Machine learning in trading: theory, models, practice and algo-trading - page 420
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
Why check anything? It's obvious to me.
Not so long ago it was obvious that the earth is flat, until you checked it.
I suggest to check everything on concrete examples, publish datasets in csv on which you got such results with ZZ and RSI
SanSanych Fomenko:
I gave the RSI-ZZ example - nothing in common, and you can build a model with less than 50% error.
The commonality is the data.
ZZ - sees forward and backward by many candles and contains momentum which is then predicted by RSI(MA etc), this is a sweat, and the test dataset does not solve the problem, because on it you also mixed the chip and targeting through ZZ and get as if it were statistically significant results.
PS: The simplest but true target is the color of a future candle, or returnee by some step in front(Open(t+N) - Close(t+1))/Close(t+1) it certainly does not see the calculations on the indicators from the past, please check what accuracy error will come out on such a target.
Not so long ago it was obvious that the earth is flat, until checked.
I propose to check everything on concrete examples, publish datasets in csv on which you got such results with ZZ and RSI
General - data.
ZZ - sees forward and backward by many candles and contains momentum which is then predicted by RSI(MA etc), this is a sweat, and the test dataset does not solve the problem, because on it you also mixed the chip and targeting through ZZ and get as if it were statistically significant result.
PS: The simplest but correct target is the color of a future candle, or returnee by some step in front(Open(t+N) - Close(t+1))/Close(t+1) it does not see the calculations on the indicators from the past, please check what accuracy / error will come out on such a target.
In this thread I have published a lot of material on "predictive ability" of the predictor. I think this is the most important thing in datamining.
Seeing your post, it just came to mind.
I haven't done MO for some time now.
PS.
You are wrong about RSI and ZZ. You can draw trends by hand. RS automates this process. Another thing is that ZZ itself cannot be the target variable, since the trend does not determine the whole shoulder of ZZ, but only the beginning of the shoulder and the end of the previous shoulder. But for such a target, more properly, I have not been able to pick predictors that have predictive power for it.
PSPSS
For predicting increments, there is a mighty set of models called GARCH
You're wrong about RSI and ZZ.
So let's check it objectively, let's take a real and random price, generate chips and targeting, weld a model and see what the problem is? Just in the air to say "you're wrong" IMHO not the gentleman, so you can say anything.
In fact it is not so important how exactly you train your model, if you understand what's what, but it should predict at least the sign of the future increment, preferably a module, and of course on the chips do not look ahead.
SanSanych Fomenko:
There is a mighty set of models called GARCH for predicting increments.
GARCH - volatility forcasts, it's simple, and ARMA etc. is a rudiment, the best is the previous one or mo windows of the previous ones, and in general they are all linear
I don't want to upset anyone, but alas, most of you don't know how to prepare targets correctly. All these inspiring results (75-80% accuracy) on slow candlesticks (>10min), in reality, are just a sweat. The accuracy of 55% is enough to make Sharpe Ratio higher than 2, and the accuracy of 60% on slow data is the same grail, about which there are legends, Sharpe Ratio 3-4, no one trades so on the real, only HFT-people, but they have a different scale of trading costs, there is less SR <2 unprofitable.
In short...
DO NOT SEE THE TARGET(target)!
That is, when calculating the target, you can't use data that is EVERYWHERE used in the calculation of features, otherwise the result will be with a poke around. For obvious reasons, such "sleight of hand" as ZZ to hell, it interpolates between extrema far into area where features are calculated, the result is exorbitant, at least 90% accuracy without problems, but it's a fake. This is the basis for obscurantist discussions about how "the forecast is not important", we should still develop the TS and so on. So in fact these "90%" all the same "favorite" 50%.
Be reasonable :)
A bold, rather self-righteous statement.
Since all predictors use quotes one way or another (OHLCV) and you require not to use them to determine the target - what can / should be used for the target? Phases of the moon? Seasons from the lunar calendar? What?
The whole thing about ZZ is nonsense.
It's just a bunch of nonsense.
Give me some examples with calculations to back up your sentiments.
I don't think you're in the loop.
Learn the theory and give examples of calculations to support your claims. Otherwise you're just babbling.
Good luck
So let's check it objectively, let's take a real and random price, generate chips and targeting, weld a model and see what the problem is? Just in the air to say "you're wrong" IMHO not the gentleman, so you can say anything.
In fact it is not so important how exactly you train your model, if you understand what's what, but it should predict at least the sign of future increments, preferably the module, and of course the features not looking ahead.
GARCH - volatility forcasts, it's simple, and ARMA etc is a rudiment, the best is the previous or mo windows of the previous, and in general they are all linear
You strongly suggest that someone check it. Check and give me the results.
Interesting to see and discuss the results.
You can't see the target(features)!
In other words, when calculating target, you can't use data that is EVERYTHING you use when calculating features, otherwise the result will be with a gawk.
I said something similar a couple hundred pages ago. But to prove the point I did not see and you do not advise it now, people like when 10% error in the test, let them indulge, here mostly near-marketers, they need such figures to advertise their sinkers. What will be the grail with an honest 49% error prediction?)
What kind of grail would be with an honest 49% prediction error???))
You may work very well with 49%. By the way, it is an excellent grail).
No, not enough, spread and commission will give all the profits.
No, not enough, spread and commission will give all the profits.
I don't know, for my Target, I'm sure he's not peeking anywhere. Well, no, he does. But only into the future and not into the past. :-)