{ "id": 330444, "title": "Experiences from the field: A few things to remember about credit risk models in practice", "url": "/competitions/amex-default-prediction/discussion/330444", "author_name": "", "post_date": "2022-06-12T11:36:21.721857Z", "votes": 40, "comment_count": 12, "views": 0, "content": "
    \n

  1. The usual variables you use for credit risk modelling cant be used in practice
    \nUsual variables like income, Gender, education, profession etc cant be used to determine an individual's credit-worthiness for ethical and compliance reasons
    \nThis post by @wesamelshamy details out one act, there are a few more in place.

    2. \n

  2. Before you jump into it understand how this competition is different from other credit risk competitions
    \nMost credit risk models are used to make approval/decline decisions ( when a customer sends in an application ).
    \nThis one is different since it refers to people as customers, implying they may already be an AmEx customer. The monthly profiles for 18 months are updated to reflect any change in creditworthiness so the model for this competition is most likely to be used to cross-sell/upsell, Line extension or upgradation. The default definition here is 120 days past due ( DPD

    3. \n

  3. Take the time to understand the evaluation metric
    \nThe notebook here gives a good idea. If it scares you, I've given a crash course on interpreting any metric wrt tabular datasets here
    \nIt cares only about the rank ordering, but it may not be the best idea to optimize solely on it

    4. \n

  4. Data lag is real in finance
    \nDelinquencies and risk variables may be sporadic in nature, so does spend variables. Be sure to understand the data flow process before building these models.
    \nIt may not update every month, and then update at once. Its quite possible for risk variables of this month to have correlations with lags of previous spends

    5. \n
\n

Thanks for reading.
\nAll the best

", "messages": [ { "id": "1818198", "postDate": "06/12/2022 11:36:21", "content": "
    \n

  1. The usual variables you use for credit risk modelling cant be used in practice
    \nUsual variables like income, Gender, education, profession etc cant be used to determine an individual's credit-worthiness for ethical and compliance reasons
    \nThis post by @wesamelshamy details out one act, there are a few more in place.

    2. \n

  2. Before you jump into it understand how this competition is different from other credit risk competitions
    \nMost credit risk models are used to make approval/decline decisions ( when a customer sends in an application ).
    \nThis one is different since it refers to people as customers, implying they may already be an AmEx customer. The monthly profiles for 18 months are updated to reflect any change in creditworthiness so the model for this competition is most likely to be used to cross-sell/upsell, Line extension or upgradation. The default definition here is 120 days past due ( DPD

    3. \n

  3. Take the time to understand the evaluation metric
    \nThe notebook here gives a good idea. If it scares you, I've given a crash course on interpreting any metric wrt tabular datasets here
    \nIt cares only about the rank ordering, but it may not be the best idea to optimize solely on it

    4. \n

  4. Data lag is real in finance
    \nDelinquencies and risk variables may be sporadic in nature, so does spend variables. Be sure to understand the data flow process before building these models.
    \nIt may not update every month, and then update at once. Its quite possible for risk variables of this month to have correlations with lags of previous spends

    5. \n
\n

Thanks for reading.
\nAll the best

", "rawMarkdown": "1. **The usual variables you use for credit risk modelling cant be used in practice**\nUsual variables like income, Gender, education, profession etc cant be used to determine an individual's credit-worthiness for ethical and compliance reasons\n[This post](https://www.kaggle.com/competitions/home-credit-default-risk/discussion/57028) by @wesamelshamy details out one act, there are a few more in place.\n\n2. **Before you jump into it understand how this competition is different from other credit risk competitions**\nMost credit risk models are used to make approval/decline decisions ( when a customer sends in an application ).\nThis one is different since it refers to people as customers, implying they may already be an AmEx customer. The monthly profiles for 18 months are updated to reflect any change in creditworthiness so the model for this competition is most likely to be used to cross-sell/upsell, Line extension or upgradation. The default definition here is 120 days past due ( DPD\n\n3. **Take the time to understand the evaluation metric**\n[The notebook here](https://www.kaggle.com/code/inversion/amex-competition-metric-python) gives a good idea. If it scares you, I've given a [crash course on interpreting any metric wrt tabular datasets here](https://www.kaggle.com/competitions/tabular-playground-series-jan-2022/discussion/302195)\nIt cares only about the rank ordering, but it may not be the best idea to optimize solely on it \n\n4. **Data lag is real in finance**\nDelinquencies and risk variables may be sporadic in nature, so does spend variables. Be sure to understand the data flow process before building these models.\nIt may not update every month, and then update at once. Its quite possible for risk variables of this month to have correlations with lags of previous spends\n\nThanks for reading.\nAll the best", "votes": null }, { "id": "1818240", "postDate": "06/12/2022 12:23:38", "content": "

