Econometrics: one step ahead forecast - page 15
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
Don't bother. I've already figured out that you're a fundamentalist. You won't read mathematical calculations.
I'm not even Jewish :((
I'm not even Jewish :((
What a wonderful ending to the topic!
Looking forward to further predictions! (we should set some goals, at least for more than 5 points:)) (for now we need to make firewood, straw)
Looking forward to further predictions! (we should set goals for at least more than 5 points:)) (for now we have to make firewood, straw).
I will. I'm thinking of making a multi-currency EURUSD forecast and comparing it with the existing lag model.
Even the most beautiful and plausible predictions have a 50% chance of coming true
I'd like to make a wish, if I may and to all be heard: something on the merits of the topic and with concrete proofs, so far one itches.
Be my guest:
" ... The model is an arbitrary function (regression) of the form y = f(x1, x2, .... xn). Function y is e.g. EURUSD or any other currency pair. xi - function's arguments (independent variables, regressors) are any other quotes available in the terminal ...".
1. (Not the main point) What do you mean by "arbitrary function"? Any regression uses some polynomial for interpolation.
2. (Main) Strict definition of "Function" try to apply to the case under discussion.
Thank you, finally a substantive question.
1. (Not the main one) What do you mean by "arbitrary function"? Any regression uses some polynomial for interpolation.
2. (Main) Strict definition of "Function" try to apply to the case under discussion.
Strictly I can't, but I will make some clarifications. Let's go from the existing example.
kotir hp1(-1 to -4) hp1_d(-1) hp1_d(-2)
1. we have a linear function with two lag variables (HP - Hedrock-Prescott indicator for kotir and hp_d - residue between the indicator and kotir). Note that we do not use lag variables of the cotier itself.
2. There can be many and different variables in the function, for example, other quotes.
3. We distinguish two types of linearity: linear by variables and linear by parameters. Our function (regression) is completely linear.
4. A function non-linear in variables (arguments, independent variables, regressors - synonyms in my case) can have a rather arbitrary form within the framework of mathematical functions allowed in EViews. For example: log(x) - natural logarithm, 1/x, etc. (see documentation) This type of non-linearity is reduced to a linear version by replacing it with an appropriate formula, e.g. xx = 1/x and then using xx instead of x
5. Much worse if the equation is non-linear in parameters, e.g.: y = x*(c^2), where the constant is squared). The trouble is that the MNC doesn't work.
6. Even worse is this: the parameters of the equation are estimated, i.e. they are random variables. The last column of the equation estimate is the probability of the corresponding coefficient being zero. Everything is fine, if the distribution law of a random variable called "coefficient" is normal, and if it is not non-normal, then the estimate cannot be trusted at all - it is just numbers.
Conclusion: all TA is bullshit, as all indicators are formulas, but nobody ever analyses these formulas like above.
I should point out that this is not my thoughts at all - a retelling of what I have read, which relates to the basics of econometrics.
Back to the forecast. There was a forecast for Friday
Forecast: 1.3548 - short. Result: Close price = 1.3753, long - forecast incorrect. Error = 205 pips.
When forecasting I noted that the forecast error is very large = 249 pips and the resulting error is within the calculation, but it is not satisfactory at all. Of course, out of three forecasts - two correct and one not correct, but that's not about it either. There is no work to be done.
The direction of further work on the equation is obvious: we need to change the equation to reduce the error to more acceptable values. At the same time the obtained achievement - strictly reject the hypothesis that the equation coefficient is equal to zero - should be kept . I am waiting for suggestions.
Recall that the equation is given by the form, for example:
kotir hp1(-1 to -4) hp1_d(-1) hp1_d(-2)
The leftmost one is the function - the independent, predicted variable in our case EURUSD, and the rightmost one is the function arguments. This schematic equation corresponds to the equation in its more familiar form:
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)
The specific value for C(i) will be obtained by estimation and for our equation has the form:
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)
which is used in calculations and forecasting. I should note that this indicator matches the quote by 97% - a great thing for lovers of mash-ups. I don't want to rest on Vinin's laurels with his toys.
I'll make another forecast on Monday afternoon as I'm forecasting on opening prices, which will give an opportunity to take the gap into account when going into next week.
Back to the forecast. There was a forecast for Friday
Forecast: 1.3548 - short. Result: closing price = 1.3753, long - forecast incorrect. Error = 205 pips.