It's all wrong, friends. - page 4

 
Neutron писал (а) >>

But I could not get anything out of it, for the simple reason that the DC, as soon as the indicator predicted a noticeable move, immediately quoted EUR/GBP! There was no time to open a position.

I have just reached this point in my post here.

 

I see, Neutron. Yes, instrument independence is not easy to ensure - on Foreh. Or at least uncorrelated.

About the first balance difference - here is a histogram of results of trades of Ch-07 winner in pips:

Well, there is a certain hint of normality, but the tails are not so thick. On the other hand, if the portfolio has a lot of uncorrelated instruments (say, more than ten), then individual distributions are not so important, even if they are fat-tailed. Distribution of sums tends to what it should be anyway, i.e. to a Gaussian curve.

P.S. Vita, what do you think the pdf of the first differences of a simple waving machine with a not too shallow period is? Gaussian!

 
Mathemat писал (а) >>

On the other hand, if the portfolio contains many uncorrelated instruments (say, more than a dozen), the individual distributions are not so important, even if thick-tailed. Distribution of the sum tends to what it should be anyway, i.e. to the Gaussian curve.

Where does it follow from? If by virtue of the limit theorem, then imho it does not - the independence of the instruments is under huge question. The dependence/independence of instruments is also the NE, and the correlation coefficient is an average estimate of it. Those instruments which were uncorrelated for a long time, may do it one day and not only in pairs. The portfolio therefore does not guarantee the normality of the returns. Proof of that can be seen in a series of bankruptcies of investment houses in the states. They took huge losses in the market, and they of course know/use Markowitz theory and its more advanced variants. There are simply no eternal solutions in the market.

 
Vita писал (а) >>

For me personally, it's like a model of a spherical horse in a vacuum - ideal conditions that are not met in reality. Can we have an example of independent tools?

What actually caused the scepticism?

Yes, the world is not perfect, but that does not prevent "ideal" mathematical models from describing it correctly! There are many methods that allow one to accurately approximate a model to a real object.

In my example, an unreal case of the TS built on instruments that do not correlate with each other is considered. This allows you to understand the logic of reasoning and see the basic principle. Nothing prevents us from introducing the matrix of correlation coefficients between the tools into this task and solving the particular task precisely...

 
Neutron писал(а) >>

What actually caused the scepticism?

Yes, the world is not perfect, but that does not prevent "perfect" mathematical models from describing it correctly! There are many methods that allow us to accurately approximate the model to the real object.

In my example, an unreal case of the TS built on instruments that do not correlate with each other is considered. This allows you to understand the logic of reasoning and see the basic principle. Nothing prevents us from introducing the matrix of correlation coefficients between the instruments into this task and solving the particular task precisely...

The scepticism here is "the instruments are independent and the first difference of the balance curve is normally distributed" - these conditions are not met, hence there is no way to apply conclusions based on the assumption that they are met. Models that describe a non-ideal world always indicate that if something is not "model-like", then the result is wrong. In our case the tools are dependent and the first difference is abnormal, it is not "model-like", hence the conclusions are not applicable.

There are a lot of tools, models, theories and sciences for each of us to "be tempted" to apply our bright knowledge to the market, to take a well-known and understandable tool and to decompose the market into molecules and atoms. For example, to take and apply parametric statistics to the market. In this case I see that all that is left is to make the market fit the parametric statistics so that conclusions based on our knowledge and application of parametric statistics become true. Otherwise it is only an illusion based on the significance of our knowledge of the instrument, unsupported by evidence that our tools are appropriate.

 

Slava, I'm not very good at limit theorems. But I heard somewhere that the last strongest versions of these theorems do not require independence, and the number of individual variables in the sum does not have to be so large to get Gaussianity.

Again, a practical argument: look at the first differences of a simple dash (with a period of, say, 13). They are quite normal - unlike the thick-tailed differences for bars.

 

Vita писал(а) >>

The scepticism here is "the instruments are independent and the first difference of the balance curve is normally distributed" - these conditions are not met,

I would probably agree with you if the world were binary - either yes or no. But fortunately it isn't, and the non-normality you speak of is weak. Its influence on the final result is weak and does not distort it much. The degree of this distortion and its sign is not difficult to estimate. The error of the estimate is not difficult to obtain... What more do you need?

To play along with you, you have to admit that Heisenberg uncertainty does not allow to say anything definitively about any phenomenon in this world at all. What, so now we don't use counting at all?

Absurd, isn't it? So why do you allow yourself to take this position in the discussion?

Why?

 
Mathemat писал(а) >>

Once again, a practical argument: look at the first differences of a simple dash (with a period of, say, 13). They are quite normal - unlike the thick-tailed differences for bars.

I want to say: let's say, and?
But the doubt is that under a thirteen times microscope we'll see something abnormal. Or would we not?

 
Neutron писал(а) >>

I would probably agree with you if the world were binary - either yes or no. But fortunately it isn't, and the negativity you speak of is weak. Its influence on the final result is weak and does not distort it much. The degree of this distortion and its sign is not difficult to estimate. The error of the estimate is not difficult to obtain... What more do you need?

To play along with you, you have to admit that Heisenberg uncertainty does not allow to say anything definitively about any phenomenon in this world at all. What, so now we don't use counting at all?

Absurd, isn't it? So why do you allow yourself to take this position in the discussion?

Why?

Because I believe that it is necessary to start from the properties of the market (non-normality of the distribution) and not from the properties of the theory (let's assume that the distribution is normal). Then the result will follow the market and not the theory. That's when you estimate the error, then you will see that the profit is not there. Or whatever it is you are after. As you can see, I am making myself quite clear and I don't need Heisenberg to help me. If you take the wrong premise, you get the wrong result. You can't be more definite than that. But "not much distortion", "not hard to estimate" and "non-Gaussianism is weak" are really vague, like "well, profit is close to hand". So there is no profit there. And it's quite a definite statement, I dare hope.

 

I've taken a closer look at Neutron's reasoning. In fact we are only operating with balance curves here - or am I wrong, Sergei? Well, balance curves are something which has, to put it mildly, other statistical characteristics than quote curves. Then why talk about bar statistics by referring to the non Gaussianity of bar returns?