Econometrics: one step ahead forecast - page 57
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
Let's break down this model:
1. trend - which trend are we talking about, as there are many of them;
2) Noise - it depends on the parameters of the trend in question and often the noise itself has a trend;
3. Periodicity - sine is inevitable, but it should be kept in mind that two consecutive Gamma functions also yield almost ideal full-period sine, which means that it is not yet clear;
4. Outliers - unpredictable, but apparently their corridor can be delineated.
Trend is a household word. Right - smoothing.
Noise: not only is there a trend, there is practically always a trend. We should take the model for the initial quotient and subtract it from the model. We will get the model noise. Check this noise for autocorrelation and if it is present, smoothing again and so on until we lose the pulse.
Repeat model properties table. Obtained by optimizing the model by the profit factor:
The most interesting and from my point of view essential properties of the model:
R-squared - correspondence of the model to the initial quotient. It should tend to 1. There are very low values. Moreover, there are negative values - the model does not correspond to the quoted price at all! But it is profitable!!! Here is the truth about the tester.
LM ACF - Lagrange autocorrelation - shows probability of no autocorrelation, i.e. trends - no need for smoothing.
ARCH - two tests. Shows the probability of no heteroscedasticity. This test together with LM gives grounds to assert that the residual is stationary
Max Prob c - maximum probability among regression coefficients means the probability that the equation coefficient is not equal to zero.
The last two columns are the number of lags in the regression equation. The first number in brackets is for NR, the second number is for the residual. We see: a) a shift by bar changes the model; b) the first smoothing did not produce a stationary residual and required a second level; c) after the second level of smoothing, the residual is stationary as can be confirmed by LM and ARCH
For Mathematician: the residual is stationary and cannot be used - R squared is just creepy and very often the probability of the regression coefficient equal to zero is very high (it just didn't hit here).
why all this...if you can't even predict the "trend" ))))
Trend is a household word. The correct word is smoothing.
Do not generalize and criticism should start with you
You did not make graphical predictions in the form of dots?
I.e. - take a candlestick chart ...
1 - a line that forecast ( it is possible to put it backwards - just as an ideal forecast)
2 - forecast points ( coloured ) for shorts for example red for long blue ( any, only clearly visible )
the table of 40 observations, let's see for interest ...
if you have the desire and time of course...
i have not made any forecasts graphically in the form of dots ?
I.e. - take a candlestick chart ...
1 - a line that forecast ( it is possible to put it backwards - just as an ideal forecast)
2 - forecast points ( coloured ) for shorts for example red for long blue ( any, only clearly visible )
the table of 40 observations, let's see for interest ...
if you have the desire and time of course ...
By the way. Just now noticed.
The S.E. regression column is the standard regression error. For 07.09.2011 the error is 653 pips. These are the results of optimization by the profit factor. In other words, the wrong model was fitted to the quote and this model quite accidentally made a profit! Such an accident cannot be seen in the tester. Profit factor testing is a necessary thing, but only for "correct" models
See my article. Even the code is laid out
If you're asking for a line, send it... drawing - also ... I just don't have the time or inclination to mess with someone else's... + a completely different software is used...
but the most important thing would be to see for yourself...apply and look at the history...
that's what i'm talking about - see the screenshot...
the main thing is to predict possible reversals ! - You call it a direction ... (circled in red interesting areas + arrows) ... I.e. - if you make a prediction for one day - then roughly it turns out that the model predicts the colour of the candle ... it sets a point in the future - and depending on whether it is higher or lower than the opening of the day (or the previous forecast) you can speculate about the direction of the movement ... there is no reversal - you trade in the same direction from the day opening as the previous day ... if there is a reversal...
if the model is unable to catch the reversal ! - it has no value at all and = a waste of time...
i.e. the well-known principle - today is like yesterday and tomorrow will be like today... that's why it makes no sense to talk about the Profit Factor until there is a little bit of a normal model...
on the screenshot -
white = teacher (what I would like to see)
yellow = forecast for 1 bar (oos)
you're going somewhere )))) if you want a line, send it over ... sketch too... I just don't have the time or inclination to mess around with someone else's... + a totally different software is used...
but the most important thing would be to see for yourself...apply it and look at the history...
that's what i'm talking about - see the screenshot...
the main thing is to predict possible reversals ! - You call it a direction ... (circled in red interesting areas + arrows) ... I.e. - if you make a prediction for one day - then roughly it turns out that the model predicts the colour of the candle ... it sets a point in the future - and depending on whether it is higher or lower than the opening of the day (or the previous forecast) you can speculate about the direction of the movement ... there is no reversal - you trade in the same direction from the day opening as the previous day ... if there is a reversal...
if the model is unable to catch the reversal ! - it has no value at all and = a waste of time...
i.e. the well-known principle - today is like yesterday and tomorrow will be like today... that's why it makes no sense to talk about the Profit Factor until there is a little bit of a normal model...
on the screenshot -
white = teacher (what I would like to see)
yellow = forecast ( oos )
You even put a link in your beak - you're too cool.
You're too cool... ...then you write that -
I.e. the wrong model was fitted to the quote and this model quite accidentally made money!
You may even send me to look something up ))))), or you may analyze the original tracking indices in one article and wonder why they don't work )))...
For some reason it seems to me lately that stationarity should not be looked for there, i.e. not directly in the residuals from the regression on the series of quotes, but in something else.
But it must be found in any case. Otherwise, the use of statistics is doomed.
We must abandon the idea of searching for stationarity. More precisely: under the conditions of a highly non-stationary series the stationarity areas are a rare exception to the general rule.
I have long observed -- and not only on this forum -- long and futile attempts to find this very stationarity.... But what is it for? Is it for the sake of sport? Or for profit? But the initial series -- which we need to deal with -- was non-stationary, and remains so. So it is of no practical use to find this stationarity.