addressalign-toparrow-leftarrow-rightbackbellblockcalendarcameraccwcheckchevron-downchevron-leftchevron-rightchevron-small-downchevron-small-leftchevron-small-rightchevron-small-upchevron-upcircle-with-checkcircle-with-crosscircle-with-pluscontroller-playcrossdots-three-verticaleditemptyheartexporteye-with-lineeyefacebookfolderfullheartglobegmailgooglegroupshelp-with-circleimageimagesinstagramFill 1light-bulblinklocation-pinm-swarmSearchmailmessagesminusmoremuplabelShape 3 + Rectangle 1ShapeoutlookpersonJoin Group on CardStartprice-ribbonprintShapeShapeShapeShapeImported LayersImported LayersImported Layersshieldstartickettrashtriangle-downtriangle-uptwitteruserwarningyahoo

Use of Logisitc regression and Linear regression for Credit Risk Underwriting

Ram will discuss how the financial industry uses data analytics to predict default likelihood and the dollar amount at risk by the defaults before and after recovery.  

The financial industry relies heavily on the use of data to maximize profitability. Despite the proliferation of financial products and asset classes, loans still ranks as a very important asset class for the industry, and having an accurate credit risk underwriting model is critical for those in the business of making loans.  In this talk, Ram will go through the approval/pricing framework used in an auto loan use case. Among the topics that he will go through include features/variables selection and their expected impact on the outcome, dimensionality reduction, model development and validation, and the types of algorithms used for each step within the framework. He will also discuss other possible use cases/industries that can benefit from this framework.  The discussion will come from a statistics perspective although the mapping to ML will be noticeable to the audience.  He will also welcome comments from those who have used machine learning for credit underwriting purposes. 

Bio: Ramkumar Ravichandran is Senior Business Analytics Manager of the Product Analytics & Strategy team at Move, Inc (owner of Realtor.com). He works with Product teams to solve business problems(Strategic and Tactical) via Business Analytics and Predictive Modeling. Product teams leverage the insights to drive decisions on Product performance, future product rollouts & Business Strategy. Prior to Move, he had led and executed Business & Advanced Analytics initiatives for Banking, Financial & Insurance firms in North America and Europe. For more background please refer to his Linkedin page. 

Lastly, Ram’s company is looking to hire people.  Job openings can be found here.




Join or login to comment.

  • Emre

    1 · December 4, 2013

    • ram

      Thanks

      December 4, 2013

  • Ankit

    Can we have the slides from the presentation?

    December 4, 2013

  • Karl A.

    December 3, 2013

    • ram

      Yes, the problem was very similar. They had some interesting ML techniques (combo of multiple ML techniques instead of one) to solve.

      December 4, 2013

  • A former member
    A former member

    Ram, thank you for an excellent presentation. Your presentation helped us understand more about how this problem is modeled and tested by a statistician. This can help educate data mining team members from different disciplines to understand each others terminology and work better as a cohesive team.

    1 · December 4, 2013

    • ram

      Thanks...

      December 4, 2013

  • Gregory C.

    Good stuff!

    1 · December 4, 2013

    • ram

      Thanks

      December 4, 2013

  • ram

    Some reference materials to answer the questions from the talk:

    1. Cost Function - How to decide on most parsimonious model (least variables with best prediction accuracy): http://www.modelselection.org/aic/
    2. Population Stability Index - How to identify which historical time window behaves similar to your analysis window: http://support.sas.com/resources/papers/proceedings10/288-2010.pdf

    Hope that helps. If you have additional questions, please reach out to me - would be glad to help.

    December 4, 2013

  • A former member
    A former member

    The presentation was very hard to follow since it was constantly interrupted with questions. It would have been much better to ask people to hold their questions until the end. The presentation also assumed a bit to much statistical background and dumbing it down a bit would have helped a lot.

    December 4, 2013

  • Dan B.

    Meetupers,

    I just finished a tech-tip about using Logistic Regression to predict short-term moves of the stock market:

    http://www.bikle.com/techtips/libsvm#logreg

    I look forward to attending this Meetup and talkin-shop!
    -- Dan

    1 · December 2, 2013

    • Song C.

      Hi Ram, I have read your slides. Very informative and clear explanation. Just a question, how often do you need to re-train your parameters and update the training samples? The reason I ask this question is that I see there is no modeling on the time-drifting factor. My intution is that finance data are time dependent data. If the analytics model has no time-dependent parameters, probably we need to re-train the parameters after some days. I just wonder does this theory apply to your case as well? Please correct me if I am wrong as I am not in a finance expert. See you tonight!

      December 3, 2013

    • ram

      You are right. Model performances are tracked every month - Actual vs. Predicted Default % by buckets. Typically models deteriorate after 6 months and need to be revised (either entire remodeling from scratch or adjustments of weights). We can talk more today after the presentation.

      December 3, 2013

  • David H.

    I have to change my reservation as something came up. Anyone recording this? R, Excel, Matlab or Phyton examples?

    December 2, 2013

    • ram

      If you are not able to download from there - I can mail them to you

      December 2, 2013

    • David H.

      I got it. Thanks.

      December 3, 2013

People in this
Meetup are also in:

Sign up

Meetup members, Log in

By clicking "Sign up" or "Sign up using Facebook", you confirm that you accept our Terms of Service & Privacy Policy