Andrija Djurovic
Senior Manager, Risk Advisory - Central Europe at Deloitte
- Report this post
๐๐ฌ๐ญ๐ข๐ฆ๐๐ญ๐ข๐ง๐ ๐ญ๐ก๐ ๐๐๐ซ๐๐ฆ๐๐ญ๐๐ซ๐ฌ ๐จ๐ ๐ญ๐ก๐ ๐๐๐ฌ๐ข๐๐๐ค ๐๐ข๐ฌ๐ญ๐ซ๐ข๐๐ฎ๐ญ๐ข๐จ๐งAs a follow-up to my previous post about the Vasicek distribution, I built a small Shiny application to fit the distribution to observed default rates. The application allows users to choose specific estimation methods and compare results. This can be done with a provided dummy dataset or by uploading a particular dataset following the instructions on the application's landing page.Those interested can check out the app via the link in the comments section.In the upcoming posts, I'll compare the biases of different estimation methods and explore potential modifications to the underlying distribution of the Vasicek model.
85
2 Comments
Andrija Djurovic
Senior Manager, Risk Advisory - Central Europe at Deloitte
2d
- Report this comment
Shiny app:https://andrijadj.shinyapps.io/vasicek_distribution/The Vasicek Distribution post:
2Reactions 3Reactions
Marc Labat, CFA
2d
- Report this comment
Thanks for sharing
1Reaction 2Reactions
To view or add a comment, sign in
More Relevant Posts
-
Andrija Djurovic
Senior Manager, Risk Advisory - Central Europe at Deloitte
- Report this post
๐๐ก๐ ๐๐๐ฌ๐ข๐๐๐ค ๐๐ข๐ฌ๐ญ๐ซ๐ข๐๐ฎ๐ญ๐ข๐จ๐งThe Vasicek Distribution holds a special place in credit risk modeling, finding applications in various areas including, but not limited to, IRB modeling, capital requirement calculations, IFRS9 modeling, and measuring concentration risk. Like other statistical distributions, the Vasicek distribution follows a specific functional form with parameters requiring estimation.Check out my new presentation, which briefly introduces the Vasicek distribution.In addition to this presentation, I am preparing to publish a Shiny application dedicated to fitting the parameters of the Vasicek distribution to observed data. Those interested can follow the GitHub link in the comments section to stay updated on upcoming announcements.
310
3 Comments
Like CommentTo view or add a comment, sign in
-
Andrija Djurovic
Senior Manager, Risk Advisory - Central Europe at Deloitte
- Report this post
๐๐จ๐จ๐ญ๐ฌ๐ญ๐ซ๐๐ฉ ๐๐ฒ๐ฉ๐จ๐ญ๐ก๐๐ฌ๐ข๐ฌ ๐๐๐ฌ๐ญ๐ฌ ๐ข๐ง ๐๐ซ๐๐๐ข๐ญ ๐๐ข๐ฌ๐คBootstrapping finds application in various domains of credit risk modeling, but it is especially beneficial in hypothesis testing when there is no consensus on the estimates' standard error, or the testing procedure is absent. In this context, bootstrapping demands special attention, particularly when practitioners aim to calculate the p-value, as it does not produce data conforming to the null hypothesis.Check out my new presentation, showcasing one of the frequently utilized approaches and metrics for validating credit risk models.The GitHub page is available via the link in the comment section.
255
3 Comments
Like CommentTo view or add a comment, sign in
-
Andrija Djurovic
Senior Manager, Risk Advisory - Central Europe at Deloitte
- Report this post
๐๐ข๐ค๐๐ฅ๐ข๐ก๐จ๐จ๐ ๐๐ฉ๐ฉ๐ซ๐จ๐๐๐ก๐๐ฌ ๐ญ๐จ ๐๐จ๐ฐ ๐๐๐๐๐ฎ๐ฅ๐ญ ๐๐จ๐ซ๐ญ๐๐จ๐ฅ๐ข๐จ๐ฌOne method for estimating the PD for Low Default Portfolios (LDP) is Alan Forrest's likelihood approach. His proposal to switch from the probability of the class of data outcome (as presented in the Pluto-Tasche or BCR papers) to the likelihood and likelihood ratio is crucial as it establishes a direct connection with the classical theory of statistical inference and its well-known approximations, which are also valid for high default cases. I extended the above framework to the portfolio-level multi-year period design, which is often considered a prevalent method in PD estimation. In essence, I introduced year-to-year correlation in the simulation of systemic factors, leading to a change in the calculation of expected likelihood from the proposed analytical approach to Monte Carlo simulation.Check out my new presentation for detailed information on the proposed adjustment. Additionally, I will consider incorporating the adjusted approach into one of the new versions of my ๐๐๐ญ๐จ๐จ๐ฅ๐ค๐ข๐ญ R package.Lastly, I want to thank Alan Forrest for his comments and motivating feedback, which encouraged me to continue working in this field.
