Statistics as a way of looking into the future! - page 4

 
m_a_sim писал(а) >>

Decided to do a bit of statistical work. I bought a book and built a multiplicative model :) using Excel. I have built regression, defined seasonal component (1 season - 24 hours, one hour archive of quotes was used) and built prediction function.

Regression equation has the following form: Y=b0+b1*t+b2*t^2+b3*X1+b4*X2, где

Yi=CLOSEi-1(gold), t-time, X1=CLOSEi-1(gold)/CLOSEi-1(usd), X2=CLOSEi-1(gold), b0...b4- regression coefficients. (I hope it's clear)

So I got the following picture

Personally I'm fascinated by it, when you "see" what will happen. I want to write an indicator.

Who has an opinion about it?

Correct me if I'm wrong that the model is most suitable for predicting future portfolio volatility, assuming a normal distribution of portfolio returns?

If so, then predicting future rates with a model unsuitable even for predicting rate volatility proper should be treated as a marketing ploy by experts from science. At one time, a dissertation stating that if Lenin had been shot long before 17, the revolution would have been made by Stalin was an example of a proper marketing move. Nowadays the thesis predicting the course is also a correct marketing move. And really, why stop at some useless, not making money right away, volatility of the portfolio? All it takes is a couple or three unacceptable assumptions and we are on target - we are looking into the future.

 
bstone писал(а) >>

Mm. Where is it good, if you ideally want a straight line at 45g?


P.S. I, for example, on the original chart already see the trade fatal error on the training period. What to say about the prediction then?

If we had the line at 45g we would live in the Canaries. But as it is, we have to choose something easier!

Seriously, the angle of slope will immediately indicate the ability of the algorithm to cover the existing spread on the symbol, and the width of the cloud will determine the minimum risk and hence the optimal MM of the TS.

 
m_a_sim писал(а) >>

tg=0.3945 angle 22 degrees

The product of the tangent of the slope by the instrument volatility value for the TF where the forecast has been made will give the average profitability of the algorithm. It must be compared with the brokerage companies' commission per transaction.

 
Neutron >> :

The product of the slope tangent by the instrument volatility value for the TF where the forecast was made will give the average profitability of the algorithm. It should be compared to the brokerage companies' commission per transaction.

How to determine the value of volatility of the instrument?

 
m_a_sim писал (а) >>

how do you determine the volatility of an instrument?

Calculate the standard deviation for a series of first differences.

 

The sum of the squares of the increments divided by the number of increments and all under the root.

 
built a new regression for gold of the form y=b0+b1*t+b2*Close(USD index). As you can see there is no gold price in the formula at all. The dollar index is taken, because no other instrument reacts to the dollar like gold does. I wrote the indicator Blue line - this regression. Here is the strategy, you can see with the naked eye how the price of gold "trails" the regression
 

For variance estimation I would certainly do the same (mean square of the increments).

But for volatility estimation I would probably sum the increment moduli and divide by their number (based on higher plausibility of exponential function pdf returns compared to normal one). But it's just trifles, a theoretical show-off...

There has long been a debate about the adequacy of the variance as a measure of dispersion. Markowitz seems to have equated it to volatility.

 
Mathemat писал(а) >>

For variance estimation I would certainly do the same (mean square of the increments).

But for volatility estimation I would probably sum the increment moduli and divide by their number (based on higher plausibility of exponential function pdf returns compared to normal one). But it's just trifles, a theoretical show-off...

It's been argued for a long time about the adequacy of the variance as a measure of dispersion. It was Markowitz, I think, who equated it with volatility.

I agree with every word!

m_a_sim wrote >>
built a new regression for gold of the form y=b0+b1*t+b2*Close(USD index). As you can see there is no gold price in the formula at all. The dollar index is taken, because no other instrument reacts to the dollar like gold does. I wrote the indicator Blue line - this regression. Here is the strategy, you can see with the naked eye how the price of gold "trails" the regression

So I don't understand, are we going to evaluate the profitability of the algorithm or look at the pictures?

 
Neutron >> :

>> I agree with every word!

can Close[i]-Open[i] be considered as an increment ?