about optimization

[Patrizio Begni]

Hi all, in you opinion, what's the best for a 15/30 min trading system :

1) choose the parameters that best fits in last 2 years

2) chosse the parameters that best fits last year

3) choose the parameters that best fits all the historical data (5+ years)

4) make an average of the previous windows 

I'm asking because in almost all TS i'm trying, the best parameters in TP and SL , can change  significally changing the window of backtest.

 

thank you

[Marco vd Heijden]

Patrizio Begni:

Hi all, in you opinion, what's the best for a 15/30 min trading system :

1) choose the parameters that best fits in last 2 years

2) chosse the parameters that best fits last year

3) choose the parameters that best fits all the historical data (5+ years)

4) make an average of the previous windows 

I'm asking because in almost all TS i'm trying, the best parameters in TP and SL , can change  significally changing the window of backtest.

 

thank you

It will depend on the mechanism of gathering those parameters.

[Patrizio Begni]

the parameters in this case is SL e TP, but maybe (for example) a period of ema or Bollinger

[Marco vd Heijden]

Patrizio Begni:
the parameters in this case is SL e TP, but maybe (for example) a period of ema or Bollinger

Okay then in my opinion the best for a 15/30 min trading system is a 1 - 5 year demo run.

What i meant was did you want good back test results or good live trading results as these are two different things.

[Patrizio Begni]

Marco vd Heijden:

Okay then in my opinion the best for a 15/30 min trading system is a 1 - 5 year demo run.

What i meant was did you want good back test results or good live trading results as these are two different things.

yes i know , a solid backtest is not the holy bible and is necessary a non over-optimization of it .... but without it u never know if a model is valid or not.

I also know that a TS is not a right science but an art ...

The point is to make the choices that are most logical/correct with the information that we have; so in order to non over-optimize i'm thinking to make an average of the SL e TP parameters, got from 2 time-windows (1 year and 5 years)

[Marco vd Heijden]

Patrizio Begni:

but without it u never know if a model is valid or not.

Of course you can only know if a model is valid by throwing some of your capital at it.

So, that's the exact opposite of what you are saying.

Patrizio Begni:

The point is to make the choices that are most logical/correct with the information that we have;

And i just gave you that information:

Marco vd Heijden:

What i meant was did you want good back test results or good live trading results as these are two different things.

Now it's up to you to pick a direction.