{
  "id": 286772,
  "title": "Autoregressive models",
  "url": "/competitions/g-research-crypto-forecasting/discussion/286772",
  "author_name": "",
  "post_date": "2021-11-10T15:26:22.807930600Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>There has been some confusion in the discussion section as to what we are actually predicting. In my understanding, the API serves us with rows in an iterative fashion starting, say at <code>t=0</code>, and continuing until the three months have passed. For each row, which is an asset <code>a</code> at time <code>t</code> we are supposed to forecast the residualized log-returns that will be realized 15 minutes into the future for this <code>a</code>.  </p>\n<p>What is unclear is which granularity of data will be provided by the API. Can we assume that it will minutes as in <code>train.csv</code>? This might be relevant for autoregressive models.</p>",
  "messages": [
    {
      "id": "1578042",
      "postDate": "11/10/2021 15:26:22",
      "content": "<p>There has been some confusion in the discussion section as to what we are actually predicting. In my understanding, the API serves us with rows in an iterative fashion starting, say at <code>t=0</code>, and continuing until the three months have passed. For each row, which is an asset <code>a</code> at time <code>t</code> we are supposed to forecast the residualized log-returns that will be realized 15 minutes into the future for this <code>a</code>.  </p>\n<p>What is unclear is which granularity of data will be provided by the API. Can we assume that it will minutes as in <code>train.csv</code>? This might be relevant for autoregressive models.</p>",
      "rawMarkdown": "There has been some confusion in the discussion section as to what we are actually predicting. In my understanding, the API serves us with rows in an iterative fashion starting, say at `t=0`, and continuing until the three months have passed. For each row, which is an asset `a ` at time `t` we are supposed to forecast the residualized log-returns that will be realized 15 minutes into the future for this `a`.  \n\nWhat is unclear is which granularity of data will be provided by the API. Can we assume that it will minutes as in `train.csv`? This might be relevant for autoregressive models.",
      "votes": null
    },
    {
      "id": "1578122",
      "postDate": "11/10/2021 16:57:15",
      "content": "<p>The API will continue to provide minute resolution data, yes.</p>",
      "rawMarkdown": "The API will continue to provide minute resolution data, yes.",
      "votes": null
    },
    {
      "id": "1651880",
      "postDate": "01/16/2022 06:35:48",
      "content": "<p>For AR model, we need the past time step TARGET value as current time step INPUT, this is possible? <a href=\"https://www.kaggle.com/sohier\" target=\"_blank\">@sohier</a> </p>",
      "rawMarkdown": "For AR model, we need the past time step TARGET value as current time step INPUT, this is possible? @sohier",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1578122,
      "author_name": "sohier",
      "author_url": "",
      "post_date": "11/10/2021 16:57:15",
      "content": "<p>The API will continue to provide minute resolution data, yes.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1651880,
      "author_name": "feitengli",
      "author_url": "",
      "post_date": "01/16/2022 06:35:48",
      "content": "<p>For AR model, we need the past time step TARGET value as current time step INPUT, this is possible? <a href=\"https://www.kaggle.com/sohier\" target=\"_blank\">@sohier</a> </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1578042": "There has been some confusion in the discussion section as to what we are actually predicting. In my understanding, the API serves us with rows in an iterative fashion starting, say at `t=0`, and continuing until the three months have passed. For each row, which is an asset `a ` at time `t` we are supposed to forecast the residualized log-returns that will be realized 15 minutes into the future for this `a`.  \n\nWhat is unclear is which granularity of data will be provided by the API. Can we assume that it will minutes as in `train.csv`? This might be relevant for autoregressive models.",
    "1578122": "The API will continue to provide minute resolution data, yes.",
    "1651880": "For AR model, we need the past time step TARGET value as current time step INPUT, this is possible? @sohier"
  },
  "source": "meta"
}