Expecting signal correction based on past performance and current sample - page 2
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
I think the doubt is mainly related to the fact that some people think that the probability of having a winning trade depends of results of previous trades, that is surely the worst approach in any trading strategy: also in betting this do not pay out in long term.
Even if your strategy is shaped to have 95% of winning ratio into the past 10 years, it do not means that you cannot have 5 losses in a row in your 5 real trades, or 500 profitable trades in a row. Statistics are just statistics.
You have the proof of all EA on the market that have 99% of winning ratio, and most of them wipe out account after less than 10 real trades :)
I think the doubt is mainly related to the fact that some people think that the probability of having a winning trade depends of results of previous trades, that is surely the worst approach in any trading strategy: also in betting this do not pay out in long term.
Even if your strategy is shaped to have 95% of winning ratio into the past 10 years, it do not means that you cannot have 5 losses in a row in your 5 real trades, or 500 profitable trades in a row. Statistics are just statistics.
You have the proof of all EA on the market that have 99% of winning ratio, and most of them wipe out account after less than 10 real trades :)
Yeah , you understand how it does not make sense in betting "fundamentally" .
Is that what the z-score counts ? how "sequential" series are ?
Exactly , that is why new and used have the same value in the market .
Is that what the z-score counts ? how "sequential" series are ?
Not sure if I ever understood correctly the sense of z-score but I think that it represents the possibility for a value to differs from its mean, so a value of z-score near 0 should be preferable... I repeat, not sure if I understood it correctly.
Most of times I prefer to focus on how a strategy react to a market in different scenarios rather than maths only...
Not sure if I ever understood correctly the sense of z-score but I think that it represents the possibility for a value to differs from its mean, so a value of z-score near 0 should be preferable... I repeat, not sure if I understood it correctly.
Most of times I prefer to focus on how a strategy react to a market in different scenario rather than maths only...
Yeah , another problem here , related to "different senarios" you mentioned , is the sample . Is the entire history the sample or the region with the "new" conditions , which can also not be defined .
What if we swap the "consecutive wins/losses" with "wins inside a window , losses inside a window" though ?
In fact the cheapness is because of the previous owner's mistreatment.
true
Yeah , another problem here , related to "different senarios" you mentioned , is the sample . Is the entire history the sample or the region with the "new" conditions , which can also not be defined .
What if we swap the "consecutive wins/losses" with "wins inside a window , losses inside a window" though ?
Usually I consider "different scenarios" as an out of sample period, than can be a period into the past or more preferable a forward test in real time.
About "inside a window", all statistics are always strictly related to the test period... It's easy to find differences of more than 20% if we backtest the same strategy in 2 different periods, the hardest thing is to understand if despite all differences, the strategy can adapt well to new markets behaviors, like increased voaltility or different patterns that will comes in into the future.
In all of this, a good part of the success will be lucky dependent, there are a lot of strategies that even if well shaped into the past, they perform bad since the start on real time monitoring (or more often real time trading :D) This is also difficult, understand when a strategy need to be put in "standby" because it's not a good period for that...
Usually I consider "different scenarios" as an out of sample period, than can be a period into the past or more preferable a forward test in real time.
About "inside a window", all statistics are always strictly related to the test period... It's easy to find differences of more than 20% if we backtest the same strategy in 2 different periods, the hardest thing is to understand if despite all differences, the strategy can adapt well to new markets behaviors, like increased voaltility or different patterns that will comes in into the future.
In all of this, a good part of the success will be lucky dependent, there are a lot of strategies that even if well shaped into the past, they perform bad since the start on real time monitoring (or more often real time trading :D) This is also difficult, understand when a strategy need to be put in "standby" because it's not a good period for that...
Yeah , in theory you can do that in the past too , assuming the tick data is correct and you inject worse conditions than the real ones ? I come across this often (the live forward test) , it's because of unreliable past simulation i guess .
You also don't want luck on your side when you are developing the strategy :)
Yeah , in theory you can do that in the past too , assuming the tick data is correct and you inject worse conditions than the real ones ? I come across this often (the live forward test) , it's because of unreliable past simulation i guess .
You also don't want luck on your side when you are developing the strategy :)
I agree, you don't need luck when you develop a strategy, but you need a little bit of luck when you switch to use it with real money.
It's not rare to see a well developed strategy, to not work good on real even if all backtests are made with all best practice possible...
About the out of sample into the past, yes consequences on strategy will be similar, but I always prefer to use the real time as out of sample instead of 5-10 years ago when markets were very different than now