Backtesting at tick level

[NinjaTrader_Bertrand]

Hi moneyman, please forward me your log and trace files to support at ninjatrader dot com with Attn Bertrand in the subject line and I will check those out. Can you please clarify what you mean by getting 'inaccurate results' when backtesting on the tick level? Thanks!

[kcsystemtrader]

moneyman,

Market Replay is not backtesting. If you want to add a tick time frame you need to do Add(PeriodType.Tick, 100) for instance. Tick data is very resource intensive. Please try running on less days back and ensure you are running on an as strong as possible machine.

Wondering what you mean by this. I am developing a 1 minute strategy and it processes with calculateonbarclose = false. So in a sense, in real time (or in market replay), it is a multi-time frame strategy with the primary bars being 1 min and the secondary bars being tick by tick. It appears that in Market Replay, it functions just like it would in real time (i.e. the market replay recorder has captured every tick and will trade intrabar as it usually does when simulating it live).

So - this may just be semantics here, but what I am doing with market replay in this case would be essentially the closest I could get to an accurate backtest, correct? I have found that I can just run a replay on 500X speed and it will basically simulate a days worth of trades (as if I had been letting it run in real time all day) in about 5 minutes. Not perfect, but then I can run it again with different parameters to pseudo-optimize the strategy. I'd like to be able to use the real backtest/optimization functionality of NT, but have been unsuccessful thus far.

I've used BarsInProgress and have added Add(PeriodType.Tick, 1) but have not been able get it to backtest the same way as when I simulate with market replay or with simulate in real time.

All this to ask --- has anyone successfully backtested a 1 minute primary bars strategy with intrabar granularity? Thanks,

Parker

[NinjaTrader_Ben]

Hello,

I know there are many traders who have done it successfully.

[bluelou]

ctrlbrk,
I use NT to trade tick-based strategies but NT is not the right product to backtest tick-based strategies. I work with 40- to 400-tick bars. I seem to be able to backtest about 4 months worth of tick data for the 40-tick bars and less for the larger bars. No way to do optimization w/tick bars in NT though b/c of the poor memory utilization.

The problem here, from my understanding, is that NT loads every single tick into memory and then creates tick-bars for use in backtesting. This is a waste of resources. Other development platforms allow you to create bars objects and store them until needed. In the case of 400-tick bars, this reduces the size of the tick data loaded into memory by a factor of 400 since you're only loading 1 bar instead of 400 individual ticks into memory.

Tick-based backtesting does not have to be resource intensive as NT claims. But, the way NT does it, yes, it's a terrible waste of memory.

I know of 3 platforms at the retail-level that use bars objects. They all have other drawbacks compared to Ninja though. PM me if you want the platform names.

[ctrlbrk]

PM sent. Looking forward to finding something that works with tick data.