- single instrument- SPY, i.e. SPDR S&P 500 ETF
- single timeframe – 15 second bars
I have observed the following:
A)
If I run backtest over a whole year of data, the strategy trades M times. However, if instead, I run the backtest over smaller periods of time (say two weeks), and repeat the process 26 times to cover the same whole year period as above (i.e. 26 x 2 weeks = 52 weeks = a year), the strategy trades a total of N times, where N > M by about 20%. So over the same period of time, and with the same data, I get a different number of trades depending on how I treat the test time period.
B)
I also notice that in the single 52-week backtest case, the charts for some days are missing whole sections of 15-second bars (which I guess will be related to whatever reason is causing fewer trades too in this case!). In the case of the two week backtests on the other hand, all charts on all days are fine, without missing bars as far as I can see.
My guess)
My guess is that the issue is related to my testing computer being overwhelmed by SPY tick data. SPY shares are, I believe, on a daily basis among the single most actively traded securities anywhere in the world, on any exchange.
In my personal experience, a typical trading day (04h00 – 20h00 EST) will see between 500,000 and 2,000,000+ SPY ticks. So a lots of ticks! A day’s SPY tick data in .txt format will be a file typically between 10 MB and 20+ MB.
So, two weeks of SPY data will be up to 200 MB of data (10 x 20 MB).
A full year, 5 GB of data (250 x 20 MB)
A full two years, 10 GB
Etc ….
My questions)
1. Does NT load all tick data required for a backtest into memory, all at once?
2. If so, then what specification in terms of memory should my testing computer have in order for me to be able to run backtests using multiple years of SPY tick data?
3. Does NT 7 offer any enhancements over NT 6.5 in this regard?
Comment