{
  "id": 1100,
  "title": "Multiple liquidity shocks per line of training / test sets?",
  "url": "/competitions/AlgorithmicTradingChallenge/discussion/1100",
  "author_name": "",
  "post_date": "2011-11-27T19:18:09.300Z",
  "votes": null,
  "comment_count": 1,
  "views": 1854,
  "content": "<p>&nbsp;</p>\r\n<p>If there are multiple liquidity shocks per line of the sets given (defined as a Quote following a Trade event with the same timestamp and increased spread) should we ignore any of those shocks preceding the t49-t50 event in the prediction of the bid ask\r\n values from t51 onwards?</p>\r\n<p>&nbsp;</p>\r\n<p>Thanks</p>",
  "messages": [
    {
      "id": "6819",
      "postDate": "11/27/2011 19:18:09",
      "content": "<p>&nbsp;</p>\r\n<p>If there are multiple liquidity shocks per line of the sets given (defined as a Quote following a Trade event with the same timestamp and increased spread) should we ignore any of those shocks preceding the t49-t50 event in the prediction of the bid ask\r\n values from t51 onwards?</p>\r\n<p>&nbsp;</p>\r\n<p>Thanks</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "6824",
      "postDate": "11/28/2011 04:06:48",
      "content": "<p>You shouldn't ignore them, they may be able to help you predict future price movement.</p>\r\n<p>On the other hand, the metadata doesn't pertain to them - these shocks may have information in a nearby-record (many of the records overlap), but in the test set adjacent records will not overlap, so you shouldn't try to use the tricks the training data\r\n allows.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 6824,
      "author_name": "jacobj",
      "author_url": "",
      "post_date": "11/28/2011 04:06:48",
      "content": "<p>You shouldn't ignore them, they may be able to help you predict future price movement.</p>\r\n<p>On the other hand, the metadata doesn't pertain to them - these shocks may have information in a nearby-record (many of the records overlap), but in the test set adjacent records will not overlap, so you shouldn't try to use the tricks the training data\r\n allows.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "6819": "",
    "6824": ""
  },
  "source": "meta"
}