PD Modeling

We offer a suite of methodologies for PD model development ranging from expert judgment based methods t0  purely statistical techniques.

Our approach for PD modeling coupled with accelerators at each step helps in efficient model building. We cover complete model development cycle starting from data preparation to model documentation. Shown are the typical steps in a model development exercise done by us.  

Our in-house developed accelerators cover the key areas of model development such as, and help in fast tracking the development process:
  • Exploratory Data Analysis
  • Candidate model generation for model performance optimization
  • Model assessment

Low Default Portfolio (LDP) Modeling

In cases where default data is scarce, we use LDP modeling techniques. LDP portfolio needs to be considered differently from portfolio where we have lot of data. Aptivaa has specifically developed methodologies to deal with the situation where much of the default data is not available.

For LDP portfolios our approach involves:
  • Expert judgment models
  • Hybrid models by combining data and expert inputs
  • Perception rating based approach in which surveys are designed to include expert inputs
  • Pluto Tasche approach for PD calibration

LGD Modeling

We have successfully developed and implemented LGD models in several banks around the world. Our selection of the ideal model generally depends on:
  • Data availability for the key indicative factors for predicting the recovery rates
  • Validation of the methods to determine the best fit for the bank’s financing portfolio
Our LGD model development framework incorporates the following build strategies:
  • A hybrid Expert and Data-driven model that relies on the confidence level on the quality and richness of LGD data. This technique is widely referred as Bayesian approach for LGD model
  • Workout Model-Involves discounting of recovery cash flows received after the loan has defaulted in order to compare the net present value of recovery streams with the EAD
  • Additive Linear Model-Designed to estimate an outcome directly. This development requires the availability of qualitatively and quantitatively flawless data
  • Regression Tree-A non-parametric and non-linear predictive model in which the predictive value of the target variable is obtained through a series of sequential logical if-then conditions

EAD Modeling

As with other credit models, we have successfully developed and implemented EAD models through a tried and tested framework.
  • We employ a model development process that is comprehensive to address all elements that contribute to exposure forecast modeling
  • We analyze requirements of the bank, based on which we develop a customized model landscape, determining the number of EAD models to be developed
  • Our EAD Model Framework is aligned with the LGD Model given the close relationship between the risk models