A. Objectives

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. To achieve its operational target, the CBUAE implemented a floor interest rate corridor system with excess reserves, placed at the Overnight Deposit Facility (ODF), remunerated at the base rase. The choice of the floor system was 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. Following previous IMF TA advice,³ the CBUAE has set the level of the base rate in line with the US Federal Reserve’s Interest on Excess Reserves (IOER) rate since July 12, 2020.⁴ 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.

7. As a secondary objective, the CBUAE aims to promote market development, increasing incentives for banks to manage liquidity more actively. The floor system has been complemented by a regular program of central bank M-Bill’s issuances to support the primary and secondary money market development. The operational framework requires from banks an active allocation of their reserves across the different instruments as well as an estimation of their precautionary unremunerated reserves.

8. Financial stability considerations have remained a subsidiary objective. The CBUAE designed a Contingent Liquidity Insurance Facility (CLIF) to provide liquidity for potentially a long maturity during stress that could be idiosyncratic or market wide. This facility is collateralized with broader range of collateral including foreign sovereign, corporate and banks bonds. Depending on the level of stress the liquidity would be provided by uniform priced auctions or bilaterally to all DTI. The CLIF is scheduled to be activated by end of 2021.

9. Other objectives have been pursued to contain the repercussions of the COVID-19 pandemic without endangering the stability of the peg. A Targeted Economic Support Scheme (TESS), offering payments holiday to banks customers, has been put in place to ease stress in the short-term funding markets while supporting economic growth during the COVID-19 pandemic.⁵ The allocated AED 257 billion volume was set in line with the monetary base coverage rule.⁶ The take-up for this facility was about 37 percent which led the CBUAE to extend the deferrals of payments to end of December 31, 2021, and those related to working capital loans to June 30, 2022.

B. Features of the Operational Framework

10. The key features of the new monetary policy implementation framework include the following four pillars: standing facilities, OMOs, liquidity insurance, and reserve requirements. Most of the banks have indicated that they have been using the CBUAE instruments as part of their liquidity management activities and are participating regularly in the OMOs (Figure 1).

View Full Size

Banks’ Usage of CBUAE Instruments

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF Mission Banks’ Survey

• Download Figure
• Download figure as PowerPoint slide

Standing Facilities

11. The interest rate corridor is bounded by two standing facilities (Figure 2). A third Intraday Liquidity Facility (ILF) has been also introduced. The width of the corridor was initially set at 100 bps and was subsequently reduced to 50 bps on January 11, 2021 (date of the launching of M-Bills).

View Full Size

Key Short-term Interest Rates in the UAE (in percent)

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: CBUAE and IMF staff

• Download Figure
• Download figure as PowerPoint slide

• The ODF, introduced on July 12, 2020, sets the floor rate for overnight money market rates signaling the monetary policy stance. The rate applied to the ODF is the main policy rate (base rate) was initially anchored on the IOER and was then aligned to the IORB since July 29, 2021. This facility has been well accepted by the market participants and most banks used daily (Figure 3).

• The Overnight Lending Facility (OLF)/Collateralized Murabaha Facility sets the ceiling for short-term money markets rates. The OLF and REPO facility have been consolidated into a single overnight lending facility. Since the banking system in the UAE continues to be in a structural excess liquidity, only few banks have been using the OLF. The eligible collateral for OLF could be 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.

• The ILF provides collateralized intraday liquidity at zero cost that allow banks to manage unforeseen payments disruptions or gridlocks, avoiding delays in payment submission and settlement. The collateral for ILF could be Monetary Bills (M-Bills)/Islamic CDs issued by CBUAE or T-Bills issued by federal or local governments.

View Full Size

Banks’ Usage of the ODF

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF Mission Banks’ Survey

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Banks’ Usage of the OLF

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF Mission Banks’ Survey

• Download Figure
• Download figure as PowerPoint slide

Open Market Operations

12. The structural OMOs play an important role in the implementation of the monetary policy framework draining the structural surplus of liquidity and smoothing short-term liquidity imbalances that may influence the interest rates. The CBUAE has launched a new tradable securities issuance program (M-Bills) in replacement of the Certificate of Deposits. It announced an issuance calendar with the dates and the tenors of M-Bills that will be allocated through auctions for primary dealers (PDs).⁷ The CBUAE will continue running the Islamic CDs program,⁸ until the launch of the Islamic M-Bill. The CBUAE is working on the development of the Islamic M-Bill that may have an environmental, social and governance (ESG) framework, and it is expected to be released in a couple of years according to the conversations with the CBUAE. The CBUAE is replacing a percentage of the CD that mature with issuance of M-Bills. Currently the CBUAE performs a market intelligence with the dealers eight days before the auction to estimate the potential demand; based on this estimation it allotted a percentage below 100 percent to promote competition in the auction (Figure 6). In addition, the CBUAE may conduct buy-backs or repurchase operation for liquidity provision.

View Full Size

ICD and M-Bills Allottments per Tenor, June 2021

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: CBUAE and IMF staff

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Percentage Allotted in the M-Bills Auctions

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: CBUAE and IMF staff

• Download Figure
• Download figure as PowerPoint slide

13. The flattening of the M-Bill curve is an indication of an excess liquidity beyond the optimal liquidity surplus in the system. The decomposition of the M-Bill primary yield curve gives an indication of the term, liquidity, and other premiums in the UAE market. In this case the flattening of the curve implies that banks are less risk averse in their search of yield.

View Full Size

CBUAE’s M-Bill Yield Curve Decomposition as of June 14, 2021

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: CBUAE and IMF staffNote: The M-Bill rate is descomposed in the risk free rate (effective fed funds rates), the term premium for tenor t that is the difference between the US T-Bill for tenor t and the risk free rate, and finally the liquidity and other premiums are the difference of the M-Bill rate for the tenor t minus the risk free rate and the term premium for tenor t. If the tenor of the M-Bill and the US T-Bill does not match the US T-Bill rates are interpolated to get a proxy for the tenor needed.

• Download Figure
• Download figure as PowerPoint slide

14. The CBUAE fine-tuning OMOs will be used to manage unexpected liquidity fluctuations that can increase the volatility of domestic money market interest rates. So far, fine-tuning operations have not been conducted as accurate liquidity forecasts are not yet available. Once these would be available, the purpose of these operations is to ensure that there is an appropriate amount of liquidity in the banking system correcting any deviation in the liquidity forecast that can take place during the RMP. Fine-tuning OMOs will also adjust for Banks’ overbidding or underbidding behaviors which cannot be reallocated in the interbank market. The CBUAE may use the following operations according to the market conditions and with high degree of flexibility: (i) repurchase transactions at different tenors; and (ii) matched transactions of securities and FX transactions including FX swaps. Appendix I describes the CBUAE’s proposed framework.

FX Swap Facility

15. FX Swap Facility was introduced by the CBUAE during the 2008–2009 financial crisis to provide liquidity to the Deposit Takers Institutions (DTIs) from seven days up to 12 months. It has been used during stress periods in the financial markets. Recently only three banks are capturing 80 percent of the facility’s volume. It is perceived by the DTIs as useful since it puts a cap for the FX Swap market especially in stress times such as during the COVID-19 turmoil in 2020. It is important to highlight the FX Swap market implied rates,⁹ which are used as reference to quote other instruments such as M-Bills.

