Credit Risk Scorecards: Developing And Implementing Intelligent Credit Scoring. Author: Naeem Siddiqi. Publication: · Book. Credit Risk Scorecards. Credit Risk Scorecards: Developing and Implementing Intelligent Credit Scoring. Editor(s). Naeem Siddiqi. First published September As the follow-up to Credit Risk Scorecards, this updated second edition NAEEM SIDDIQI is the Director of Credit Scoring and Decisioning with SAS® Institute.
|Published (Last):||11 November 2011|
|PDF File Size:||14.21 Mb|
|ePub File Size:||13.76 Mb|
|Price:||Free* [*Free Regsitration Required]|
But where did it all start?
What was the first business application of predictive analytics? It is hard to pin down one particular application but one of the earliest and highly successful applications is certainly credit risk models and retail credit scorecards.
Credit scorecards help banks assess the credit worthiness and future repayment capability of their borrowers. Credit scorecard development is a highly structured and well defined analytical process. It is a good idea for every analytics and data science professional to be familiar with this process. Developing and Implementing Intelligent Credit Scoring. You have authored one of the most influential books for the practitioners of retail credit risk measurement.
The 2 nd edition of this book is expected to be published later this year. What kept you busy since you published the first edition over 10 years ago? Lots of work and travel. I have learned so much more in the past 10 years, and hope to incorporate some of that knowledge into the 2 nd edition of my book.
While there are a lot of regional nuances for scorecsrds banks lend, the basics are pretty much the same. Between the two editions of your book on credit scorecards, inthe planet has seen the worst credit crisis in its history.
Credit Risk Scorecards : Naeem Siddiqi :
Where do you think credit scorecards failed in that entire fiasco? Credit scorecards, or models, did not cause the credit crash. Models are tools that are very useful when used judiciously, recognizing both their strengths and weaknesses. The credit crisis was a complex event that included failures of risk management, distorted incentives and some fraud. That credlt a good point Naeem all predictive models require judicious and responsible usage and credit scorecards are no different.
What are the new directions in the retail credit industry for risk measurement since you wrote the first edition of your book in ?
Siddiqi Naeem. Intelligent Credit Scoring: Building and Implementing Better Credit Risk Scorecards
There is certainly more emphasis being placed on governance and model risk. In addition, there is on-going discussions on new algorithms. However, the vast majority of banks continue to use simple, transparent techniques such as logistic regression and scorecards.
In many countries, credit bureaus have started, which provides new data sources for lenders. In others, lenders are looking at alternate data sources such as utility and cell phone bill payments, as well as social media data.
I also see more banks creating large corporate data warehouses. These have a more comprehensive customer view, and help build better naeme. However, do you see a possibility for a repeat of crisis in the future? What measures do you think are required to avoid such an event?
Better governance, stronger and independent risk management functions, and sensible incentives will help. I also think better communications in terms of explaining the strengths and weaknesses of models, credih how to use them properly will be key.
Bankers need to do their jobs and exercise conservatism.
Are scorecardss banks and financial institutions better prepared now to avoid a crisis like that? Basel II has helped quite a bit in creating truly independent risk functions, and many non-Basel II have adopted its recommendations as best practices. I see more focus on risk management, and creating better infrastructures for the development and maintenance of models.
All those add up to create healthier risk management environments at the banks and certainly more oversight for credit scoring. How is credit underwriting different for this new industry?
How does one build robust scorecards for P2P lending? We should still rely on lending principles such as looking at character, capacity, collateral and conditions. I would hope that P2P lenders are using these prudent risk management principles to lend money, including the use naeek scores as well as policy rules.
Credit Risk Scorecards : Developing and Implementing Intelligent Credit Scoring
Building good scorecards is possible with large volumes of good clean data. Some P2P lenders may not be in a position to build these models as they may have very low volumes.
In such cases, I would suggest using generic bureau scores and some judgment. Otherwise, the same rules of scorecard development apply for P2P lenders as for bankers. Data science and analytics have evolved to a new level in the last decade with the explosion of big data technologies. How do you see credit risk scoring change in this new environment? Big Data has allowed banks to do things such as more frequent scoring.
