使用滚动窗口创建预测

Question

我正在尝试使用滚动窗口预测来计算已实现的波动率预测。我的目标是使用前 500 个观测值来预测第 501 个观测值，然后使用观测值 2 到 501 将窗口向前移动一步来预测第 502 个观测值。我希望重复此操作，直到获得完整的伪样本外预测。

我的 df 包含 23 个变量的 2411 个观测值，这是一个小例子：

Date	EUR/GBP_RV t+1	EUR/GBP_RV	EUR/GBP_RV_5	EUR/GBP_RV_22
2012-02-13	0.004019947	0.004467109	0.004995980	0.005150602
2012-02-14	0.004077322	0.004019947	0.004704742	0.005099209
2012-02-15	0.004666244	0.004077322	0.004564522	0.005020790
2012-02-16	0.004582351	0.004666244	0.004532930	0.005011078
2012-02-17	0.003951071	0.004582351	0.004362595	0.004999634
2012-02-21	0.005549329	0.004824547	0.004467109	0.004954358

df 的头部可以用

structure(list(Date = structure(c(15383, 15384, 15385, 15386, 
15387, 15391), class = "Date"), `EUR/GBP_RV t+1,t+22` = c(0.00437808494654599, 
0.00439764887779732, 0.00441997330738947, 0.00437799055732722, 
0.00435021150904479, 0.00434881306458859), `EUR/GBP_RGV t+1,t+22` = c(0.00310674728131117, 
0.00311502965608506, 0.00313756204195853, 0.00312923997431178, 
0.00310846773074243, 0.00309755540903969), `EUR/GBP_RBV t+1,t+22` = c(0.00304961980133602, 
0.00306866297401415, 0.00307794319983677, 0.00302801249035602, 
0.00300946121851254, 0.00302272755554915), `EUR/GBP_RV t+1,t+5` = c(0.00425938711995146, 
0.00442030721926592, 0.00471470868108043, 0.00487869292442023, 
0.00476849421943949, 0.0046067439660438), `EUR/GBP_RGV t+1,t+5` = c(0.00287718093809447, 
0.00311308998902836, 0.00344398536439729, 0.00376911066284235, 
0.00377425759242542, 0.00353909902900148), `EUR/GBP_RBV t+1,t+5` = c(0.00313771439501486, 
0.00310626895475871, 0.0031789431316576, 0.00305962424370556, 
0.00288671182187571, 0.00291630955829613), `EUR/GBP_RV t+1` = c(0.00401994700051122, 
0.00407732180013827, 0.00466624428620966, 0.00458235106134663, 
0.00395107145155152, 0.00554932910921084), `EUR/GBP_RGV t+1` = c(0.00280067842079775, 
0.00275522601341373, 0.00296142025653418, 0.00320716195737542, 
0.00266141804235125, 0.00440970289025837), `EUR/GBP_RBV t+1` = c(0.00288377777753365, 
0.00300554199386584, 0.00360608177982326, 0.00327292734850446, 
0.0029202430753471, 0.00336891287836027), `EUR/GBP_RV` = c(0.00446710941914797, 
0.00401994700051122, 0.00407732180013827, 0.00466624428620966, 
0.00458235106134663, 0.00482454749708351), `EUR/GBP_RV_5` = c(0.00499598016427624, 
0.00470474196716823, 0.00456452150727461, 0.00453292965488229, 
0.00436259471347075, 0.00442030721926592), `EUR/GBP_RV_22` = c(0.00515060219897263, 
0.00509920878086075, 0.00502079002092667, 0.00501107755050095, 
0.00499963428727425, 0.00495435807685843), `EUR/GBP_RBV` = c(0.00324667045338159, 
0.00288377777753365, 0.00300554199386584, 0.00360608177982326, 
0.00327292734850446, 0.00272655057625291), `EUR/GBP_RBV_5` = c(0.0034985439252634, 
0.00318731350301598, 0.00319529774619179, 0.0032799330165155, 
0.00320299987062176, 0.00310626895475871), `EUR/GBP_RBV_22` = c(0.0036581887941976, 
0.00362298829344015, 0.00355498498935768, 0.0035465419934449, 
0.00355752455210586, 0.00351669973034298), `EUR/GBP_RGV` = c(0.00306825643155514, 
0.00280067842079775, 0.00275522601341373, 0.00296142025653418, 
0.00320716195737542, 0.00398022367546721), `EUR/GBP_RGV_5` = c(0.00353290218424469, 
0.00345194162523714, 0.00325362896209417, 0.00312087362386985, 
0.00295854861593525, 0.00311308998902836), `EUR/GBP_RGV_22` = c(0.00359974659692405, 
0.00356342531619103, 0.00352224165959556, 0.00351722666683608, 
0.00349184335907274, 0.00346394410318992), `EUR/GBP_TotalCount` = structure(c(-0.179518680545716, 
-0.0757259172924681, 0.333514120677483, 0.759064450015801, -0.419724789788948, 
-0.360414639358521), .Dim = c(6L, 1L)), `EUR/GBP_BloombergCount` = structure(c(-0.701694550941621, 
-0.669811381311996, -0.574161872423118, -0.644304845608295, -0.784590791978649, 
-0.631551577756444), .Dim = c(6L, 1L)), `EUR/GBP_SocialMediaCount` = structure(c(-1.68861759789696, 
-1.66829994393287, -1.5910928588693, -1.59515638966212, -1.6967446594826, 
-1.63172816679749), .Dim = c(6L, 1L)), `EUR/GBP_NewsCount` = structure(c(0.76560533722738, 
0.879161625683434, 1.33717198912285, 1.90306082659552, 0.487392430510049, 
0.487392430510049), .Dim = c(6L, 1L))), na.action = structure(c(`6` = 6L, 
`38` = 38L, `74` = 74L, `101` = 101L, `140` = 140L, `165` = 165L, 
`189` = 189L, `197` = 197L, `236` = 236L, `256` = 256L, `323` = 323L, 
`351` = 351L, `393` = 393L, `423` = 423L, `442` = 442L, `455` = 455L, 
`490` = 490L, `510` = 510L, `551` = 551L, `576` = 576L, `605` = 605L, 
`646` = 646L, `676` = 676L, `696` = 696L, `708` = 708L, `742` = 742L, 
`762` = 762L, `793` = 793L, `829` = 829L, `858` = 858L, `904` = 904L, 
`929` = 929L, `950` = 950L, `961` = 961L, `996` = 996L, `1016` = 1016L, 
`1089` = 1089L, `1114` = 1114L, `1159` = 1159L, `1184` = 1184L, 
`1207` = 1207L, `1216` = 1216L, `1238` = 1238L, `1243` = 1243L, 
`1253` = 1253L, `1278` = 1278L, `1315` = 1315L, `1346` = 1346L, 
`1372` = 1372L, `1416` = 1416L, `1441` = 1441L, `1473` = 1473L, 
`1532` = 1532L, `1599` = 1599L, `1626` = 1626L, `1669` = 1669L, 
`1694` = 1694L, `1718` = 1718L, `1726` = 1726L, `1766` = 1766L, 
`1786` = 1786L, `1827` = 1827L, `1853` = 1853L, `1881` = 1881L, 
`1922` = 1922L, `1952` = 1952L, `1971` = 1971L, `1984` = 1984L, 
`2019` = 2019L, `2039` = 2039L, `2075` = 2075L, `2106` = 2106L, 
`2135` = 2135L, `2181` = 2181L, `2206` = 2206L, `2227` = 2227L, 
`2238` = 2238L, `2273` = 2273L, `2293` = 2293L, `2325` = 2325L, 
`2366` = 2366L, `2391` = 2391L, `2436` = 2436L, `2461` = 2461L, 
`2483` = 2483L, `2493` = 2493L), class = "omit"), row.names = c(1L, 
2L, 3L, 4L, 5L, 7L), class = "data.frame")