16. In the next phase, the CBUAE FX swap facility will be absorbed by the OLF in which the banks can get liquidity using US dollars as collateral. This operation will be equivalent to a synthetic overnight FX swap, settled at T+0. On average during 2021, the implied rate in the FX swap facility was around 22 bps that is lower than the OLF rate which could disincentivize the use of the OLF.

View Full Size

FX Swap Facility: Amount Alloted in Million Dollars by Bank (2020–21)

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: CBUAE

• Download Figure
• Download figure as PowerPoint slide

View Full Size

CBUAE FX Swap Facility AED Implied Rate

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF Mission Calculations

• Download Figure
• Download figure as PowerPoint slide

Reserve Requirements

17. The CBUAE imposes reserve requirements on all deposit-taking licensed financial institutions.¹⁰ The deposit base also includes government deposits. The RR ratio is lower for time deposits (1 percent) than for demand deposits (7 percent). To prevent distortions, the same RR ratio as on local currency deposits applies to foreign currency deposits which account for around 30 percent of total deposits (see Figure 10, upper left panel).

18. The penalty for non-compliance is significant. Banks face a penalty rate of the CBUAE’s base rate plus 400 bp on any shortfall in required reserve—well above the lending facility. The penalty incentivizes a conservative reserve management by banks.

View Full Size

Reserve Requirements in the UAE

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: CBUAE and IMF Mission Calculation

• Download Figure
• Download figure as PowerPoint slide

19. The reserve requirements saw three important changes in 2020:

• Lower RR ratio: As part of the COVID-19 measures, the CBUAE lowered the required reserve ratio on demand deposits from 14 percent to 7 percent on April 5, 2020. The RR ratio remained at 1 percent for time deposits.

• Introduction of the ODF: Since July 12, 2020, banks have the option to hold excess reserves in the ODF which is remunerated at the IOER.

• Extension of the reserve maintenance period: Within the reform of the monetary framework, the CBUAE has extended 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 is allowed over the reserve maintenance period, aiming to simplify liquidity management.

• Fulfillment of the FX reserve requirement in Dirham: With the introduction of the new monetary policy framework, the fulfillment currency for the reserve requirements has changed to dirham for both local and foreign currency required reserves.

20. Required Reserves are unremunerated, but the recent introduction of the ODF offers remuneration on excess reserves at the IOER. Under the new system, banks manage their reserves at the CBUAE between two accounts—the Reserve Account and the ODF. Only reserves held overnight in the ODF are remunerated whereas only those in the Reserve Account count towards fulfilling the RR. Banks choose how much to transfer to the ODF by 8:00 pm each day, one hour before the closing of the payment system. Nevertheless, the transfer to the ODF is usually a bank’s last transaction of the day as liquidity needs for payments are limited at this time of the day.

21. Banks take a conservative approach to meeting the RR and generally hold ample reserves. The overall level of reserves far exceeds the requirement and remains stable over the maintenance period. Bank-level data since the extended RMP does show neither front nor back-loading of reserve fulfillments. However, many banks do not shift funds into the ODF on Saturdays as trading desks are closed that day, over-fulling the RR which allows them to move more funds into the ODF on weekdays. Figure 10, lower right panel, highlights the fulfillment pattern across the maintenance period.

22. The distribution of reserves reflects the heterogeneity of the UAE’s banking system. A few large banks, especially First Abu Dhabi Bank, naturally also play an outsized role in the reserve distribution due to their size. Some smaller banks on the other side, do not appear to manage their reserves very actively on a day-to-day basis, often maintaining large amounts on the reserve account forgoing the remuneration on excess reserves to be obtained in the ODF. Equally, the reserve position varies widely across banks with some banks facing tighter liquidity despite the overall surplus.

C. Recommendations OLF and FX SWAP Facility

23. The mission supports the decision to discontinue the FX Swap Facility. The current one-way SWAP facility is priced at a lower level than the OLF which can create distortions since liquidity provision for the O/N tenor is being priced differently (see Figure 11). Moreover, the pricing of the FX SWAP for the remaining tenors requires from the CBUAE to set the term interest rate differential rather than letting the market price it through the conduct of a price-discovery auction system. In the interim period, the SWAP facility shall remain as a liquidity-provision tool. The CBUAE should announce to the market at least three months before the discontinuation of this facility and once the OLF accept dollars as collateral.

24. The eligible collateral for the OLF should be broadened to include US dollars. Some operating banks in the UAE have long positions in USD and may be at times short in Dirham. The CBAUE may pay an interest for this type of collateral taking these into consideration:

• The rates at which the CBAUE may invest these amounts overnight minus the adequate basis points to compensate the CBAUE for the related operational cost.

• The total rate will be lower than other alternatives for DTIs to prevent the use of this facility as an USD investment vehicle.

View Full Size

Overnight FX Swap versus OLF Forward Points (in basis points)

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: CBUAE and IMF Mission Calculation

• Download Figure
• Download figure as PowerPoint slide

Structural OMOs

25. The mission recommends improving M-bills auctions by:

• Announcing the volume of M-Bills to be auctioned ahead according to the projections of the autonomous factors and allocate it fully,¹¹ without surprising the market. A minimal discretion to reject an out-of-market bid can be retained. However, this should be set through an internal rule without being disclosed to the public. The mission suggests to reject the bids which fall beyond the thresholds defined by the last price or mid price before the auction plus or minus (depending if they are draining or injecting liquidity) the bid-ask average plus two standard deviation divided by 2.

• Evaluating the feasibility of pre-registering and getting the ISIN for the M-Bills one quarter ahead to reduce the eight-day window between the collection of the order book and auction. Currently, the CBUAE monitors the size of the order book eight days before the auction which can alter the quality of the estimations, especially regarding those of the PD’s clients. This piece of information will remain relevant and will need to be crosschecked with the liquidity forecasts.

• Avoiding non-competitive, bilateral operations to the extent possible in order to avoid moral hazard and the undermining of price-setting in the primary auctions. A relaxation of the current 45 percent cap of holdings by a single PD per issuance line should be considered to better reflect the skewedness of liquidity holdings across PDs. Such a relaxation should be conditional on inventories or past market-making activities to properly align incentives.

• Aligning the calendar of M-Bills/ICDs with RMP by allowing the settlement of the M-Bills/ICDs on the initial day of the RMP (see Figure 12).¹²

• Continuing the reoffering of auctioned M-Bills as these operations increase their liquidity in particular for M-Bills with longer tenors (please refer to paragaph 142) .

• Consider increasing the issuance of the three-month tenor M-Bills specifically as it creates an important money market benchmark that also falls within the HQLA classification of many banks, including the current Basel III prudential liquidity requirements.

View Full Size

Current and Suggested Calendar for M-Bills Auction Within Two Consecutive RMPs

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: CBUAE and IMF Mission

• Download Figure
• Download figure as PowerPoint slide

26. The mission recommends maintaining the primary dealer arrangement subject to a tightening of the underwriting obligation. Although this is not a common practice for central bank securities, the benefit of a primary dealers’ program is the development of the secondary market through the obligation of making a two-way quotation. Currently, PDs do not seem to add any markup on their clients’ bids in the primary market Nevertheless, to address the potential risks stemming from an increased two-tier banking system, the underwriting obligation should be made further explicit, for example, by specifying a certain percentage for secondary market operations. The mission welcomes also the planned assessment of the performance of each PD after the inaugural year and suggests covering the difference between the secondary and primary quotes as well as the bid-ask spread of secondary market operations among the assessment criteria.

