Market etiquette or good manners in a minefield - page 65
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
And how did you get to 1 on the first countdown so soon? And why do you have zero spread there. Are you not randomising the weights and starting from zero?
I don't know. I randomise the weights:
Increased randomisation(0,1)
Got it. I got one in the first step as soon as I corrected a small error:
Now it is:
Now here's what you get:
What do you think? The divergence of the average errors appears to be tending towards a constant.
Can anyone explain to me the theoretical justification for the idea that there is an autoregression of prices (say, opening or closing), with its equation at time t being different from that at time t+1 (which requires network retraining) and using the last few bars as inputs? All other systems, indicators, etc. are based on the assumption that the market is controlled by some mechanisms, which behave more or less equally, if the conditions are similar. And that the current market movement is a superposition of trends of different temporal structure.
Yes, I once dabbled with autoregression myself and obtained for y = ln(High[i] / High[i+1]) or ln(Low[i] / Low[i+1]) as a function of ln(Close[i+1] / Low[i+1]) and ln(Close[i+1] / High[i+1]) R2 value about 0.45 - but I obviously lack theoretical basis :)
Why would you need a theoretical justification when there is practice clearly demonstrating that yesterday's market is not like today's and will not be like tomorrow's. It would be the other way round nets would not be needed at all. It follows directly that in order for the knowledge of the grid to correspond to the current moment it must be retrained on every countdown. Most indicators are not based on what you think (not long ago I thought the same thing). Now I realise that all the indices actually exploit a completely different mechanism - the human mind's tendency to build deterministic models of reality like TER or anti-TER. I.e. (I translate from Russian into plain English): laziness and stupidity.
The need to use some optimally short stretch of history is dictated by the same non-stationarity of market VR. I.e. what works now - may not work in an hour. The mechanisms behave in the same way - you're right here. The market behaves differently.
I use a vertical breakdown of the price series for the forecast.
Is that something new? I don't know about that. What is it?
What do you think? The divergence of average errors appears to be tending towards a constant.
Yeah, what can I say? Let's go!
Anything new? I don't know about that. What is it?
Read Pastukhov. His PhD thesis.
You'd be surprised.
Why would you need a theoretical justification when there is practice clearly demonstrating that yesterday's market is not like today's and will not be like tomorrow's. It would be the other way round nets would not be needed at all. It follows directly that in order for the knowledge of the grid to correspond to the current moment it must be retrained on every countdown. Most indicators are not based on what you think (not long ago I thought the same thing). Now I realise that all the indices actually exploit a completely different mechanism - the human mind's tendency to build deterministic models of reality like TER or anti-TER. I.e. (I translate from Russian into plain English): laziness and stupidity.
The need to use some optimally short stretch of history is dictated by the same non-stationarity of market VR. I.e. what works now - may not work in an hour. The mechanisms behave in the same way - you're right here. The market behaves differently.
There are transformations with which you can switch to stationary data. As I understand the series is non-stationary for you simply because of the volatility growing over the years.
The man was asking about a change in the cause-and-effect relationship over time, if there is one, you can hardly do anything at all...
And about the practice: "Why do you need a theoretical proof, if you have a practice, that yesterday's market is not like today's market and will not be like tomorrow's" - show me an example...(you may not show me an example in principle, because there is a counter-example for every example...)
- show an example...(you don't have to show an example in principle, because for every example there is a counterexample...)
Why not? You know it very well yourself: gigabytes of losing trading robots. Do you need any more examples?
Well, why wouldn't you? You know it yourself: gigabytes of losing trading robots. Do you need more examples?
It's gigabytes of inadequate models... do not say that today's market is not like tomorrow's, etc.