Machine learning in trading: theory, models, practice and algo-trading - page 3567
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 created a prediction algorithm on the hourly and even the minutes.
It is actually possible to forecast 24 points a day ahead.
Imho, a very decent result. It remains to invent, create, criteria to evaluate the quality of the forecast. Like correlation of trend and residuals, for example. And based on this, we should think about which trading systems will work best here.
Imho, a very decent result. It remains to invent, create, criteria to evaluate the quality of the forecast. Like correlation of trend and residuals, for example. And based on this, we should think about which trading systems will work best here.
Well, the ANC will not work, I think something like correlation or cointegration or something in between will work.
I would start by checking the TS - deviation from equilibrium. The price forecast is used to build a muve with a period of 2*n, where n is the depth of the forecast, in this case 2*24. Due to the relatively large period, the error in predicting the value of the muve will be small. Accordingly, if there is a deviation down from the muva, for example, by some value on the 1H TF and there is a deviation down on the minutes, then a buy is opened. And vice versa.
This is too primitive, "head-on" this does not work, you can see that the forecast repeats the shape of prices more than absolute values.
You need some clever preprocessing to synchronise prices with forecasts.
==========
I added several TFs to the forecast, it seems to be improving with more inputs...
But one resulting forecast is considered more than 5 minutes, it is unrealistically long for any normal experiments.
EUR m1, 15 min ahead forecast is more or less good.
eura m1, 15 min ahead forecast is more or less good.
global inverse correlation is also evident
there is also a global inverse correlation
and the inverse correlation is more characteristic of forecasts on small TFs.
If we take large TFs on the same data, the forecast already has a positive correlation.
Conclusions.
Markets are complex.
And we don't know a damn thing about them...
Confidence intervals are quite similar
Confidence intervals are quite similar
It's not bad at all here
It's not bad here at all
I think they've made auto arima in python too, go for it.