Econometrics: one step ahead forecast - page 69
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
And you use a linear one-layer forward propagation network with Hodrick-Prescott as input and prediction as output
EURUSD = C(1)*EURUSD_HP(1) + C(2)*D(EURUSD_HP(1)) + C(3)*D(EURUSD_HP(2))
https://en.wikipedia.org/wiki/Feedforward_neural_network
So your statement should read as follows: "I don't believe in non-linear multilayer networks".
and even more accurately, you could put it something like this: "I don't believe in anything that isn't in Eviews" ;)))
There are no linear multilayer networks as they are mathematically reducible to single-layer networks. So faa1947's statement should go like this: "I don't believe in non-linear networks". Pretty narrow-minded.
There are.
Simple, convergent - but as long as they're not convergent, they happen :)
Vladimir: SanSanych's outlook is not narrow, but his task is specific, it seems to me. imho, of course.
And he has a bulldog grip...
Nets get understudied.
I'm sure it is.
Still trying to get the topic back on track. My indications of the goodness of the model lead to an increase in its predictability?
Here's Avals introducing trendiness. If at the end of the sample this sample has his trendiness score, then maybe this will give predictability?
But this is not known, one thing is clear: you teach the black box something and believe it. Then the tester and the faith grows, grows and future depot draining is proportional to this faith, because there is no statistics on usage. How many publications on NS I have seen and not once have I seen R-squared in the text.
By the way, this is by no means a black box.
Try to output the results of all the calculations you are interested in via the indicator. This could be sums in neurons, outputs after activation function, etc.
You will get illustrative pictures and useful knowledge.
And as a result you will get a clear idea of how the black box works, but now completely transparent.
By the way, this is by no means a black box.
Try to output the results of all the data you are interested in through the indicator. This could be sums in neurons, outputs after the activation function, etc.
You will get illustrative pictures and useful knowledge.
And as a result you will get a clear idea of how the black box works, but now completely transparent.
Well, of course (no netting!): 13 separate operations, the positions are different.
Thanks for trying to make sense of the figure.
Now, once again, slowly and carefully: we take a big moment in time and begin to sell the muv in small portions. The period of the mouve choose in advance - say, 13. When we sell a muv, we have already decided in advance that the buyback operation will only take place when the price is above the muv. But we do not know what will happen to the price after 13 bars. Let's assume that at the 13th bar, the price is still above the muv. Yep, we have missed (this is a typical situation, not an exception.)
We continue to patch the mouve until the price is again below the mouve. How do we do the follow up? In this case, we close the small position we opened on the left vertical line, and open a new microposition on a new bar. That is, our muv is relevant again, i.e. we have exactly the muv on the current bar. This is the "accompaniment" of the muv. And now we do this until the price is again below the muv.
This is the point at which we catch the price in order to buy back the entire price. But by the way, we do not need to redeem the price if our equity is higher than the balance (in such a case, the "I" operation is completed). We only redeem if the equity is lower than the balance.
Suppose the price had to be redeemed. Then we continue to hold the move until the price is higher (as highlighted in blue above). We may have to do this 50-60 or more times. But the muv won't stay on one side of the price forever, right? That's as soon as it will be on the other side of the mouve - finita la comedy, we cover all open positions and fix total profit of operation "Y".
All this reasoning seems complicated until you grasp the main point: it's stat.arb between muv and price. The only real problem is the need for the initial accumulation of muv and its accompaniment.
P.S. I looked at your drawing and realized that you are redeeming the muv, not selling it. So the situation in this case is just favorable, and we can safely sell the price. And we can wait to sell the price later.
OK, let's not develop the topic here, as it disturbs the topic of the topic-starter.
Closer to the R-square, please
I tend to believe that by teaching on the flow of price data, this is not achievable/receivable at all.
Here's my take on it.
Of course (no netting!): 13 separate operations, the poses are different.
Thanks for trying to make sense of the pattern.
Now, once again, slowly and carefully: take a fondue moment in time and start selling the muv in small portions. We choose the period of the mouve in advance - say, 13. When we sell a muv, we have already decided in advance that the buyback operation will only take place when the price is above the muv. But we do not know what will happen to the price after 13 bars. Let's assume that at the 13th bar, the price is still above the muv. Yep, we have missed (this is a typical situation, not an exception.)
We continue to patch the mouve until the price is again below the mouve. How do we do the follow up? In this case, we close the small position we opened on the left vertical line and open a new microposition on a new bar. That is, our muv is relevant again, i.e. we have exactly the muv on the current bar. This is the "accompaniment" of the muv. And now we do this until the price is again below the muv.
This is the point at which we catch the price in order to buy the entire price. But by the way, we do not need to redeem the price if our equity is higher than the balance (in such a case, the "I" operation is completed). We only redeem if the equity is lower than the balance.
Suppose the price had to be redeemed. Then we continue to hold the move until the price is higher (as highlighted in blue above). We may have to do this 50-60 times or more. But the muv won't stay on one side of the price forever, right? That's as soon as it will be on the other side of the mouve - finita la comedy, we cover all open positions and fix the total profit of operation "Y".
All this reasoning seems complicated until you grasp the main idea: it's stat.arb between muv and price. The only real problem is the need to initially accumulate muv and accompany it.