This section discusses the regulatory and business background for the treatment of specialized lending exposures. Pursuant to Article 147(8) of the CRR, specialized lending exposures constitute a unique subclass of corporate exposures under the internal ratings-based (IRB) approach. The regulation defines specialized lending exposures as

• •

  the exposure to a special purpose entity (SPE) created for the sole purpose of financing and operating physical assets,

• •

  the exposure that gives the credit institution a substantial degree of control over the assets and the income generated by these assets, and

• •

  the exposure with the primary source of repayment being generated by the financed assets and not the independent capacity of the obligor.

The above definition is in line with the Basel III view on specialized lending outlined in the BCBS’s publication on finalizing post-crisis reforms (Basel Committee on Banking Supervision 2017). The Basel III mechanics of the IRB approach categorize the specialized lending exposures into five subclasses of the corporate class: project finance, object finance, commodities finance, income-producing real estate (IPRE) and high-volatility commercial real estate (HVCRE). At this point, we note that land financing and acquisition, development and construction (ADC) are used in banks as HVCRE (Li and Chen 2013).

This paper focuses on a case of a slotting model for specialized lending exposures in real estate. Table 1 provides an overview of the two types of real estate exposures (IPRE and HVCRE) and details their key differences.

This paper recognizes two problematic areas stemming from the lack of detailed regulatory guidance for the categorization of the specialized lending exposures. First, distinguishing real estate projects that should not be classified as specialized lending (Giannotti et al 2011) remains especially problematic. Second, it is difficult to assign a real estate specialized exposure to either the IPRE category or HVCRE category. These difficulties arise from the fact that banks often issue a completion guarantee that provides the buyer of a property under construction with an assurance that the work will be completed should the developer default (Skibniewski 2014). The completion guarantee is not an investment loan, but it is treated as a credit line.

To aid practitioners in determining which real estate projects should be treated as specialized lending, Table 2 presents several scenarios in which a real estate project can be easily misclassified as specialized lending. These scenarios serve as lessons for credit risk analysts. An especially problematic case may arise from the situation where a borrower (an SPE) runs several projects. Although there is no regulatory rule for the determination of the specialized lending status in this case, any bank should carefully assess the degree of diversification of the projects and the capacity to replace the income generated by the obligor’s assets with other incomes of the obligor.

Summarizing Table 2 in light of the limited regulatory guidance, we suggest the following determinants of the specialized slotting categorization.

Materiality of the obligor’s cashflows:

a bank should assess if other activities of the obligor are significant and if the obligor can substitute any cashflow. We point to the 30% substitution coverage threshold used in the industry. Therefore, the ability of the obligor to sufficiently substitute any failed cashflow (above 30%) as a source of payment should be taken into account when determining the corporate status.

Materiality of the bank financing:

a bank should check whether the obligor uses more than one line of credit from the bank, as several credit products are commonly offered to the same obligor. At this point, the predominant nature of the credit product is key in determining the exposure categorization. The obligor can be assigned only to one category based on the predominant credit line.

To determine the real estate specialized exposure type (IPRE versus HVCRE), we suggest following the purpose of the credit. For the HVCRE type, the following types of credit are appropriate: land financing without a general plan of the layout; and real estate with a particular layout plan; real estate with a construction permit. For the IPRE type, the following type of credit is appropriate: investment loans to income-producing real estate.

This section presents a standard risk-weight-based slotting approach to modeling the IPRE real estate exposures. Pursuant to Article 153(5) of the CRR, the risk weights are based on the remaining maturity of the exposure and the strength of the real estate project indicated by a slotting category, as shown in Table 3.

The IPRE real estate exposures are assigned to a single category upon calculating the scores of the following regulatory-prescribed factors.

Financial strength.

This factor looks at market conditions, financial ratios, loan-to-value (LTV) ratios, cashflow predictability and the ability of the project to survive stress conditions (cashflow reduction and interest rates increase).

Political and legal environment.

This factor analyzes whether the jurisdiction supports the repossession and enforcement of contracts as well as the political risk and transfer risk.

Transaction and/or asset characteristics.