Thanks for the post! Although this competition is a credit risk competition, it has little relation with work-assignments

", "rawMarkdown": "Thanks for the post! Although this competition is a credit risk competition, it has little relation with work-assignments", "votes": null }, { "id": "1818386", "postDate": "06/12/2022 16:16:28", "content": "

Keep sharing your experience, dear friend.
\nYou've a great background experience in this area.

", "rawMarkdown": "Keep sharing your experience, dear friend.\nYou've a great background experience in this area.", "votes": null }, { "id": "1818418", "postDate": "06/12/2022 16:56:58", "content": "

@ravi20076 glad you liked the post.
\nIt's a competition based on real data that I've worked with in the past, hence the pointers hold true if you closely inspect the data
\nConstructive criticism is always welcome.

\n

Thanks

", "rawMarkdown": "ravi20076 glad you liked the post.\nIt's a competition based on real data that I've worked with in the past, hence the pointers hold true if you closely inspect the data \nConstructive criticism is always welcome.\n\nThanks", "votes": null }, { "id": "1818419", "postDate": "06/12/2022 16:57:11", "content": "

Thank you Grandmaster @mpwolke

", "rawMarkdown": "Thank you Grandmaster @mpwolke", "votes": null }, { "id": "1818529", "postDate": "06/12/2022 19:50:15", "content": "

Interesting, thanks!

", "rawMarkdown": "Interesting, thanks!", "votes": null }, { "id": "1819328", "postDate": "06/13/2022 16:41:38", "content": "

It's my 1st day to be here, thx for your experiences😄

", "rawMarkdown": "It's my 1st day to be here, thx for your experiences😄", "votes": null }, { "id": "1819378", "postDate": "06/13/2022 17:38:07", "content": "

wonderful insights.

", "rawMarkdown": "wonderful insights.", "votes": null }, { "id": "1820857", "postDate": "06/15/2022 04:14:00", "content": "

Thanks for sharing your insights! 🙌

", "rawMarkdown": "Thanks for sharing your insights! 🙌", "votes": null }, { "id": "1821005", "postDate": "06/15/2022 07:14:09", "content": "

Great Insights!

", "rawMarkdown": "Great Insights!", "votes": null }, { "id": "1821030", "postDate": "06/15/2022 07:45:36", "content": "

Thanks very much for sharing. Having the lagged information really should help with optimizing features.

", "rawMarkdown": "Thanks very much for sharing. Having the lagged information really should help with optimizing features.", "votes": null }, { "id": "1903972", "postDate": "08/17/2022 20:04:22", "content": "

@mpwolke, I dont understand the downvote brigade

", "rawMarkdown": "mpwolke, I dont understand the downvote brigade", "votes": null }, { "id": "1904034", "postDate": "08/17/2022 22:26:03", "content": "

Don't try to understand them. I've big issues in my life that make spare my neurons and not thinking about cancellators/downvoters.

", "rawMarkdown": "Don't try to understand them. I've big issues in my life that make spare my neurons and not thinking about cancellators/downvoters.", "votes": null } ], "comments": [ { "id": 1818240, "author_name": "ravi20076", "author_url": "", "post_date": "06/12/2022 12:23:38", "content": "

Thanks for the post! Although this competition is a credit risk competition, it has little relation with work-assignments

", "votes": null, "replies": [ { "id": 1818418, "author_name": "kritidoneria", "author_url": "", "post_date": "06/12/2022 16:56:58", "content": "

@ravi20076 glad you liked the post.
\nIt's a competition based on real data that I've worked with in the past, hence the pointers hold true if you closely inspect the data
\nConstructive criticism is always welcome.

\n

Thanks

", "votes": null, "replies": [] } ] }, { "id": 1818386, "author_name": "mpwolke", "author_url": "", "post_date": "06/12/2022 16:16:28", "content": "

Keep sharing your experience, dear friend.
\nYou've a great background experience in this area.