我已经为每个预测范围（t+1、t+5 和 t+22）构建了 9 个回归模型来重现，并且我正在使用预测函数来我的预测。我已将其编码在 R 循环中，其中我还尝试将每个预测分配给一个矩阵。我的代码如下：

last_window = nrow(EURGBP_final) - window_size
number_of_forecasts = dim( EURGBP_final)[1] - window_size

EURGBP_forecast.1 <- matrix( NA, nrow=number_of_forecasts, ncol=9 )
EURGBP_forecast.5 <- matrix( NA, nrow=number_of_forecasts, ncol=9 )
EURGBP_forecast.22 <- matrix( NA, nrow=number_of_forecasts, ncol=9 )

EURGBP_actuals.1 <- matrix( NA, nrow=number_of_forecasts, ncol=9 )
EURGBP_actuals.5 <- matrix( NA, nrow=number_of_forecasts, ncol=9 )
EURGBP_actuals.22 <- matrix( NA, nrow=number_of_forecasts, ncol=9 )

forecasts = for (i in (1:last_window)){
  ## data splitting ###
  data.estimate <- EURGBP_final[i:(w_size + i - 1), ] # = change to X[1:(w_size + i - 1), ] for expanding window
  data.forecast <- EURGBP_final[w_size + i, ]
  