Fine-Tuning OMOs

27. The mission recommends carrying fine-tuning OMOs at the end of the RMP and/or on any day of the RMP,¹³ whenever deemed necessary (Figure 13). Fine-tuning operations should take place at the opening of the market and the announcement could be done in the afternoon of the previous day in order to allow DTIs to estimate their needs properly. Once banks have gained some operational experience and improved their internal liquidity forecasting process, fine-tuning operations could be announced couple of hours before execution (see Appendix III). The transactions should settle on the same day in order to have an immediate traction on the interbank rates.

View Full Size

Suggested Calendar for Fine-Tuning Auctions

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF Mission

• Download Figure
• Download figure as PowerPoint slide

28. The CBUAE may use FX swaps and Repurchase agreements (Repo) on AED to conduct fine-tuning operations. Several central banks use these instruments for liquidity management (see Appendix III for details). The following considerations should prevail:

• The monetary base coverage rule by foreign reserves should be the primary guide for the choice between FX swap and repo market fine-tuning operations. Repo operations would expand the CBUAE balance sheet while the FX swap would be recorded off balance sheet.

• If this rule is not constraining, the CBUAE should 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 on the amount of liquidity to inject or withdraw from the system and the tenors, the CBUAE should distribute the allocation according to the received bids. At the time of the mission, the operations in the repo market were very scares and the CBUAE may encourage the use of repo transactions to develop this important market building block.

• The auction mechanism for these operations should be transparent and consistent to maximize participation, ensure efficient allocation and promote price discovery. Multiple price auctions generally work better and are more popular in practice. They are more robust to collusion and misinterpretation risks that are more prevalent in developing or less competitive markets, and in practice are very unlikely to generate a spread in prices that would be a cause for concern. On the other hand, single price auctions theoretically reduce the “winner’s curse” and increase the incentive to bid competitively but may have serious drawbacks with increased incentives to collude.

• The CBUAE has developped a framework that provides significant transparency to the market and should continue for the case of the auctions for FX swaps and Repurchase agreements as ex-ante and ex-post transparency are essential. Ex-ante transparency implies the publication of auction rules and access criteria, procedures for the refusal of bids and the size, timing and composition of operations is important to foster an environment that encourages maximum consistent participation from all eligible parties. It is also important in ensuring the results of the operation are accepted by the market as reflective of market conditions. Ex-post transparency includes the publication of the results that include some granular information that could be a reference for market participants for price discovery in underdeveloped financial markets (see Appendix III).

• In the case of CBUAE FX Swap auctions the DTIs will bid in terms of forward points for the term of the swap and in terms of interest rates for the repo auctions. The spot rate for this instrument should be the same as the rate offered in the CBUAE spot facility considering the long leg of the swap. In the case of the Repo transactions all the rates allotted should be in the corridor. Finally, the CBUAE should keep the discretionary right to reject bids that do not represent the market conditions. The main terms of these instruments are provided in Table 2.

^{Table 2.}

United Arab Emirates: OMO Instruments

Source: CBUAE and IMF Mission

^{Table 2.}

United Arab Emirates: OMO Instruments

	Structural OMO	Regular Fine-tuning OMO Instruments
	M-Bill Issuances	Repo	FX swaps
Counterparty	Primary dealers	DTIs	DTIs
Collateral	N/A	A and C classes	USD and AED
Pricing	Auction determined Discounted price	Auction determined Rate quotes	Auctions determined Swap point quotes
Maturity	<= 12 months	<= 7 days	<= 14 days
Frequency of Operation	Fortnightly	Weekly	Weekly
Settlement	T+2	T+0	T+0

Source: CBUAE and IMF Mission

View Table

^{Table 2.}

United Arab Emirates: OMO Instruments

	Structural OMO	Regular Fine-tuning OMO Instruments
	M-Bill Issuances	Repo	FX swaps
Counterparty	Primary dealers	DTIs	DTIs
Collateral	N/A	A and C classes	USD and AED
Pricing	Auction determined Discounted price	Auction determined Rate quotes	Auctions determined Swap point quotes
Maturity	<= 12 months	<= 7 days	<= 14 days
Frequency of Operation	Fortnightly	Weekly	Weekly
Settlement	T+2	T+0	T+0

Source: CBUAE and IMF Mission

D. Reserve Requirements

Reserve Averaging Reduces the Interbank Rate Volatility

35. The extension of the reserves maintenance period to two weeks simplifies liquidity management for banks. The longer maintenance period gives more flexibility to banks to smooth out liquidity shocks. Reserves can act as a buffer and banks do not accommodate swings in liquidity in the market immediately. Reserve averaging effectively flattens the reserve demand curve (see Figure 14). This cushions the interbank rate against high-frequency imbalances in supply and demand for reserves and thus supports the anchoring of the policy rate. However, given the still-short sample and parallel changes to the overall framework, empirical evidence on the extent of the reduction in volatility in the interbank rate is limited. Nevertheless, since the introduction of the ODF shortly followed by the extended RMP, the volatility in the interbank rate has fallen sharply. Moreover, the importance of averaging may become more salient once structural operations reduce the liquidity surplus more.

View Full Size

Demand for Reserves With and Without Averaging

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

36. In the medium-term, the CBUAE could consider extending the RMP even further up to 28 days. Cross-country experience suggests that reserve maintenance periods of four to five weeks are most effective to reap the full benefits of reserve averaging. Once the system has fully adapted to the new operational framework, a gradual extension of the RMP could give banks even more flexibility in response to liquidity shocks. However, an RMP that is too long could undermine the prudential function of the RR, especially for FX RR.

A. Staffing and Organizational Arrangements

39. A useful way to organize the liquidity forecasting process is to assign the forecasting responsibility to a section within the CBUAE Monetary Management Department. This is to ensure, on the one hand, a close link between liquidity forecasting and liquidity management and, on the other hand, establish a clear accountability for the forecasting exercise. The forecasting section/unit is the “center” in which the information about the demand and supply components of bank reserves come together. The information has to be provided by different sources within and outside the CBUAE (e.g., the federal Ministry of Finance, the CBUAE’s foreign exchange operations and accounting departments). The assignment of a liquidity forecasting division would thus comprise: (i) communicating with the different information sources and ensuring the timely receipt of the data; (ii) aggregating the analytical central bank’s balance sheet to obtain the autonomous factors (see Appendix V); (iii) forecasting the individual liquidity components; (iv) supervising the consistency of the forecasted components; (v) preparing an overall liquidity projection to be updated on a daily basis; and (vi) assessing forecasting errors.

40. The CBUAE is severely understaffed, and is not able to forecast liquidity using modern methods and to manage liquidity appropriately. The IMF team has provided the colleagues of the CBUAE with a full-fledged liquidity forecasting framework, using state-of-the-art methodology with a user-friendly interface. However, it is crucial that the CBUAE has in-house econometrics and statistical expertise to fine-tune and adjust the model to new market conditions and structural breaks. 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.

41. The CBUAE needs to upgrade its data management system in order to ensure that databases can be accessed in a timely manner and on a high frequency basis. The liquidity forecasting exercise and calibration of OMOs necessitate the availability of the CBUAE balance sheet on a daily basis, in addition to the detailled monetary operations. The granular time-series should be stored in databases for ease of compilation of the autonomous factors and overall reconciliation (see Appendix VI).

