Testing real-time forecasting systems - page 47
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
I like your forecasts, in my opinion it is quite right that the forecast is constantly "mutating" by digesting fresh data.
Let me ask you a couple of questions:
- Have you managed to automate trading on these forecasts?
- How does the accuracy of forecasts look like depending on TF ("scale" of input data), forecast depth/window? Did you manage to find the golden mean? It's a stumbling block for me...
- And in two general words, what is an input of NS?
- Did you manage to automate trading on these forecasts?
Let's just say it's clear how to do it, i.e. put it on automatic. The forecast calculation is now in MahtCAD, there are still some optimization problems to solve. The trading itself based on the forecast (it doesn't happen often when I'm super confident) is done in semi-manual mode. This is possible thanks to the MT-MathCAD interface technology by Andrew (komposter). Once again I want to thank him for that :o).
- How does the accuracy of the forecast look like depending on the TF ("scale" of input data), forecast depth/window? Have you managed to find the golden mean? It's a stumbling block for me...
Very small statistics so far, only about 100 runs. It's because of the long calculation time (already complained :o) and the semi-automated mode of the process itself. The detailed answer will be given later, when I will be able to properly plan and run the experiment. Conceptually (I mean theoretically) I investigated the prediction interval using fractal analysis. (If interested, A.A. Potapov "Fractals in Radiophysics and Radar" Topology of Sampling, p. 47, section "prediction interval of chaotic systems". I have a hard copy, not available electronically :o(
- And if I may, in a nutshell and roughly, what serves as the NS input?
er, it's a bit more complicated in terms of conspiracy. Let me just say: if you read the above - it is based on self-organising stochastic systems with a random structure. That is, the market is presented as a set of "mini-models for all occasions" (just the models). At the time of price formation the transition between models is performed, sometimes completely random, sometimes not so random, I mean correlated. Well, the inputs come in:
There are several networks. The initial probability density function looks like this (I may use a picture I gave earlier here https://forum.mql4.com/ru/6100/page31):
Have you tried stock prediction? It would be interesting to see
Just for fun, which CFD can you try to predict from here? Or from Alpari, FXstart?
Have you tried to calculate the stock? It would be interesting to see
I haven't tried it yet, but I don't think anything good will come of it. The thing is that the System is tuned to certain series, namely currency quotes. Shares have fundamentally different characteristics and nature and it is almost impossible to predict them (IMHO, of course), unless of course you have an insider. For example, predicting the collapse of Enron was impossible in principle, not even with such a rare crap as an Elliot wave. :о)
marketeer писал(а) >>
grasn, it is questionable whether it is necessary to try to make a forecast for more than 1 day. IMHO, it is a waste of resources. Instead of doing a weekly forecast, it would be better to do a more accurate 1 day forecast with more models. And if we keep thinking that the quality of forecasts should improve with the shorter timeframe, then we should make only 1 hour forecast (if it can cover 1 hour in real time, of course). Then you can literally ride the market wave, trying to catch all the trends and corrections, and not guess about tomorrow and beyond.
You are probably right, but only an experiment can refine the theoretical calculations on the boundaries. I am preparing for it. I would only point out that a 1-2 bar prediction is meaningless (IMHO, of course). This level of nesting is the most unreasonable, for the largest, fastest and most chaotic moves occur here. No model will accurately make that prediction.
I haven't tried it yet, but I don't think anything good will come of it. The point is that the System is geared to certain series, namely currency quotes. Shares have fundamentally different characteristics and nature and it is virtually impossible to predict them (IMHO, of course), unless of course you have insider knowledge. For example, predicting the collapse of Enron was impossible in principle, not even with such a rare crap as an Elliot wave. :о)
Debatable assertion. There is less stochasticity in stock prices, so in general they are easier to predict. But it's probably stochastic methods that aren't appropriate there. And unpredictable collapses and surges happen in forex as well. Currencies are influenced by many more factors than a particular stock. The same news releases sometimes initiate some pretty big disturbances (look for example at June 5 14:30 - I especially like EURJPY ;-) ). Also IMHO.
You are probably right, but only an experiment can refine the theoretical calculations on the boundaries. I am preparing for it. I would only point out that a 1-2 bar prediction is meaningless (IMHO, of course). This level of nesting is the most unreasonable, for the largest, fastest and most chaotic moves occur here. No model will accurately fulfill such forecast.
I agree, 1-2 bars is nonsense. But the scale does depend on the TF. "I'm much longer in parrots". A dozen of M15 bars - that sounds like operational planning. But of course you know better - you know the system.
Controversial statement. There is less stochasticity in stock prices, so in general they are easier to predict. But it's probably the stochastic methods that aren't appropriate there. And unpredictable collapses and surges happen in forex as well. Currencies are influenced by many more factors than a particular stock. The same news releases sometimes initiate some pretty big disturbances (look for example at June 5 14:30 - I especially like EURJPY ;-) ). Also IMHO.
I don't doubt it's debatable. It seems a bit simpler to me, let's open for example EURUSD (without dates) and find at least one crisis (by eye). I'm sure that without dates at the bottom of the chart, we won't find the "collapse" that Soros had in his time. Quotes are solidly made up of such "disasters". In other words - such perturbations are everywhere.
a bit of philosophy. I have a very simple approach, and I find confirmation of that all the time. If very, very briefly, the philosophy boils down to one sentence: "there are no unmanageable processes". No one wants an "uncontrollable car", an "uncontrollable washing machine" etc. Likewise (and better still) no one needs "unmanageable dollar", "unmanageable ruble", "unmanageable stock". With equities, though, it seems harder to me, if you don't understand the "rules of the game" it's better to stay out of it. Few people seem to understand them. Speculative capital is present in these markets in large quantities, I think much more than the value of the companies themselves.
I agree, by 1-2 bars is nonsense. But the scale does depend on the TF. "In parrots I'm much longer". A dozen M15 bars sounds like operational planning already. But of course you know better - you know the system.
Of course, the scale is important. But so far I am not even looking beyond H4.
For the sake of sport, what CFD can you try to predict from here? Or from Alpari, FXstart?
The MICEX, the RTS, or Sberbank, for example
ps the data can be prepared in any convenient format