Where is the line between fitting and actual patterns? - page 15
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
Where is the line between fitting and real patterns?
Looking at the market we see that possibly existing patterns cannot be parametrically constant. Every system has a level of fit and a level of regularity of one or more events.
Somehow the regularities themselves are left out of the picture. The more detailed the TS describes the regularities, the more specific patterns we are looking for, and it is not certain that the pattern will repeat with the accuracy of our description, the greater the chance of a fitting in the bad sense of the word.
Here's an example of TS, EURUSD (not only by the way), daily chart, previous down candle we buy, previous up candle we sell, the system is reversed. It is not that the system does not sink, but it persistently grows upwards without significant breaks since the creation of the Euro. Why? The reason is that the regularity of buying low and selling high is something that lies in the basis of any trading and does not require training. It is worth to add 1-2 parameters to the system, for example min. size of the pre-candle, we get growth in the period of training, and deterioration of the OOS.
And the direct selection of the set-up seems to me to be a secondary matter and not so important.
Yes, to find out what the MoD is. And then test what you can squeeze out of it
Mostly yes. Especially convenient with lot=0.1 to compare with spread at once. And other initial selection criteria may be used. If for example PF<1.5, of course MM can correct the situation, but MM is an additional adjustment, and the smaller it is, the calmer it is)))) Although it may be a matter of taste what to fit :)
Mostly yes. Especially convenient with lot=0.1 to compare with spread at once. And other initial selection criteria may be used as well. If for example PF<1.5 MM of course may help, but MM is an additional adjustment, and the less it is, the calmer it is)))) Although it may be a matter of taste what to adjust :)
With a lot of PF trades, the fact that the PF is small is not very important.
What if it is less than 1, but the number of trades is very large? Or 1.01 and mo few pips? :)
Use regularisation algorithms and you'll be happy. :))
And if you could elaborate...
and if you could elaborate...
I reckon it's something nerve-wracking again.
Yes, you're right - there's already a thread on normalisation - https://www.mql5.com/ru/forum/131347 :-)))
Yes, you're right - there's already a thread on normalisation - https://www.mql5.com/ru/forum/131347 :-) ))
I don't think that's what you meant
Can you elaborate...
Regularisation algorithms are algorithms that prevent the over-learning effect of NS. If you Google it, you can find a sufficient number of implemented and working ones.
Thus with these algorithms, the very topic as 'the line between fit and regularity' simply disappears.
Although the question of network architecture remains open.
And don't confuse normalisation and regularisation.
Normalisation is about bringing differently scaled data to the same scale.
And lastly, not all types of NS work well in financial markets.