Free lectures from Stanford University - page 8
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
Is your name Victor Mikhailovich, by any chance?
Profile.
I used to study TA. At MESI, I think there are three departments teaching econometrics and mathematical statistics, I have not studied. I am not aware of any profitable mathematical methods. I know the mathematical methods that in skilled hands can make a profit, many times referred to. With tables of contents, lists of books. Mat packages .... Take EViews or R if you don't want to google or watch my profile. But it's all a tool, like a list of indicators.
I don't have the grail. I have my personal skill in using TA and math statistics. I will not teach this skill to anyone.
I think I have answered in detail.
Profile.
I used to study TA. At MESI, I think there are three departments teaching econometrics and mathematical statistics, I have not studied. I am not aware of any profitable mathematical methods. I know the mathematical methods that in skilled hands can make a profit, many times referred to. With tables of contents, lists of books. Mat packages .... Take EViews or R if you don't want to google or see my profile. But it's all a tool, like a list of indicators.
I don't have the grail. I have my personal skill in using TA and math statistics. I will not teach this skill to anyone.
I think I've answered this in a comprehensive way.
First you say "TA is for patsies, so they can drain the depot with a look of importance". And now "I have my own personal skill in using TA and mathematical statistics. I'm not going to teach it to anyone."
All right. Don't get hung up on words. Here's my personal opinion on neural networks. I started studying them in 2006, trying to apply them to the market. And in 6 years, I've come to the conclusion that the networks themselves are not as important as the preparation of the input data. Moreover, input data should be transformed by the same TA to reduce its dimension and be invariant to distortions on time and price axis. Besides, we should know beforehand what we want from the net, which input and output system it should implement. For example, if we want the net to trade on rebounds/breaks of support/resistance levels, the input data should be presented accordingly. But after the necessary input data have been collected for the net to be trained and to make decisions for us, we do not need the net at all, because we already know how to enter and exit. The attempt to put prices on network inputs, hoping that the network will transform them and figure out how to use them, leads to disastrous results.
Then the question arises: are nets really necessary and worth studying? Everyone has his or her own answer. For example, I do not think I have wasted my time learning about networks. They are useful for my work. Besides, our brain is the same neural network. So denying its usefulness in trading is just like denying the usefulness of the brain. The problem here is that we still don't understand how our brain takes the same price series, focuses on the important points, abstracting away the details, and makes a decision. A simple textbook neural network cannot simulate this behaviour. Even if we were able to simulate this behaviour, the learning and operating speed of such a "biological" network would be much slower than our brain and not applicable to trading.
First you say "TA is for patsies to drain the depot with an important look". And now "I have my own skill in using TA and mathematical statistics. This is a skill I will not teach to anyone else".
.....Mathematical methods that can make a profit are not known to me. Math.methods which in skilled hands can bring profit are known to me, many times called......
The problem here is that we still don't understand how our brain takes the same price range, concentrates on the important points, abstracts from the details, and makes a decision .
like this...
Hypothesis formation
Data mining and collection
Data preparation (filtering, transformations)
Model selection, selection of model parameters and training algorithm
Model training (automatic search for remaining model parameters)
Analysis of training quality
Analysis of identified patterns
all this is called Data Mining...
A network in this machine (if it's used)... about properly prepared data I totally agree... if the data is normal then a simple linear regression is enough... that's what you should aim for...
First you say "TA is for patsies, so that they drain the depot with an important look". And now "I have my own skill in using TA and mathematical statistics. That's a skill I'm not going to teach anyone."
All right. Don't get hung up on words. Here's my personal opinion on neural networks. I started studying them in 2006, trying to apply them to the market. And in 6 years, I've come to the conclusion that the networks themselves are not as important as the preparation of the input data. Moreover, input data should be transformed by the same TA to reduce its dimension and make it invariant to distortions on time and price axis. Besides, we should know beforehand what we want from the net, which input and output system it should implement. For example, if we want the net to trade on rebounds/breaks of support/resistance levels, the input data should be presented accordingly. But after the necessary input data have been collected for the net to be trained and to make decisions for us, we do not need the net at all, because we already know how to enter and exit. The attempt to put prices on network inputs, hoping that the network will transform them and figure out how to use them, leads to disastrous results.