185
7 Comments
Like CommentTo view or add a comment, sign in
-
Andrija Djurovic
Senior Manager, Risk Advisory - Central Europe at Deloitte
- Report this post
๐๐จ๐ฐ ๐๐๐๐๐ฎ๐ฅ๐ญ ๐๐จ๐ซ๐ญ๐๐จ๐ฅ๐ข๐จ๐ฌ - ๐๐จ๐ง๐ฌ๐๐ซ๐ฏ๐๐ญ๐ข๐ฏ๐ ๐๐ฌ๐ญ๐ข๐ฆ๐๐ญ๐ข๐จ๐ง ๐จ๐ ๐๐๐๐๐ฎ๐ฅ๐ญ ๐๐ซ๐จ๐๐๐๐ข๐ฅ๐ข๐ญ๐ข๐๐ฌEstimating the Probability of Default (PD) for low default portfolios typically poses a significant challenge for practitioners. The scarcity of defaults undermines PD estimates' reliability and statistical validity based on historical default experience. This issue is also a primary concern for regulators, given that PD based solely on simple historical averages or judgmental considerations may underestimate the bank's capital requirements.My latest presentation illustrates the Benjamin-Cathcart-Ryan approach for conservative PD estimation in low default portfolios. The presented simulation focuses on a prevalent design in practice: multi-year period estimation, which considers asset and year-to-year correlation. However, other methods are also utilized for this purpose. I encourage readers and practitioners to explore other approaches, and I will showcase some of them in one of the upcoming posts.The accompanying files for the presentation are on my GitHub page (link in the comment).
471
29 Comments
Like CommentTo view or add a comment, sign in
-
Andrija Djurovic
Senior Manager, Risk Advisory - Central Europe at Deloitte
- Report this post
๐๐ญ๐๐ญ๐ข๐ฌ๐ญ๐ข๐๐๐ฅ ๐๐ข๐ง๐ง๐ข๐ง๐ ๐จ๐ ๐ง๐ฎ๐ฆ๐๐ซ๐ข๐ ๐ซ๐ข๐ฌ๐ค ๐๐๐๐ญ๐จ๐ซ๐ฌWhile binning is not a mandatory step in developing credit risk models, practitioners commonly employ it because it offers some advantages over using continuous risk factors. With time, certain principles have emerged as the best practices in the binning process.Different software offer diverse solutions for the binning process. I have authored several packages specifically designed for this purpose, such as monobin, monobinShiny, and PDtoolkit in R and monobinpy in Python. These packages encompass the most frequently employed methods I have encountered in credit risk practice. I always encourage readers and users to explore and compare different options as alternative packages are available.Check out my latest presentation for basic insights into the binning process.
217
11 Comments
Like CommentTo view or add a comment, sign in
-
Andrija Djurovic
Senior Manager, Risk Advisory - Central Europe at Deloitte
- Report this post
๐๐๐๐ฌ๐ฎ๐ซ๐ข๐ง๐ ๐๐จ๐ง๐๐๐ง๐ญ๐ซ๐๐ญ๐ข๐จ๐ง ๐๐ข๐ฌ๐คConcentration risk represents a significant source of risk in banking, particularly in emerging and small economies. Banks are required to independently assess and allocate capital buffers to mitigate concentration risk as part of Pillar 2 models. Quantifying this type of risk often presents significant challenges to the modelers.Check out my new presentation, which demonstrates a partial portfolio approach as a solution to address concentration risk.
302
8 Comments
Like CommentTo view or add a comment, sign in
-
Andrija Djurovic
Senior Manager, Risk Advisory - Central Europe at Deloitte
- Report this post
๐๐๐๐๐๐ญ๐ข๐ฏ๐ ๐๐ง๐ญ๐๐ซ๐๐ฌ๐ญ ๐๐๐ญ๐The effective interest rate is a commonly employed financial term in the domain of loans, signifying the genuine expense of borrowing for a borrower over a specified period. It encompasses the nominal interest rate and additional fees and charges associated with the loan. The attached example demonstrates calculating the effective interest rate using R and Python, assuming the overall loan cost is deducted from the initially approved amount. While real-world scenarios may differ, this example can serve as a solid basis for adjusting to specific cases.The link to a GitHub page containing the same example is available in the comment section.
86
2 Comments
Like CommentTo view or add a comment, sign in
5,524 followers
- 178 Posts
- 4 Articles
View Profile
FollowMore from this author
- IRB models uncertainty and conformal inference Andrija Djurovic 7mo
- IRB model calibration Andrija Djurovic 7mo
- Selective use of machine learning for IRB models Andrija Djurovic 8mo
Explore topics
- Sales
- Marketing
- Business Administration
- HR Management
- Content Management
- Engineering
- Soft Skills
- See All