Now we can for example say: backtest the results for an entry with RSI(14) < 30 and close above EMA200 and optimize this for an exit with x percent and a stoploss at x percent.
But is it possible to search (neurolab + the latest AI?) for good entry points based on the following idea: what if you ask what the positions of some indicators are at the moment where ROC(x)+n > y.
For example: ROC(12) + 10 days forward is greater than 10 percent.
So in backtesting:
you look in the future (or you take the current date and use ROC -n which is the same) and when the ROC 12 days is for example 10 percent, you wanna know why the price is 10 percent higher than 12 days ago.
Probably there is a pattern, an ideal combination of the indicators that gives a high probability for a coming rise in the next 12 days.
But is it possible to search (neurolab + the latest AI?) for good entry points based on the following idea: what if you ask what the positions of some indicators are at the moment where ROC(x)+n > y.
For example: ROC(12) + 10 days forward is greater than 10 percent.
So in backtesting:
you look in the future (or you take the current date and use ROC -n which is the same) and when the ROC 12 days is for example 10 percent, you wanna know why the price is 10 percent higher than 12 days ago.
Probably there is a pattern, an ideal combination of the indicators that gives a high probability for a coming rise in the next 12 days.
Rename
This question:
Is a perfect match for (and completely answered by) the new finantic.IndicatorSelection extension.
(https://www.wealth-lab.com/extension/detail/finantic.IndicatorSelection)
Steps to answer your question:
You code up a strategy that enters every day and exits n days later.
These trades become your "Target Series" i.e. the "thing" you want to predict.
Then you use the Ranking Tool (part of finantic.IndicatorSelection) to find among all available indictors (currently about 4000) the "Top 10" which predict such a move best.
In this case "predict" means, that there is a correlation between the indicator values the day before the move starts and the move itself. A high correlation is said to mean that the indicator values are able to predict the move.
It is then a second step to feed these "best indicators" as inputs into NeuroLab or some other AI.
It is then the job of the AI to find a good combination formula for the already selected indicators.
All experiments to find good indicators with some AI directly failed so far. I guess the reason is that the input data is way too noisy. AI algorithms are not designed to handle such a huge amount of noise.
QUOTE:
what the positions of some indicators are at the moment where ROC(x)+n > y
Is a perfect match for (and completely answered by) the new finantic.IndicatorSelection extension.
(https://www.wealth-lab.com/extension/detail/finantic.IndicatorSelection)
Steps to answer your question:
You code up a strategy that enters every day and exits n days later.
These trades become your "Target Series" i.e. the "thing" you want to predict.
Then you use the Ranking Tool (part of finantic.IndicatorSelection) to find among all available indictors (currently about 4000) the "Top 10" which predict such a move best.
In this case "predict" means, that there is a correlation between the indicator values the day before the move starts and the move itself. A high correlation is said to mean that the indicator values are able to predict the move.
It is then a second step to feed these "best indicators" as inputs into NeuroLab or some other AI.
It is then the job of the AI to find a good combination formula for the already selected indicators.
All experiments to find good indicators with some AI directly failed so far. I guess the reason is that the input data is way too noisy. AI algorithms are not designed to handle such a huge amount of noise.
QUOTE:
All experiments to find good indicators with some AI directly failed so far. I guess the reason is that the input data is way too noisy.
Thank you for that post. That clarifies the market reality.
I think you first need to reduce some of the "noise" with what decorrelates well with IndexLab indexes. But you are still left with a great deal of real noise.
This brings us to another mystery about some indicators. For sometime I wondered why RSI must diverge from Price prior to buying? Why is divergence so important? I have a theory....
I "think" RSI acts as an averaging filter over time to remove the trading noise in a bias way. The Price action doesn't have this bias, so it follows what traders think the market is doing over time. Eventually there's enough divergence between the two that something has to give as a "reality check" in the market. That also identifies a value opportunity to buy. If you buy when the divergence is highest (prior to the reality check), then you capture this opportunity.
The trick is timing this "reality check" while "playing chicken". If you wait too long, you miss buying before the reality check correction.
Neuro Networks aren't setup to characterize divergences. That's why we have the PeakTroughCalculator class in WL.
@DrKoch, that finantic.indicatorSelection extension seems indeed the closest that probably exists for WL. It looks amazing. Everytime I see something like that, I think: if the guy who programmed this cannot decipher the market, why I should even try it :-)
WL already can do awesome stuff. Why would it even with the help of AI cannot find the best values of let say 3 indicators when the profit is the highest since the entry 10 days ago?
WL already can do awesome stuff. Why would it even with the help of AI cannot find the best values of let say 3 indicators when the profit is the highest since the entry 10 days ago?
@superticker:
The divergence gives a higher chance of a turn because it suggests that the down momentum is fading.
Normally there is a steep decline, a consolidation and another decline but less powerful. That fits also in the Elliott Wave theory. The new low with less momentum is most likely a fifth wave.
The divergence gives a higher chance of a turn because it suggests that the down momentum is fading.
Normally there is a steep decline, a consolidation and another decline but less powerful. That fits also in the Elliott Wave theory. The new low with less momentum is most likely a fifth wave.
@Dirk3000, that's exactly what I've requested in one of my threads. In manual mode it can be done with a combination of Analysis Series and Indicator Profiler, however I hadn't any success with them. DrKoch's solutions seems to be the best at the moment.
As for the built-in functionality, one of the possible options is to make Strategy Evolver complement ready-made strategies, then it will be possible to find both intraday strategies(that's already been asked for in another thread) and all possible combinations of queries, as in your post. Unfortunately this initiative has received very few votes to be realized asap.
As for the built-in functionality, one of the possible options is to make Strategy Evolver complement ready-made strategies, then it will be possible to find both intraday strategies(that's already been asked for in another thread) and all possible combinations of queries, as in your post. Unfortunately this initiative has received very few votes to be realized asap.
Your Response
Post
Edit Post
Login is required