TotalView participants vs. actual trades

Discussion in 'Order Execution' started by surfer25, Mar 18, 2011.

  1. surfer25

    surfer25

    Hi,
    I am just wondering about something:

    Let's say you are looking at TotalView for SPY which is 127.88 bid and 127.89 asked, you see perhaps 160,000 cumulative bid at 127.84 to 127.88 and 150,00 cumulative asked at 127.89 to 127.94.

    SPY will often move to either 127.83 or 127.95 with far less shares being sold or bought than the cumulative totals. I know that bids and offers are being pulled and refreshed constantly, but how useful is TotalView for determining how many shares have to be gone through to get to a certain price level?

    Thanks.
     
  2. drp7804

    drp7804

    That's a really good question... how big a grain of salt should you take with the displayed size? What percentage of the currently displayed size will disappear over the next X seconds vs. the percentage that will stay put? (I say "disappear", but it might be that some of it just shifts a tick up or down). How consistent are those percentages from timeframe to timeframe, day to day, etc.

    A lot of this is stuff that could be backtested, but one guess is that the longer a limit order has been on the book, the more likely it will hang around and let itself get hit when exposed as the best bid / offer (thinking of the guy who places limit orders before work and then sees what fish he caught at the end of the day... his orders ain't moving). Conversely, the orders that were placed very recently might have a higher likelihood of cancellation or retreating back from the NBBO (thinking HFT, scalping, etc).

    I know TotalView-ITCH labels each order with an ID so that you know when a specific order gets placed on your book and when it gets removed or changed (via fill, cancellation, etc). I'm assuming you'd be able to access that info(?)

    So you could run a backtest on that to see if there are some predictive qualities there. For the sake of argument, assume that there are (I don't actually know). So then you have a breakdown of probabilities by order lifetime: (eg) 10+ minutes = 90% chance of holding firm until the next timeframe, 5-10 minutes = 75% chance, 3-5 minutes = 57%, etc, etc.

    From that, you can then take a TotalView snapshot of the current book and estimate the amount of volume that will actually stay and trade vs. the volume that will flee the scene. Going with our example, if there are 5000 shares-worth of orders which have been on the book for 10+ minutes, then you can estimate that 4500 shares of that will hang around until the next timeframe... the other 500 shares will flee (90%). Continue with the rest of your breakdown until you have a complete picture. The resulting "predicted book" is your estimate of how many shares will no-shit stay and welcome execution vs. how many will cancel/retreat.

    That "order lifetime" idea is just one way to approach this (hope it makes sense). This is something I'd like to take a look at, as well. Another item on the to-do list :)