Once an ATS has been proven stable in terms of doing what it's programmed to do without glitches, how long do you recommend testing it in simulated accounts before going live? If this has already been answered somewhere please link me up, thx.
I am very conservative, so I require deep backtesting, min 3000 sample size, and forward testing of three months. Nothing ever seems to sim test forward like the backtests, but if after three months it is fairly close, I run a small account with it. I have three day trading programs and one very long term program automated. It is funny how many errors pop up sometimes in the three months that never came up in the backtesting.
Can you give some examples so that we may learn from your mistakes? My answer to the OP would be 6 months. More importantly: trade it for another 6 months with small size to see what kind of slippage to expect.
If the system has been debugged and QA'ed properly (ie. at the level that a real software product would be QA'ed, including corner cases, etc), then it seems like you should take the system live immediately. There are a lot of things that I've run into after bringing a system live that I never would have found out via sim-testing. For example, if you use limit orders, there's a decent likelihood that your target is hit without you getting filled. How do you account for this? Do you use Market-if-touched and endure the slippage? What about things like loss of internet connection, or your server dies? Are the orders placed at the exchange, or are they simulated locally? I used Ninjatrader, and one time in the middle of the night, my Internet dropped and I had to scramble to figure out how to get connected. It turns out that NT simulates OCO orders locally, so I had a buy AND sell order in that could both be triggered. What about partial fills and then the market moves in the direction of your trade? Will your ATS automatically cancel any outstanding entry orders once you reach a certain point? I've also encountered timing conditions where the markets move too quickly. I use stop-limit orders to enter my trades in order to eliminate slippage, but I've found that if the markets move too quickly at my stop price, when the stop limit is placed, it could be below the market price, which causes the orders to get rejected. I've had issues where I sent a sell order to the Eurex and it never got accepted, so I had a hanging position far past where I wanted to sell. There are a lot of things that can go wrong besides just trading logic problems. What I would suggest is if you are satisfied with the way the system behaves, then bring it live immediately (with small size as suggested by the others), but watch it like a hawk, and under various market conditions until you're satisified.