Definitions and Interpretations

The following terms shall have the meaning assigned to them for the purpose of interpreting these Standards and the related Guidance:
1.	Board: As defined in the CBUAE’s Corporate Governance Regulation for Banks.
2.	Causality (written in lower case as “causality”): Relationship between cause and effect. It is the influence of one event on the occurrence of another event.
3.	CBUAE: Central Bank of the United Arab Emirates.
4.	Correlation (written in lower case as “correlation”): Any statistical relationship between two variables, without explicit causality explaining the observed joint behaviours. Several metrics exist to capture this relationship. Amongst others, linear correlations are often captured by the Pearson coefficient. Linear or non-linear correlation are often captured by the Spearman’s rank correlation coefficient.
5.	Correlation Analysis (written in lower case as “correlation analysis”): Correlation analysis refers to a process by which the relationships between variables are explored. For a given set of data and variables, observe (i) the statistical properties of each variable independently, (ii) the relationship between the dependent variable and each of the independent variables on a bilateral basis, and (iii) the relationship between the independent variables with each other.
6.	CI (Credit Index): In the context of credit modelling, a credit index is a quantity defined over (-∞,+∞) derived from observable default rates, for instance through probit transformation. CI represents a systemic driver of creditworthiness. While this index is synthetic, (an abstract driver), it is often assimilated to the creditworthiness of specific industry or geography.
7.	Default (written in lower case as “default”): The definition of default depends on the modelling context, either for the development of rating models or for the calibration and probabilities of default. For a comprehensive definition, refer to the section on rating models in the MMG.
8.	Deterministic Model (written in lower case as “deterministic model”): A deterministic model is a mathematical construction linking, with certainty, one or several dependent variables, to one or several independent variables. Deterministic models are different from statistical models. The concept of confidence interval does not apply to deterministic models. Examples of deterministic models include NPV models, financial cash flow models or exposure models for amortizing facilities.
9.	DMF (Data Management Framework): Set of policies, procedures and systems designed to organise and structure the management of data employed for modelling.
10.	DPD (Days-Past-Due): A payment is considered past due if it has not been made by its contractual due date. The days-past-due is the number of days that a payment is past its due date, i.e. the number of days for which a payment is late.
11.	DSIB (Domestic Systemically Important Banks): These are UAE banks deemed sufficiently large and interconnected to warrant the application of additional regulatory requirements. The identification of the institutions is based upon a framework defined by the CBUAE.
12.	EAD (Exposure At Default): Expected exposure of an institution towards an obligor (or a facility) upon a future default of this obligor (or its facility). It also refers to the observed exposure upon the realised default of an obligor (or a facility). This amount materialises at the default date and can be uncertain at reporting dates prior to the default date. The uncertainty surrounding EAD depends on the type of exposure and the possibility of future drawings. In the case of a lending facility with a pre-agreed amortisation schedule, the EAD is known. In the case of off-balance sheet exposures such as credit cards, guarantees, working capital facilities or derivatives, the EAD is not certain on the date of measurement and should be estimated with statistical models.
13.	EAR (Earning At Risk): Refer to NII.
14.	ECL (Expected Credit Loss): Metric supporting the estimation of provisions under IFRS9 to cover credit risk arising from facilities and bonds in the banking book. It is designed as a probability-weighted expected loss.
15.	Economic Intuition (written in lower case as “economic intuition”): Also referred to as economic intuition and business sense. Property of a model and its output to be interpreted in terms and metrics that are commonly employed for business and risk decisions. It also refers to the property of the model variables and the model outputs to meet the intuition of experts and practitioners, in such a way that the model can be explained and used to support decision-making.
16.	Effective Challenge: Characteristic of a validation process. An effective model validation ensures that model defects are suitably identified, discussed and addressed in a timely fashion. Effectiveness is achieved via certain key features of the validation process such as independence, expertise, clear reporting and prompt action from the development team.
17.	EVE (Economic Value of Equity): It is defined as the difference between the present value of the institution’s assets minus the present value of liabilities. The EVE is sensitive to changes in interest rates. It is used in the measurement of interest rate risk in the banking book.
18.	Expert-Based Models (written in lower case as “expert-based models”): Also referred to as judgemental models, these models rely on the subjective judgement of expert individuals rather than on quantitative data. In particular, this type of model is used to issue subjective scores in order to rank corporate clients.
19.	Institutions (written in lower case as “institution(s)”): All banks licenced by the CBUAE. Entities that take deposits from individuals and/or corporations, while simultaneously issuing loans or capital market securities.
20.	LGD (Loss Given Default): Estimation of the potential loss incurred by a lending institution upon the default of an obligor (or a facility), measured as a percentage of the EAD. It also refers to the actual loss incurred upon past defaults also expressed as a percentage of EAD. The observed LGD levels tend to be related to PD levels with various strength of correlation.
21.	Limits and limitations (written in lower case as “limits” and “limitations”): Model limits are thresholds applied to a model’s outputs and/or its parameters in order to control its performance. Model limitations are boundary conditions beyond which the model ceases to be accurate.
