Econometrics: one step ahead forecast - page 109
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
Excel 2002 won't open a file with extension xlsx - it doesn't have such an extension in the list of file formats.
So one more repeat for the particularly gifted econometricians:
Is it possible to detail, as the data in general is not too much, formulate RESULTS in form of table for fitting and separately for forecasts as: date and time of ZZ turning?
For you personally in any quantities
According to the tables, everything seems to add up to a 4 hour bar. No catch has been noticed, it seems.
What can I say? Get your wife to sew a money bag and buy a good shovel at the hardware shop to shovel the money in. If your regression models can really predict ZZ peaks so accurately, then all sorts of Soros and other Ganns and Nostradamuses rest easy.
I had a case. Built a neural network, trained it. Ran it outside the training sample. The results were 100%. On one hand I was glad, but on the other hand I understood that miracles never happen, so I started to check it again. It turned out that the program that had been preparing the training examples to the NS, by mistake one of the inputs was fed a value that was intended for the output. Well, it was duplicated at the output too. A miracle did not happen.
According to the tables, everything seems to add up to a 4 hour bar. There don't seem to be any catch.
What can I say? Instruct your wife to sew a bag for money and buy a good shovel at a hardware shop to shovel the same money. If your regression models can really predict ZZ peaks so accurately, then all sorts of Soros and other Ganns and Nostradamuses rest easy.
I had a case. Built a neural network, trained it. Ran it outside the training sample. The results were 100%. On one hand I was glad, but on the other hand I understood that there were no miracles, so I started double-checking. It turned out that the program that had been preparing the training examples to the National Computer, by mistake one of the inputs was feeding the value intended for the output. Well, it was duplicated at the output too. The miracle did not happen.
Unfortunately our opinion is the same, I will not sew the bag - I will save on threads.
As I wrote above, I don't understand the result in principle. Attempts to read anything about probit models have led nowhere. But would really like to model reversals.
Unfortunately our opinions are the same, I won't sew the bag - I'll save on thread.
Economy and economics are different terms, although they are similar in sound. Economy is most often a consequence of overly progressive ideas in economics.
As I wrote above, I do not understand the result in principle. Attempts to read something about probit models have led to nothing.
And why understand it? If the result is confirmed in practice, then there is nothing and no time to understand - it is necessary to cut the cabbage.
And if not confirmed, the lack of results is also a result.
Economy and economics are different terms, albeit similar in sound. Economy is most often a consequence of the application of excessively progressive ideas in economics.
Why understand? If the result is confirmed in practice, then there is nothing and no time to understand - it is necessary to cut the cabbage.
And if it is not confirmed, the lack of results is also a result.
Good day!
Let me add a comment for Mr. faa1947. Try to simulate the trade by reversal inputs. I think this is where the bullshit lies. In the sense that maybe they are useless for real trading (for example, moving average forecasting also gives good results in terms of accuracy, but nothing for trading).
Another thought, I write: try modelling trades of duration from pivot to pivot. I wonder if there will be fish?
I happen to be here...
... (for example, predicting a moving average also gives good results in terms of accuracy, but gives nothing for trading).
It does not, for the simple reason that the so-called Slutsky-Yule effect is present in this curve. Its meaning is that false correlations appear on a flat spot. Essentially, you take a fixed-length segment and shift it by one sample. Such shifted samples are 99% the same (differ only by one new value) and of course some statistical values, such as averages, will be very strongly correlated, but in fact there is no such relationship as a class. One can even check on a sov.random series and get a very high correlation for MA on such series.(i.e. high MA correlation implies that the MA is similar to itself)
For successful trading, such MA must be predicted with a great accuracy but it is impossible in principle. So it is like that.
PS: since you're here, I'll give you a piece of advice - faa, don't fuck around, ZZs are not predicted - they are places where the price practically never happens, they are random, and completely random. No regression model is adequate for this case.
I'm here by chance...
(i.e. high MA correlation says that MA is similar to itself)
For successful trading such MA must be predicted with a great accuracy, which is impossible in principle. So it is like this.
Why understand? If the result is confirmed in practice, then there is nothing to understand and no time to chop the cabbage.
There is a fundamental difference in our approaches:
NS is a black box for people with good diction.
Econometrics (read statistics) does not recognise the figures obtained unless there is a verbal explanation for them. The main feature of statistics is distrust of everything and everyone. They call it scientific - probability, confidence interval, etc.
I'm closer to statistics, which is why I don't accept NS. Too often I've seen people who get a number and start waving it around. One of the basic concepts of statistics is correlation and it is in the base that people fall into the terrible fornication - getting connections to Saturn rings, coffee, etc. Any figure has to be proven meaningfully first and then mathematically.
Good afternoon!
Allow me to insert a comment for Mr faa1947. Try to simulate trading on pivot inputs. I think this is where the bullshit lies. In the sense that maybe they are useless for real trading (for example, moving average forecasting also gives good results in terms of accuracy, but nothing for trading).
Another thought, I write: try modelling trades of duration from pivot to pivot. I wonder if there will be fish?
The breakout model contains two numbers for the dependent variable: 0 and 1. And what are the reversal inputs?
The modelling process itself is as follows: a sample is taken (me 500 bars) and the coefficients of the equations are calculated from it. Then there are two modes. 1 - we take the coefficient and make a forecast for the next bar forward in the sample. We add the next bar inside the sample and again make forecasts. I posted the result above. This is classic fitting.
You can make a forecast outside the sample. Previously, when predicting cotier level is the next bar. Here it is not clear. The next reversal is definitely not the next bar. The distance between reversals is many bars and variable (random). I don't understand how to predict, that's the problem.
The regressions you construct are ideologically no different from NS: in fact they are the same games with numbers without any intelligible internal content.
The probit model contains two numbers for the dependent variable: 0 and 1
Tell me in popular language what this "probit model" is and why it is so good for econometricians.
Just don't link to the wiki. Tell us in your own language with examples.