From theory to practice - page 5

 
Yury Kirillov:

A couple of pips on ticks is millions on the day. This is what attracts.


Well, the reality is that such short-term signals almost nowhere can be traded without infrastructure, and most likely with it too

It means you need to understand what you can and cannot trade

 
СанСаныч Фоменко:

I would still like to know your opinion on the quote above:

Stationarity and non-stationarity of the market is the basis for all subsequent reasoning.

Or are you proving that all your reasoning applies to a non-stationary market and then it has value, otherwise it is just another bicycle of someone who understands nothing about financial markets at all.


The price movement itself is certainly non-stationary. And, in the future, you will see that I will use NON-PARAMETRICAL statistics to describe it. At this point I have made the assumption that it isthe process of price increments that is stationary and the probability distribution does not change over time for a particular currency pair.

If this is not the case, the only thing would be to abandon the WMA as a measure of central tendency - I agree, that would be a definite mistake.

Can anyone help me - take for any pair tick quotes of the same volume over different years and compare the statistics of the distributions??? Time is catastrophically short...

 
Alexander_K:

The price movement itself is certainly non-stationary. And, in the future, you will see that I will use NON-PARAMETRICAL statistics to describe it. At this point I have made the assumption that it isthe process of price increments that is stationary and the probability distribution does not change over time for a particular currency pair.

If this is not the case, the only thing would be to abandon the WMA as a measure of central tendency - I agree, that would be a definite mistake.

Can anyone help me - take for any pair tick quotes of the same volume over different years and compare the statistics of the distributions??? Time is catastrophically short...

The assumption of stationarity - this leads to the most widespread mistake in science: solving the right problem with the right methods NOT the right problem. In our case, you are creating a model of a process that does not exist in the financial markets. Understand finally: there is no stationarity, not even in a flat market.


At the moment there are two paths:

1. We recognise that "History repeats itself" and use TA or more advanced - machine learning in the classification part.

2. We recognize that the market is NOT stationary, begin to penetrate into the statistical meaning of "non-stationary" and step by step begin to simulate this non-stationarity in the hope that eventually we will get a residual from the model, which will be approximately(??) stationary. At the same time, this residual will be so small that you can't expect any trickery from it. A lot of work has been done along the way (GARCH). These models also have distributions like you, but there all the problems are acknowledged and there are special tests for stability of the parameters of the whole model.

 
СанСаныч Фоменко:

The assumption of stationarity leads to the most widespread mistake in science: solving with the right methods NOT the right problem. In our case, you are creating a model of a process which does not exist in the financial markets. Understand finally: there is no stationarity, not even in a flat market.


At the moment there are two paths:

1. We recognise that "History repeats itself" and use TA or more advanced - machine learning in the classification part.

2. We recognize that the market is NOT stationary, begin to penetrate into the statistical meaning of "non-stationary" and step by step begin to simulate this non-stationarity in the hope that eventually we will get a residual from the model, which will be approximately(??) stationary. At the same time, this residual will be so small that you can't expect any trickery from it. A lot of work has been done along the way (GARCH). These models also have distributions like yours, but there acknowledges all the problems and there are special tests for the stability of the parameters of the whole model.

Hmmm... So you are sure that the return process is just as unsteady? It unambiguously at least has both mode and median and mean = 0. The only question is whether the variance changes with time.

About GARCH - I don't deny that it's not a bad model but the results are not very good... Otherwise, I wouldn't bother with scientific research - I don't have time for anything. I hope you don't think I have nothing better to do.

 
Yury Kirillov:

A couple of pips on ticks is millions on the day. That's the attraction.

Who's going to give them?

The shop will close in a day at the most.

 

Has anyone figured out what period of WMA we're taking? Did I miss something...

 
Dennis Kirichenko:

Has anyone figured out what period of WMA we're taking? Did I miss something...

Denis, be a friend - check the tick data of some pair in Excel for example for November 2017 and November 2016 - namely returns increments. Do they have different statistics? We will draw conclusions from this important experiment.
 
Alexander_K:

Greetings Vladimir, I am already looking forward to your comments - you can feel the approach of a real engineer.

Yes, I must admit that I am not entirely sure about the specific analytical formula for probability density function, because it is given for continuous distributions, while we have a discrete one. I do use tabular data in my calculations. But it takes a lot of time to compile tables - that's what I'm doing now. My task was set by myself - TS must deal with all currency pairs at a time. My broker has 36 of them. At the moment I've processed data for 18 pairs only. I think I will manage it till New Year.

1. If analysis is so slow for different pairs, maybe you should not do it. All the more, the question of the number of degrees of freedom is very important for you. Actually, if you take 28 pairs formed by the most traded 8 currencies USD EUR GBP CHF JPY AUD NZD CAD, then this system of 28 quotes has only 7 degrees of freedom, always EURGBP = EURJPY/GBPJPY. If you take the average exchange rate, which is equal to the middle between Bid and Ask.

Move on to currency strength analysis - it will be faster, and probably more interesting.

2. what is returns? Everybody here seems to know, but I absolutely do not understand. What is the beast?

 
Vladimir:

1. If it's so slow to analyse the different pairs, maybe you shouldn't do it. Especially since the question of the number of degrees of freedom is very important to you. Actually, if we take 28 pairs formed by the most traded 8 currencies USD EUR GBP CHF JPY AUD NZD CAD, then this system of 28 rates has only 7 degrees of freedom, always EURGBP = EURJPY/GBPJPY. If you take the average exchange rate, which is equal to the middle between Bid and Ask.

Going to a strength analysis of currencies is faster, and probably more interesting.

2. What is returns? Everyone here seems to know it, but I don't understand it at all. What kind of beast is it?

This is the difference between the current and previous price values.

Vladimir, how fair is it to work with the average between Bid and Ask? I checked the spread histogram - it's very far from bell-shaped - it looks like a Humbel distribution. It seems not quite correct to work with such an average?

 
Vladimir:

1. If it's so slow to analyse the different pairs, maybe you shouldn't do it. Especially since the question of the number of degrees of freedom is very important to you. Actually, if we take 28 pairs formed by the most traded 8 currencies USD EUR GBP CHF JPY AUD NZD CAD, then this system of 28 rates has only 7 degrees of freedom, always EURGBP = EURJPY/GBPJPY. If you take the average exchange rate, which is equal to the middle between Bid and Ask.

Move on to currency strength analysis - it will be faster, and probably more interesting.

2. what is returns? Everybody here seems to know, but I absolutely do not understand. What is this beast?

Am I right to understand that, logically, it is important to take the geometric mean (the root of the product ofBid and Ask), or can I not bother with such subtleties?