42. The current draft Memorandum of Understanding (MoU) between the CBUAE and the DoF should be extended to the Federal and other local Treasuries from the different Emirates. MoUs should cover the regular exchange of granular information about future Treasuries cash flows, also covering public institutions, and be implemented within one year at latest. Although the net position of the government at the central bank is much smaller than other autonomous factors, it is important for the CBUAE to be regularly informed on the Treasury future cash-flows. The CBUAE has to manage liquidity to neutralize the potential disrupting volatility in the autonomous factors, and forward-looking information from the Treasury is key. The exchange of information should also include Treasury forecasts and methodology, as well as any relevant information for liquidity management purposes.

43. The Federal Treasury should operate a Treasury Single Account (TSA) at the CBUAE to optimize its cash management and enhance the CBUAE’s liquidity management. Currently, the Federal Treasury manages its liquidity in large part through commercial banks, complicating substantially liquidity management for the CBUAE. A TSA would help the Treasury to streamline and strengthens its cash-flows management, while improving the CBUAE’s forecasts of liquidity. The customer’s account management of the CBUAE is equally efficient as in commercial/state-owned banks. The CBUAE should stand ready to remunerate outstanding balances on the government’s account at the ODF rate. If advances were to be granted for non-cash management purposes, those should be granted at the OLF rate. It would also be advisable to apply a TSA for other local treasuries too.

B. Forecasting Framework

44. The IMF team has designed a state-of-the-art, full-fledged forecasting framework for each of the three main autonomous factors, as well as a reconciliation strategy to forecast the sum of the autonomous factors with an estimation of the confidence interval. Over the last 20 years, the statistical literature has made substantial progress over the standard ARIMA model. It has proposed a wide range of different modeling approaches to model seasonality more accurately. Reflecting on these new methodologies, the IMF team provided a complete forecasting framework. This framework tests different families of forecasting models, with an integrated selection of the best performer based on information criteria and out-of-sample performance. It also allows to automatically reparametrize the models to account for structural breaks and new data developments.

45. The mission provided a full-fledged software infrastructure to the authorities using a modern, open-source programming language. The forecasting framework has been entirely programmed under R, the leading free and open-source statistical software used in the forecasting industry. The IMF experts have designed a user-friendly infrastructure to help the authorities produce their forecasts smoothly and quickly. The programming infrastructure is entirely automatized and generates charts and tables on the fly for easy inspection and interpretation. The code is fully documented, using best programming practices to limit errors and bugs. Finally, the IMF experts can provide extra support and training to the authorities on how to use the software, ensuring that the CBUAE can readily produce forecasts for operational purposes.

46. This section is organized as per the following: It first presents the data, then discusses the empirical strategy and the models, for each autonomous factor (currency in circulation (CiC), State Account Balance (SAB), and net foreign assets (NFA)). The section also discusses advanced treatment, such as model averaging and dynamic selection, performance metrics, inference testing, forecast reconciliation, etc. It concludes by recommending a full-fledged empirical strategy for forecasting liquidity in the UAE.

Data Series

47. The data series on UAE currency in circulation is characterized by trends and other non-stationarity features, with strong weekly and monthly patterns and spikes around major religious and other public holidays. Such data are similar to the patterns found in other countries in the region.

View Full Size

Currency in Circulation Data

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: CBUAE and IMF staff

• Download Figure
• Download figure as PowerPoint slide

48. Despite an increasing uptake of cashless transactions, the CiC exhibits an upward trend, which can possibly be attributed to an increase in tourism in the UAE. The weekly and monthly patterns are attributable to pay days, in particular many employees are paid on the 26th of the month. Also, a structural break appears in early 2020 at the onset of the COVID crisis. This may be related to individuals holding a higher level of cash as a precautionary measure or a policy of printing new banknotes due to fears of a higher risk of transmission for older notes.

49. Net foreign assets exhibit strong volatility without clear pattern for the mean. The mean is difficult to predict but some conditional heteroskedasticity is usually present. There is a clear outlier at the beginning of 2016 which can either be excluded from the analysis or handled using a dummy variable.

View Full Size

Net Foreign Assets Data (Top Panel Shows the First Difference and the Bottom Panel Shows the Level)

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: CBUAE and IMF staff

• Download Figure
• Download figure as PowerPoint slide

50. State account balance data exhibit some weekly and monthly seasonality, particularly in the latter half of the sample. Also noticeable are some outliers in early 2020 associated with the onset of the COVID-19 pandemic. These can easily be controlled for using dummy variables.

View Full Size

State Account Balance

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: CBUAE and IMF staff

• Download Figure
• Download figure as PowerPoint slide

51. The IMF team has tested three main families of models to forecast the conditional mean of the autonomous factors: ARIMA models, ETS models, and TBATS models. Standard ARIMA is fitted alongside seasonal ARIMA and ARIMA with regression. For the ETS family models both with and without seasonality are considered. For the net foreign assets data, models for conditional heteroskedasticity are used. The models are presented in details in Appendix VII, including with preliminary fit on the UAE data.

Holiday Effects

53. A significant challenge is modeling the effect of important holidays. In particular, cash in circulation tends to rise in the lead up to holidays. The figure below presents the case of Eid-al-Fitr.

View Full Size

Currency in Circulation Around Eid-al-Fitr

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

While an approach based on dummy variables could be used here, given the smooth rise and fall in cash in circulation around public holidays the following approach is considered. A variable DaysToHoliday (DtH_t) is constructed. Using the example of Eid-al-Adha which fell on August 11 in 2019, the variable DtH_t takes a value of 0 on August 11, a value of 1 on August 12, a value of -1 on August 10, and so on. The Eid-al-Adha effect can be defined as

$$Ea{A}_t=\max (0,(1-Dt{H}_t^2)/7)$$

This formulation creates a smooth hump for seven days around the holiday. Similar variables could be constructed for other major holidays, namely Eid-al_Fitr, UAE National Day, and New Years’ Day. The top panel of Figure 19 shows the contribution of the holiday effect when modelled in this way.

View Full Size

Contribution to Fitted Values of A) Holiday Effects Modelled With a Quadratic Term (Top Panel), B) Monthly Effects Using Fourier Terms (Middle Panel), and C) Weekly Effect Using Dummies (Bottom Panel)

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

Capital Flows

54. In the case of net foreign assets, capital flows data were considered as covariates. These are shown in Figure 20. It should be noted that the mean of net foreign assets is difficult to predict and these covariates did not improve forecasting performance.

View Full Size

Capital Flows Data (From Top to Bottom: Incoming Flows, Outgoing Flows, Incoming Number of Transactions, and Outgoing Number of Transactions)

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

Model Selection (Regressions)

55. The covariates to use in forecasting each autonomous factor were selected by fitting all possible regression specifications. In each case the order of the ARIMA model was selected using the Hyndman and Kandakhar algorithm described in Hyndman and Khandakar (2008). The best model was chosen according to the Akaike Information criterion. The models selected for each autonomous factor and for the total are summarized in the table below:

Regressor	CIC	SAB	Total
Day of week (Dummies)	Yes	Yes	Yes
Day of Month (Fourier)	No	Yes	Yes
Day of Quarter (Fourier)	No	No	No
Day of Religious Year (Fourier)	No	No	No
Day of Civic Year (Fourier)	No	No	Yes
Holiday Effects	Yes	No	Yes
COVID-19 Dummy	Yes	Yes	Yes

View Table

Regressor	CIC	SAB	Total
Day of week (Dummies)	Yes	Yes	Yes
Day of Month (Fourier)	No	Yes	Yes
Day of Quarter (Fourier)	No	No	No
Day of Religious Year (Fourier)	No	No	No
Day of Civic Year (Fourier)	No	No	Yes
Holiday Effects	Yes	No	Yes
COVID-19 Dummy	Yes	Yes	Yes

For NFA, the random walk model was adequate for the mean, the more forecastable component of this time series is the variance. Also, for currency in circulation, and state account balance the best specification was a seasonal ARIMA with m = 7 while for the total, standard ARIMA led to the lowest AIC.

