Econometrics: one step ahead forecast - page 13

New comment
 
tara:

Is leverage a leverage, or am I confused again? If not, I would like to know more about clean and random, but without leverage ...

In EViews leverage is a property of cotierre to make reversals after strong moves. I have probably translated it wrong. Suggest it, I would appreciate it.
 
Mathemat:

Well actually not at all well and good. That's an odd phrase.

faa, please comment on why you don't like equity compared to balance so much.

I didn't write that I don't like it. My point is different: before you step on the gas, you should be sure that the engine will not jam, both at the stage of car's design and its operation. And equity or balance is an outcome that is usually negative and why is not known.
 
faa1947:

Using these articles I propose the following: to work collectively on creating econometric models for predicting the quotes of currency pairs one step ahead. The size of one step corresponds to the timeframe to which the Expert Advisor outlined in article #2 is attached.

Using econometric models for quotes makes no sense in my opinion. Maybe, the application of econometrics to some GDP or unemployment levels makes sense - I don't know. But quotes, that's a completely different object.
 
HideYourRichess:
Applying econometric models to quotes makes no sense in my view. Maybe it makes sense to apply econometrics to some sort of GDP or unemployment rates - I don't know. But quotes, that's a completely different object.
The hell it is. Up to and including H1 it seems, as the prediction error is comparable to the length of the candle. So far I see this as the problem. The longer the period, the quicker you can get a stable residual, which is reassuring and the error is much smaller than the candle length.
 
Mathemat:

It's already been dealt with here. A colleague accidentally confused martin[gale] with martingale, it happens...

I would like to express a wish, if I may be allowed and heard: something on the substance of the topic and with concrete evidence, so far it's just an itch.
 
faa1947:

The most interesting ones are state space models

Give me an example of a model -- let it be a case study -- to get to the bottom of econometric models.

By the way, if state-space models are used in econometrics, they are borrowed from control theory and nothing else.

 

The result of yesterday's forecast - came true

Making a new forecast for the current day.

Here is the end of kotier

2011.11.06 00:00,1.3828
2011.11.07 00:00,1.3816
2011.11.08 00:00,1.3766
2011.11.09 00:00,1.383
2011.11.10 00:00,1.3524

2011.11.11 00:00,1.361

We take the old model, because we have no claims to it so far

kotir hp1(-1 to -4) hp1_d(-1) hp1_d(-2)

and estimate the coefficients. We get the following coefficients:

KOTIR = C(1)*HP1(-1) + C(2)*HP1(-2) + C(3)*HP1(-3) + C(4)*HP1(-4) + C(5)*HP1_D(-1) + C(6)*HP1_D(-2)

Substituted Coefficients:
=========================

KOTIR = -11.2283410255*HP1(-1) + 35.6907876956*HP1(-2) - 34.1403883033*HP1(-3) + 10.6774253876*HP1(-4) - 0.662636180868*HP1_D(-1) - 0.897124355018*HP1_D(-2)

Evaluation result:


No reason not to trust it.

Prediction: 1.3548 - short

Forecast statistics:

Very alarming is an average absolute prediction error (mean Absolute Error) = 229 pips! Although in % it is=0.29%.

We are waiting for Saturday.

 
avtomat:

Give an example of a model -- let it be a case study -- to get into econometric models.

Two models have already been made public in this thread. A very old idea is being exploited: isolate the deterministic component as the last 4 values of NR and take into account error = the difference between NR and quotient (two values).

By the way, if state space models are used in econometrics, they are borrowed from control theory, and nothing else

Naturally. In this sense, Matlab is very interesting. It only refers to ARCH models as econometrics. But separately there are other toolboxes, from which EViews can be assembled, only in a more posh form in all senses. There the kinship can be seen with the naked eye.

 
faa1947:

Give an example of a model - let it be a case study - you need to get into econometric models.

Two models have already been made public in this thread. A very old idea is being exploited: isolate the deterministic component as the last 4 values of NR and take into account error = the difference between NR and quotient (two values).

By the way, if state space models are used in econometrics, they are borrowed from control theory, and nothing else

Naturally. In this sense, Matlab is very interesting. It only refers to ARCH models as econometrics. But separately there are other toolboxes, from which EViews can be assembled, only in a more elegant form in all senses. There the kinship can be seen with the naked eye.

no... that's not...

I mean an example of a meaningful applied econometrics problem.

 
avtomat:

no... that's not...

I mean an example of a meaningful applied econometric problem.

I don't get it. Clarify. The models given are quite meaningful, much more meaningful than any set of indicators. What do you mean by that?
1...67891011121314151617181920...139
New comment