{
  "id": 477476,
  "title": "An example of HomeCredit model failure after COVID outbreak",
  "url": "/competitions/home-credit-credit-risk-model-stability/discussion/477476",
  "author_name": "",
  "post_date": "2024-02-16T09:51:16.989266200Z",
  "votes": 6,
  "comment_count": 2,
  "views": 0,
  "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F494641%2F1d2610af182262a58dc3a13f44425cd6%2Fnewplot%20(19).png?generation=1708076886452255&amp;alt=media\"></p>\n<p>It seems that since April 2020 HomeCredit model continues to predict lower default probability for weekends, based on the trend from before COVID (see my post <a href=\"https://www.kaggle.com/competitions/home-credit-credit-risk-model-stability/discussion/477470\" target=\"_blank\">here</a>)</p>\n<p>But the mean default rates do not follow the same pattern any more.</p>\n<p><a href=\"https://www.kaggle.com/jetakow\" target=\"_blank\">@jetakow</a> I am interested in your view on this. Did I get this right?</p>",
  "messages": [
    {
      "id": "2654619",
      "postDate": "02/16/2024 09:51:16",
      "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F494641%2F1d2610af182262a58dc3a13f44425cd6%2Fnewplot%20(19).png?generation=1708076886452255&amp;alt=media\"></p>\n<p>It seems that since April 2020 HomeCredit model continues to predict lower default probability for weekends, based on the trend from before COVID (see my post <a href=\"https://www.kaggle.com/competitions/home-credit-credit-risk-model-stability/discussion/477470\" target=\"_blank\">here</a>)</p>\n<p>But the mean default rates do not follow the same pattern any more.</p>\n<p><a href=\"https://www.kaggle.com/jetakow\" target=\"_blank\">@jetakow</a> I am interested in your view on this. Did I get this right?</p>",
      "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F494641%2F1d2610af182262a58dc3a13f44425cd6%2Fnewplot%20(19).png?generation=1708076886452255&alt=media)\n\nIt seems that since April 2020 HomeCredit model continues to predict lower default probability for weekends, based on the trend from before COVID (see my post [here](https://www.kaggle.com/competitions/home-credit-credit-risk-model-stability/discussion/477470))\n\nBut the mean default rates do not follow the same pattern any more.\n\n@jetakow I am interested in your view on this. Did I get this right?",
      "votes": null
    },
    {
      "id": "2654846",
      "postDate": "02/16/2024 13:32:38",
      "content": "<p>Nice finding, I have to disclose that kagglers didn't get all the data we have inhouse, it was specifically undersampled so it has some properties before we published the data. You can of course utilize the finding that during weekends the risk is lower. It may be the case later in test sample, or it may not. </p>",
      "rawMarkdown": "Nice finding, I have to disclose that kagglers didn't get all the data we have inhouse, it was specifically undersampled so it has some properties before we published the data. You can of course utilize the finding that during weekends the risk is lower. It may be the case later in test sample, or it may not.",
      "votes": null
    },
    {
      "id": "2655003",
      "postDate": "02/16/2024 16:27:47",
      "content": "<p>Yeah, based on comparisons vs the past, I had the feeling the data in <code>applprev</code> files in 2020 is heavily undersampled. </p>",
      "rawMarkdown": "Yeah, based on comparisons vs the past, I had the feeling the data in `applprev` files in 2020 is heavily undersampled.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2654846,
      "author_name": "jetakow",
      "author_url": "",
      "post_date": "02/16/2024 13:32:38",
      "content": "<p>Nice finding, I have to disclose that kagglers didn't get all the data we have inhouse, it was specifically undersampled so it has some properties before we published the data. You can of course utilize the finding that during weekends the risk is lower. It may be the case later in test sample, or it may not. </p>",
      "votes": null,
      "replies": [
        {
          "id": 2655003,
          "author_name": "narsil",
          "author_url": "",
          "post_date": "02/16/2024 16:27:47",
          "content": "<p>Yeah, based on comparisons vs the past, I had the feeling the data in <code>applprev</code> files in 2020 is heavily undersampled. </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2654619": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F494641%2F1d2610af182262a58dc3a13f44425cd6%2Fnewplot%20(19).png?generation=1708076886452255&alt=media)\n\nIt seems that since April 2020 HomeCredit model continues to predict lower default probability for weekends, based on the trend from before COVID (see my post [here](https://www.kaggle.com/competitions/home-credit-credit-risk-model-stability/discussion/477470))\n\nBut the mean default rates do not follow the same pattern any more.\n\n@jetakow I am interested in your view on this. Did I get this right?",
    "2654846": "Nice finding, I have to disclose that kagglers didn't get all the data we have inhouse, it was specifically undersampled so it has some properties before we published the data. You can of course utilize the finding that during weekends the risk is lower. It may be the case later in test sample, or it may not.",
    "2655003": "Yeah, based on comparisons vs the past, I had the feeling the data in `applprev` files in 2020 is heavily undersampled."
  },
  "source": "meta"
}