使用midasr进行预测时的误差(包括可重复的例子)

Question

代码是自包含的，除了下面链接的数据集之外。

代码中使用的.csv文件，请先下载此文件： https://drive.google.com/?authuser=0#folders/0B1ciW4R5hjUCRFpjQlJKZGFqcVU

library(midasr)
library(zoo)


 yvellaregdata <- read.table("~/Desktop/attempt1/ymonthlyjackson.csv", quote="\"")
 yvellareg <- ts(yvellaregdata, start=c(2008,7), frequency=12)

 xvellareginit <- read.table("~/Desktop/attempt1/xdailyjackson.csv", quote="\"")
 xvellaregzoo <- zoo(xvellareg)
 xvellareg <- as.numeric(xvellaregzoo) #i had to convert to numeric for it to work

#yvellareg is the monthly y variable
#xvellareg is the daily x variable
 betareg <- midas_r(yvellareg ~ mls(yvellareg, 1, 1) + mls(xvellareg, 3:25, 30), start=NULL)
 summary(betareg)


#Defining data for forecasting
 xdailyfulldataread <- read.table("~/Desktop/attempt1/xdailyfulldatajackson.csv", quote="\"")
 xdailyfulldata <- zoo(xdailyfulldataread)
 xdailyfulldata <- as.numeric(xdailyfulldata)

 ymonthlyfulldataread <- read.table("~/Desktop/attempt1/ymonthlyfulldatajackson.csv", quote="\"")
 ymonthlyfulldata <- ts(ymonthlyfulldataread,start=c(2008,7), frequency=12)


fulldata <- list(xx=xdailyfulldata,
                   yy=ymonthlyfulldata)
insample <- 1:length(yvellareg)
outsample <- (1:length(fulldata$yy))[-insample]

#errorhere
avgf<-average_forecast(list(betareg),
                       data=fulldata,
                       insample=insample,
                       outsample=outsample)
sqrt(avgf$accuracy$individual$MSE.out.of.sample)

Answer 1

由于您已经准备好了R之外的样本和全样本数据，所以没有必要将其转换为时间序列对象。

下面是代码的清理版本，它假定数据文件位于R工作目录中：

library(midasr)

yvellareg <- scan("ymonthlyjackson.csv")
xvellareg <- scan("xdailyjackson.csv")

#yvellareg is the monthly y variable
#xvellareg is the daily x variable
 betareg <- midas_r(yvellareg ~ mls(yvellareg, 1, 1) + mls(xvellareg, 3:25, 30), start=NULL)
 summary(betareg)


#Defining data for forecasting
xdailyfulldata <- scan("xdailyfulldatajackson.csv")
ymonthlyfulldata <- scan("ymonthlyfulldatajackson.csv")


fulldata <- list(xvellareg=xdailyfulldata,
                   yvellareg=ymonthlyfulldata)
insample <- 1:length(yvellareg)
outsample <- (1:length(fulldata$yvellareg))[-insample]

#errorhere
avgf<-average_forecast(list(betareg),
                       data=fulldata,
                       insample=insample,
                       outsample=outsample)
sqrt(avgf$accuracy$individual$MSE.out.of.sample)

但是这仍然会引发一个错误，因为您的数据是不一致的。包midasr期望每个低频周期有相同数量的高频周期。在你的情况下这是30岁。但我们有

> length(xdailyfulldata)
[1] 1230
> length(ymonthlyfulldata)
[1] 42
> 1230/42
[1] 29.28571

自42*30=1260以来，你似乎有更多的每月观察比每日观察。删除一个月观察会使代码运行时不会出现错误：

fulldata <- list(xvellareg=xdailyfulldata,
                   yvellareg=ymonthlyfulldata[-42])
insample <- 1:length(yvellareg)
outsample <- (1:length(fulldata$yvellareg))[-insample]

#errorhere
avgf<-average_forecast(list(betareg),
                       data=fulldata,
                       insample=insample,
                       outsample=outsample)
sqrt(avgf$accuracy$individual$MSE.out.of.sample)
[1] 1.118709