This factor looks at the quality of the location, design and condition of the property as well as the financial structure (amortization schedule and refinancing risk).

Strength of the sponsor and developer.

This factor looks at the financial capacity of, and willingness by, the sponsor/developer to support the project together with their management experience, as well as the relationships between relevant real estate actors (eg, promoters).

Security package.

This factor looks at the nature of lien (a credit institution having the first claim on the debt) and the quality of the insurance coverage for the project.

As noted by Pitschke and Bone-Winkel (2006), the slotting model for real estate specialized lending exposures falls under the remit of the IRB framework, which is currently being reviewed by the European Banking Authority (EBA). Therefore, the final regulatory-prescribed standards for the slotting approach may, in the near future, be subject to changes that reflect the ongoing efforts to minimize the interference of the credit risk models with IFRS 9. The EBA has already expressed its expectations for the maximization of the synergy between the IRB credit risk models and the IFRS 9 requirements (European Banking Authority 2016a). At this point, a slotting model that does not produce PD estimates is in conflict with the regulatory push for minimizing the divergence between the IRB and IFRS 9 frameworks.

According to Jankowitsch et al (2007), this type of credit risk model is used when the PD cannot be estimated easily. The slotting approach determines regulatory capital based on the final slotting score assigned to the real estate project. The EBA specifically dictates the way of assigning risk weights to specialized lending exposures for which an institution is not able to estimate PDs. With this in mind, Scannella (2013) argues that this regulatory-prescribed way of developing a slotting model constrains the methodological choices, and Akkizidis and Kalyvas (2018) point to the elevated levels of risk-weighted capital stemming from the use of the regulatory-prescribed methodology. In a study covering 3442 UK IPRE projects, Frodsham and Gimblett (2012) pointed out that in some simulated cases the risk-weighted capital mandated for the “strong” and “good” slotting categories is well in excess of downturn LGD. Focusing on the UK IPRE projects, Frodsham and Gimblett also pointed to the fact that the implementation of the EBA’s slotting methodology by the local regulator is not consistent with the leasing norms in the United Kingdom.

In summary, the slotting approach outlined by the EBA has been heavily criticized by both practitioners and academics for its methodological weaknesses. Pinsent Masons (2012) suggested that the use of slotting models might discourage lending to real estate projects. Finally, Ranson (2006) stated that the limited number of slots and the lack of predefined weights for the slotting factors do not capture credit risk adequately. Thus, the slotting approach has been avoided by credit institutions and has not received significant attention in academic studies, which focus primarily on PD modeling. However, we recognize that the potential attractiveness of using the slotting approach in IPRE and HVCRE projects lies in the simplicity of the model implementation. As noted by Abdou and Pointon (2011), a real estate slotting model is often implemented as a Microsoft Excel scorecard filled in by the analysts using their expert judgment. Thus, it is easier to validate or backtest this type of model. Table 4 presents a general methodology of a real estate slotting approach indicating where the expert judgment is used.

This paper is built on the assumption that the most material problems of the slotting approach refer to the IFRS 9 requirements rather than the limited regulatory guidance on factor risk weights or methodological inconsistencies. We posit that the slotting model has to be developed in line with the regulator-dictated methodological choices (EBA’s regulatory technical standards (RTS)) that limit pursuing alternative methodologies. Therefore, challenging the regulatory choices remains counterintuitive for a credit institution compelled to follow specific methodological pathways. A bank cannot choose not to include certain regulatory-prescribed factors or choose to use variables that are not aligned to the factor descriptions provided by the EBA.

IFRS 9 introduces the concept of staging when assessing a significant increase in credit risk. The concept is outlined in Paragraph 5.5.5 of the IFRS 9 regulation, which requires the measurement of the loss allowance on a 12-month expected credit loss basis if there is no significant increase in credit risk. Determining the significant increase in risk is done at each reporting stage (Ozdemir 2018). Pursuant to Paragraph 5.5.9 of the IFRS 9 regulation, the bank is then required to assess whether the credit risk on a financial instrument has increased substantially since the initial reporting. The IFRS 9 framework specifically stipulates that the assessment should be based on the changes to default risk occurring over the expected life of the financial instrument (Chawla et al 2016). According to Memmel et al (2015), this approach becomes problematic for models that do not provide a granular distinction between default rates. Against this backdrop, our paper provides a possible solution for the assessment of any significant increase in the credit risk of the slotting model without undue cost or effort.

As indicated in Paragraph B5.5.5 of IFRS 9, any significant increase in credit risk should be assessed of on the basis of shared credit risk characteristics with the objective of facilitating an analysis designed to identify a significant increase in credit risk on a timely basis (Beerbaum 2015). Although the IFRS 9 rules allow both qualitative and nonstatistical quantitative information to be used to determine whether a significant increase in credit risk occurred, Miihkinen (2012) argues against using expert judgment in the assessment. In a nutshell, the following criteria can be used for the staging process.

Stage $\mathrm{1}$:

credit risk has not increased significantly since the initial recognition. The loss allowance is equal to the 12-month ECL.

Stage $\mathrm{2}$:

credit risk has increased significantly since the initial recognition. The loss allowance is equal to the lifetime ECL. It can be measured by a significant downgrade in the rating master scale since the initial recognition, which refers to denotching.

Stage $\mathrm{3}$:

credit risk has increased to the point that credit impairment has occurred.

As a quantitative measure of the significant increase in credit risk, we point to the methodology based on the internal rating downgrades (denotching). This approach follows the guidelines outlined in Paragraph B5.5.16 of IFRS 9. Thus, the measurement of any significant increase in credit risk involves comparing ratings or slotting scores at different points in time. As noted by Vlachostergios (2017), other factors considered in the IFRS 9 staging process relate to default indicators such as the forbearance and past due days. Leventis et al (2011) highlights emerging problems linked to the financial asset classification using the IFRS 9 staging process. These problems are related to the measurement of a significant deterioration in credit risk, which implies that credit risk models should reconsider the risk of default measured as the probability of default (PD). Figure 1 shows the IFRS 9-aligned staging process.

The key differences between the principles of the IFRS 9 and the EBA’s RTS lie within the regulatory objectives of the two components. The IFRS 9 is focused on ensuring accuracy in the estimation of the risk-weighted assets (RWA). In doing so, IFRS 9 would allow for individually tailored approaches to modeling specialized lending exposures. On the other hand, the overarching principle of the EBA’s RTS is to ensure the harmonization of risk management and calculation of risk-weighted exposures across credit institutions. In this vein, the slotting approach does not allow additional flexibility and is centered on retaining the conservatism in the risk estimates at the cost of the accuracy.

Complementing Figure 1, we point out some common problems dominating the slotting models. Due to the limited number of regulatory-prescribed slotting categories for credit risk, it remains difficult to justify the denotching levels that would indicate a significant rating downgrade leading to stage 2 of the IFRS 9 loss recognition.

Onali and Ginesti (2014) point to the fact that the IFRS 9 accounting framework is based on principles introducing a logical model for classification and measurement of financial assets and liabilities. This is regarded by Ball (2006) as an important step toward simplifying the complex rules of International Accounting Standard 39. Further, as noted by Pool et al (2015), while changing the ways of hedge accounting, the IFRS 9 rules address the need for a forward-looking expected loss impairment model. Table 5 presents the IFRS 9 articles that relate specifically to the area of credit risk measurement and modeling.

Prorokowski (2018) points to the major differences between the IRB and IFRS 9 concepts relating to the PD/LGD framework. The slotting models have as yet received little academic attention. The IRB PD models are measured on a through-the-cycle basis, which reflects the cyclical nature of economic conditions (Carlehed and Petrov 2012). According to Edwards (2014), the IFRS 9 models should be point-in-time. However, in this paper we argue that the slotting approach satisfies the point-in-time criterion despite using some factors that may be regarded as idiosyncratic or lagged economic indicators. This is due to the fact that the slot categorization is made by the analyst based on the currently available data without the use of the through-the-cycle averaged factors. All in all, a review of existing studies suggests that the scholarly discussion on differences between the IRB and the IFRS 9 frameworks is limited to flagging conceptual discrepancies.

As indicated in Table 5, the ECL estimation considers the following aspects.

Probability weighted amount:

the ECL cannot be based on either the worst- or best-case scenarios, or limited slots indicating the strength of the obligor and the project. The ECL should reflect the probability of a loss occurring during the lifetime of the financial instruments.

Time value of money:

the ECL should be discounted to the reporting date.

Required forward-looking information:

the ECL should use reasonable and supportable information that is accessible without undue cost. However, under the IRB framework, banks should not avoid excessive and unnecessary costs associated with obtaining the forward-looking information.

According to Miu and Ozdemir (2016), the conceptual differences boil down to the methodological approaches between the IFRS 9 compliant models and the IRB modeling. Interestingly, there is also a string of academic studies that investigate whether the existing IRB models can be reused for IFRS 9 purposes despite the compliance gaps. The current paper is partly rooted in this string of scholarly literature. In particular, we recognize the study of Reitgruber (2015), which argues that having the same suite of models satisfying both the IRB and IFRS 9 requirements allows for the consistency in the use of risk models to be retained. However, as noted by Marlin (2017), the process of adapting the existing credit risk models to the IFRS 9 requirements remains challenging due to the need for significant adjustment of the data underpinning the PD and LGD models. With this in mind, we assume the adjustments extend beyond the data to the model outputs in the case of the slotting approach to specialized lending.

The model used in this exercise is a slotting scorecard for specialized lending exposures developed and implemented in October 2018 by a universal bank (ie, a domestic systemically important bank) domiciled in the Czech Republic. The model constitutes a significant component of the bank’s rating process for the IPRE and HVCRE projects. The scorecard generates a supervisory category rating (score) that is used to assign a real estate project to the slotting categories. The model and the corresponding rating system passed the initial validation in November 2018. The initial validation covers the following areas.

Model design.

This focuses on the scope, consistency of the perimeter, model history, regulatory compliance checks and documentation review.

Impact analysis.

This focuses on the EAD and RWA impact assessment and result improvements.

Portfolio analysis.

This focuses on model materiality and portfolio characteristics as well as missing values and data errors.

Default definition.

This focuses on the consistency of the implemented default in the systems.

Databases and systems.

This focuses on the data quality process and BCBS 239 (Basel Committee on Banking Supervision 2013) compliance.

Model methodology.

This describes the modeling choices and replicating the model.

Model performance.

This focuses on benchmarking and robustness checks as well as calibration and discrimination tests.

Implementation.

This focuses on the process for implementation of the scorecard.

Use test.

This focuses on model use within the bank.

Governance.

Gathering stakeholders’ feedback and providing a model risk rating.

The model was developed with the aim of replicating the EBA recommendations outlined in European Banking Authority (2016b) with the designation for the commercial real estate loan origination business (IPRE and HVCRE projects). As already discussed in this paper, the EBA’s final draft RTS document specifies how credit institutions should incorporate the credit risk factors into their model when assigning risk weights to specialized lending exposures. Therefore, the model consists of five specific factors characterizing a real estate exposure, and each factor contains several subfactors of a qualitative and quantitative nature.

Complementing Table 6, we acknowledge that it is necessary to have a multitude of factors and subfactors in order to capture all the major risk drivers. We also note that the specification of the factors and subfactors has no major deviations from the EBA’s RTS. Each factor leading to the final slotting score is assigned an individual weight at the bank participating in this study by submitting its slotting model.

The rating is the final slotting score, which allows a real estate project to be assigned to one of five distinct categories. All rating inputs are based on the expert judgment of the real estate analyst populating the scorecard. Each subfactor has underlying variables that are used by the real estate analyst, who has substantial knowledge about the evaluated real estate project and the obligor. The quantitative variables are based on the most recent financial statements of the obligor and the project, external independent property valuation reports and other publicly available information (eg, market reports). The ratings are then validated and approved by the credit officer under the “four-eyes principle” described by Hiebl (2015).

Table 7 presents the slotting categories, which are divided into five rating categories following the compliance with the EBA’s RTS. We note that the assignment of a project to the fifth slotting category (“default”) does not depend on the scorecard results, and hence does not follow the standard rating process, but is done manually by the credit risk analyst.

Complementing Table 7, we note that each factor leading to the final slotting score was assigned an individual weight by the bank participating in this study. The factors are aggregated to arrive at the final slotting score, which is used to determine the slotting category.

The data was provided by the bank participating in this exercise. It contains the metadata items around the model output (Table 8).

Average maturity:

the average remaining maturity of the real estate specialized lending project aligned to Article 153(5) of the CRR.

Target RWA:

the regulatory-prescribed risk weight threshold aligned to Article 153(5) of the CRR.

Target EL:

the ratio of the expected loss (EL) assigned to the slotting categories by the bank participating in the study.

PD master scale:

the rating scale for midcorporate PDs used by the bank participating in the study. The PD master scale contains regulatory PDs and PDs used by the bank for the midcorporate portfolio of obligors.

As shown in Table 8, the slotting approach to modeling the credit risk of real estate specialized lending pushes banks (including the bank participating in this study) to develop a discrete classification of specialized exposures. In doing so, banks need to comply with the regulatory framework of assigning risk weights and EL values as directed by Articles 153(5) and 158(6) of the CRR. In the PD mapping methodology proposed in this paper, the following data items are considered:

• •

  target RWA,

• •

  target EL and

• •

  average maturity.

This section describes the methodology for deriving the correspondence between the four slotting categories (the “default” category is ignored) and the rating scale that quantitatively reflects the default risk of a particular obligor. To maximize practical implementation, the methodology uses a realistic rating scale for midcorporate obligors applied by the bank participating in this study. Further, the proposed methodology is based on the regulatory-prescribed formulas for risk weights and EL. We recognize the following benefits of the aforementioned methodological assumptions:

1. (1)

   The methodology yields a more granular scale for evaluating credit risk at an obligor-level and enables a more practical implementation of the IFRS 9 denotching mechanism.

2. (2)

   The methodology satisfies IFRS 9 requirements for the measurement of cumulative default probabilities over a given forecasting horizon.

3. (3)

   The methodology satisfies IFRS 9 requirements for the assessment of a significant increase in credit risk, which is an essential component of the IFRS 9 framework.

The proposed mapping methodology is a sequential process. At the starting point, both the target RWA and target EL are considered. These perimeters are derived from the regulatory-prescribed formulas and are constructed around the primary credit risk indicators: the probability of default and the loss given default. The proposed methodology serves to identify PD and LGD values by solving the following simultaneous equations:

where $R(\mathrm{PD})$ is the correlation coefficient and $b$ is the maturity adjustment factor. Both $R$ and $b$ are endogenous variables dependent on the PD. $M$ is the maturity of the exposure to the real estate project. The above coefficients are derived from the regulatory-prescribed formulas:

The target RWA and target EL are derived from the following formulas:

Since the real estate specialized lending exposures are assigned to four distinct slotting categories, we should solve (3.1) for each slotting category. Moreover, real estate projects are differentiated by the maturity thresholds in the regulatory framework. Therefore, the mapping determines two sets of solutions, with the first being for the fixed average maturity of 1.25 years and the second for the fixed average maturity at 3 years.

First, on solving (3.1), the PDs and LGDs are obtained for each final slotting score. This procedure tests the following LGD values to obtain the PDs: 50%, 45%, 40%, 35%, 30%, 25%, 20%, 15%, 10%, 5%.

The first step involves fixing the average maturity to 1.25 years and 3 years, testing for the fixed LGDs and setting the targets for the RWA and EL values (as prescribed by the regulator). Tables 9 and 10 show the different PDs assigned to the slotting categories. The aim of this exercise is to satisfy the target RWA (${\mathrm{RWA}}_{\mathrm{T}}$) with the lowest calibration error (CE) possible and then to satisfy the following conditions for the EL:

where VaR denotes value-at-risk.

A visual inspection of these tables reveals that the conditions for the target RWA and the minimum difference between VaR and the EL are satisfied for the subsequent PDs, presented in Table 11.

We now make several important remarks summarizing Tables 9–11. First, in all cases, the $\mathrm{LGD}=50$%, which makes it a constant parameter. Second, the granularity of the ratings is limited and there are several ratings missing between the values provided. Third, using the PD master scale for the midcorporate portfolio yields more realistic results and is also justified by the fact that the real estate specialized lending exposures usually stem from the midcorporate obligor class (Mondal 2011).

Finally, solving for (3.1) requires the conditions for the target RWA and target EL to be satisfied simultaneously. Table 12 shows the optimal PDs mapped to the slotting categories for a maturity $M=1.25$ years. A similar process is carried out for $M=3$ years.

Solving for the optimal PDs reveals that the optimal PDs differ from the mapped PDs at the fixed LGD level ($\mathrm{LGD}=50$%). At this point, fixing the LGD at 50% leads to an increase in the calibration error for the EL in slotting categories 2 and 3 for both cases with average maturity. Figure 2 shows the combination of PDs and LGDs satisfying the target RWA.

As mentioned in the previous section, solving for (3.1) is inefficient, as it leaves limited a granularity of the ratings. The case of ratings missing between the reported ratings should be addressed. This involves retrieving the scores associated with the missing ratings by using the resolution of a third-order polynomial. Once the PD values have been obtained for the slotting categories, we seek the correspondence with the PD master scale. This is done by starting with the PD value and finding the corresponding final slotting score:

where $x$ ($x\in (1,4]$) denotes the slotting score. By using the bank’s PD master scale for midcorporate obligors in combination with the third-order polynomial, the missing scores can be obtained using Excel Solver.

In the first step, for the LGD fixed at 50%, the process of reconciling the content of Table 9 with the polynomial fitting using the function for the matrix product of two arrays is initiated. This leads us to obtain interim base PDs corresponding to the slotting scores 1, 2, 3 and 4, respectively. Appendix A online shows the results of solving for the polynomial fitting. The following correspondence is established.

• •

  For $M=1.25$: $-5.67$ for a slotting score of 1; $-5.03$ for a slotting score of 2; $-3.78$ for a slotting score of 3; and $-1.62$ for a slotting score of 4.

• •

  For $M=3$: $-5.69$ for a slotting score of 1; $-5.12$ for a slotting score of 2; $-4.46$ for a slotting score of 3; and $-1.92$ for a slotting score of 4.

Having the interim PDs as a starting point and using the third-order polynomial function (4.3), we find the following slotting scores. This process is the inverse of the function for obtaining PDs, as the resolution starts with a given PD value.

Appendix A online shows the results for the corresponding slotting scores. At this point, the interim PDs should be converted into final PDs. This is done using the exponential function,

where $x$ is the interim PD obtained by the polynomial fitting and matched with a corresponding slotting score. Using the results in Table A4 of Appendix A online, more granular PD parameters can be assigned to the corresponding slotting scores, as shown in Table 13.

On achieving a more granular rating correspondence between the slotting scores and mapped PDs, the last step involves defining both the lower and upper limits for each slotting category (Table 14).

Complementing this table, we should note that the lowest mapping boundary is set to 1 and the highest mapping boundary is set to 4, owing to the alignment with the slotting model. Table 15 summarizes the rating assignments to different slotting categories.

As highlighted in Section 1, the main aim of this paper is to map the slotting scores to PD estimates in order to deliver adequate inputs for the computation of the ECL. Specifically, our proposed solution meets the objective of measuring the IFRS 9 cumulative PD estimates:

The availability of historical default data is limited for the newly developed slotting model. Thus, building the point-in-time parameters conditional on macroeconomic indicators remains counterintuitive. However, the cumulative PDs can be used by relying on basic assumptions specified in (4.5). Hence, a cumulative $T$-year PD estimate ${\mathrm{PD}}_T$ is derived from the one-year PD together with the PD estimate, denoted ${\mathrm{PD}}_1$.

Converting the slotting scores into the ratings aids the denotching process under the IFRS 9 framework. As described by Honohan (2008), denotching is an automatic penalization of one notch for the estimate. This process is triggered if the obligor is not reevaluated within the 12-month period following the last rating/grading date. Without the increased granularity, the denotching process would be overly punitive for a bank relying on the four-point scale of the regulatory-prescribed slotting approach. However, with the mapped ratings in place, the impact of the denotching on the reported RWA and the EL is ameliorated, as evidenced in Table 16 for the example of an exposure with a maturity of 1.25 years.

As shown in the table, the effect of the reassignment to a lower slotting category on the RWA/EL does not always materialize. The nonmateriality of the denotching process in certain cases has significant implications for the staging process under the IFRS 9 framework, as it does not result in a “significant increase in credit risk”. Without the mapped ratings, every instance of denotching would result in a significant increase in credit risk under the IFRS 9 framework.

The solution presented in this paper is particularly useful for the IFRS 9 staging process, which is based on the internal rating master scale. The IFRS 9 staging process classifies exposures in accordance to their default risk. At this point, under the IFRS 9 framework there are three distinct stages. As mentioned earlier in the paper, an exposure would be reclassified from stage 1 to stage 2 if a significant increase in credit risk is observed. Going forward, the exposure would be reclassified from stage 2 to stage 3 if a credit-impaired status was established. The mapping of the slotting scores to the ratings helps establish specific thresholds for the IFRS 9 staging, as shown in Table 17.

An interesting point with reference to the IFRS 9 accounting framework is the LGD trend in the process of minimizing the mapping errors (target RWA and target EL). As shown in Table 9 for the example of the maturity, $M$, set to 1.25 years, the LGD levels are inconsistent across slotting categories. Under the IFRS 9 accounting framework, LGDs are expected to change consistently across the IRB standardized approach slotting categories. We can assume that the LGDs will either decrease or increase for each slotting category. As shown in Table 9, problems occur when attempting to satisfy a target RWA of 250% for a slotting category of 4. In this case, the LGD increases again to a level of 50% and the monotonic trend of falling LGDs cannot be conserved. However, at this point, a retained PD of 19.81% is more realistic than the PD of 30.7%. Therefore, we challenge the assumption that LGDs should decrease for the lower slotting categories. We note that both the PDs and ELs should increase for real estate specialized lending projects that fall into worse slotting categories. Then, given the fact that the EL is a function of PD and LGD ($\mathrm{EL}=\mathrm{PD}\times \mathrm{LGD}$), it remains counterfactual to expect LGDs to decrease for increasing PDs.

To account for the lack of consistency in LGDs, in this paper we argue against fixing the LGD at 50% for maturities of 1.25 years and 3 years. A visual inspection of Tables 9 and 10 reveals that fixing the LGD at 50% does not minimize the error for the files falling into slotting categories 1 or 3, with the latter suffering from elevated levels of error and significant differences in the PDs, which decreased from 9.2% to 2.29%. Against the background of IFRS 9, fixing the LGDs would constitute a built-in bias to the derived PDs and noncompliance with the new reporting standards. Table 18 shows the differences in PDs stemming from fixing the LGDs at 50%. As shown in the table, the gap is evident for slotting category 3.

Another interesting point with reference to IFRS 9 compliance is the fact that the proposed mapping does not generate ratings that suggest a strong repayment capacity (ie, A$+$, A, A$-$) of the obligor. In real estate projects, there is often a good scope for borrowing and access to other cashflows to cover the company’s assets. Further, the BBB and BBB$-$ ratings should not refer to two different slotting categories (SC1 and SC2), as they are assigned to obligors of good repayment capacity, comparable with normal standards in the sector. Although some weaknesses can be identified, these would likely manifest themselves in times of financial distress. Finally, slotting category 2 would include both projects classified as “good” (where the company has borrowing capacity and there are additional cashflows to cover assets) and projects classified as “weak” (where clear weaknesses are identified and the company has no room for maneuver in its repayment capacity). The above points should be taken into consideration by practitioners attempting to map the slotting scores to the corresponding PDs and master scale ratings.