Point Forecasts Comparison

The figure below shows both the mean absolute error (MAE) and the root mean square error (RMSE) for each autonomous factor and the total and for the best model within each model class.

View Full Size

Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) for Forecasting Models on the Total of the Autonomous Factors (Left Panels Show MAE and Right Panels Show RMSE)

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

These figures show how the RMSE naturally increases at longer forecasting horizons. The results for RMSE are also displayed in tabular form for the total of all autonomous factors in the table below. Here all RMSEs are shown relative to the RMSE of the model average, numbers below 1 indicate that a model outperforms the model average.

ARIMA	ETS	ETS (No Seas)	Reg. ARIMA	Seas. ARIMA	TBATS	Mod. Av.
1.035	0.966	0.98	1.013	1.029	1.047	1
1.018	1.025	0.977	1.02	1.04	1.048	1
1.047	0.984	0.989	1.182	1.027	1.125	1
1.023	1.064	1.048	1.028	1.052	1.083	1
1.081	1.116	1.069	0.925	1.088	1.06	1
1.091	1.051	1.099	0.981	1.077	1.046	1
1.168	1.23	1.22	0.821	0.984	1.043	1
1.079	1.244	1.249	1.006	0.905	0.978	1
1.047	1.241	1.189	0.928	1.192	0.811	1
1.027	1.266	1.2	0.911	1.19	0.837	1

View Table

ARIMA	ETS	ETS (No Seas)	Reg. ARIMA	Seas. ARIMA	TBATS	Mod. Av.
1.035	0.966	0.98	1.013	1.029	1.047	1
1.018	1.025	0.977	1.02	1.04	1.048	1
1.047	0.984	0.989	1.182	1.027	1.125	1
1.023	1.064	1.048	1.028	1.052	1.083	1
1.081	1.116	1.069	0.925	1.088	1.06	1
1.091	1.051	1.099	0.981	1.077	1.046	1
1.168	1.23	1.22	0.821	0.984	1.043	1
1.079	1.244	1.249	1.006	0.905	0.978	1
1.047	1.241	1.189	0.928	1.192	0.811	1
1.027	1.266	1.2	0.911	1.19	0.837	1

59. In general, model averaging is very robust at all horizons. Both ARIMA with regression and TBATS can outperform the model average, particularly at longer horizons. There are some instances where ETS models outperform the model average at short horizon. While these results are shown for the total above, similar conclusions can be drawn when looking at the results for CIC and SAB. It should be noted though that using the Nemenyi test for multiple comparisons indicate that even in instances where another model outperforms the model average, this is not statistically significant. To give an indication of this, the figure below summarize the results for one-day ahead forecasts of SAB and 14-day ahead forecasts of CIC respectively. Although an ETS model and ARIMA with regression model are the best models in these two specific cases, the superiority in forecasting performance is not significantly better than any other method whose corresponding interval lies within the grey band. In both cases the model average (as well as other models) lies in this band.

View Full Size

Nemenyi Plot for RMSE for One-Day Ahead Forecasts for SAB

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Nemenyi Plot For RMSE for 14-Day Ahead Forecasts for CiC

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

Interval Forecast Comparisons

60. To evaluate interval forecasts, the IMF team considers two metrics. The first is the hit rate, defined as the proportion of times that the actual value of the forecast falls outside of the prediction interval. The second is the Winkler score defined above.

The hit for the 80 percent prediction interval for different series and methods are shown in the table below. Numbers close to 0.2 indicate a better calibrated interval.

h=1	h=2	h=3	h=6	h=7	h=8	h=9	h=10	h=13	h=14
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

View Table

h=1	h=2	h=3	h=6	h=7	h=8	h=9	h=10	h=13	h=14
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

61. The results indicate that prediction intervals tend to be on the conservative side, particularly for very short-term forecasts. The TBATS model tends to lead to under-coverage. Similar conclusions can be made for the 95 percent prediction intervals.

The Winkler scores for the 80 percent prediction interval (left panels) and the 95 percent prediction interval (right panels) are shown from top to bottom for the CIC, NFA, SAB and Total.

62. When considering the forecast intervals, the ARIMA with regression performs well, particularly at longer horizons, for CIC, SAB and the total sum. For the NFA, all GARCH specifications lead to similar performance. The model average does not perform as well as ARIMA with regression particularly at longer horizons and for the total. The Nemenyi plot for a 95 percent prediction interval for the total of all autonomous factors shows the superior performance of ARIMA with regression relative to the model average is statistically significant.

View Full Size

Winkler Scores for the 80 Percent Prediction Interval (Left) and the 95 Percent Prediction Interval (Right) and From Top to Bottom for the CIC, NFA, SAB, and Total (Each Plot Shows the Performance of Different Forecasting Methods Over All Horizons)

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Nemenyi Plot for Winkler Scores for the 95 Percent Prediction Interval and 14-Day Ahead Forecasts

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

Forecast Reconciliation

63. For the forecast reconciliation, ARIMA with regression is used for Total, CIC and SAB while the eGARCH was used for NFA. For brevity, results comparing different forecast reconciliation methods are presented for the total only.

64. The most robust and best performing reconciliation method is OLS with MinT also yielding similar results. For the point forecasts the difference in forecasting performance is not statistically significant between any of the methods, while for longer horizons the bottom up method of simply adding up the forecasts of individual autonomous factors is significantly worse than all other alternative. This is shown in the figures below.

65. For the prediction intervals, OLS reconciliation significantly dominates all methods at short horizons and all methods bar MinT at longer horizons. This result is presented for the Winkler score for 80 percent prediction intervals.

View Full Size

Nemenyi Plot for RMSE of Different Reconciled and Unreconciled One-Day Ahead Forecasts of the Total

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Nemenyi Plot for RMSE of Different Reconciled and Unreconciled 14-Day Ahead Forecasts of the Total

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Nemenyi Plot for Winkler Score of 80 Percent Prediction Interval of Different Reconciled and Unreconciled One-Day Ahead Forecasts of the Total

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Nemenyi Plot for Winkler Score of 80 Percent Prediction Interval of Different Reconciled and Unreconciled 14-Day Ahead Forecasts of the Total

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

C. Recommendations

67. The CBUAE should forecast all the autonomous factors separately, as well as the net liquidity injection, using the forecasting framework designed by the IMF. Liquidity forecasting relies on projections for all the autonomous factors, and not only currency in circulation. Although NFA and SAB are often more complex to forecast, they can be modeled with the appropriate tools, which will still produce better forecasts than ignoring them or naively use past values.

68. The IMF team recommends using an ARIMA model with regression for forecasting CiC and SAB, and to use an EGARCH model for forecasting NFA. Out-of-sample performance evaluation show that these models dominate the alternatives for the horizon relevant for liquidity management.

69. Model dynamic selection and averaging should be used to improve the forecasting accuracy and reduce the modeling risk. The IMF software infrastructure allows to easily fit a wide range of models and select and average them on the fly while following the best standards in the forecasting literature. Although more complex than univariate forecast based on simple models, the team has shown that this approach offers superior accuracy and reduces the risk of misspecification.

