I have been backtesting/coding for about a month on Multicharts 64 bit (about 6 months total on the 32 bit version) and I am seriously considering a license at this point. This really is an exciting piece of software. I am surprised that more traders aren't flocking to this...the backtesting capabilities are pretty incredible. Now they just need CUDA support. I am probably going to have to plop $3000-$5000 on a new server for backtesting. 48 cores sure would be nice!
I have been backtesting on roughly 2-3 years of tick data, mostly CL, some ES, using a variety of different strategies. The built in indicators/signals are enough to keep you busy for a while, and there is plenty of EasyLanguage code floating out there to help piece together some nice systems.
I optimize with 1 tick of slippage on each side, and also build in slippage. What I have seen is that this forces the system to take less trades, and shoot for higher profits.. (percent win is anywhere betwen 30-43%). I run bar magnifier (1 tick) for trade accuracy.
My most recent system, takes about 650 trades in 3 years, profits about 28,000 a year on average, and has a drawdown of around $6500. That might not seem like a lot, but after doing as much backtesting as I have, this is the best I have seen it do over a 3 year period on CL with one contract at a time.
I left out about 25% of the data for OOS, and the system does well over those remaining periods.
I have tried to consume as much information as I can in regards to over fitting strategies/curve-fitting. I am using a ton of tick data here, but 650 trades really isn't that much in the grand scheme of things. I would be more comfortable if it was like 1500 trades. But with slippage, commission, this is a lot tougher than I had originally thought. The system has an ATR trailing stop ( I wanted to avoid any hard, static, stops, although I do have a protective stop), with 2 variables, and another 4 variables for the entries.
Question: Is 650 trades too small over a 3 year time period? Is this something that I could potentially trust?
Re-optimizing: I have also tried to read on this topic, because as I move forward, I know that the system will change over time. How often, if at all, is it ok to re-optimize your system once you are live? My thought would be to always backtest over the most recent 2-3 year period, and avoiding large changes in your system, only incremental changes. Once you have a specific set of settings, forward testing it on new market data, for a period of 2-3 months before changing the system. Only actually change settings maybe once every 2-3 months.
Question 2: For those with live automated strats, do you re-optimize? If you do, how often and what type of strategy do you employ?
I have also been pondering what it might look like to have some redundancy once you are live. For instance, you enter a position, and your server goes down in CHI..do you have another server thats running as a backup? Is Multicharts capable of syncing the system and positions seamlessly? For those of you that might trade live:
Question 3: Do you employ any type of redundancy with your servers? What does it look like?