Question

我正在尝试在R中实施DCC（动态条件相关）预测。但是，当我尝试一些测试值时，与Pearson相关性相比，预测似乎不合适，特别是在预测未来两年以上的相关性时。我对R很新，所以我真的非常感谢任何帮助！

我正在使用rmgarch包（在此处找到：https://cran.r-project.org/web/packages/rmgarch/index.html）。

这是我的代码：

library(quantmod)
library(rmgarch)
library(bizdays)
library(timeDate)

forecast <- function(ticker1, ticker2, start, end1, end2)
{
  startingYear = as.numeric(format(as.yearmon(end1), "%Y"))
  endingYear = as.numeric(format(as.yearmon(end2), "%Y"))
  hol = as.POSIXlt(holidayNYSE(startingYear))
  hol = as.character(hol)

  year = startingYear + 1
  while (year < endingYear)
  {
    temp = as.POSIXlt(holidayNYSE(year))
    temp = as.character(temp)
    hol = c(hol, temp)
    year = year + 1
  }

  #cal <- Calendar(holidays = hol, start.date = end1, end.date = end2, weekdays = c("saturday", "sunday"))
  #n.ahead = bizdays(end1, end2, cal)
  n.ahead = 252 * (endingYear - startingYear)

  dat1 = getSymbols(ticker1, src = 'yahoo', from = start, to = end1, auto.assign=F)
  dat2 = getSymbols(ticker2, src = 'yahoo', from = start, to = end1, auto.assign=F)
  dat1 = dat1[,4]
  dat2 = dat2[,4]
  eps <- data.frame(s1=dat1, s2=dat2)

  spec = ugarchspec()
  uspec = multispec(replicate(2, spec))
  spec = dccspec(uspec)
  fit = dccfit(spec, eps)
  result = dccforecast(fit, n.ahead)
  return(result)
}

con <- file("test.log")
sink(con, append = TRUE)
sink(con, append = TRUE, type = "message")
source("script.R", echo = TRUE, max.deparse.length = 10000)

startYear = 1998
endYear = 2014
trainingPeriod = 5
forecastPeriod = 5
for (i in startYear:endYear)
{
  result = forecast('^GSPC', '^RUT', paste(i-trainingPeriod-forecastPeriod, '-01-01', sep=""), paste(i-forecastPeriod, '-01-01', sep=""), paste(i, '-01-01', sep=""))
  print(result)
  result = forecast('^GSPC', '^RUT', paste(i-trainingPeriod-forecastPeriod, '-04-01', sep=""), paste(i-forecastPeriod, '-04-01', sep=""), paste(i, '-04-01', sep=""))
  print(result)
  result = forecast('^GSPC', '^RUT', paste(i-trainingPeriod-forecastPeriod, '-07-01', sep=""), paste(i-forecastPeriod, '-07-01', sep=""), paste(i, '-07-01', sep=""))
  print(result)
  result = forecast('^GSPC', '^RUT', paste(i-trainingPeriod-forecastPeriod, '-10-01', sep=""), paste(i-forecastPeriod, '-10-01', sep=""), paste(i, '-10-01', sep=""))
  print(result)
  print("====================================================================================")
}

sink()
sink(type="message")

有谁知道会出现什么问题？我意识到DCC对于更长的预测范围变得越来越不准确，但我不确定这是否会导致问题。非常感谢任何建议。

R（rmgarch）中的DCC预测实现似乎不准确？

0 个答案: