions and cash flows. - Begin analysis of forecasting errors. - Status: Partially implemented as the identification has taken place and a draft MoU is already available. However, resources have not been yet allocated, and forecasting model is under development. - Reform the overnight deposit facility by: - Open the facility for the entire trading day. - Set the price equal to the Interest on Excess Reserves Rate (IOER) of the US Federal Reserve. - Mission view: The mission considers appropriate the current operational hours. The facility closes at 8:00 pm one hour before the payment systems. Few late operations takes place afterwards. - Status: Implemented. - Reconsider the proposed primary dealer system and implement the new liquidity management framework and allow access to all banks. - Status: Due to the initial stage of the secondary market the mission considers appropriate to maintain for some time the primary dealer system to develop the market. - Settle all fine-tuning OMO instruments for value T+0. - Status: Not implemented yet. - Discontinue practice to use bilaterally negotiated operations outside of overnight standing facilities. - Status: Partially implemented. - Simplify implementation of the M-Bill auctions by removing the non-competitive bidding and re-offer options. - Status: Partially implemented. - Adopt the reforms in the following sequence to reduce implementation risks, and status for each: 1. Introduce overnight standing facilities. - Status: Implemented 2. Reform reserve requirements by expanding the use of averaging. - Status: Implemented from 7-14 days. 3. Establish and communicate the regulatory treatment of the new M-Bills. - Status: Implemented. 4. Complete development of the liquidity forecasting function. - Status: Not implemented yet. 5. Shift settlement of FX transactions with the CBUAE to T+2. - Status: Partially implemented. The settlement convention with financial institutions is T+2, while FX operations with the Abu Dhabi Department of Finance are settled at T+0. 6. Introduce the new M-Bill program while retaining the outstanding CD issues and price the associated early redemption option in line with new M-Bills. - Status: Implemented. 7. Review the costing exercise (the impact analysis on CBUAE’s P&L) performed for the new monetary framework. - Status: Not implemented yet. *Source: 2018 and 2021 IMF Missions* --- ### Examples of Announcements Related to Fine-Tuning Auctions (Illustrative) - Example announcement header: Central Bank's Intervention in the Money Market — July 8, 2021. - Expected change in the total balance of the credit institutions' current accounts: 80,000. - Concept: Expected change in the total balance of the credit institutions' current accounts resulting from the maturity of previous operations, federal government transactions, and currency and coins' withdrawals or deposits. - Central Bank July 9, 2021, 9:00 AM Auctions (illustrative table captured as text): - Maturity / Amount / Settlement / Instrument / Operation - 3 / 10,000 / July 9, 2021 / FX swap/Reverse Repo / Withdraw - 5 / 15,000 / July 9, 2021 / FX swap/Reverse Repo / Withdraw - 7 / 5,000 / July 9, 2021 / FX swap/Reverse Repo / Withdraw - 10 / 25,000 / July 9, 2021 / FX swap / Withdraw - 14 / 25,000 / July 9, 2021 / FX swap / Withdraw - Ex-post transparency example: Central Bank Open Market operations Results — July 09, 2021 - Operation: Withdraw - Total amount demanded: 100,000 - Total amount allocated: 80,000 - Number of participating institutions: 30 - Number of participating institutions that received Allocation: 25 - Auction 1 Allotted Bids (excerpt): - 7 days FX Swap Forward points 0.02 — Tender 2,000 — Allotted 2,000; Reverse Repo Rate % 7 days .16 — Tender 900 — Allotted 900 - 7 days FX Swap Forward points 0.03 — Tender 3,000 — Allotted 2,000; Reverse Repo Rate % 7 days .17 — Tender 500 — Allotted 100 - 7 days FX Swap Forward points 0.05 — Tender 5,000 — Allotted 0 *Source: IMF Mission* --- ### CBUAE’s FX Swaps and Repurchase Transactions — International Comparison (Selected Features) - OMO FX Swap — selected country features (as summarized): - Maturity: Up to 3 months (Australia); 1 day—3 months (Switzerland); Discretionary (India); Up to 6 months (Singapore). - Frequency: Several times per month (Australia); As needed (Switzerland); Discretionary (India); Daily, as needed (Singapore). - Pricing Method: Bilateral with counterparties (Australia); N/A (Switzerland); Auction (India, Singapore). - Access: Limited By/To FX counterparty panel (Australia); N/A (Switzerland); Authorized dealer category-1 banks / Primary dealers (India, Singapore). - Functions: Liquidity injection & withdrawal; Liquidity management. - Example hour: 9:45 am (Australia entry). - Repo/Reverse Repo — selected country features (excerpt): - Australia: Maturity up to 1 year; typically around 1 month; Frequency at least daily; Pricing Method Auction; Access Most members of Reserve Bank’s RTGS system; Function Liquidity injection and withdrawal. - Brazil: Maturity Up to 6 months; Frequency As needed; Pricing Method Auction; Access Primary Dealers and Financial institutions; Function Reinforce target rate, Liquidity management. - Canada: Maturity 1 day; Frequency As needed; Pricing Method Single price auction; Access Primary Dealers, Government of Canada collateral; Function Reinforce target rate. - United States: (i) Up to 65 business days (ii) O/N; RP: Discretionary (i) Discretionary (ii) Daily; RP pricing RP: Fixed-quantity multi-price auction; RRP: Fixed-price or fixed-quantity; Access RP: primary dealers; RRP: primary dealers plus some US government-sponsored enterprises, money market mutual funds, and depository institutions; Functions: (i) Liquidity injection and withdrawal (ii) Reduces reserves, supports interest rate control as administered rate. - Eurosystem: (i) 1 week and 3 months (ii) Varies; Frequency (i) Weekly for 1-week operations; monthly for 3-month operations; Pricing Method (i) Fixed rate; Access Monetary financial institutions fulfilling eligibility criteria; Function Basic refinancing (increased emphasis on intermediation) and Fine tuning. *Source: BIS Markets Committee Compendium 2019; IMF Mission* --- ### Purposes of the Reserve Requirements (RR) in the UAE - Monetary policy tool: - RR contributes to the defense of the fixed exchange rate arrangement by creating a minimum demand for local currency reserve to prevent rapid drops in reserve demand that could lead to important FX sales by the CBUAE. - By supporting the demand for local currency reserves, the RR accelerates the increase of the domestic interest to reach a new equilibrium to stop outflows and defend the fixed exchange rate. - Macro-prudential tool: - RR decreases systemic risks by: (i) increasing system-wide resilience against systemic liquidity shocks; and (ii) smoothing the credit cycle. - A raise of the RR moderates credit growth and can reduce procyclicality of credit. - RR should be used for macroprudential purposes only if alternative instruments are not available or likely to be less effective. - Changes in RR influence monetary policy stance, may be circumvented (e.g., re-classification of liabilities), and are usually less targeted than other macroprudential tools. - Note: Many central banks use tighter RRs on FX liabilities to reduce dollarization, but the current regulatory stance in the UAE is neutral with no differentiation in RR ratios between local and foreign currency deposits. - Micro-prudential tool: - RR ensures banks hold a minimum amount of a specific HQLA as backing for deposits — particularly relevant for RR on USD liabilities since the CBUAE cannot play lender of last resort in USD. - Liquidity requirements such as the LCR and ELAR in the UAE have increasingly taken on that function in many systems. - Liquidity management tool: - RR complements other tools for calibrating appropriate system liquidity and has several advantages over OMOs: does not rely on banks’ participation in auctions (reduces operational risk); reduces amount of liquidity to be absorbed by OMOs (facilitates OMO conduct); reserve averaging provides an additional channel to limit interbank rate volatility and allows the central bank to reduce frequency of OMOs. --- ### Example of Analytical Central Bank Balance Sheet and Compilation of Autonomous Factors of Liquidity - Analytical identity (balance sheet aggregation): - Assets: Net Foreign Assets (NFA) + Claims on Government (CG) + Liquidity Provision Operations (LP) + Other items. - Liabilities: Currency in Circulation (CiC) + Government Deposits (GD) + Liquidity Absorption Operations (LA) + Banks’ Reserves (R) + Other Factors Net (OFN). - Accounting identity: TOTAL ASSETS = TOTAL LIABILITIES. - Aggregated equation: NFA + CG + LP = CiC + GD + LA + R + OFN. - Reserve identity: R = NFA + CG + LP – CC – GD – LA – OFN. - Decomposition: R = AF (NFA + CG – CC – GD – OFN) + CB Operations (LP – LA). - Change in reserves: ΔR = ΔAF + ΔCBO. *Source: Andrea Schaeter, Implementation of Monetary Policy and the Central Bank’s