Machine learning in trading: theory, models, practice and algo-trading - page 1508

 
mytarmailS:

How is advisor Ilya?

I am disappointed in the grail. There is a strong deviation from my theoretical calculations and assumptions. I was inspired by Maxim's article of the same name to study the phenomenon of so-called long memory orlong-range dependence .I wrote an indicator based on ARFIMA. According to the authors it has a better predictive ability thanARIMA.

Autoregressive fractionally integrated moving average - Wikipedia
Autoregressive fractionally integrated moving average - Wikipedia
  • en.wikipedia.org
In statistics, autoregressive fractionally integrated moving average models are time series models that generalize ARIMA (autoregressive integrated moving average) models by allowing non-integer values of the differencing parameter. These models are useful in modeling time series with long memory—that is, in which deviations from the long-run...
Files:
R_ARFIMA.zip  20 kb
 
Ilya Antipin:

Disappointed. There is a strong deviation from my theoretical calculations and assumptions. Maxim's article of the same name has inspired me to take up the phenomenon of so-called long-range memory orlong-range dependence...

There is no such thing, you don't have to worry about it.
 
Ilya Antipin:

I am disappointed with the grail. There is a strong deviation from my theoretical calculations and assumptions. Maxim's article of the same name inspired me to study the phenomenon of so-called long memory or long-range dependence .I wrote an indicator based on ARFIMA. It has a better forecasting capability than ARIMA, according to the authors.

LRD floats in waves, it makes sense to filter out areas with high en, then the predictions make sense, for a small perspective. And pick a window.

Are the chains bad for clustering? I've almost got it figured out, I'll rewrite them in mql for fun, let them be in my arsenal
 
Ilya Antipin:

I am disappointed in the grail.

So the indicator that I threw, have you tried it with chains? Or only with prices?

 

Impressed, pampered psychologists better not watch


 
Maxim Dmitrievsky:

Impressed, pampered psychologists better not watch


How about that! Drobyshevsky from systematization of the skulls of homenids, moved to philosophizing on the subject of the contents of these (skulls, in fact). Unexpectedly! I watched it with pleasure, thanks.

Pyssy: It seems that he does not rest on the laurels of Anokhin and Chernigovskaya.
 
sibirqk:

How about that! Drobyshevsky from systematization of the skulls of homenids, moved to philosophizing on the subject of the contents of these (skulls in this case). Unexpected! I watched it with pleasure, thanks.

Pyssy: It seems that he does not rest on the laurels of Anokhin and Chernigovskaya.

I can't take seriously people who when they don't understand something, they invariably come to God, or they so grovel before the feelings of believers. Anokhin seems to be not a stupid guy, but when he begins to think about high ... I'll say nothing about the second one :))

 
Maxim Dmitrievsky:

Impressed pampered psychologists better not watch


Well, I'm surprised, I didn't know that humanoid ancient apes/humans are now considered ancestors of modern apes (58:18).

From the material, it feels like a fit of history :) Unfortunately, the reasoning is weak and contradictory, but as a point of view, let it be.

In my opinion, such researches should be supported by the analysis of changes in the DNA of subspecies, with the allocation of new mutations that served the development of the species, and then this would be more interesting for me.

And as for "conditioned reflexes", on the basis of which a person often unthinkingly makes decisions in his everyday life, including on important issues, Robert Cialdini's book called "Psychology of Influence" is interesting. The topic there is also not fully disclosed, but the collected techniques used by knowledgeable and interested people for their own purposes are interesting.

 
Well, I told you so :( psychologists demand DNA extraction from fossilized remains, as ancient as dinosaur's d.
 
Maxim Dmitrievsky:
What about the circuits, did you run out of performance?

Everything is ambiguous with chains. I took a EURUSD/M20 logreturn, "guessed" it with the help of "lambertW" ( see pictures here). I trained the model with two states (the better one, the author says 5). Here is the code and pictures.

require(LambertW)
set.seed(12358)
y1 <- diff(log(pr$close), 3)*100
out <- Gaussianize(y1, return.tau.mat = TRUE)
x1 <- get_input(y1, c(out$tau.mat[, 1]))  #  same as out$input
#---
 States <- 2 L
#set.seed(12358)
 param0 <- matrix(c(0.1, 0.05, -1, -0.1, -0.05, 1), States, 3 L, byrow = TRUE)
gamma0 <- ldhmm.gamma_init(m = States)
h <- ldhmm(m = States, param = param0, gamma = gamma0, stationary = TRUE)

mod <- ldhmm.mle(h, x1)
dc <- ldhmm.decoding(mod, x1)
post <- t(as.matrix(dc@states.prob))
#predStates <- cbind(post[,1:States])
prStat <- apply(post, 1, function(x) which.max(x))
Stat <- dc@states.global

mod@param
mod@delta
floor(mod@gamma*10000)/100
ldhmm.ld_stats(mod)


> mod@param
               mu      sigma    lambda
[1,] -0.001182310 0.09119015 0.4625385
[2,]  0.001808756 0.04114724 0.6977260
> mod@delta
[1] 0.4189323 0.5810677
> floor(mod@gamma*10000)/100
      [,1]  [,2]
[1,] 93.79  6.20
[2,]  4.47 95.52
> ldhmm.ld_stats(mod)
             mean         sd kurtosis
[1,] -0.001182310 0.05256228 2.142325
[2,]  0.001808756 0.02542165 2.465665

Let's plot the signal and the quotes

par(mfrow = c(2,1))
matplot(tail(post, 300), t = "l", col = c(1,2,4,5))
abline(h = 0.5, col = 2)
plot(tail(pr$close, 300), t = "l")
par(mfrow = c(1,1))

signal