70. The forecast reconciliation model designed by the IMF expert should be used to forecast the net liquidity injection and estimate the associated confidence interval. An important contribution of the IMF modeling framework is to use a forecast reconciliation approach to refine the forecast of the aggregate autonomous factors, instead of naively summing them. This approach also allows to estimate the confidence interval of the net liquidity injection, which can then be used to calibrate the required reserves.

71. The CBUAE could enhance the forecasts by incorporating further the expertise and granular knowledge of the CBUAE experts. This expertise can improve further the construction and selection of indicators for special events and policy events in the models. Furthermore, the IMF team has followed an agnostic approach, building on information that practitioners could extract solely from the data. Contextual data and expertise can further refine modeling choices to improve forecast further.

72. The CBUAE should publish the whole liquidity forecast and associated methodology to help anchoring market expectations and support liquidity management of market participants. The CBUAE has now 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 help strengthening the liquidity management by market participants and help the market to form consistent expectations about short-term liquidity changes.

73. The CBUAE should staff a small team of econometricians to estimate and fine-tune the liquidity forecasting framework. Given the importance of liquidity forecasts for efficient liquidity management, and the increasing complexity of the UAE financial system, it is crucial that the CBUAE has in-house expertise in liquidity forecasting.

74. 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.

75. 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 as an international best practice for Treasury cash-flow management and central bank liquidity management.

A. Conceptual Framework: Optimal Liquidity Surplus in a Floor

76. In a floor system, the central bank should maintain enough liquidity surplus to anchor the interbank rate at the policy rate, which is the 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 that banks are able to settle their payments. However, the aggregate precautionary demand from banks is not necessarily the optimal level of reserves, as the level of incompressible demand for reserves doesn’t guarantee that the interbank rate will be anchored to the policy rate (i.e., the overnight deposit facility). To anchor the interbank rate at the policy rate, the central bank should supply enough reserves to stimulate the deposits of commercial banks at the overnight deposit facility, to stabilize the interbank rate at the floor and to avoid excessive volatility to occur.

77. The central bank faces a tradeoff between anchoring the interbank rate at the policy rate and supporting money market development and building the sovereign yield curve. While an excessive amount of structural liquidity generally anchors the interbank rate well, it has a detrimental impact on interbank money market trading. As banks are more and more uniformly excessively liquid, less and less banks need to borrow reserves on the market. Consequently, banks cannot place their excess reserves on the market and should either deposit them at the ODF or purchase central bank securities. This situation currently occurs in the UAE; as the CBUAE has been operating with an increasing level of excess reserves, the interbank market turnover has shrunk over the last two years by around two thirds of the historical average. Also, sterilizing liquidity through the issuance of central bank securities is important in a country lacking Treasury instruments to build the sovereign yield curve. Therefore, the central bank should sterilize enough liquidity via securities to support capital market development and yield curve pricing.

View Full Size

Excess Reserves and Interbank Turnover (All Tenors)

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff and the CBUAE

• Download Figure
• Download figure as PowerPoint slide

View Full Size

UAE Survey: Placement of Banks’ Liquidity

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff and the CBUAE

• Download Figure
• Download figure as PowerPoint slide

78. The market dynamic of the interbank rate spread with the policy rate and the objectives of the central bank should determine the optimal amount of liquidity surplus. The IMF team proposes a conceptual framework relying on: (i) a model of the demand and supply of central bank’s reserves; and (ii) a formalization of the central bank monetary objectives. The conceptual framework serves as a basis for the empirical analysis presented below.

79. The model of precautionary demand for reserves is based on Poole (1968) and is customized to the specificities of the UAE, in particular, as an oil producing country. In Poole (1968), the demand for excess reserves is part of a bank’s liquidity management decision. Banks want to hold excess reserves to avoid overdraft or reserve deficiency penalties on their account at the central bank when facing uncertain flows of funds. The model’s predictions are twofold: (i) the quantity of excess reserves demanded varies inversely with the opportunity cost of holding reserves; and (ii) demand of excess reserves increases with uncertainty about reserves balances. It is usually assumed that uncertainty increases with payment activity (Furfine, 2000) and falls with the level of reserves requirements (Dow, 2000).

80. Building on this literature the behavior of excess reserves over a maintenance period of length t = 1,…., T days can be described as follows. Commercial banks demand central bank reserves $E_t^d$ for precautionary purposes, mostly to settle payments with other financial institutions, while arbitraging opportunities, domestically and internationally:

Hence $E_t^d$ sdepends on:

• The money market interbank spread with the policy rate $(i_t-\overline{\iota })$. In the case of the UAE, the policy rate is equal to the US IOER.¹⁷

• A set of stochastic exogeneous factors ω_t such as oil prices, seasonal effects, etc.

• The fulfillment of statutory reserves R over the maintenance period, depending on the averaging strategy m() used by the banks

• The spread between the US T-Bills and the policy rate $(r_t-\overline{\iota })$ capturing an alternative investment opportunity that banks have.

$E_t^d\ge E_t^i$ is the incompressible demand for reserves. The reserve requirement is met for the maintenance period if the cumulative sum of actual reserves on the last day of the maintenance period at least equals the sum of required reserves over the period ${\Sigma }_{t=1}^T{R}_t\ge T*\overline{R}$.

On each day of the maintenance period excess reserves are calculated as reserves account balances in excess of the required amount

81. In line with specification (1), Furfine (2000) develops a model of bank reserves management that incorporates payment flows. In his model, payment flows are positively correlated with reserves balance uncertainty. Uncertainty generates a precautionary demand for reserves. As a result, interbank rates are higher on days characterized by higher payment flows. This result is empirically confirmed by Acharya and Merrouche (2012) in the United Kingdom context. In terms of specification (1), it implies that the spread should vary positively with ω_t.

82. As shown in Acharya and Merrouche (2012) any factor that causes banks to hoard liquidity for precautionary reasons would push rates and volatility upward. And any factor that pushes the undesired level of excess reserves upward, including a rise in capital inflows (AF) together with insufficient interventions by the central bank to absorb enough excess liquidity via the issuance of securities (B) is expected to cause rates to go down.

83. Further the model includes a full range of calendar effects. An extensive literature has documented that the interbank rate does not follow a martingale. Rates vary predictably on different days of the maintenance period (see Hamilton, 1996) and Bartolini et al. (2002). And volatility is higher at the end of the maintenance period. Whitesell (2006) shows that period-average requirements create incentives for banks to actively manage liquidity contributing to lower spreads and volatility, but this effect attenuates toward the end of the maintenance period. In a system with abundant liquidity, it is expected that these patterns in the interbank rate over the maintenance period are not observed and that the benefit of reserve averaging are not as strong unless reserves are unevenly distributed across banks.

84. The results of a banks’ survey conducted by the CBUAE and the IMF mission confirm that excess reserves are held in priority to meet unpredictable payment flows. The penalty charged for shortfalls in reserves requirements is among the top four reasons for holding excess reserves (also ELAR). Another factor that leads banks to hold more reserves is that reserves count toward the fulfillment of liquidity requirements.

View Full Size

Survey Results: Reasons for Holding Excess Reserves

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff and the CBUAE

• Download Figure
• Download figure as PowerPoint slide

85. The supply of banks’ reserves is determined by the autonomous factors and the policy position of the central bank. The net foreign assets, net position of the government, other items net inject liquidity while currency in circulation absorbs liquidity. Then, the central bank can adjust the structural liquidity by conducting open market operations, as well as operating standing facilities and ELA.