Then the question arises: are nets really necessary and worth studying? Everyone has his or her own answer. For example, I do not think I have wasted my time learning about networks. They are useful for my work. Besides, our brain is the same neural network. So denying its usefulness in trading is just like denying the usefulness of the brain. The problem here is that we still don't understand how our brain takes the same price series, focuses on the important points, abstracting away the details, and makes a decision. A simple textbook neural network cannot simulate this behaviour. Even if we were able to simulate such behavior, the speed of learning and operation of such a "biological" network would be much slower than our brain and inapplicable to trading.
There is only book knowledge on networks.
In econometrics TS is passed as one of classification tools. But classification is not enough to build a normal econometric model, NS may be part of the model, and not the most important. Modelling begins with learning and mastering estimation methods. Without having estimation methods one cannot analyze a quotient, one cannot evaluate a model that we have built on the results of the analysis, one cannot evaluate the results of model application. To all of the above, NS has nothing to do.
Your post confirms my perplexity. You have managed to figure out a very not simple idea called NS, which can be applied in trading after all. But for some reason you haven't spent a single minute studying a much simpler concept called regression. And understanding regressions and knowing how to use them would have dramatically changed your attitude towards indicators in particular and TA in general. And there would be one more person on the forum who would claim "TA is for patsies on the field of miracles".
something like this...
Hypothesis formation
Data retrieval and collection
Data preparation (filtering, transformations)
Model selection, model parameter selection and training algorithm
Model training (automatic search for the remaining model parameters)
Analysis of training quality
Analysis of identified patterns
all this is called Data Mining...
A network in this machine (if it's used)... about properly prepared data I totally agree... if the data is normal then a simple linear regression is enough... that's what you should aim for...
I agree that we know the steps to build the model. We do not yet know how to create a neural network embodying these steps. Maybe in the distant future we will learn how to create such neural networks. For now, one neural network (the brain of the developer) does all the important preparatory work of processing data, identifying patterns, selecting a model, and optimizing this model, while another network (in the program) does only computing based on the data and structure given to it by the developer. Adding self-optimising weights to this second network does not make it smarter, but only reduces modelling error.
I agree that the steps to build a model are known to us. It is not yet known how to create a neural network embodying these steps. Maybe in the distant future we will learn to create such neural networks. For now, one neural network (the brain of the developer) does all the important preparatory work of processing data, identifying patterns, selecting a model, and optimizing this model, while another network (in the program) does only computing based on the data and structure given to it by the developer. Adding self-optimising weights to this second network does not make it smarter, but only reduces modelling error.
it is possible to make such a construction...in ready-made statistical packages or several...to accompany everything with scripts and macros (i.e. to automate completely)
one of the options -
Hypothesis formation - use any charting tool and train net or ha or etc. on max return
- get either BP which will give max return or Booleans (as signals)... in future you can use target function
on max return and etc. if net or etc is self-written and normal...
Searching and collecting data - downloading everything we have automatically...
Data preparation (filtering, transformations) - classification, clustering, factor analysis, whitening of inputs
can also be done partly by the grid... you can do with ready-made algorithms...
(This is the most important section and cunning transformations may very well improve the results)
Model selection, choice of model parameters and learning algorithm - several models are tested by 1 learning error and the best one for the available data is chosen (there are such ready-made modules in statistical packages)...
Analysis of identified patterns - here you can just feed them into the network and look at the weights in % ratio or so on...
Such construction I would probably the closest to a suit of intelligence - as the target is selected by the model itself to begin with and everything is automatic from raw data to final cut...
In general, of course, all this is very time consuming and inefficient... and the interesting thing is that in practice it won't differ much from optimization of normal (and) TA indicators for max return or econometric problems (unless a normal predictor flashes in the data))) as all properties of BP will anyway penetrate into the model and get registered... with all subsequent consequences...
So while Sanych will run with eq ruler and say that econometrics is the best... The more ingenious ones will try to cut with TA... with retrain at a certain step (as one of the options), catching BP dynamics that continues for some time and ignoring non-stationarity...))