$1400? Is there anything less? I saw tickdata.com or something said they sell 1 symbol of data for $20. Is their format compatible with all backtesting software?
I'm not familiar with tickdata.com. Opentick.com is cheap, but I found their data to be unreliable. I have heard though, that they have improved substantially. I think Amibroker with InteractiveBrokers is the cheapest. It's a few hundred dollars for the software and you get data at no additional cost from IB. If you want longer histories, get IBcollector for $150 and collect whatever data histories you want from IB, import into Amibroker, and backtest away.
even in realtime, IB aggregates their tick data, such that it is not true tick by tick . it's agg'd on the time/sales if you watch and compare it to data from TS or Esignal. i use IB as a broker, but just be aware of that caveat. fine for most timeframes, but could be problematic for some systems , especially those that consider tick delta's etc. that's why i use TS for my realtime tick by tickdata. their brokerage (they actually use obrien iirc) sux.
Getting data out of IB is like pulling hen's teeth. I found that I can use free Ninjatrader and make an account setting for Opentick and get tons of data. For the YM, the instrument of interest here currently, it looks to be fine. From Ninjatrader you can output it in .csv format and import it to other packages. I fully agree with hypo's statement about simplicity. If it isn't really simple it probably is not going to work all that well over time.
So it sounds like I will start with NinjaTrader because it's free. And Amibroker emailed me back and said their free demo lets you backtest, too. Which of those two is the easiest to code/use? My coding experience is 1 year of C in college, and 1.5 years of some proprietary BS language at my first job out of college. I'm good at simple stuff, but 3-dimensional arrays and pointers and stuff is where I start to get confused.
I'm not sure there is a right answer that takes the form of X days. I think it's important to test on short and long time frames in order to get a sense of how the strategy waxes and wanes, as it invariably will. Then you can work back to those periods and try to guess what market forces favour and hurt your program. As for backtesting and real time results - I've found that in my case it has been spot on. And I've run my programs for years now. So .... I think the moral of my story is that backtesting can be accurate, but you have to be really careful about the various cognitive biases you impose on the process. We all like to focus on good results and throw out bad ones, so we're always unaware of how much bias selection pressure we're brought to the table when we point to the model and say "eureka!". How many failed models did we discard? If you ran through 100 combinations before finding your one successful one, even if it yielded 2 sigma results, it would be worthless.
Download Investor/RTs demo. There isn't any actual "coding" involved, in the way that you'd think about coding. It's an excellent backtesting/charting/trading package. They have a monthly lease program. You'll need to also pay for data, although they have an IB connection for free IB data if you want to go that route. www.linnsoft.com
I don't mind eSignals back testing once you get all the neuances down. It's pretty fast too. And I love the tick replay. Geoff
According to their website it looks like you can only backtest their $DEMO data and not real data. But I think I'll still download it to get a feel for how it works.