is walk forward testing always required?

Aug 26, 2008
10
1
13
Washington
#1
the title is my question.

if the system uses pure, raw data and has fixed rules is there any value in walk forward testing? notice here that i am NOT saying backtesting, just walk forward testing.

the system could be like 'if the current day's high is lower than the previous day's high, buy the next morning long, sell 3 days later at the close'. there is no variable or parameter here. it seems like the walk forward test would give the same answer but with a lot more work. it seems like the backtest could be run over the entire time period all at once.

am i missing something here?

thanks,
david
 
Apr 4, 2016
1,959
96
58
#2
Yes live testing is always required to find out the guy you are trying to take money from will say no. He doesn't care how many years of data you have. You can't have his money and that's final.

Put yourself in his shoes and you will do exactly as he does.

Imagine some idiot walk up to you in the street and says: look here rich guy, i have 200 years worth of data right here in my rucksack that says your money is my money, now give me all your money right now. You will either whack the guy on the nose or call the police.
 
Last edited:

Pat494

Well-known member
Mar 27, 2004
13,037
1,221
223
#4
Use both.
The great advantage of forward testing is that any glitches that are, will be more apparent. For instance sl and tp levels, TF etc.
I learnt not to leave it over night !
 
Nov 25, 2016
149
24
28
#5
I would say you should forward test it or start trading it with minimal possible stakes, not the size you intend to trade for real.

Backtesting can be insufficient for a number of reasons. Your results might be due to curve fitting, and so you haven't tested out of sample. Your backtesting software might not be reliable, or not account for something that you would notice doing a forward test. Also it's whether you can actually trade the thing. A backtest might work, taking 27 trades across 15 different instruments from hours of 3am to 11pm, but you can't actually trade that unless you automate. You need to get a feel for what it is like to execute. Ideally you would want to forward test by trading live without any money. It won't be quite the same in terms of fills, but will help you on your execution.

Regardless of all of that, you'll forward test it one way or another, it's just whether you forward test with money or not. Forward testing with money could reveal non-systematic things, so there could be value in that.

I think it's worth asking why you wouldn't want to test something more that could protect you from losing capital.
 

Jason101

Well-known member
Oct 9, 2008
1,357
212
73
#6
the title is my question.

if the system uses pure, raw data and has fixed rules is there any value in walk forward testing? notice here that i am NOT saying backtesting, just walk forward testing.

the system could be like 'if the current day's high is lower than the previous day's high, buy the next morning long, sell 3 days later at the close'. there is no variable or parameter here. it seems like the walk forward test would give the same answer but with a lot more work. it seems like the backtest could be run over the entire time period all at once.

am i missing something here?

thanks,
david
dvdadie,
You joined in 08, almost 10 yrs ago. And this is your first Post !!
(seems like a newbie type question) What you been up to?
 

Quantt

Active member
Jul 23, 2017
945
57
38
#7
the title is my question.

if the system uses pure, raw data and has fixed rules is there any value in walk forward testing? notice here that i am NOT saying backtesting, just walk forward testing.

the system could be like 'if the current day's high is lower than the previous day's high, buy the next morning long, sell 3 days later at the close'. there is no variable or parameter here. it seems like the walk forward test would give the same answer but with a lot more work. it seems like the backtest could be run over the entire time period all at once.

am i missing something here?

thanks,
david
You can also froward test from some point into your data as well...
 
Aug 26, 2008
10
1
13
Washington
#9
thanks_truth

thanks truth and quantt, good stuff.

after some more pondering, the walk forward testing could provide information on how 'stable' / 'robust' the system is from year to year.