Back testing turboprop2 in a Trading Strategy

[michael]

Here is my assumption of what i think is happening , ( i could be very wrong)

First i put the TP2 into a predictor to find out how many bars of data i should use , as on other backtests in TS with TP2 seemed to always take a long time with the default of 200 bars. Predictor gave me 88 bars. Running predictor for hours .

So now i have loaded the buy and sell TP2 into a trading strategy .

Leaving the defaults the same except bars back at 88, and optimizing over entire chart.

i think the tp2 calculates the first bar and solves for max return on account,

then moves to the next bar when set for retrain on each bar.

So the final result i see on the max return on account is the sum of all those individual bars.

That is what I hope is happening , so that when a new bar is added in real time data stream, it calculates only that individual bar with the 88 bars of data and it is not looking at the final max return for all the previous data . It needs to be what would be the final max return for that bar only.

So i now have only 2 days of data at 3 minute intervals, 780 minutes divide by 3 = 260 bars.  Only operating exchange hours 8:30 am to 4 pm.

But this is still taking hours to run , about 7-8 hours . so if i take 7 hours to run for 260 bars means that 7*60=420 minutes  and divide that by 260 bars = 1.61 minutes per bar .

Or is the TP2 comparing max return account across all the bars at the same time and then changing one parameter input at a time ?

Any ideas gratefully accepted

Regards

Michael.

 

 

 

[michael]

What i am planning on doing after the TS has finished is to go into the trading rules and see what values it has come up for the inputs and then see if i can narrow some of the ranges on the inputs . Eg bars back on an input restrict to say 1 to 5 instead of 1 to 10 , etc .

[michael]

Here is an interesting idea ,

In the predictor you can stop the model training at any stage and keep that model, but with the tp2 running in the TS , if the TP2 is calculating each bar and it is taking a long time , therefore if i stop it early it has not calculated each bar only some of them. As in stopping it early in a walk forward it has only completed the ones that have been optimized.

 

So if i stop a TP2 running early in the TS , I am only seeing results from however many bars were done in that time frame . So if i look at the results eg GP/GL , Ratio win/loss avg win, % win etc . So if i leave the TS running with the TP2 for say 6 hours and then stop it . I am wondering if i could say that approximately i have the ratios that i want to compare ?? As the ratios are not changing or they have not changed for a long time .?