The market is a controlled dynamic system. - page 17
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
NS is nothing more than a smoothing algorithm, and it is completely black and poorly controlled. NS does not solve the problem of model stability, which is the main thing. There are many steps involved in getting a stable model (or at least estimating it) (see my article). It's like building a building: first get the excavation right, then the foundation, then the walls, etc. You correct and correct the model, and in the end it turns out that the foundation is not good enough for the result. It is a process. That's why your "clearly... sorry" what does that mean?
I have models with up to 10 pips prediction error. But a very narrow range of model formulation ideas. EViews (as well as other packages) is a toolkit, and a huge one at that, but it's not a textbook econometric model builder. One guy here built a sensible econometric model and became a Nobel Prize winner.
No, why... you can use the NS in any way you like... and as a search for patterns etc... about the stability of the model - I agree...
ok - sorry - applies - would like to see the model forecast... can i give a prediction at my discretion for the next step ?
if yes which TF is preferable ?
faa1947 and avtomat the conversation is going on a bit nervously...let's not attack each other...
I'm very interested to see the forecasts of faa1947
The first thing to do is to find an R-market function that can initially handle the artificial sine waves and/or sums of them you have specified. Suggest variants of the R-function. I have one of its variants, give me a sinusoid of any complexity - we will describe it with satisfactory accuracy (up to 10%).
How do you build sine waves from a graph - from a spectrum? A set of spectra from which you get sine waves and which you want to extend into the future at least up to 3000 years is stationary only in the given area, further it will float (and you know it yourself)
No, what I mean is that we already have sine waves... they are not extracted from anywhere... why extract them - get them from somewhere... just initially adapt them to the market a bit... but I said that - as one option... in fact, I am not interested in them... as they will not work...
Here is a set of successive changes in values - a sine wave of any complexity. You can sum the values and get a trend. Describe it.
clear- sorry - applies - would like to see a model prediction... can i give at my discretion the end of quotes for the next step of which you will give the prediction ?if yes, which TF is preferable ?
It doesn't matter for demonstration - H1, H4, D1. But which model to take ? For example,
EURUSD = C(1)* EURUSD(-1) + C(2)* EURUSD (-2)
or
EURUSD = C(1)* NR(-1) + C(2)* NR (-2)
where NR is the Hendrick-Prescott filter
What is the number of candles?
Formulate it and I will post the result.
For demonstration purposes it doesn't matter - H1, H4, D1. But which model to take? For example,
EURUSD = C(1)* EURUSD(-1) + C(2)* EURUSD (-2)
or
EURUSD = C(1)* NR(-1) + C(2)* NR (-2)
where NR is the Hendrick-Prescott filter
What is the number of candles?
Formulate it and I'll post the result.
Well then, whichever is more convenient.... it's important that the sections are more complex...
like on the screen for example - try to predict the decline...
several sections preferably - if it doesn't take too much time...and you're willing...
we can do it tomorrow... there's no rush...
Well then, whichever is more convenient .... The main thing is to make the plots more difficult...
like in the screenshot for example - try to predict the fall...
a few sections preferably - if it doesn't take too long...and you're willing...
we can do it tomorrow... there's no rush...
Let me, too -- for comparison -- demonstrate the results of a couple of my models. Tell me what kind of tool it is, TF.
Well then, whatever you like .... The main thing is to make the plots more difficult...
like in the screenshot for example - try to predict the fall...
a few sections preferably - if it doesn't take too long...and you're willing...
we can do it tomorrow... there's no rush...
Let's begin.
I think you should start forecasting with simple candlesticks - there are many of them and profit among them, but unique candlesticks, such as you want, are an outlier and should be deleted. Nevertheless, let's try.
Let's formulate a model (that's where the whole dog is buried). I will take it from the article:
EURUSD = C(1)*EURUSD_HP(-1) + C(2)*D(EURUSD_HP(-1)) + C(3)*D(EURUSD_HP(-2))
i.e. we take Hodrick-Prescott smoothing and two values of the residual between the smoothing and the quote.
The candle you specified is 2011.09.21 16:00
We take the interval of 137 candles from 2011.08.22 00:00 to estimate the model
Obtain the quotes in table form. Beginning of the interval:
End of interval:
Graph:
To extract a regular component we smooth the initial quote using HP filter with the slab 10:
Let's estimate the model:
Variable Coefficient Std. Error t-Statistic Prob.
HP(-1) 0.999640 0.000195 5132.513 0.0000
D_HP(-1) -0.058770 0.097616 -0.602049 0.5482
D_HP(-2) -0.426574 0.097682 -4.366958 0.0000
R-squared 0.990507 Mean dependent var 1.407233
Adjusted R-squared 0.990363 S.D. dependent var 0.032459
S.E. of regression 0.003186 Akaike info criterion -8.637831
Sum squared resid 0.001340 Schwarz criterion -8.573269
Log likelihood 586.0536 Hannan-Quinn criterion. -8.611595
Durbin-Watson stat 1.557580.
The further part can be skipped as the probability of the coefficient at D_HP(-1) being equal to zero is 54.82%.
The model used is not good enough. You should not trust the forecast of this model.
I give you the result just to see how it looks like:
From the table view, the forecast for the candle of interest is 1.368504 - which is very far from fact.
Once again. This prediction cannot be trusted because my model did not pass the test.
Please give me suggestions for the model, I will do the math. We need to check on all the candles of the plot.
Once again. This prediction cannot be trusted as my model has not been tested
thanks interesting...but the question is why exactly Hedrick and not MA for example ? or SSA...
and how much does the score depend on the number of test samples ? i.e. 137 now and if you take 2000 for example...i.e. it's all about adjusting to the latest pair dynamics...or is it not ?
The relay states are clear from the above figures.
It should also be noted that the final decision is made taking into account the older TF Daily - the movements should be in-phase.