Econometrics: one step ahead forecast - page 49
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
The forecasts should be long and short with an assessment of the probability of the appropriate market direction.
but not by the means described here...
The market has a good chance to go long or short... it's not so bad if you don't want to go long or short... under what conditions this approach presumably works... in which conditions it doesn't...
if you look carefully at the tests - the approach is nonsense and the same results (the same forecasts) may be obtained using much simpler means... and done in the form of a simple indicator...
and prediction error analysis is an inefficient tool...
In general, this post should be called - "I read a book - I made it - it doesn't work - tell me why))) "
the proposed method has 3 flaws...2 of them are specific...
in the proposed method there are 3 faults...2 of them are specific...
Ow, gentlemen traders! Well, at least something concrete!!!!
the proposed method has 3 faults...2 of them specific...
Ow, gentlemen traders! Well, at least something concrete!!!!
What's wrong with standard methods of assessing the quality of the TS - like the profit factor, for example?
OK, let it be specific))). As you have already been told, bare statistics, without substance, are doomed to failure. I will only decipher what I personally mean by it.
Before you work with the real market, try this simple model as an input quotes. We take white noise. It may be even Gaussian. Forthe first n samples add constant M1 . Do not add anything tonext n samples . Add constant M2 to next N samples, do not add anythingto next n samples . Add constant M3 to next N counts and soon. Then we integrate obtained non-stationary white noise and take it as an input process. I.e. we obtained a martingale that contains trends. The lump-trend model )))). Let also the constants M1 , M2 , ... be large enough (compared to the variance of white noise), so that each trend can be profited from. And let the constant n be small enough compared to N . Say N = 100, n = 10. Classical regression models fly on such a process. The confidence intervals will be so wide that you just don't have time to capture the trend from n samples. Let's say on 10 counts out of 10 you will realize - yes there was a trend here. But it will not give anything for the further game.
Is it possible to earn on such series? Yes, we need to add some content to the bare statistics - to understand that there are short periodic trends here.
This is all for the sake of example. In real quotes there are no periodic trends. But there are all sorts of other local effects that can be statistically fixed only after the fact.
in the proposed method there are 3 joints...2 of them specific...
Hello, gentlemen traders! Well, at least give me something concrete !!!!
From my engineering practice.
A colleague of mine was once sent on a business trip. He was designing a vibro-dipper for 2 years. The vibro-dipper is a device, which has something like an eccentric, and it was designed for driving piles into the ground.
So, he left on a business trip with his rattling wonder. Our customers call from there: "Your specialist came, installed his rig on the pile and said - this (bleep) won't work! He took out a bottle of vodka, drained it in two gulps, and disappeared in an unknown direction. .....
The man didn't admit until the very end that his work was crap. But one day he did.
What's wrong with standard methods of assessing the quality of the TS - such as the profit factor, for example?
OK, let it be specific))). As you have already been told, bare statistics, without substance, are doomed to failure. I will only decipher what I personally mean by that.
Before working with the real market, try this simple model as input quotes. We take white noise. It may be even Gaussian. Forthe first n samples add constant M1 . Do not add anything tonext n samples . Add constant M2 to next N samples, do not add anythingto next n samples . Add constant M3 to next N counts and soon. Then we integrate obtained non-stationary white noise and take it as an input process. I.e. we obtained a martingale that contains trends. The lump-trend model )))). Let also the constants M1 , M2 , ... be large enough (compared to the variance of white noise), so that each trend can be profited from. And let the constant n be small enough compared to N . Say N = 100, n = 10. So classical regression models fly on such a process. The confidence intervals will be so wide that you simply won't have time to capture the trend from n samples. Let's say on 10 samples out of 10 you will realize that there was a trend. But it will not give anything for the further game.
Is it possible to earn on such series? Yes, we need to add some content to the bare statistics - to understand that there are short periodic trends here.
This is all for the sake of example. In real quotes there are no periodic trends. But there are all sorts of other local effects that can be registered by statistics only after the fact.
It's satisfying and practically the only one. But there are two circumstances: 1) it is not known what to change if the result is bad, and 2) the prognosis for the future is unknown.
1. either the model parameters or the model itself. You can elaborate on the criteria
2. it will always be unknown. One can only hope that the market stays the same for a while. The rest is utopia or insider
OK, let it be specific))). As you have already been told, bare statistics, without substance, are doomed to failure. I will only decipher what I personally mean by it.
Before you work with the real market, try this simple model as an input quotes. We take white noise. It may be even Gaussian. Forthe first n samples add constant M1 . Do not add anything tonext n samples . Add constant M2 to next N samples, do not add anythingto next n samples . Add constant M3 to next N counts and soon. Then we integrate obtained non-stationary white noise and take it as an input process. I.e. we obtained a martingale that contains trends. The lump-trend model )))). Let also the constants M1 , M2 , ... be large enough (compared to the variance of white noise), so that each trend can be profited from. And let the constant n be small enough compared to N . Say N = 100, n = 10. Classical regression models fly on such a process. The confidence intervals will be so wide that you simply won't have time to capture the trend from n samples. Let's say on 10 counts out of 10 you will realize - yes there was a trend here. But it will not give anything for the further game.
Is it possible to earn on such series? Yes, we need to add some content to the bare statistics - to understand that there are short periodic trends here.
This is all for the sake of example. In real quotes there are no periodic trends. But there are all sorts of other local effects that can be registered with statistics only after the fact.
It is possible to invent a lot of things.
Initially, I stated my verbal description of a quote = trend + noise. This description makes sense in terms of forecasting, as the trend is predicted.
In this thread I have raised a very narrow issue: a 1 step forward forecast. I have proposed a model and am trying to find out if the forecast can be trusted. If you can, why, and if not, why not. On this topic I would like to hear opinions and suggestions. And willing to do the dirty work of coding to test hypotheses. This is what I call specificity.
1. either the model parameters or the model itself. You can elaborate on the criteria
Here is part of the summary table:
What to change?