22.	LSI (Large and/or Sophisticated Institutions): This group comprises DSIBs and any other institutions that are deemed large and/or with mature processes and skills. This categorisation is defined dynamically based on the outcome of regular banking supervision.
23.	Macroeconomic Model (written in lower case as “macroeconomic model” or “macro model”): Refers to two types of models. (i) A model that links a set of independent macro variables to another single dependent macro variable or to several other dependent macro variables or (ii) a model that links a set of independent macro variables to a risk metric (or a set of risk metrics) such as probabilities of default or any other business metric such as revenues.
24.	Market Data: Refers to the various data attributes of a traded financial instrument reported by a trading exchange. It includes the quoted value of the instrument and/or the quoted parameters of that instrument that allow the derivation of its value. It also includes transaction information including the volume exchanged and the bid-ask spread.
25.	Materiality: The materiality of a model represents the financial scope covered by the model in the context of a given institution. It can be used to estimate the potential loss arising from model uncertainty (see Model Risk). Model materiality can be captured by various metrics depending on model types. Typically, total exposure can be used as a metric for credit models.
26.	MMG: CBUAE’s Model Management Guidance.
27.	MMS: CBUAE’s Model Management Standards.
28.	Model (written in lower case as “model”): A quantitative method, system, or approach that applies statistical, economic, financial, or mathematical theories, techniques, and assumptions to process input data into quantitative estimates. For the purpose of the MMS and MSG, models are categorised in to three distinct groups: statistical models, deterministic models and expert-based models.
29.	Model Calibration (written in lower case as “model calibration”): Key step of the model development process. Model calibration means changing the values of the parameters and/or the weights of a model, without changing the structure of the model, i.e. without changing the nature of the variables and their transformations.
30.	Model Complexity (written in lower case as “model complexity”): Overall characteristic of a model reflecting the degree of ease (versus difficulty) with which one can understand the model conceptual framework, its practical design, calibration and usage. Amongst other things, such complexity is driven by, the number of inputs, the interactions between variables, the dependency with other models, the model mathematical concepts and their implementation.
31.	Model Construction (written in lower case as “model construction”): Key step of the model development process. The construction of a model depends on its nature, i.e. statistical or deterministic. For the purpose of the MMS and the MMG, model construction means the following: for statistical models, for a given methodology and a set of data and transformed variables, it means estimating and choosing, with a degree of confidence, the number and nature of the dependent variables along with their associated weights or coefficients. For deterministic models, for a given methodology, it means establishing the relationship between a set of input variables and an output variable, without statistical confidence intervals.
32.	Model Development (written in lower case as “model development”): Means creating a model by making a set of sequential and recursive decisions according to the steps outlined in the dedicated sections of the MMS. Model re-development means conducting the model development steps again with the intention to replace an existing model. The replacement may, or may not, occur upon re-development.
33.	Modelling Decision (written in lower case as “modelling decision”): A modelling decision is a deliberate choice that determines the core functionality and output of a model. Modelling decisions relate to each of the steps of the data acquisition, the development and the implementation phase. In particular, modelling decisions relate to (i) the choice of data, (ii) the analysis of data and sampling techniques, (iii) the methodology, (iv) the calibration and (v) the implementation of models. Some modelling decisions are more material than others. Key modelling decisions refer to decisions with strategic implications and/or with material consequences on the model outputs.
34.	Model Risk: Potential loss faced by institutions from making decisions based on inaccurate or erroneous outputs of models due to errors in the development, the implementation or the inappropriate usage of such models. Losses incurred from Model Risk should be understood in the broad sense as Model Risk has multiple sources. This definition includes direct quantifiable financial loss but also any adverse consequences on the ability of the institution to conduct its activities as originally intended, such as reputational damage, opportunity costs or underestimation of capital. In the context of the MMS and the MMG, Model Risk for a given model should be regarded as the combination of its materiality and the uncertainty surrounding its results.
35.	Model Selection (written in lower case as “model selection”): This step is part of the development process. This means choosing a specific model amongst a pool of available models, each with a different set of variables and parameters.
36.	Model Uncertainty (written in lower case as “model uncertainty”): This refers to the uncertainty surrounding the results generated by a model. Such uncertainty can be quantified as a confidence interval around the model output values. It is used as a component to estimate Model Risk.
37.	Multivariate Analysis (written in lower case as “multivariate analysis”): For a given set of data and variables, this is a process of observing the joint distribution of the dependent and independent variables together and drawing conclusions regarding their degree of correlation and causality.
38.	NII (Net Interest Income): To simplify notations, both Net Interest Income (for conventional products) and/or Net Profit Income (for Islamic Products) are referred to as “NII”. In this context, ‘profit’ is assimilated as interest. It is defined as the difference between total interest income and total interest expense, over a specific time horizon and taking into account hedging. The change in NII (“?NII”) is defined as the difference between the NII estimated with stressed interest rates under various scenarios, minus the NII estimated with the interest rates as of the portfolio reporting date. ?NII is also referred to as earnings at risk (“EAR”).
