Question

我的干预分析遇到一些问题。我正在建模当英伟达（NVDA）于2016年4月宣布深度学习超级计算机时对收盘价的影响。我添加了一个虚拟变量，该变量在干预日期之前为0，之后为1，并将其添加为回归变量在我的模型中。我的想法是，我无法理解输出，因为我从干预系数中从未得到任何重要信息。我不知道我做错了什么。这是我的代码：

rm(list=ls())
library('ggplot2'); library('forecast'); library('tseries'); library('xts'); library(quantmod) ; library(lmtest)

start <- as.Date("2005-01-01")
end <- as.Date("2018-10-01")

getSymbols("NVDA", src = "yahoo", from = start, to = end)
plot(NVDA[, "NVDA.Close"], main = "NVIDIA")


# Adding dummy intervention variable

intDate<-as.Date("2016-04-05") #  sets the intervention Date  
closing = NVDA[, "NVDA.Close"]
closing$Intervention = 0
for (i in 1:nrow(closing)){
  if (index(closing[i,1]) < intDate){
    closing[i,"Intervention"] = 0
  }
  if(index(closing[i,1]) >= intDate){
    closing[i,"Intervention"] = 1
  }
}

model<-auto.arima(closing[,"NVDA.Close"], xreg = closing[,"Intervention"])

model
coeftest(model)

dates = as.Date(index(closing),"YYYY-MM-DD")
fittedVal = xts(fitted.values(model), dates)

plot(NVDA[,"NVDA.Close"], col = "blue", type = "l")
lines(fittedVal, col = "red", type = "l")

Answer 1

我花了几个小时弄乱了模型参数。我没有使用“纯跳跃”干预变量（0和1），而是尝试了200个周期内逐渐增加的效果。这有助于捕获干预效果并使所有系数显着，但是由于模型的残差不是正态分布的，因此我无法真正使用结果。这是我使用的代码：

rm(list=ls())
library('ggplot2'); library('forecast'); library('tseries'); library('xts'); library(quantmod) ; library(lmtest)

start <- as.Date("2005-01-01")
end <- as.Date("2018-10-01")

getSymbols("NVDA", src = "yahoo", from = start, to = end)
plot(NVDA[, "NVDA.Close"], main = "NVIDIA")


# Adding dummy intervention variable

intDate<-as.Date("2016-04-05") 
closing = NVDA[, "NVDA.Close"]
periods = 200
count = 0
closing$Intervention = 0
for (i in 1:nrow(closing)){
  if (index(closing[i,1]) < intDate){
    closing[i,"Intervention"] = 0
  }
  if(index(closing[i,1]) >= intDate){
    count = count + 1
    if(count<=periods){
      closing[i,"Intervention"] = count*0.025
    }
    if(count>periods){
      closing[i,"Intervention"] = periods*0.025
    }

  }
}

model<-auto.arima(closing[,"NVDA.Close"], xreg = closing[,"Intervention"])

model
coeftest(model)

dates = as.Date(index(closing),"YYYY-MM-DD")
fittedVal = xts(fitted.values(model), dates)

plot(NVDA[,"NVDA.Close"], col = "blue", type = "l")
lines(fittedVal, col = "red", type = "l")

使用R中的库存数据的ARIMA模型进行干预分析

1 个答案: