Sorry for rambling on group call…will try and explain it here…(with more rambling)
When thinking about selection bias , I tested using open<=buylimit (would require a limit on open order)
The first impact is a significant reduction in selection bias by dropping the number of possible buy orders each day.
The second impact on system metrics, which I did not explain well, was not in reducing the selection bias further but looking at the potential impact of selection bias on backtest results. This type of entry (open <=buylimit) may make the backtest results more closely aligned with live results on those days when buysignals outnumber positions available. In currect MOC systems, my understanding is that a random selection of possible stocks are selected by Amibroker when in reality (as per Nick's posts and modules) this is not the case when using buylimit orders with EOD data. If a limit on open order is used, then the live results and the backtest results should both be selecting stocks on a "random" basis on these rare days when buysignals outnumber positions available. Going forward, the two results (backtest v live) should be more closely aligned (I think ?)
The reason I brought it up in the online call in the first place was I thought it may be of interest to some who had tossed out some of their MOC systems due to problems with selection bias.
Try changing entry to:
LE = Ref(buysetup,-1) AND open<=Ref(buylimit,-1) AND NOT OnLastTwoBarsOfDelistedSecurity ;
LEPrice = open;
Maybe some of those discarded systems can be resurrected ? For those not on the call, Nick mentioned there are hurdles trying to trade this way (gap down on open) so it may not work live still
Clear as mud ?