39.	NPV (Net Present Value): Present value of future cash flows minus the initial investment, i.e. the amount that a rational investor is willing to pay today in exchange for receiving these cash flows in the future. NPV is estimated through a discounting method. It is commonly used to estimate various metrics for the purpose of financial accounting, risk management and business decisions.
40.	PD (Probability of Default): Probability that an obligor fails to meet its contractual obligation under the terms of an agreed financing contract. Such probability is computed over a given horizon, typically 12 months, in which case it is referred to as a 1-year PD. It can also be computed over longer horizons. This probability can also be defined at several levels of granularity, including, but not limited to, single facility, pool of facilities, obligor, or consolidated group level.
41.	PD Model (written as “PD model”): This terminology refers to a wide variety of models with several objectives. Amongst other things, these models include mapping methods from scores generated by rating models onto probability of defaults. They also include models employed to estimate the PD or the PD term structure of facilities, clients or pool of clients.
42.	PD Term Structure (written as “PD term structure”): Refers to the probability of default over several time horizons, for instance 2 years, 5 years or 10 years. A distinction is made between the cumulative PD and the marginal PD. The cumulative PD is the total probability of default of the obligor over a given horizon. The marginal PD is the probability of default between two dates in the future, provided that the obligor has survived until the first date.
43.	PIT (Point-In-Time) and TTC (Through-The-Cycle): A point-in-time assessment refers to the value of a metric (typically PD or LGD) that incorporates the current economic conditions. This contrasts with a through-the-cycle assessment that refers to the value of the same metric across a period covering one or several economic cycles.
44.	Qualitative validation: A review of model conceptual soundness, design, documentation, and development and implementation process.
45.	Quantitative validation: A review of model numerical output, covering at least its accuracy, degree of conservatism, stability, robustness and sensitivity.
46.	Rating / Scoring (written in lower case “rating or scoring”): For the purpose of the MMS and the MMG, a rating and a score are considered as the same concept, i.e. an ordinal quantity representing the relative creditworthiness of an obligor (or a facility) on a predefined scale. ‘Ratings’ are commonly used in the context of corporate assessments whilst ‘scores’ are used for retail client assessments.
47.	Restructuring (written in lower case “restructuring”): The definition of restructuring / rescheduling used for modelling in the context of the MMS and MMG should be understood as the definition provided in the dedicated CBUAE regulation and, in particular, in the Circular 28/2010 on the classification of loans, with subsequent amendments to this Circular and any new CBUAE regulation on this topic.
48.	Rating Model (written in lower case “rating model”): The objective of such model is to discriminate ex-ante between performing clients and potentially non-performing clients. Such models generally produce a score along an arbitrary scale reflecting client creditworthiness. This score can subsequently mapped to a probability of default. However, rating models should not be confused with PD models.
49.	Retail Clients (written in lower case as “retail clients”): Retail clients refer to individuals to whom credit facilities are granted for the following purpose: personal consumer credit facilities, auto credit facilities, overdraft and credit cards, refinanced government housing credit facilities, other housing credit facilities, credit facilities against shares to individuals. It also includes small business credit facilities for which the credit risk is managed using similar methods as applied for personal credit facilities.
50.	Segment (written in lower case as “segment”): Subsets of an institution’s portfolio obtained by splitting the portfolio by the most relevant dimensions which explain its risk profile. Typical dimensions include obligor size, industries, geographies, ratings, product types, tenor and currency of exposure. Segmentation choices are key drivers of modelling accuracy and robustness.
51.	Senior Management: As defined in the CBUAE’s Corporate Governance Regulation for Banks.
52.	Statistical Model (written in lower case as “statistical model”): A statistical model is a mathematical construction achieved by the application of statistical techniques to samples of data. The model links one or several dependent variables to one or several independent variables. The objective of such a model is to predict, with a confidence interval, the values of the dependent variables given certain values of the independent variables. Examples of statistical models include rating models or value-at-risk (VaR) models. Statistical models are different from deterministic models. By construction, statistical models always include a degree of Model Risk.
53.	Tiers: Models are allocated to different groups, or Tiers, depending on their associated Model Risk.
54.	Time series analysis (written in lower case as “time series analysis”): For a given set of data and variables, this is a process of observing the behaviour of these variables through time. This can be done by considering each variable individually or by considering the joint pattern of the variables together.
55.	UAT (User Acceptance Testing): Phase of the implementation process during which users rigorously test the functionalities, robustness, accuracy and reliability of a system containing a new model before releasing it into production.
56.	Variable Transformation (written in lower case as “variable transformation”): Step of the modelling process involving a transformation of the model inputs before developing a model. Amongst others, common transformations include (i) relative or absolute differencing between variables, (ii) logarithmic scaling, (iii) relative or absolute time change, (iv) ranking, (v) lagging, and (vi) logistic or probit transformation.
57.	Wholesale Clients (written in lower case as “wholesale clients”): Wholesale clients refer to any client that is not considered as a retail client as per the definition of these Standards.