Machine learning in trading: theory, models, practice and algo-trading - page 291
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
Can you tell me how to make your code above work with this one? :)
There's something wrong with this command in the package--
indexTZ(EURUSD) <- "UTC",
It sets it to the wrong time zone. I wrote in the comments of the file how to check it.
The final chart is the same. I've never tried it before.
There is something wrong with this command in the package -
indexTZ(EURUSD) <- "UTC",
It sets it to the wrong time zone. I wrote in the comments of the file how to check.
Yes, there is something wrong, I have constant warping :( but it seems to work as it should...
Maybe the problem with "UTC" is that rusquant streams the quotes go "as is", the last bar is not closed, maybe I have problems with "UTC +1" because of it, but I haven't tried to crash...
The final chart remains the same. My intuition says that I can probably add indicators to chart_Series(EURUSD), but I've never tried it.
Of course you can)) rusquant is quanmod but with the possibility to load "our" quotes, the functions are the same, and the package is essentially the same....
But to be honest I myself do not know how to draw everything there, so far all my efforts go to "what to visualize" and not to "how to visualize it nicely".
p.s. Thanks for your help, I will keep you posted on the results of research
Well, for everyone in the world then it is clear that the rate could not stay at 35 rubles because oil prices fell and the inflow of currency (reduced foreign exchange supply) at the MICEX fell and as a result of the well-known events increased capital outflow from Russia (increased foreign exchange demand), but you can continue to think as before!
I'm trying to protect myself from the total propaganda, which about the ruble exchange rate has nothing to do with reality at all. All of the arguments you cited relate almost 100% to propaganda and were used to make global changes to the RF economy. This was done to cover up the extremely negative social consequences of the weakening of the ruble for the population.
In order.
1. The dependence of the ruble exchange rate on oil prices is greatly exaggerated. Russia ranks in either the first or the second top ten among oil-producing countries in terms of dependence of the budget on the price of energy resources. But Russia is the only country among the major exporters of energy resources where the national currency exchange rate has halved.
2. After the imposition of financial sanctions in August 2014, verbal interventions by top Russian officials were made to weaken the ruble exchange rate, including the announcement of the Central Bank's refusal to support the exchange rate. Naturally, such conditions were not without speculators, who took the words of top government officials as a leading indicator.
3. back in 2011 Glazyev advocated the weakening of the ruble.
As a result, today we have a different macroeconomic situation in Russia: the emphasis on exports and oppressed imports. In the country itself there were structural changes in prices: for example, the currency is half the price of gasoline, half the price of real estate...
The exchange rate of the national currency is always politics. And how to present the changes in the exchange rate, painful for the population, is another matter. In Russia they did as they did, on the sly...
Yeah...
There is a category of people who find it easier to believe in reptiloids, Freemasons, behind-the-scenes government, the Loch Ness monster, the planet Nibiru, anunnaki - it's easier than reading a textbook on economics or the theory of evolution.
Hello, it's lively, it's spring!)
Help with a question, please!
I have a different number of events (bars) at any given time (on each bar)
Each event contains (for example) 4 parameters.
The task is to build the logic of inputs, to train the network.
Possible solutions:
At one moment one number of events, at another moment another number of events.
- Consequently, to enter the network it is necessary to use already ready number of allocated inputs, if the input is not involved, then put zeros.
The problem is in redundancy of inputs, since each event has its own period, and there may be thousands of periods. Consequently, allocating space for each period, the input data can be in millions (if of course the event will have thousands of parameters).
You can also use the allocated number of inputs relative to the active events, there can be (for example) 5.
Then the problem is that weights will be adjusted for a certain event, and if the network iterated on an event with period 10, then the next moment (on the next bar) the cell will contain an event with period 60. And the difference between the events is in the frequency at least. If I understand correctly, then if the input is set to one period, it should be set to that period. Or the main thing is that the data type corresponds, if the period, then it does not matter what period will be fed.
As an alternative solution:
If we use a network with long short term memory (LSTM)
And allocate four inputs and loop the number of events active at the moment (current bar), it is possible to bypass a large topology without costing, isn't it?
Hello, it's lively, it's spring!)
Help with a question, please!
I have a different number of events (bars) at any given time (on each bar)
Each event contains (for example) 4 parameters.
The task is to build the logic of inputs, to train the network.
Possible solutions:
At one moment one number of events, at another moment another number of events.
- Consequently, to enter the network it is necessary to use already ready number of allocated inputs, if the input is not involved, then put zeros.
The problem is in redundancy of inputs, since each event has its own period, and there may be thousands of periods. Consequently, allocating space for each period, the input data can be in millions (if of course the event will have thousands of parameters).
You can also use the allocated number of inputs relative to the active events, there can be (for example) 5.
Then the problem is that weights will be adjusted to a certain event, and if the network iterated on an event with period 10, then the next moment (on the next bar) in the cell will be an event with period 60. And the difference between the events is in the frequency at least. If I understand correctly, then if the input is set to one period, it should be set to that period. Or the main thing is that the data type corresponds, if the period, then it does not matter what period will be fed.
As an alternative solution:
If we use a network with long short term memory (LSTM)
And allocate four inputs and loop the number of events active at the moment (current bar), it is possible to bypass a large topology without costing, isn't it?
There is such a notion as data normalization. But I think you are just making nonsense. Seriously......!!!!!
Normalization is something I know.
Maybe, but until I feel it, I will do what I understand!
Why do you think so? Is the approach itself not correct? Or are you confused by the fact that I want to get, in principle, points of anticipated reversals?I made a mistake with the birds and got amazing graphs, the meaning of which I can not interpret.
These are histograms. They are obtained as follows.
CHFJPY gain is taken and converted to -1 and +1. It is shifted one bar to the left.
Then we take a vector of the price of a specific currency pair and divide it into two parts: one vector is related to +1, and the second one to -1
Then we plot the histograms.
I was struck by the view. Earlier it seemed to me that there should be variations of the normal or near-normal law.
See
What's not to see!
Does anyone have an interpretation of this?
Or a practical conclusion:
Maybe someone has an interpretation of this?
Because the histograms +1 and -1 coincide 100%, this operation is useless. The result will be the same as if you draw just a distribution of prices, without targets at all.
The result will be similar to the support and resistance charts. For example, the AUDUSD price in that timeframe was most skewed around 0.76 and the EURCAD was most skewed around 1.46.
I was amazed at the view. I used to think there should be variations of normal or near-normal law.