Roughly ten years ago, we had Basel II. It focused primarily on improved internal models and regulatory capital. There were incentives for developing IRB models, however these were largely around stability of models and capital using the so called Through-the-Cycle (TTC) log-run average calibrations. After the 2008 financial crisis, Stress Testing became important which focused on tail losses and now IFRS9/CECL requires new, more forward looking ways for measuring losses.
Ignoring credit cycles is akin to saying that weather patterns are completely unpredictable, there are no seasons and one cannot predict the season after six months. While seasons happen due to known distinct physical phenomena, credit cycles specific to industries and regions happen due to economic phenomena and like seasons they can be measured and can be partly forecasted. Introduction of credit cycles, makes PIT credit risk measurement more accurate in terms of current and future estimates of ECLs.
Z-Risk Engine incorporates an advanced credit analytics framework designed to Unlock Credit Cycles – the key component of accurately projecting ECL impairments. Z-Risk Engine supports ‘unbiased probability weighted forward-looking’ ECLs as required by IFRS9/CECL, through simulation of industry and regional credit cycles using mean reversion and momentum models to assess ECLs over all possible risks – these simulated ECLs are unconditional, representing all possible future scenarios. Z-Risk Engine also supports scenario based conditional forward-looking base and stressed ECL projections as required by Stress Testing, by making use of macro-economic factors to credit factor bridge models.
The Z-Risk Engine solution bring together 10+ years of research and development on a Point-in-Time (PIT) and Through-the-Cycle (TTC) dual ratings approach for commercial and corporate portfolios. This approach was developed and implemented at 2 large global banks and formally signed-off under the banks Basel Waivers. Z-Risk Engine is a completely redeveloped solution based on this approach and will be released as a SAS solution in September 2016. However, in the past year we have developed prototype IFRS9/CECL solutions at banks.
Wholesale defaults and losses vary by about an order-of-magnitude over the cycle and most legacy, PD and ratings models capture no more than 20% of this variation. Further, back tests regulatory driven models estimate realized provisions and charge-offs 50% or more (conservatism). However, IFRS9 and CECL call for unbiased estimate of losses. To do this, firms can reduce the average levels of PDs by re-calibrating them along with credit cycle indices so that they best fit realized default experience.
Best forward-looking information should incorporate market-value indicators, and for corporate book, specifically measures of market leverage over volatility (Merton approach). Credit index based models produce PIT PDs, LGDs and EADs and also explain PD, LGD, EAD correlation. Also, to be fully forward looking, need to account for predictable movements in credit indices. With regard to scenarios, for ECL, one needs to run many, joint, PIT PD, LGD, and EAD scenarios; which commonly depend on underlying credit indices.
Within the wholesale, corporate and commercial space where we focus, all models have been designed for the past 10 years to predict mid-point, i.e. through-the-cycle (TTC) like. This was due to (a) capital stability focus in Basel II (b) lack of data and modelling techniques and (c) mimicking Agency ratings which are TTC-like. In fact, two research articles that we published in 2015 in theJournal of Risk Model Validation, apeer reviewed journal shows that Agency Ratings are around 80% TTC. So within this space, institutions which anchor their ratings to Agency ratings end up creating hybrid TTC-like internal models. This is a challenge because IFRS9 calls for adjustment for current conditions and then a future forecast and such TTC-like models would not be able to comply in our opinion. For this, one needs a PIT/TTC framework to convert one type of model output to other. This is not just non-compliance risk but the loss calculations could be off by a magnitude of 10 times. We see the need for PIT/TTC methodology as the single biggest challenge.
To do so, firms can derive period-by-period EAD and LGD scenarios over a facility’s life from the associated credit indices. In doing so we assume that structural features of facilities deterministic and so only Z variations affect the outlook. This is consistent with academic research. For details refer our forthcoming paper on this topic.
Z-Risk Engine is a SAS based batch automation solution suite and is designed to integrate directly with a financial institution’s internal client static data and wholesale credit exposures by obligor/facility type and utilises each institution’s own PD, LGD, EAD credit models. These internal models are assessed for their ‘degree of PIT ness’ and then together with industry and regional credit cycles – customised to each institution’s portfolio footprint – are used to convert these PD, LGD and EAD estimates to multi-year PIT PDs, LGDs and EADs. The advanced analytics and batch processing architecture can be run in either simulation mode assessing ECLs in detail on an unconditional, or probability-weighted basis or in deterministic scenario mode to assess stress or baseline ECLs.
Z-Risk Engine utilises each institution’s own PD, LGD, EAD credit models and hence such models should exist and should be “valid models”, i.e. offer some degree of accuracy and discrimination in predicting default rates, losses or exposures at default. If however, your institution does not have “valid models” for some or part of your portfolio, then Aguais and Associates can build such models for your institution as a consultancy engagement. Our team has decades of experience building such models in large banks and achieving AIRB waivers for them.
Z-Risk Engine customization is done in several ways. It makes use of institution’s industry-region footprint to create relevant industry and regional credit cycles which drive PIT conversion. Further, it makes use of institution specific model calibrations to create PIT measures. Also, it makes use of institution specific factors such as Significant Deterioration criteria and Stress Testing input factors.
Z-Risk Engine is a SAS based batch automation solution and is deployed inside the client’s environment. The solution makes use of an institution’s SAS servers and clients, where internal and vendor data is integrated with the solution. The entire customized solution then sits inside an institution’s firewall and all processes are run on institution’s computers and servers.