Econometrics: one step ahead forecast - page 78
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
Sarcasm doesn't describe you very well. R-squared is all over the world, except in your annals. So you put me in your local annals and I'll put you in the world's. You will be more famous than me.
Well, I'm sorry, I'm sorry. Just how else to respond to your assertion that fitting is OK. Here you are writing that the model within the sample characterises itself perfectly and passes all the tests, but when you try to predict the future by at least one bar , it somehow breaks down. Any sensible person who doesn't even know the basics of statistics would understand that it's about fitting, you are trying to use the tail (your model) to wag the dog (the market), and genuinely, almost to the point of tenderness, naively wonder why the market snaps back.
You have a wonderful knowledge of econometrics and statistics, but geez... why apply them to such thoughtless ideas!
There is art and there is science.
I have not made a mistake. I am well aware that fooling the heads of people who came to the field of wonders with TA and NS will be more difficult, but I will participate and wait for the ban.
About the ban, that's not for me, it's for the moderators.
TA and NS can also be used to fool heads if one does not know that the results should be tested on a sample independent of the fit.
Here is the simplest example, we adjusted TS to a sample and got stationary in this very sample (258 trades) data: approximately the same expected payoff - profit growth and variance - drawdown. But the broker will not pay us anything for this - it's yesterday's news. We need to know how the TS behaves in the future.
Now look at what the fitted model will give us outside the sample, i.e. in the future:
And we see that the out-of-sample model is non-stationary, as starting from 259 transactions at least the variance - drawdown has changed. Though, one can clearly see that some previous out-of-sample model's properties are preserved for some time (regression remains the same) - up to 380 trades profit growth is observed, and then expectation becomes negative up to 547 trades, and then its sign changes for positive again.
From this we can conclude that expectation sign does not change (remains positive) at the stretch exceeding the optimization sample by 121 trades. I.e., if double margin of safety is taken into account, then reoptimization of TS (model recalculation) must be performed every 60 trades, as (380 - 259) / 2 = 60.5 trades.
From this we can conclude that expectation sign does not change (remains positive) in the area that exceeds the optimisation sample by 121 trades.
I don't think you've understood anything from what I've written.
Unfortunately you don't understand that primitive ns is the same regression... that in any algorithm you can use any error in training... and most importantly it doesn't really matter what algorithm is used...
You can't.
No one except you is forbidding it.
It's silly to say this based on a single example of 100 transactions. Don't you get it?
How much do you need?
It's silly to assert this based on a single example of 100 trades. Don't you get it?
I understand, that's why I apply a safety margin that limits the out-of-sample profitable area to 60 trades instead of 121. You could also take a triple safety margin, i.e. over-optimisation run every 40 trades.