View Full Size

Supply of Banks’ Reserves

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

86. The IMF team formalizes the central bank supply of reserves by using the accounting equality of the central bank stylized balance sheet. On the central bank assets’ side, NFA_t represents the net foreign assets, NLG_t the net lending to the government and ONDA_t the other net domestic assets. On the liabilities side, the central bank has currency in circulation (CIC_t), bank’s excess reserves $\left(E_t^s\right)$, required reserves (R and securities (B_t).

The notations are simplified and the autonomous factors are expressed as AF_t = NFA_t + NLG_t + ONDA_t — CIC_t. From the accounting equality, the supply of reserves can be expressed as the difference between the liquidity injected by the autonomous factors and the liquidity absorbed by the central bank:

View Full Size

Structural Liquidity Position: Autonomous Factors and Banks’ Reserves

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff and the CBUAE

• Download Figure
• Download figure as PowerPoint slide

87. The clearing market condition between demand and supply of reserves determines the interest spread dynamic. At the equilibrium, $E_t^s=E_t^d$, which implies:

Where the notation is simplified: $M\left(\underset{¯}{R}\right)=m({\Sigma }_{j=1}^t{R}_j-\underset{¯}{R})-\underset{¯}{R}$

88. The structural functions f,g,h, m are potentially non-linear, difficult to estimate and to invert. Ideally, a structural model on banks’ demand based on granular micro data should be used, but even with such data, the true demand of reserves from banks is confounded with other effects, in particular market microstructure. For these reasons, the IMF team has implemented a reduced-form approach (see below).

89. The IMF mission models the interest rate spread dynamic as a highly non-linear function articulated around three phases. f^-1 reflects:

• The incompressible demand for reserves and the banks’ settlement risk. When the reserves are below the incompressible demand, banks stop trading as they can’t even settle trades.

• An active interbank trading market with volatile interest rate. Above the incompressible demand for reserves but below the optimal level, the relative scarcity of reserves is driving the rates above the policy rate.

• The liquidity trap when there is too much excess reserves. When reserves are in excessive amount, banks have little incentives to trade and park their liquidity at the ODF. The interbank market shrinks.

View Full Size

Theoretical Interest Rate Spread Dynamic

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

90. The theoretical specification of the interest rate spread dynamic is consistent with the UAE data. Over the last five years of observations, there is indeed a strong, nonlinear relationship between the interbank spread and the level of excess reserves. When the level of excess reserves is below AED 25 bn, the DONIA spread surges quickly, and exponentially as the level of reserves approaches the level of incompressible demand.

View Full Size

Interbank Rate Spread with the Policy Rate and Excess Reserves

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

91. The optimal level of liquidity surplus derives from the central bank objectives, which are twofold. The IMF team assumes that the central bank decides upon the volume of structural liquidity to:

• Maximize the monetary policy traction (i.e., to anchor the interbank rate at the policy rate).

• Support the interbank money market development and build a sovereign yield curve.

However, anchoring the interbank rate has the highest priority as a monetary objective. Therefore, the optimal amount of reserves should reflect this priority by first making sure that the interbank rate is anchored and only then to give the maximize room possible for market trading.

The optimal amount of reserves is the minimal amount of excess reserves for which the interest rate spread is, in absolute terms, minimal:

92. Once the optimal liquidity surplus $E_t^{*}$ is known, the optimal amount of liquidity absorption is the difference between the expected autonomous factors and $E_t^{*}$:

Intuitively, the central bank should: (i) sterilize the expected liquidity injected by the autonomous factors; but (ii) leave enough liquidity to the banks to meet their incompressible demand for reserves and support market trading and limit interbank trading volatility.

B. Empirical Estimation of the Optimal Liquidity Surplus

93. The empirical strategy aims at recovering the true relationship between excess reserves and the interbank rate. The conceptual framework and the descriptive statistics suggest that the functional relationship between the excess reserves and the interbank spread is highly non-linear, with a complex exponential decay shape. For that reason, it is not possible to use standard linear regressions. The IMF team presents two approaches, relying either on: (i) unconditional fit, via cubic splines or parametrized functional form; and (ii) a conditional fit, instrumented and cross-validated. The IMF experts have a preference for parametrized, even over-parametrized models, to guarantee parametric stability, which is crucial for operational purposes, as the CBUAE needs stable calibration of its operations. Once the functional relationship has been properly estimated, it is straightforward to recover the different phases.

94. The raw data exhibits a large mass point on the bottom left of the diagram, making necessary a data treatment. The mass point makes the direct fit, either conditional or unconditional, difficult and biased without treatment. It is possible to use a special estimator to tackle this issue (for instance, a rescaled Tobit model), or aggregate the data to flatten the distribution mass over the support. The IMF team prefers the latter, as it is a simple and robust way of operation. The IMF experts have implemented two types of data aggregation: (i) max-min dispersion of the policy rate spread for each level of reserves (discretized each half AED billion); and (ii) a weighted average of the DONIA spread for the same level of reserves, weighted by the market turnover on that day. The shape of the scatter plot under both approaches is broadly similar and allows for a balanced fit over the support of excess reserves.

95. The IMF team operationalizes the CBUAE objective into a practical, data-oriented condition. The conceptual framework defines the optimal level of excess reserves as the minimum level of excess reserves for which the interbank rate is anchored to the policy rate.

View Full Size

Data Aggregation by Max-Min Dispersion and Weighted Average

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

96. In practice, the interbank rate can not be exactly at the policy rate, as there is a constant market noise. The market rate is technically hovering around the policy rate in a region delimited by a few basis points. The optimal length of the region is defined as the level for which the volatility of the interbank rate is contained within the tolerance zone for most observations (e.g., more than 90 percent). The determination of the market tolerance zone follows the empirical distribution function of the daily DONIA spread (raw data). The IMF team chooses 25 bps, which corresponds roughly to the lowest 20^th percentile of the empirical cumulative distribution function.

97. The IMF team estimates an unconditional fit of the functional relationship between the excess reserves and the interbank spread, using two methods. The first method is a non-parametric fit via cubic splines. Cubic splines are very flexible, yet smooth enough to provide a reasonable fit. The second method is based on a parametrized fit via a decreasing double exponential functional form. A double exponential composes and exponential function and a power function. More precisely, the decreasing double exponential function is specified as:

View Full Size

Empirical Distribution of the Interbank Spread on the Left Percentile

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

98. The decreasing double exponential provides a very sharp decay rate and a quick stabilized shape. It is parametrized by three variables, the location α, the scale β and the decay rate λ providing plenty of flexibility to accommodate different datasets. The estimation of the three parameters in the unconditional setup is carried out by non-linear least squares with the Levenberg-Marquardt algorithm, which is relatively robust to initial conditions and involves limited computation time compared to more intensive algorithms.

View Full Size

Calibration of the Optimal Liquidity Surplus via Unconditional Fit

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

99. Under both the unconditional non-parametric and the parametric approaches, the optimal liquidity surplus is estimated between AED 50–60 bn. The optimal liquidity surplus depends on the aggregation method as well as the fit algorithm chosen. While the cubic fit is poor on the max-min dispersion aggregation method, it works relatively well on the weighted average metric. After the AED 50–60 bn threshold, most of the observations exhibit an interbank spread within the tolerance band of 25 bps.

