Hello Ladies and Gents
I am currently backtesting an idea.
Simply said it's a volatility breakout system.
The idea is to run it on multiple currencies and on multiple pairs.
First Question
I can run only one backtest at a time and then extract the individual trades from the "Results" tab, by copying them all into Excel.
I the Exel file hat I do so far is arrange the "Type" Column to only show S/L & T/P. Note that I'm using a trailing stop, so I have quite a few S/L, which are actually in profit.
My question is the following
I'M not very good with excel)
Is there any chance I can feed multiple such reports into the excel file?
Obviously I could just copy and paste it. The idea is though to have a chart where I could see all strategies' graphs in one and then see how each one would have affected a common balance.
It is problematic so far, because my "Time" Column is not showing every point in time, but only every time the EA traded. Which means there are gaps in it.
These gaps, of course, are not the same for Report 1 compared to Report 2. So I was thinking if there could not be established a general timeline, with all days/hours/weeks or whatever, and then have the individual balance points shown of each specific report.
What I want to see there is how correlation would be between the different strategies.
I have attached an example of one (losing) test-run, so you can see what I mean.
I believe there are other possibilities to see correlations. I think the guys from StrategyQuant have a tool to do that, but spending $ 1500 is not something I'm ready doing if there is another way.
Second Question
With regards to backtesting, I'd like to hear some thoughts. Currently I have an idea that is coded but leaves a lot of open parameters.
So far I have Birt's Tickdata Suite, so I have tick data for the major pairs from early 2003 till today, which is really nice. My backtest results give me 99% modeling quality, so I have reasons to believe that I can trust it.
I have been approaching it like this:
I let an optimization run, with fixed lots, so I don't fool myself with money management. I let it run from say 2003 till 2011 and then I planned to let the best results run against data from 2011 till 2016. The timeframe I have right now running is H1. It takes quite long:
Am I doing this right? It's like a shotgun approach, saying: "Ok here is the basic idea (which is that volatility will rise at some point, after being compressed), now show me what the best parameters are here".
Am I not curve fitting it?
Am I avoiding curve fitting, by letting the best results run against out of sample data later on?
Does it make sense to give the backtest nastier conditions, like multiplying the spread by factor x and by introducing slippage? I can do that with Tick Data Suite.
Then again, that's kind of a robustness test, but since everything is running on tickdata anyway, I assume I already have tested against real spreads...
I'd really appreciate your inputs. Thank you very much.
Cheers
Nikolai
I am currently backtesting an idea.
Simply said it's a volatility breakout system.
The idea is to run it on multiple currencies and on multiple pairs.
First Question
I can run only one backtest at a time and then extract the individual trades from the "Results" tab, by copying them all into Excel.
I the Exel file hat I do so far is arrange the "Type" Column to only show S/L & T/P. Note that I'm using a trailing stop, so I have quite a few S/L, which are actually in profit.
My question is the following
I'M not very good with excel)Is there any chance I can feed multiple such reports into the excel file?
Obviously I could just copy and paste it. The idea is though to have a chart where I could see all strategies' graphs in one and then see how each one would have affected a common balance.
It is problematic so far, because my "Time" Column is not showing every point in time, but only every time the EA traded. Which means there are gaps in it.
These gaps, of course, are not the same for Report 1 compared to Report 2. So I was thinking if there could not be established a general timeline, with all days/hours/weeks or whatever, and then have the individual balance points shown of each specific report.
What I want to see there is how correlation would be between the different strategies.
I have attached an example of one (losing) test-run, so you can see what I mean.
I believe there are other possibilities to see correlations. I think the guys from StrategyQuant have a tool to do that, but spending $ 1500 is not something I'm ready doing if there is another way.
Second Question
With regards to backtesting, I'd like to hear some thoughts. Currently I have an idea that is coded but leaves a lot of open parameters.
So far I have Birt's Tickdata Suite, so I have tick data for the major pairs from early 2003 till today, which is really nice. My backtest results give me 99% modeling quality, so I have reasons to believe that I can trust it.
I have been approaching it like this:
I let an optimization run, with fixed lots, so I don't fool myself with money management. I let it run from say 2003 till 2011 and then I planned to let the best results run against data from 2011 till 2016. The timeframe I have right now running is H1. It takes quite long:
Am I doing this right? It's like a shotgun approach, saying: "Ok here is the basic idea (which is that volatility will rise at some point, after being compressed), now show me what the best parameters are here".
Am I not curve fitting it?
Am I avoiding curve fitting, by letting the best results run against out of sample data later on?
Does it make sense to give the backtest nastier conditions, like multiplying the spread by factor x and by introducing slippage? I can do that with Tick Data Suite.
Then again, that's kind of a robustness test, but since everything is running on tickdata anyway, I assume I already have tested against real spreads...
I'd really appreciate your inputs. Thank you very much.
Cheers
Nikolai
Demo only
