Econometrics: State-space model forecasting - page 19
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 I don't tell a damn thing.)
And nndo ibc...
that's it - not in the cartoons
You are redundant here. Just a few people who understand econometrics, but we deprive a bunch of well-organised coders with moderator status who graze here in the hope of a measly 100 quid for pseudo-advisors. They just hate me. Two years ago I was the first person to write the word "econometrics" on this site. Before that, a search didn't yield such a word at all. I was immediately emailed in a personal message saying, "Why would you do that?" Now you have been equated with me and your fate is sealed. That's what practice shows. You will not be allowed to post a result about the benefits of econometrics. To blab, twist, puff your cheeks and "like the older and more experienced traders" are full of virtuosos.
I haven't posted anything substantive here for a year and I urge you not to do so. We need to find another place.
PS. Although you can continue to use such a compromising word to anyone on the site "adieu" if you stop using the word "econometrics".
Yeah, SunSunich, you can't go in the hedgehog, it's too brainwashing for you.
And I think I'll put that quote in the Annals.
do the opposite - make it smart, test it and put it out there for all to see
A bit of theory.
The state space model consists of several equations, but at least two. The first one is the measurement equation, the second (subsequent) equations are the state equations.
My example in general form:
eurusd(t) = w*x(t-1) + noise(t)
x(t) = F* x(t-1) g * noise(t)
noise(t) is a random iid process. That is, if w = 0, then we have random walk.
The point of the state-space model is that the state is predicted, and then the quantity we're measuring is predicted. Naturally, we can stick other quotes, indexes, Bernanke's speeches into the states....
I have no such passions and by x I mean eurusd, i.e. I have everything reduced to autoregression. This approach impoverishes the model very much. I am pasting the result below.
There is another circumstance to the theory which is decisive in that model. w*x(t-1) is decomposed into parts: trend + irregular noise. The trend, or more correctly said "smoothing", is the part of the quotient that has an analytical form. I know several types: regression, filters, splines, wavelets. If you take that part out of a quotation, you get a random variable. I stick to the terminology "innovation", chto distinguish it from noise, which I have iid. Innovation is news, which is always random, because of it non-stationarity, fat tails and all the other delights. We model it separately.
Do the opposite - make it smart, test it, and post it for everyone to see
My model:
My model assumes that trend and innovation are multiplicative.
Very often a seasonal component is extracted from the quotes. We are looking at daily cycles and weekly cycles. I do not deal with this, although the model allows me to do so.
I will not give a specific form of state space model equations.
Additionally.
To the state space model I use a model by which I transform a point forecast into a direction forecast - a threshold autoregressive model.
Enter long - cross the upper threshold, reverse to short - cross the lower threshold.
in short - autoregression. There is no fish in autoregression.
There is also no fish in regression model construction, where the forecast for a currency pair is built from the regression model of another pair
P.S. There is no fish in seasonality either. The man in the other thread chops with seasonal trading, but mostly in commodity markets. He says - successfully
Results.
My model allows modelling of non-linear non-stationary random processes, i.e. thick tails, heteroscedasticity....
The eurusd history of 1038 bars was used. A window of 20 bars was moving along this history. On the first 20 bars initial states for the model were calculated (a very unpleasant thing in state space models). Next the results on bars 21-1038.
Calculated in pips, lot = 1, no refill.
Balance = 0.1780.
Income balance = 0.4279
Loss balance = 0.2559
Profit Factor Graph:
The profit factor is approximately 1.6.
The last one. Thresholds. Chart.
The most curious thing about the rapids. Both sills can lie either above or below the axle. Hence, the thresholds can't be replaced by a sko.
what is the average gain and average loss in pp?