{
  "id": 1063,
  "title": "A better benchmark",
  "url": "/competitions/AlgorithmicTradingChallenge/discussion/1063",
  "author_name": "",
  "post_date": "2011-11-20T05:53:34.237Z",
  "votes": null,
  "comment_count": 3,
  "views": 2670,
  "content": "<p>If you can't get an RMSE equal to 0.85067, or lower, here's a tip:</p>\r\n<p>simply repeat the bid50 and ask50 for each row, for your whole entry. No change at all.</p>\r\n<p>You will achieve this RMSE.</p>\r\n<p>This should suggest quite a bit as to how to go about the rest of the competition too.</p>",
  "messages": [
    {
      "id": "6622",
      "postDate": "11/20/2011 05:53:34",
      "content": "<p>If you can't get an RMSE equal to 0.85067, or lower, here's a tip:</p>\r\n<p>simply repeat the bid50 and ask50 for each row, for your whole entry. No change at all.</p>\r\n<p>You will achieve this RMSE.</p>\r\n<p>This should suggest quite a bit as to how to go about the rest of the competition too.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "6626",
      "postDate": "11/20/2011 07:31:51",
      "content": "<p>Thanks for the tip JacobJ</p>\r\n<p>BTW: my initial attempt was way off since I tried using avg bid and avg ask over the full 50 periods rather than values around the event.</p>\r\n<p>This showed up one quirk in the data. Particularly in cases close to market open (around 8am) there are numerous cases of bogus quotes. It seems to settle down once the first trade is recorded, but data needs to be filtered.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "6629",
      "postDate": "11/20/2011 09:58:31",
      "content": "<p>If you can't get an RMSE equal to 0.85067, plotting the data and looking at it is another good tip :)</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "6641",
      "postDate": "11/20/2011 22:52:35",
      "content": "<p>@JacobJ:<br>\r\nThanks for this information. I had assumed that this was the naive benchmark approach. But all my testing on the training data are consistent with your result - that this is better than the linear benchmark. It looks like the naive benchmark is based on continuing\r\n the bid/ask for the last transaction, not necessarily bid50/ask50.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 6626,
      "author_name": "zenbot",
      "author_url": "",
      "post_date": "11/20/2011 07:31:51",
      "content": "<p>Thanks for the tip JacobJ</p>\r\n<p>BTW: my initial attempt was way off since I tried using avg bid and avg ask over the full 50 periods rather than values around the event.</p>\r\n<p>This showed up one quirk in the data. Particularly in cases close to market open (around 8am) there are numerous cases of bogus quotes. It seems to settle down once the first trade is recorded, but data needs to be filtered.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 6629,
      "author_name": "alecstephenson",
      "author_url": "",
      "post_date": "11/20/2011 09:58:31",
      "content": "<p>If you can't get an RMSE equal to 0.85067, plotting the data and looking at it is another good tip :)</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 6641,
      "author_name": "dejavu",
      "author_url": "",
      "post_date": "11/20/2011 22:52:35",
      "content": "<p>@JacobJ:<br>\r\nThanks for this information. I had assumed that this was the naive benchmark approach. But all my testing on the training data are consistent with your result - that this is better than the linear benchmark. It looks like the naive benchmark is based on continuing\r\n the bid/ask for the last transaction, not necessarily bid50/ask50.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "6622": "",
    "6626": "",
    "6629": "",
    "6641": ""
  },
  "source": "meta"
}