{
  "id": 297592,
  "title": "a subject matter question",
  "url": "/competitions/g-research-crypto-forecasting/discussion/297592",
  "author_name": "",
  "post_date": "2021-12-28T11:05:27.588392800Z",
  "votes": 5,
  "comment_count": 1,
  "views": 0,
  "content": "<p>It seems like, we are talking/aiming at value .05-.06 for eval metric (or more)<br>\nIn absolute terms, this number doesn't reflect a high correlation with real world data.<br>\nMy question is: would a model producing output in this region be of use in real life?<br>\nAny lights you can shed on this?</p>\n<p>Note: you dont have to reveal your 'million dollar trading strategy', just clues will be OK :-)</p>",
  "messages": [
    {
      "id": "1631367",
      "postDate": "12/28/2021 11:05:27",
      "content": "<p>It seems like, we are talking/aiming at value .05-.06 for eval metric (or more)<br>\nIn absolute terms, this number doesn't reflect a high correlation with real world data.<br>\nMy question is: would a model producing output in this region be of use in real life?<br>\nAny lights you can shed on this?</p>\n<p>Note: you dont have to reveal your 'million dollar trading strategy', just clues will be OK :-)</p>",
      "rawMarkdown": "It seems like, we are talking/aiming at value .05-.06 for eval metric (or more)\nIn absolute terms, this number doesn't reflect a high correlation with real world data.\nMy question is: would a model producing output in this region be of use in real life?\nAny lights you can shed on this?\n\nNote: you dont have to reveal your 'million dollar trading strategy', just clues will be OK :-)",
      "votes": null
    },
    {
      "id": "1631435",
      "postDate": "12/28/2021 12:13:41",
      "content": "<p>This is a very good question. <br>\nI observed that for shorter prediction horizont, the Pearson correlation is considerably higher, e.g. 0.5 for a 5 Minutes prediction horizont, which could be seen as a high correlation.<br>\nBut the competition targets a 16 Minutes ahead prediction, where most of \"known\" models perform modestly.</p>\n<p>I assume that the netto winning potential is higher for a larger prediction horizont, but, at the same time the prediction quality is lower. With realistic data about trading costs it would be possible to answer your question.</p>",
      "rawMarkdown": "This is a very good question. \nI observed that for shorter prediction horizont, the Pearson correlation is considerably higher, e.g. 0.5 for a 5 Minutes prediction horizont, which could be seen as a high correlation.\nBut the competition targets a 16 Minutes ahead prediction, where most of \"known\" models perform modestly.\n\nI assume that the netto winning potential is higher for a larger prediction horizont, but, at the same time the prediction quality is lower. With realistic data about trading costs it would be possible to answer your question.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1631435,
      "author_name": "vi2018",
      "author_url": "",
      "post_date": "12/28/2021 12:13:41",
      "content": "<p>This is a very good question. <br>\nI observed that for shorter prediction horizont, the Pearson correlation is considerably higher, e.g. 0.5 for a 5 Minutes prediction horizont, which could be seen as a high correlation.<br>\nBut the competition targets a 16 Minutes ahead prediction, where most of \"known\" models perform modestly.</p>\n<p>I assume that the netto winning potential is higher for a larger prediction horizont, but, at the same time the prediction quality is lower. With realistic data about trading costs it would be possible to answer your question.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1631367": "It seems like, we are talking/aiming at value .05-.06 for eval metric (or more)\nIn absolute terms, this number doesn't reflect a high correlation with real world data.\nMy question is: would a model producing output in this region be of use in real life?\nAny lights you can shed on this?\n\nNote: you dont have to reveal your 'million dollar trading strategy', just clues will be OK :-)",
    "1631435": "This is a very good question. \nI observed that for shorter prediction horizont, the Pearson correlation is considerably higher, e.g. 0.5 for a 5 Minutes prediction horizont, which could be seen as a high correlation.\nBut the competition targets a 16 Minutes ahead prediction, where most of \"known\" models perform modestly.\n\nI assume that the netto winning potential is higher for a larger prediction horizont, but, at the same time the prediction quality is lower. With realistic data about trading costs it would be possible to answer your question."
  },
  "source": "meta"
}