", "votes": null, "replies": [ { "id": 1818419, "author_name": "kritidoneria", "author_url": "", "post_date": "06/12/2022 16:57:11", "content": "

Thank you Grandmaster @mpwolke

", "votes": null, "replies": [] }, { "id": 1903972, "author_name": "kritidoneria", "author_url": "", "post_date": "08/17/2022 20:04:22", "content": "

@mpwolke, I dont understand the downvote brigade

", "votes": null, "replies": [] }, { "id": 1904034, "author_name": "mpwolke", "author_url": "", "post_date": "08/17/2022 22:26:03", "content": "

Don't try to understand them. I've big issues in my life that make spare my neurons and not thinking about cancellators/downvoters.

", "votes": null, "replies": [] } ] }, { "id": 1818529, "author_name": "saberghaderi", "author_url": "", "post_date": "06/12/2022 19:50:15", "content": "

Interesting, thanks!

", "votes": null, "replies": [] }, { "id": 1819328, "author_name": "ultramar1nescitech", "author_url": "", "post_date": "06/13/2022 16:41:38", "content": "

It's my 1st day to be here, thx for your experiences😄

", "votes": null, "replies": [] }, { "id": 1819378, "author_name": "mmmelggg", "author_url": "", "post_date": "06/13/2022 17:38:07", "content": "

wonderful insights.

", "votes": null, "replies": [] }, { "id": 1820857, "author_name": "niekvanderzwaag", "author_url": "", "post_date": "06/15/2022 04:14:00", "content": "

Thanks for sharing your insights! 🙌

", "votes": null, "replies": [] }, { "id": 1821005, "author_name": "naiborhujosua", "author_url": "", "post_date": "06/15/2022 07:14:09", "content": "

Great Insights!

", "votes": null, "replies": [] }, { "id": 1821030, "author_name": "datajmcn", "author_url": "", "post_date": "06/15/2022 07:45:36", "content": "

Thanks very much for sharing. Having the lagged information really should help with optimizing features.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1818198": "1. **The usual variables you use for credit risk modelling cant be used in practice**\nUsual variables like income, Gender, education, profession etc cant be used to determine an individual's credit-worthiness for ethical and compliance reasons\n[This post](https://www.kaggle.com/competitions/home-credit-default-risk/discussion/57028) by @wesamelshamy details out one act, there are a few more in place.\n\n2. **Before you jump into it understand how this competition is different from other credit risk competitions**\nMost credit risk models are used to make approval/decline decisions ( when a customer sends in an application ).\nThis one is different since it refers to people as customers, implying they may already be an AmEx customer. The monthly profiles for 18 months are updated to reflect any change in creditworthiness so the model for this competition is most likely to be used to cross-sell/upsell, Line extension or upgradation. The default definition here is 120 days past due ( DPD\n\n3. **Take the time to understand the evaluation metric**\n[The notebook here](https://www.kaggle.com/code/inversion/amex-competition-metric-python) gives a good idea. If it scares you, I've given a [crash course on interpreting any metric wrt tabular datasets here](https://www.kaggle.com/competitions/tabular-playground-series-jan-2022/discussion/302195)\nIt cares only about the rank ordering, but it may not be the best idea to optimize solely on it \n\n4. **Data lag is real in finance**\nDelinquencies and risk variables may be sporadic in nature, so does spend variables. Be sure to understand the data flow process before building these models.\nIt may not update every month, and then update at once. Its quite possible for risk variables of this month to have correlations with lags of previous spends\n\nThanks for reading.\nAll the best", "1818240": "Thanks for the post! Although this competition is a credit risk competition, it has little relation with work-assignments", "1818386": "Keep sharing your experience, dear friend.\nYou've a great background experience in this area.", "1818418": "ravi20076 glad you liked the post.\nIt's a competition based on real data that I've worked with in the past, hence the pointers hold true if you closely inspect the data \nConstructive criticism is always welcome.\n\nThanks", "1818419": "Thank you Grandmaster @mpwolke", "1818529": "Interesting, thanks!", "1819328": "It's my 1st day to be here, thx for your experiences😄", "1819378": "wonderful insights.", "1820857": "Thanks for sharing your insights! 🙌", "1821005": "Great Insights!", "1821030": "Thanks very much for sharing. Having the lagged information really should help with optimizing features.", "1903972": "mpwolke, I dont understand the downvote brigade", "1904034": "Don't try to understand them. I've big issues in my life that make spare my neurons and not thinking about cancellators/downvoters." }, "source": "meta" }