Looking for patterns - page 110
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
Thank you, Alexey. What progress have you made so far? I understand one of the important issues is to understand what data to feed into the inputs, how to prepare them. What do you "feed" your brainchild with?
I have a lot of predictors - about 800. These are different predictors based on indicators, time, candlestick combinations, price position in the regression channel and ATR of the upper TF, the structure of ZZ segments (the same waves).
The first problem is the estimation methods in training - there is no estimation of the split distribution over the sample in the models I know of, resulting in short-term patterns. If who writes in C++, I invite together to make changes in CatBoost code with the purpose of the decision of this problem.
The second problem is universal for all strategies, and I already designated it - impossibility to give an estimation of result having data only about a part of the information (there is no future) that complicates selection of models/legalities.
I have a lot of predictors - about 800.
This is a very large number. I'm afraid with that number of scales through which 800 signals pass, you can describe any particular case on the chart history.
I would slash hard.
I have a lot of predictors - about 800. These are different predictors based on indicators, time, candlestick combinations, price position in the regression channel and ATR of the upper TF, the structure of ZZ segments (the same waves).
The first problem is the estimation methods in training - there is no estimation of the split distribution over the sample in the models I know of, resulting in short-term patterns. If who writes in C++, I invite together to make changes in CatBoost code with the purpose of the decision of this problem.
The second problem is universal for all strategies, and I already designated it - impossibility to give an estimation of result having data only on a part of information (absence of the future) that complicates selection of models/legalities.
No one knows the future. We all cannot even guess what kind of trick fate will throw at us tomorrow).
There are still some hidden reserves to predict the price, but these are projected results for which there is little hope.
Our hope is always with us, in case we get lucky. Our only hope is probability.
No AI or neural networks know exactly what will happen tomorrow.
We rely only on probability.
I am relying on wave and trailing chart analysis. This is a computer daily chart of EURUSD.
I look at it this way too. It is the impact of waves on the price structure. There are many regularities looking at waves.
It occurs to us...
What if we do not use the usual binary scheme to assess the trend: an up trend is +1 and a down trend is -1.
This kind of indicators shows the past and at best the present. That is, at the moment when the price updates an extremum, we know for sure that the trend is going in that direction. I have marked these places with a red marker. This is where we know that the trend is in the present. And when price moves away from the point where the last extremum was updated, we know that the trend was certain before the extremum, but there is now some uncertainty. Here we already know for sure that the trend has existed only in the past.
But even seeing a trend update right before our eyes, and knowing that it exists at that moment, we still have no assurance that it won't start to fall backwards in the next instant.
So, if we are still dealing with uncertainty, then we need to build an indicator that will show the probabilistic existence of the trend in the next step to the right. A step that we don't know yet. Then this indicator will look as follows: when the price approaches an extremum update, the probability will increase up to its maximum but decrease during the prolonged update, indicating the approach of the reversal. At the moment of a pullback the probability should in some cases be even higher than during a long update of the extremum.
Knowing the probability of continuation of the trend we can, for example, use different lot sizes.
Just a thought for now...
This is a very large number. I'm afraid that this number of scales through which 800 signals pass can describe any particular case on a graph.
I would be cutting a lot.
That's why I use an estimate of each leaf of the tree. This allows me to select leaves with a consistent pattern across the entire sample according to a number of criteria. Each leaf consists of about 6-8 predictors, which is clearly not enough to describe the whole sample. But, the methodology is slow, which is why I want to make edits to CatBoost as one of the fastest algorithms, in order to account for calculations not only for the number of correct classifications, but also for other criteria.
We only rely on probability.
What makes you think that AI can predict the future? It's all about probability on history.
The advantage is that you can more quickly find the right combinations that give you a statistical advantage.
Personally, I made predictors based on my knowledge of the market, just trying to tell/show them to the algorithm, and it already tries to generalise them and identify patterns.
I'm not saying it's a panacea, but it's a way to get ideas.
Let's test your arrows on history - are they useful?
What makes you think that AI can predict the future? It's all about probability on history.
The advantage is that you can more quickly find the right combinations that give a statistical advantage.
Personally, I made predictors based on my knowledge of the market, just trying to tell/show them to the algorithm, and it already tries to generalise them and identify patterns.
I'm not saying it's a panacea, but it's a way to get ideas.
Let's check your arrows on history - are they useful?
I am currently working on a short-term entry strategy. In fact everything is ready to work. I've built my own neural network. It does not show arrows but immediately places an order or, if in doubt, gives me the button I need to manually place an order under my magician for further processing.
No one knows the future.
Let's say. Although in some cases predictions can be made.
Our only hope is probabilities.
It sounds as if you know these probabilities. Even if not always and they are far from 1.0, but at least more than 0.5 is something.
No AI, no neural network knows exactly what will happen tomorrow.
Why is that? Because these probabilities, unless they are based on purely subjective factors and can be formalized, can be fed to AI and neural networks and receive a forecast. It may not be absolutely accurate, but it gives a certain statistical advantage.
Although these probabilities can be used for their intended purpose and without AI and NS. The question is whether they actually exist or it is a mirage. What are they based on I'm afraid to even ask. )
I am still interested in the results. I have just discovered another brokerage company, which is struggling with arbitrage on demo accounts. I will show you the demo results, which are unexpectedly high for market execution. My balance grew from 10 to 104 thousand during 2 days. The account was opened exactly at 22:43 the day before yesterday according to their server time. The statement was recorded today at 22:48 MSK, their server time is 20:48.
I don't really understand why you are doing this? If the DC is so poor and/or greedy that it cannot afford to buy a plugin to buy arbitrage transactions, do you really think it will allow you to withdraw at least a hundred quid earned on the real, on these transactions? I think it's a pointless waste of time and effort.
Granted. Although in some cases predictions can be made.
It sounds as if you know these probabilities. They may not always be 1.0, but at least more than 0.5 is something.
Why is that? Because these probabilities, of course, if they are not based on purely subjective factors and can be formalized, you can feed the AI and neural networks and get a forecast. It may not be absolutely accurate, but it gives a certain statistical advantage.
Although these probabilities can be used for their intended purpose and without AI and NS. The question is whether they actually exist or it is a mirage. What are they based on I'm afraid to even ask. )
Tomorrow today will be history. Tomorrow we will know exactly what happened today. It follows that we may assume that nothing will happen and build a forecast with some probability on that.
It is true that not all TS are built on this principle. There are a lot of variants for TS creation.
Aleksey Stepanenko searches for possibilities to build TS on the mythical trend. And there is no special reasoning explaining what exactly the trend is.)