100. Important factors impact the reaction function of the interbank spread to the excess reserves, suggesting that the fit should be conditional instead of unconditional. The bubble plot presents the relationship between the two main variables—interbank spread and excess reserves—as well as prevailing interbank market turnover and the three-month realized volatility of the DONIA. The figure suggests that the very large amount of excess reserves is associated with tiny interbank market turnover and low volatility, reflecting low trading activity as explained above. On the contrary, the observations with the largest spreads have occurred both under large turnover and high volatility. These elements suggest that conditional fit is crucial to properly recover the true functional shape of the reaction of the interbank spread to the excess reserves.

View Full Size

Interbank Spread, Volatility, and Liquidity

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

101. The price quantity relationship between the excess reserves and the DONIA spread is a typical example of endogeneity simultaneity and should be instrumented. Since the price (the interbank rate) partially determines the quantity (excess reserves available on the market) and vice-versa, the relationship is likely endogenous. Given the lack of natural experiment or experimental set-up, the only viable solution is to instrument the excess reserves by an exogeneous variable. The autonomous factors are natural candidates: the team uses both the level and the volatility of the sum of the autonomous factors, to design an over-identified regression set-up.

102. The IMF team estimates a conditional fit model, controlling for seasonal effects, market microstructure variables, uncovered interest parity and exogeneous risk sentiment variables. As suggested by the conceptual framework, these variables enter either the demand or the supply of reserves (or both), and therefore impact the determination of the interbank spread with the policy rate. Therefore, the mission has used the following variables as control variables:

• Seasonal factors: religious and secular holidays, regular calendar seasonality.

• Market turnover: capture market microstructure.

• Fed IOER: capture the uncovered interest parity and reflects international arbitrage.

• Oil prices (both in level and volatility): capture both the precautionary demand and the drivers of the autonomous factors.

• VIX to capture international risk sentiment.

• The Dubai stock market realized volatility and the Donia realized volatility to capture domestic risk sentiment.

103. The instrumented conditional fit estimation relies on an IV-2SLS algorithm with cross-validation.¹⁸ A non-linear least-square set-up—as used for the conditional fit—would be very inconvenient in the context of a two-step instrumentation procedure. Instead, it is easier to impose the parametric form on the level of excess reserves directly, and to use a bijective transformation of the regressor instead of the regressor itself. The specification is therefore:

While most coefficients can be directly estimated via a standard IV-SLS procedure, the decay coefficient λ cannot. Hence, to circumvent this issue, the IMF team uses a cross-validation two-step approach:

• 1. Estimate via IV-2SLS α, β, γ for a given level of the parametric decay rate λ

• 2. Repeat the process over a grid of potential λ and select the best one on maximizing an information criterion or minimizing the residuals sum of squares.

The cross-validation is key to be able to estimate a complex non-linear functional form with a standard linear IV-2SLS estimator.

104. The cross-validated IV-2SLS regressions exhibit very decent in-sample fit—with R2 between 65 and 78 percent for max-min dispersion and weighted average respectively, and a substantial the regressors are statistically significant at the 5^th or at the 10^th percentile threshold. Besides, the first stage of the IV-2SLS has an F-statistic near 10, suggesting a relatively decent fit on the first stage and adequate endogeneity control. The estimated results of the regressions are presented in the Python regression table below.

View Full Size

Estimated Coefficients of the CV IV-2SLS for the Max-Min Spread Dispersion

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Estimated Coefficients of the CV IV-2SLS for the Weighted Average Dispersion

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

105. The IMF team presents the conditional fit in a univariate set-up, using the estimated residuals of the DONIA spread on the exogeneous regressors instead of the raw level of the DONIA. Technically, while the optimal liquidity surplus can be estimated directly from the estimated coefficients of the cross-validated IV-2SLS regression, it is more intuitive to present the univariate non-linear fit of the excess reserves on the residuals of the dependent variable on the other regressors, so that the impact of excess reserves can be clearly represented. This is one potential application of the Frisch-Waugh theorem.¹⁹ The approach is valid as the Frisch-Waugh theorem is a bijective mapping on the residual space, so that the optimal level of excess reserves in the two-step fit is the same as under the one-step approach.

View Full Size

Liquidity Calibration via Cross-Validated Instrumented Conditional Fit (for Max-Min Spread and Weighted Average Aggregation)

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff

• Download Figure
• Download figure as PowerPoint slide

106. Under both aggregation approaches, the optimal liquidity surplus estimated through the conditional cross-validated IV-2SLS fit is around AED 50 bn. The optimal liquidity surplus is slightly lower than the one estimated via the unconditional fit, reflecting the importance of omitted variables in the unconditional estimation. Both methods show that at AED 50 bn, the interbank rate is anchored to the policy rate with minimal interbank rate volatility. Above this level, there is no room to anchor further the interbank rate to the policy rate but leaving excessive liquidity on the market would be detrimental for market development and yield curve building.

107. Through unconditional and conditional fitting approaches, applied on two different data aggregation methods, the IMF experts have determined that the CBUAE should operate a parsimonious floor with excess liquidity in the magnitude of AED 50 bn. However, it is still unclear how the CBUAE should sterilize excess liquidity to reach a structural surplus of AED 50 bn. How should the CBUAE calibrate the mix between the different instruments—the required reserves, the issuance of long-term securities and the issuance of short-term securities? The IMF team has designed a conceptual and empirical framework to answer this question, detailed in the next section.

C. Optimal Operational Sterilization Mix

108. The CBUAE has different instruments to sterilize liquidity, each of them with specific features. It can use three instruments to sterilize liquidity: (i) the level of statutory reserves R with averaging gives the possibility to banks to smooth their liquidity across the reserves maintenance period, thereby reducing the interbank market volatility (see Bindseil 2014 for a discussion on the impact of reserves averaging on the interbank rate); (ii) long-term (up to one year) central bank securities $B_t^{LT}$ are key to develop the capital markets and benchmark the yield curve, with regular issuance in sufficient amount; and (iii) short-term instruments $B_t^{ST}$ should quickly reflect the evolution of liquidity conditions, with minimal duration risk. In practice, the CBUAE issues different types of instruments (M-Bills, Islamic CD) at different maturities and has a remaining stock of outstanding CDs which have been discontinued, but for the sake of simplicity, the IMF team only differentiates between long-term (structural) issuances and short-term issuances. By definition, the total level of CBUAE securities issued is equal to $B_t^{LT}+B_t^{ST}=B_t$.

View Full Size

Current UAE Structural Liquidity Position

Citation: IMF Staff Country Reports 2022, 236; 10.5089/9798400215759.002.A001

Source: IMF staff and CBUAE

• Download Figure
• Download figure as PowerPoint slide

109. The optimal level of statutory reserves should depend on the volatility of the autonomous factors. In practice, the CBUAE decides its securities issuances based on its forecasts of the autonomous factors, and not on the realized (unknown at time t-5) autonomous factors. Hence, from the accounting equality, the optimal level of securities that the CBUAE will issue is equal to:

Where ℑ(AF_t) represents the forecasted autonomous factors. However, the optimal amount of liquidity absorption depends on the true expected values, and not the forecasted ones, potentially noisy. Hence, based on the clearing condition, where the interbank spread with the policy rate is 0 on average around the optimal liquidity surplus $E_t^{*}$:

Combining the CBUAE sterilization objective with the optimal sterilization level gives the optimal level of required reserves the central bank should impose on banks: