In our earlier blog, we discussed PD terminology and PD calibration approaches as applicable to the IFRS 9framework. IFRS 9 has mandated computation of Impairment Losses, approach for which has beendiscussed in our 6th blog post. For computation of Expected Credit Loss (ECL), IASB expects organizationsto consider forward looking information including macroeconomic factors that are relevant to the exposurebeing evaluated and that must go beyond historical and current available data. BCBS strongly endorsedthis requirement in its paper published on 18th December 2015 (Please refer to our white paper aroundBCBS Guidelines on IFRS 9) and has very clearly defined its expectations from banks in terms ofconsideration of supportable forward-looking information into ECL estimates. Regardless of the impairmentapproaches considered by a bank, (Please refer to our 6th blog post) i.e. Roll Rate, Vintage analysis,Expected Loss Model or Discounted cash flow, the ECL estimate should incorporate the expected impactson account of forward-looking information, including macroeconomic forecasts. In this blog, we cover howPDs (point in time) can be adjusted for the mandatory "forward-looking" requirements prescribed by IFRS 9.
While banks have been following similar forwardlookingmacroeconomic adjustments of PD for quitesome time now under Basel IRB Probability ofDefault (PD) modeling, stress testing and CCAR,IFRS 9 has introduced additional complexities in theform of Lifetime ECL and Lifetime PD for Stage 2and Stage 3 exposures. Lifetime here refers to lifeof the loan, the effective maturity in other words.While regulatory stress testing norms require banksto assess impact of macroeconomic factors on PDover 4 to 9 quarters (under CCAR guidelines),however, under IFRS 9 they are required to extendthe macroeconomic adjustment of PD for the life ofthe loan, which can extend to over 20 quarters (longterm project financing loans, for instance).
While banks can use their existing methodologies for forward looking macroeconomic adjustments of PD (1year horizon) for Stage 1 exposures, they will be required to develop methodologies to estimate the impactof macroeconomic adjustments on PD term structure for Stage 2 and Stage 3 exposures. Forecastingmacroeconomic scenarios for such a long horizon and developing a stable relationship between thesescenarios and PD estimates are two of the major challenges that banks may face.
The graph below shows the impact of potential economic changes on PD, especially when 12-monthhorizon fails to capture the potential economic change. This economic change over the life of instrument iseffectively captured by lifetime PD, as evident by the right hand side graph (grey line).
At present, banks have already adopted several methodologies to mitigate these challenges. We will look atthe suitability of these methodologies and an approach to improve them to comply with the IFRS 9requirements.
Methodologies for Macroeconomic Adjustment of PD: Macroeconomic adjustment of PD can be carried out through 3 key approaches as discussed in thissection. 
1) Macroeconomic adjustments of portfolio Central Tendency (CT): Under this approach, historical data of portfolio Point-in-Time (PIT) PD is used to arrive at 1-year forward-lookingcentral tendency (CT) for the portfolio and a link between forward looking macroeconomicparameters and 1-year forward-looking CTs are established. The derived macroeconomic adjusted CT isthen used to calibrate PIT PDs for each rating grade.
Macroeconomic adjustment of portfolio Central Tendency (CT) approach is more relevant when a ratingmodel is based on TTC Rating philosophy. Under TTC Rating Philosophy, rating models include onlyidiosyncratic factors, as a result, the rating grade generated by these models do not reflect changes inmacroeconomic factors and the transition matrices developed for this type of model is likely to remainstable through the business cycle. However, due to change in macroeconomic factors, the likely defaultfrequency in each rating grade will change. Adjusting the portfolio CT for macroeconomic scenarios andrecalibrating the model's rating grade to the new PIT PDs, allows to incorporate the impact ofmacroeconomic factors in PIT PD estimates.
2) Markov Chain (Z Score) Approach: Under this approach, linkage is established between 1-year transition matrix and forward-lookingmacroeconomic parameters. Historical transition matrices are converted into z-scores, which in turn arelinked with forward-looking macroeconomic parameters.
3) Macroeconomic Adjustments of Rating Grade Migration Approach: Under this approach, linkages are established between 1-year migration probabilities of each rating gradewith macroeconomic parameters.
Rating Grade Migration i-j = f (macroeconomic factors)
Where, Rating Grade Migration i-j is the migration probability of a rating grade i going to rating grade j within 1-year time horizon.
Both Markov Chain (Z Score) and Macroeconomic Adjustments of Rating Grade Migration methodologiesare more applicable when the internal-rating model of the bank is developed based on PIT RatingPhilosophy. Under PIT rating philosophy, rating models include both idiosyncratic and macroeconomicfactors. As a result rating grade generated by these models will change due to macroeconomic factor andconsequently, the probability of migration of borrowers from one rating grade to another (captured intransition matrix) will vary depending on the macroeconomic scenarios (business cycle).
Though both approaches mentioned above establish relationship between macroeconomic parameters andthe rating movement (upgrade/downgrade), Macroeconomic Adjustments of Rating Grade Migrationapproach is likely to be more sensitive to macroeconomic factors and closer to real life situation, since theeffect of macroeconomic factors are assessed at individual rating grade level migration. In the MarkovChain (Z Score) approach, the entire transition matrix is converted into a single dependent variable andeffect of macroeconomic parameters are assessed on that variable. As a result, the migration effect ofmacroeconomic parameters will be uniform across rating grade.
Though the second approach seems more accurate, it requires more data and needs substantial modelingefforts. For instance, for a 8 point Internal Rating Grade, under first approach a bank will be required todevelop one model; however as per the second approach, a bank will be required to develop 64 models(8x8).
We believe that banks may use any of the above 3 approaches for macroeconomic adjustments of PD (PIT)under IFRS 9, depending on the rating philosophy followed for development their internal rating model.However, if banks are currently doing these macroeconomic adjustments for PD (TTC), they will be requiredto repeat the exercise for PD (PIT) adjustments for IFRS 9 compliance.
Methodologies for Development of PD Term Structure: 1) Binomial Approach: Under the Binomial approach, credit deterioration is assumed to be a two state process: Default and NonDefault. The approach does not recognize the deterioration of credit quality to intermediate rating grades.Under this methodology, PD term structure can be created using cumulative PD methodologies.
PDCumm(i)= PDFD(i-1) + {(1-PDFD1) * ......(1-PDFD(i-1))}*PDFDi
Where,
PDCumm(i) = Cumulative PD at the end of year i PDFDi = Forward PD in the year i (1-PDFD(i-1)) = Non Defaulted Portfolio percentage at the beginning of year i
To create PD term structure using Binomial method, forward PDs need to be estimated by makingmacroeconomic adjustments to portfolio Central Tendency (CT) accounting for future macroeconomicscenarios, and then recalibrating PDs of various rating grades based these macroeconomic adjusted CTs.Once the forward PDs are estimated, the same can be used in Binomial approach to the create PD termstructure.
Some banks, instead of estimating forward PDs based on future macroeconomic scenarios, only use 1-yearPD estimates to create cumulative PD using Binomial approach. The basic assumption in this approach is thatforward PDs will remain same as current 1 year PD. This assumption is valid in case the banks are developingcumulative PD for PD (TTC) term structure, as 1 year PD (TTC) is likely to remain stable across the businesscycle. However, forward PD (PIT) will change with future macroeconomic scenarios and hence, to generate PD(PIT) term structure using Binomial approach based on 1 year PD (PIT) estimate may not be right approach.Banks would need to estimate forward PIT PD based on future macroeconomic scenarios. 2) Markov Chain (Z Score) and Rating Grade Migration Approach: PD (PIT), adjusted for macroeconomic parameters using either Markov Chain (Z Score) or Rating GradeMigration approach, can easily be used to develop PD (PIT) term structure by applying matrix multiplicationtechniques.
Banks may estimate the effect of macroeconomic factors on transition matrix in the 1st year and use thesame to create PIT PD term structure. Like the Binomial approach, to generate PD (PIT) term using 1-yearrating migration (transition matrix) information may not be right approach, as rating migration probabilitieswill change over time based on future macroeconomic conditions. To create PD (PIT) term structure usingabove approaches, it is important that future 1 year rating migration probabilities are estimated and used inmatrix multiplication process.
3) Mapping to External Rating Agency Term Structure: An alternative that many banks are considering over the extensive data-driven computation methodologiesdiscussed earlier, is to use the PD term structure provided by external global rating agencies. Thisalternative approach requires banks to map their internal rating grades with rating grades provided byglobal rating agencies and adopt the PD term structures corresponding to rating grades benchmarked withglobal rating agencies. While this approach may sound logical, it does entail a few issues.
Global rating agencies follow TTC rating philosophy, and term structure published by them is PD (TTC) termstructure. If the bank's internal rating model also follows TTC rating philosophy, then this is a goodapproach to create PD (TTC) term structure.
If bank's internal rating model follows a PIT rating philosophy, then the PD mapping between banks Internalrating grade and external rating grade will not be stable over time and mapping exercise needs to beconducted frequently (at minimum once a year). At a more fundamental level, mapping of 1 year PD (PIT)estimate of the internal rating model to PD (TTC) of external rating agency's rating grade, then estimate PD(TTC) term structure of the internal rating grade using PD (TTC) term structure of external rating agency willunder or overestimate PD (TTC) term structure of the internal rating grade, depending on the current stateof the business cycle when mapping is conducted.
Thus before using external rating agency's PD (TTC) term structure as it is for IFRS 9 purpose, banks willhave to find a way to convert rating agency PD (TTC) term structure into PD (PIT) term structure andincorporate macroeconomic adjustments to it. Several options to make such transition possible are beingresearched across industry. One of such research talks about transforming TTC PD term structure to PITPD term structure in two steps: first, use Nelson-Siegel function to estimate TTC PD term structure usinghistorical default rates; second, apply one-factor Merton model and transform TTC PD term structure to PITPD term structure with macroeconomic adjustments. Moreover, global rating agency's PD (TTC) termstructure is mostly based on its default experience of large publicly rated borrowers mainly located indeveloped economies. Thus, its applicability in case of portfolio consisting of Small and Medium Enterpriseborrowers and borrowers from emerging market economy is questionable.
In addition to complexities associated with adjustment of PD (PIT) for forward looking macroeconomicscenarios and development of PD (PIT) term structure, banks also need to be cognizant of the challengesin developing methodology for forecasting of macroeconomic scenarios. To mitigate these challenges,IFRS9 has given following flexibilities:
While these flexibilities help in mitigating challenges of long term macroeconomic forecast, establishing astable relationship between macroeconomic scenarios and PD (PIT) will still remain a big obstacle.
Don't miss this roundup of our newest and most distinctive insights