I'm looking to backtest performance of an automated high-frequency strategy on 100+ stocks, going back for a year of tick data. Is there any retail platform that will accomodate this, and includes the tick data? Thanks, Russell
I've used or I'm using NinjaTrader, OpenQuant, and TradeStation. None of these are really suitable for backtesting on high-frequency data. Ninja is the least bad of the 3 but that's not quite good enough for what you need. I'm not aware of anything suitable at the retail level. However, there is a new entrant, Marketcetera, that deserves a close look. It's open-source and the target mkt is institutional, The firm appears to have top-tier management and backers. It appears to be suitable for high freq equity strategies. The strategy development languages are Java and Ruby. They may add C# in about a year.
i'll just say this... i have a fully loaded dell workstation and can't even backtest using 5min data in NT, let alone tick data. NT is great for trading futures via Zen-Fire but a real resource hog when it comes to backtesting. if your only options are with retail packages, i'd recommend TS for the lighter stuff, OQ for the heavier stuff.
With NT, I can backtest futures using tick data but only on about 4 months worth of data at a time. This is just for a single test run, no optimization. Two tricks to speed up the backtests/ minimize memory leaks are 1) make sure you're not connected to a live feed while testing and 2) close and then reopen the Strategy Analyzer before each test.
Tick-data of 100+ stocks for an entire year - have you calculated how many GBs of data that is? That would only work with some sort of streaming technology that doesn't require to load all data into memory before simulation (like most retail software does). The only software i know that could probably do it is Neoticker's SimServer. The SimServer streams data to the platform just like a realtime datafeed would, and discards older data that is not required anymore. The caveat is that its not possible to do optimization with that and that its slower than a normal tick-replay (which is the standard way to do tick-precise simulations in NT).
i wrote a NT indicator that writes ticks out to a flat text file, and for ES, the file is around 5MB per day with the format hh:mm:ss.fff;price;vol. an example tick would look like: 10:11:03.345;768.5;32. the nice part about text files is they compress really well.. the 5MB text file compressed is only around 500KB. now that i think about it though, high volume shares probably have many many more ticks during the day, and the space probably increases dramatically...