For example, many credit card companies now score customers on a nightly basis, as compared to monthly in the past. What used to be done monthly is now being done daily, and what was done daily is now being calculated in real time.
Many banks have invested in very large integrated data warehouses where they can now use data sources that were not available to them in the past. At SAS, our customers use for example, transactional data from ATM usage, savings and checking accounts in their behavior scorecards.
Banks are also starting to build models on full populations, instead of using sampling, simply because they now have more powerful machines to do such tasks. Do you see a role for artificial intelligence or deep learning in credit scoring?
This is certainly possible in some parts of credit risk management, such as fraud analytics. There are regulatory requirements that impose a high standard of transparency, interpretability and general openness on risk models. Banks tend naeej choose the methods that make compliance and audit easier. What is your opinion about using non-traditional data sources like social riso for the development of credit scorecards? I would suggest a lot of caution in using those sources.
There are major concerns around privacy, ethics, reputational risk, dubious causality of the data, and of course, reliability. Credit scoring is becoming a more widely known topic, and as people become aware, they can easily alter their profiles to fit what they think is good credit risk behavior. There is nothing stopping anyone sckrecards creating a fake profile or altering their own to like these things.
In my views, much of the social media data eventually relates back to the traditional data such as income and ability to pay debt servicewhich is a lot more reliable and can be better explained. In your opinion, what are the future directions for retail lending and credit scorecards? I certainly see more usage of real-time data in areas such as collections and authorizations.
I think the usage of more complex algorithms such as machine learning are also inevitable, but will depend on changes in the regulatory and model validation functions. We are also seeing integration naem credit scoring and the finance function in terms of calculating expected losses. The alternate lenders will get bigger, but that growth will depend on how they get regulated. You started your screcards over 20 years ago as a risk analyst, and have become a prominent ceedit in the credit risk community since then.
At what point of time in your life did you get interested in risk analysis?
I graduated with my MBA and was looking for a job. I had naeeem to a bunch of different places and a credit risk analyst position was the first one that came through. I thought credit scoring was a good application of a mixture of technical stats, maths and business knowledge, so I stayed in it. Every single position has been new. I quite like creating my own path, and establishing everything from scratch.
I have also been extremely lucky to have very good mentors, managers, colleagues, and clients who have generously shared their knowledge with me. I try to pass some of it forward. Could riwk please tell us about some of the projects or assignments you have worked on in your career that you enjoyed the most? There have been many. We improvised a lot. Creating analytics based champion-challenger strategies for authorizations and credit limit management back in was eye opening for me, as most banks did that sort of thing manually back then.
Helping create an industry-leading credit scoring solution at SAS has been a great journey. There was a lot of inertia at first, but the idea has caught on across the world.
Credit Scoring is truly global! What advice do you have for young professionals who want to start their careers in risk analytics and credit risk modeling? Get a good quantitative degree, and then immerse yourself in the business.
If you are working in a bank, building scorecards is a business activity, not an academic exercise, so adapt and vredit accordingly. Talk to practitioners and try to understand, for example, the rjsk of lending money, managing risk or collections etc. That practical advice will help you create better, more useful models. There is demand for credit scoring professionals in every single country that I have visited, so you have a lot of choice and bright career prospects.
I thought it was about time. Actually, it was probably time many years ago, but my hectic travel and work schedule never allowed me the time to sit down and write. So, I wrote some shorter papers in the meantime. But late last year I finally decided to bite the bullet and set myself a deadline. Also I have learned a lot in the past 10 years, and I want to pass on that knowledge to others.
Rism I mentioned earlier, I have been lucky to have had great mentors who have shared their knowledge with me. The book is my way of passing the favor forward. What are the major changes in this latest edition of your book from the previous version? There will be a lot more practical examples throughout the book, including an end-to-end example using real data. The main theme of the book remains the same i.