Constructing Stress.cdr - Crisil

2 downloads 188 Views 156KB Size Report
CRISIL Global Research & Analytics (GR&A) is the world's largest and top-ranked provider of high-end research an
A CRISIL GR&A IMPACT

Background

CONSTRUCTING STRESS-TESTING MODEL TO ESTIMATE PDS

Challenges

Solution

Benefits

Background A US bank wanted assistance with time-series modelling to estimate stressed Probability of Default (PD). CRISIL Global Research & Analytics (GR&A) was engaged to develop a prototype model for the bank's Commercial & Industrial portfolio, which could be suitably modified for other portfolios.

Challenges n

Some of the key challenges were: - Limited internal data on Pds - Non-stationary characteristics of certain macro-economic variables, which made their direct use in a model difficult - Incorporating suitable macro-economic variables in the model

Solution About CRISIL Global Research & Analytics (GR&A)

n

CRISIL Global Research & Analytics (GR&A) is the world's largest and top-ranked provider of high-end research and analytics services.

n

We are the world's largest provider of equity and credit research services. We are also the foremost provider of end-to-end risk and analytics services to trading and risk management functions at world's leading financial institutions and corporations. We operate from research centers in Argentina, China, India and Poland, working with our clients across several time zones and in multiple languages.

n

n

n Being part of CRISIL enables us to attract and retain top quality talent. We have over 2,300 employees, 75% of whom hold advanced degrees in finance, accounting and management. We employ the largest number of CFAs and CFA aspirants in India.

n

n We won top honors at the NASSCOM Exemplary Talent Practices Award (NExT Practices) for the second year in a row in 2012. The award recognizes us as a firm that has the vision to proactively invest in its people and get them future-ready.

Benefits n

We are committed to delivering cutting-edge analysis, opinions, and solutions. This underscores our proposition of being the best people to work with. n For more details visit: www.crisil.com/gra

A variant of the Auto Regressive model was used. The Commercial & Industrial portfolio was further divided into Investment, SubInvestment and Problem Grade portfolios, thereby accounting for their varying performance. Some of the variables (including their lags) used were unemployment growth, quarterly GDP growth, Case Shiller Index growth and Oil Price growth. Detailed testing of various variables, including their lags, was performed to determine predictive variables. External default data from one of the rating agencies was used, and to ensure a richer default data, downgrades were also considered. First differencing was used to ensure that the data was stationary. To account for limited data, one model regresses the external default/downgrade data to economic variables and a separate model to fit the external downgrade/default data with that of the bank. The documentation on model methodology was comprehensive and included detailed tests to show that external data used was comparable to the internal data.

n

By using two models instead of one model, and by making suitable data modifications such as first differencing, some of the data issues were addressed and the model showed reasonable performance. The model included the relevant macro-economic variables and the model showed strong sensitivity to input parameters under stressed conditions. The model can be suitably modified for other products too.

The best people to work with