Balance Sheet, IMF WP/01/149; IMF staff* --- ### Data and IT Infrastructure Requirements for Liquidity Forecasting - Overarching needs: - Liquidity forecasting requires accurate, updated, and comprehensive data, plus adequate software and hardware infrastructure. - Modeling and technical capacities should align with industry standards and practices of central banks in major financial centers (e.g., Hong Kong Monetary Authorities, Singapore Monetary Authorities). - A complete and modern data and IT infrastructure supports pricing CBUAE instruments, calibrating operations, managing collateral, monitoring markets, and implementing early-warning signals. - Data coverage — daily and granular datasets recommended for: - Currency in Circulation: Daily data on currency issued by the central bank and currency deposited by commercial banks; decomposition by banknotes and coins denomination recommended. - State Account: Treasury should share future expenditure and revenues, particularly large and exceptional items, and granular revenues/expenditure data. - Net Foreign Assets: CBUAE should have access to FX settlement data to predict FX movements up to T+2 and receive information about large FX payments—particularly oil and gas. - General Financial and Economic Data: Access to Bloomberg Market Data Feed™, Haver DLX™, EPFR Global Data™, Thomson Reuters Datastream™, and Refinitiv™ recommended; diversify providers to increase coverage. - UAE-specific Market Developments: Collect and automatize data-reporting on indicators typically unavailable through commercial sources (e.g., interbank market turnover, traded prices, secondary instruments trading). - Database management and reporting: - Maintain a central database repository with adequately managed access rights (e.g., Oracle SQL database); monetary policy department should have direct access for automation, querying, cleaning. - External provider data should be reported automatically and carefully checked; statistics department should run automatic data control process (e.g., detect movements more than one standard deviation) and raise warnings. - Harmonize time series to ensure sufficient historical coverage and consistency over time. - Software requirements: - Install Python and R on CBUAE computers; complement with Matlab or Eviews if needed. - R: specialized statistical language with packages for forecasting, statistical inference, Bayesian estimation. - Python: leading language for data management, visualization, simulated methods, machine learning, big data, kernel estimation. - Use interoperability packages (e.g., reticulate, rpy2) when needed. - Install integrated development environments: RStudio and Python Anaconda complete distributions. - Use versioning software such as git with an internal Github repository to track changes and share code. - Note: RStudio, Python Anaconda, and Git/Github are open-source and widely used. - Hardware requirements: - Equip CBUAE with high-performance computing server infrastructure accessible by team from laptops (remotely if possible) with substantial CPU and GPU power and fast-access memory. - Install Python Anaconda and RStudio on servers and connect servers to data repository. - Cloud computing access recommended to scale large projects efficiently. - IMF example hardware (for illustration of capacity levels): - On-premises: Intel Xeon 6246 3.30 GHz CPU with 24 cores physical, 48 cores hyperthreaded with 768 GB of RAM and 6TB Nvme SSD; some servers with NVIDIA® Tesla™ V100 GPU Computing Accelerator—32GB HBM2—PCIe 3.0 x16. - On the cloud: Access to Amazon AWS and Microsoft Azure HPC; VMs sized to computational needs. *Source: IMF staff; IMF TA mission examples* *Italic: Source document — APPENDIX II. UPDATE ON STATUS OF PREVIOUS TA MISSION’S RECOMMENDATIONS (IMF PDF).* ### APPENDIX VIII. STATISTICAL FORECASTING MODELS ### APPENDIX VIII. STATISTICAL FORECASTING MODELS ### Overview - Presents specifications of statistical models used for forecasting the autonomous factors. - Details about treatment of structural breaks and choice of covariates are available in the main text. ### ARIMA Models - Definition: (1−휙(퐵))(1−퐵)^{푑} 푦_{푡} = (1+휃(퐵))휖_{푡} - 퐵 is the backshift operator (퐵푦_{푡}=푦_{푡−1}). - d is typically 1 (rarely 2) for nonstationary series and 0 for stationary series. - (1−휙(퐵)) = 1 − 휙_{1}퐵 − 휙_{2}퐵^{2} − ⋯ − 휙_{푝}퐵^{푝} is the AR polynomial of order 푝. - (1+휃(퐵)) = 1 + 휃_{1}퐵 + 휃_{2}퐵^{2} + ⋯ + 휃_{푞}퐵^{푞} is the MA polynomial of order 푞. - 휖_{푡} is an innovation term. - Nomenclature: ARIMA(p,d,q). Example: ARIMA(2,1,2). - Illustration: Forecasting results shown for ARIMA(1,1,1) using training period March 6, 2016 to March 5, 2019 and forecast period March 6, 2019 to March 19, 2019. - Visualization elements: actual (black), fitted (orange), forecasts (blue), prediction intervals as fan chart with 80 percent (darker) and 95 percent (lighter) prediction intervals. - Note: Prediction intervals should guide appropriate OMOs during the reserves maintenance period. ### Model Selection (ARIMA Order) - Order choices: p, d, q, P, D, Q must be selected. - Estimation uses stepwise algorithm of Hyndman and Khandakar (2008): 1. Find d using KPSS test. 2. Estimate four initial models and choose the best of these four models. 3. Expand candidate model set by considering models with S or q differing from current best by 1. 4. Iterate until no improvement is made. - Selection criterion: Akaike Information Criterion corrected for small sample size (AICc). - Implementation: auto.arima function in the forecast package in R. ### Seasonal ARIMA (SARIMA) - General form: (1−휙(퐵))(1−훷(퐵^{m}))(1−퐵)^{d}(1−퐵^{m})^{D} 𝐴 푦_{푡} = (1+휃(퐵))(1+휃(퐵^{m}))휖_{푡} - P, D, Q are seasonal AR order, seasonal differencing, seasonal MA order. - Nomenclature: ARIMA(p,d,q)(P,D,Q)[m]. Example: ARIMA(1,0,0)(0,1,1)[5] equivalent to 푦_{푡} = 푦_{푡−5} + 휙(푦_{푡−1} − 푦_{푡−6}) + 휖_{푡} − 휖_{푡−5}. - Example: SARIMA(4,1,1)(2,0,0)[5] forecasted a decline in currency in circulation during evaluation period. - Single-seasonality limitation: SARIMA captures one form of seasonality explicitly. - Candidate choices for m: - 7 Weekly - 30 Monthly - 354 Yearly (Civic Calendar) - 365 Yearly (Hijri Calendar) - ARIMA can incorporate covariates to capture multiple seasonalities. ### Seasonal ARIMA with Regression - Dummy variables: - Example day-of-week dummy: D_{t}(Mon) = 1 if Day t is a Monday, 1 otherwise. (Note: source shows this dummy definition; intended meaning is 1 if Monday, 0 otherwise.) - Dummies included in covariate vector x'_{t} and model specified on transformed series y_{t} − x'_{t}β. - Correct specification for AR(1) with covariates: - y_{t} − x'_{t}β = 휙( y_{t−1} − x'_{t−1}β ) + 휖_{t} - This avoids using stationary predictors as covariates for non-stationary data and allows easier interpretation of coefficients. - Fourier terms (for large m and smooth seasonality): - s_{t}(j) = sin(2π j D / m) - c_{t}(j) = cos(2π j D / m) - Choice J = 2 considered for monthly and yearly effects (both civic and religious calendars). ### ETS Models (Error Trend Seasonality) - Simple exponential smoothing: - ŷ_{T+1} = α y_{T} + α(1−α) y_{T−1} + α^{2}(1−α) y_{T−2} + ⋯ - Latent level l_{t} representation: ŷ_{T+1} = l_{t}; l_{t} = α y_{t} + (1−α) l_{t−1} - Trend components: - ŷ_{T+h} = l_{t} + h b_{t} - l_{t} = α y_{t} + (1−α)(l_{t−1} + b_{t−1}) - b_{t} = β^{*}(l_{t} − l_{t−1}) + (1−β^{*}) b_{t−1} - Damped trend with damping parameter φ: - ŷ_{T+h} = l_{t} + (1 + φ + φ^{2} + ⋯ + φ^{h}) b_{t} - l_{t} and b_{t} updated with φ included as shown in source. - Seasonality (additive form): - ŷ_{T+h} = l_{t} + h b_{t} + s_{t+h−m(k+1)} - l_{t}, b_{t}, s_{t} update equations provided with smoothing parameters α, β^{*}, γ. - Seasonality (multiplicative form) equations provided. - Error structures: additive or multiplicative errors can be included to create statistical ETS models. - Example simple additive-errors representation: y_{t} = l_{t−1} + ε_{t}; l_{t} = l_{t−1} + α ε_{t} - ETS model components combine: - Error: Additive (A) or Multiplicative (M) - Trend: No trend (N), Additive trend (A), or Additive damped trend (Ad) - Seasonality: No seasonality (N), Additive seasonality (A), or Multiplicative seasonality (Ad) - Nomenclature: ETS(.,.,.), e.g., ETS(A,N,A). - Model selection via AICc. - Disadvantage: Covariates cannot be easily incorporated compared with ARIMA models. ### TBATS Model - TBATS features: - Handles seasonality and trend via exponential smoothing using trigonometric terms for seasonality. - Uses Box-Cox transformation and ARIMA innovations. - Allows seasonality to change over time and can handle multiple calendars (e.g., civic Gregorian and religious Hijri calendars). - TBATS was initially proposed for electricity demand modeling in Turkey. ### Volatility Models (for Net Foreign Assets) - Three classes fitted: GARCH, eGARCH, GJR-GARCH. - GARCH: - Variance equation: σ_{t}^{2} = ω + Σ_{i=1}^{p} β_{i} σ_{t−i}^{2} + Σ_{i=1}^{p} α_{i} e_{t−i}^{2}. - Example figure shows net foreign assets residuals and estimated/forecast