Forward test with odd lots?

Discussion in 'Automated Trading' started by 6pst6, Dec 11, 2009.

  1. 6pst6


    Is there any ECN that will allow 1 share lots to forward test automated strategies for realistic fills or is considered taboo?
  2. dloyer


    Most brokers allow odd lots, but I have found that the fills you get are not as good as you will get if you trade > 100 shares.

    You will not find odd lots < 100 shares on the time and sales tape. They dont have to be reported and a limit order for < 100 shares will not move the national best bid or ask.

    Also, most brokers have minimum charges, so your cost will be much higher on a per share basis.

    Still, not a bad way to test execution logic. When I take a new system live, I usually start with small positions, a few $k or so and intervean if needed until I get the logic worked out.
  3. 6pst6


    Thanks for the reply.

    Yeah, I was thinking that even if they tripled the ECN cost per share...I'm still talking pennies.

    Getting worse fills makes sense, and would makes for an even better forward test as the results really need to stand out to make up for it.
  4. I have found just the opposite. I used to have a bottom-fishing strategy with < 100 shares that occasionally had fills BELOW the posted low for the day. These were limit orders though. Market orders may be a different story.
  5. promagma


    Don't do it on ARCA. NASDAQ and BATS are technically ok but many brokers still won't like it, especially if you are daytrading.
  6. rosy2


    what good is one lot going to do? how can you test partial fills or averages
  7. 6pst6


    The strategy wouldn't be share heavy (max 400, more than likely ~200) with several dozen open positions at once. Partial fills and averages won't have a large impact and are easier to calculate.
  8. my personal experience has been that odd lots make a huge different (I auto-trade with IB)
    If you backtest with highly liquid stocks (> 10 mill shares/day) and you assume 2 cents slippage, you can double that slippage as soon as you introduce odd lots.

    That doesn't mean that you can't use limit orders, ny experience has been that when I use odd lots my chances for getting filled on these limit orders are very low (even when I put a limit at the Ask to go long I wouldn't get filled as often as I wanted).

    Obviously highly depends on the strategy, my testing was all based on day trades in the first/last 90 minutes of the market. If you backtest for swing trades, 4 cents may not make such a big difference.

    Just my 2 cents.
  9. dloyer


    Engine: Good point and useful data.

    I use TradeStation for execution and notice that the fills also depend on the route. The worse routes are EDGX and EDGA. Tradestation by default will route to these exchanges because they tend to have the lowest execution costs (for TradeStation).

    On these routes, I observe lower fill rates than others.
  10. Obviously from our different experiences with odd lots the answer is "it depends". I have found that a better approach is to run your system real-time but instead of actually executing trades just log it along with time & sales include bid and ask. Then review each trade and judge for yourself the likely-hood of a favorable outcome.

    For example say your system always looks to buy one tick above the current high for the session. Time & sales show that typically prices bounce off that high, pull back and try to break through again and again. Finally one little trade occurs at +1 tick and BANG, that trigger many simultaneous buy orders. Time & sales shows that at the moment you would have placed your order there were successive trades at ever higher prices almost instantaneously until it finally settles many ticks above where your simulator expected a fill.

    What you can count on is the computers near the exchange got filled first, the people in the same city were filled next and somewhere way down the line your order filled. Depending on how much time and money you are devoting to "being there first" you were in the group filled several ticks higher. This kind of analysis is good in any case, and it lets you screen out strategies you don't even need to bother test-trading in realtime.
    #10     Dec 13, 2009