volatility. - 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}). - g(ε_{t}) = θ ε_{t} + λ( |ε_{t}| − E(|ε_{t}|) ). - Advantage: asymmetric effects of sign and magnitude of innovations on variance. - GJR-GARCH: - Specification: σ_{t}^{2} = ω + δ σ_{t−1}^{2} + α ε_{t−1}^{2} + φ ε_{t−1}^{2} C_{t−1} - C_{t−1} = 0 if ε_{t−1} ≥ 0 and C_{t−1} = 1 if ε_{t−1} < 0. - Allows asymmetric effects similar to eGARCH. ### Software Documentation (Appendix IX) — Implementation Notes - Language: R (information at https://www.r-project.org/). - Recommended environment: RStudio (https://www.rstudio.com/). - Recommended R version: 3.6.3 to match library versions. - IMF provides a folder containing necessary libraries for Windows; place this folder with the R source code. Files provided: - R scripts - Config file - Documentation (this file) R scripts and usage: - All R scripts use the config file via source("config.R"). - Four R scripts to be run in order: - step_000_dataclean.R - step_001_features.R - step_002_model.R - step_003_summary.R Step 0: Cleaning Data (step_000_dataclean.R) - Reads specified Excel or csv from rawdatafile parameter. - Sets start and end dates and removes bad dates per config. - Adds column datatype with values data, holiday, and imputed. - Removes weekends based on exclude_days parameter. - Linearly interpolates missing dates. - Saves output as dataclean parameter. - Creates and saves a plot of the target variable as PDF. - Input format expectation: first column dates; target variable column specified by targetcol (e.g., targetcol = 2 for simplest two-column file). Step 1: Adding Features (step_001_features.R) - Reads cleaned csv from step 0. - Generates: - Daily biases for each day of the week. - Trigonometric biases for each day of the month. - Trigonometric biases for each day of the year. - Adds structural breaks from structural_breaks parameter. - Adds bias terms around the 20th of each month and bias terms for the quarter. - Combines biases with data and saves as dataFeat parameter. - Plots all features and saves as PDF files. - Notes: - If N weekdays used (typically N = 5), there will be N − 1 biases so each weekday gets a unique bias. - Structural breaks parameter allows modeling rapid value changes for given dates. Step 2: Fitting the Model (step_002_model.R) - Reads csv from step 1. - Creates matrix of features. - Runs forecastLoop function over a range of dates. - Saves forecasted results as an Excel file. - forecastLoop inputs: - list of index values, data Y, validation set length testLength, forecast horizon horizon, dates, feature matrix X, and models modelLabels. - Iterates over index list and generates forecasts for each date using all models in modelLabels. - Available models: - Naïve: random walk benchmark using most recent value. - SNaive: Seasonal Naive using base frequency (typically five days per week). - ETS: ExponenTial Smoothing. - ARIMA: auto.arima selection within max.d, max.p, max.q. - ETSfeat: ETS with external features (seasonal biases from step 1). - ARIMAfeat: ARIMA with external features (seasonal biases). - ARIMAtrigB: ARIMAfeat variant using state-space ARIMA. - TBATS: De Livera, Hyndman, and Snyder model—exponential smoothing state-space with Box-Cox. - TBATSfeat: Two-step approach—exponential smoothing fit then TBATS on residuals. - Forecast results saved as Excel. Step 3: Summarizing Results (step_003_summary.R) - Reads Excel forecasts. - Calculates errors and RMSE for each model. - Constructs composite models: - Model-selection composite: choose model with best rolling error over several dates leading up to forecast date. - Mean-of-models composite: mean of several models. - Writes RMSE results as csv. - Plots RMSE values and forecast vs original data; saves PDFs. - Merges forecast plots and data into one large PDF with pages for different forecast origins. Contacts for help with code: - Kei Moriya (kmoriya@imf.org) - Romain Lafarguette (rlafarguette@imf.org) ### User R Guide (Appendix X) — Running the Code in RStudio - RStudio panels overview: source code (upper left), console (lower left), variables (upper right), files/packages (bottom right). - Running code: - Edit config file if needed; R scripts include source('config.R'). - Execute lines with Ctrl + Enter or whole script with Ctrl + A then Ctrl + Enter or Source button. - Outputs generated: csv, Excel, PDF files for model understanding and diagnostics. *Source: IMF staff* --- _Source: https://www.imf.org/-/media/files/publications/cr/2022/english/1uaeea2022003.pdf_