  # --------------------------------------
  # Forecast horizon 1
  # --------------------------------------
  
  EURGBP_HAR_RV_train <- lm(data.estimate$`EUR/GBP_RV t+1`~data.estimate$`EUR/GBP_RV`+data.estimate$`EUR/GBP_RV_5`+data.estimate$`EUR/GBP_RV_22`)
  EURGBP_HAR_RV_NT_train <- lm(data.estimate$`EUR/GBP_RV t+1`~data.estimate$`EUR/GBP_RV`+data.estimate$`EUR/GBP_RV_5`+data.estimate$`EUR/GBP_RV_22`+data.estimate$`EUR/GBP_TotalCount`)
  EURGBP_HAR_RVB_train <- lm(data.estimate$`EUR/GBP_RBV t+1`~data.estimate$`EUR/GBP_RBV`+data.estimate$`EUR/GBP_RBV_5`+data.estimate$`EUR/GBP_RBV_22`)
  EURGBP_HAR_RVB_NT_train <- lm(data.estimate$`EUR/GBP_RBV t+1`~data.estimate$`EUR/GBP_RBV`+data.estimate$`EUR/GBP_RBV_5`+data.estimate$`EUR/GBP_RBV_22`+data.estimate$`EUR/GBP_TotalCount`)
  EURGBP_HAR_RVG_train <- lm(data.estimate$`EUR/GBP_RGV t+1`~data.estimate$`EUR/GBP_RGV`+data.estimate$`EUR/GBP_RGV_5`+data.estimate$`EUR/GBP_RGV_22`)
  EURGBP_HAR_RVG_NT_train <- lm(data.estimate$`EUR/GBP_RGV t+1`~data.estimate$`EUR/GBP_RGV`+data.estimate$`EUR/GBP_RGV_5`+data.estimate$`EUR/GBP_RGV_22`+data.estimate$`EUR/GBP_TotalCount`)
  EURGBP_HAR_RV_Bloomberg_train <- lm(data.estimate$`EUR/GBP_RV t+1`~data.estimate$`EUR/GBP_RV`+data.estimate$`EUR/GBP_RV_5`+data.estimate$`EUR/GBP_RV_22`+data.estimate$`EUR/GBP_BloombergCount`)
  EURGBP_HAR_RV_SocialMedia_train <- lm(data.estimate$`EUR/GBP_RV t+1`~data.estimate$`EUR/GBP_RV`+data.estimate$`EUR/GBP_RV_5`+data.estimate$`EUR/GBP_RV_22`+data.estimate$`EUR/GBP_SocialMediaCount`)
  EURGBP_HAR_RV_News_train <- lm(data.estimate$`EUR/GBP_RV t+1`~data.estimate$`EUR/GBP_RV`+data.estimate$`EUR/GBP_RV_5`+data.estimate$`EUR/GBP_RV_22`+data.estimate$`EUR/GBP_NewsCount`)
  
  EURGBP_forecast.1[, 1 ] <- predict( EURGBP_HAR_RV_train, newdata=data.forecast )          
  EURGBP_forecast.1[, 2 ] <- predict( EURGBP_HAR_RV_NT_train, newdata=data.forecast )           
  EURGBP_forecast.1[, 3 ] <- predict( EURGBP_HAR_RVB_train, newdata=data.forecast )         
  EURGBP_forecast.1[, 4 ] <- predict( EURGBP_HAR_RVB_NT_train, newdata=data.forecast )          
  EURGBP_forecast.1[, 5 ] <- predict( EURGBP_HAR_RVG_train, newdata=data.forecast )         
  EURGBP_forecast.1[, 6 ] <- predict( EURGBP_HAR_RVG_NT_train, newdata=data.forecast )          
  EURGBP_forecast.1[, 7 ] <- predict( EURGBP_HAR_RV_Bloomberg_train, newdata=data.forecast )
  EURGBP_forecast.1[, 8 ] <- predict( EURGBP_HAR_RV_SocialMedia_train, newdata=data.forecast )
  EURGBP_forecast.1[, 9 ] <- predict( EURGBP_HAR_RV_News_train, newdata=data.forecast )
  
  EURGBP_actuals.1[ , 1] <- data.forecast$`EUR/GBP_RV t+1`
  
  # --------------------------------------
  # Forecast horizon 5
  # --------------------------------------
  
  EURGBP_HAR_RVt5_train <- lm(data.estimate$`EUR/GBP_RV t+1,t+5`~data.estimate$`EUR/GBP_RV`+data.estimate$`EUR/GBP_RV_5`+data.estimate$`EUR/GBP_RV_22`)
  EURGBP_HAR_RVt5_NT_train <- lm(data.estimate$`EUR/GBP_RV t+1,t+5`~data.estimate$`EUR/GBP_RV`+data.estimate$`EUR/GBP_RV_5`+data.estimate$`EUR/GBP_RV_22`+data.estimate$`EUR/GBP_TotalCount`)
  EURGBP_HAR_RVBt5_train <- lm(data.estimate$`EUR/GBP_RBV t+1,t+5`~data.estimate$`EUR/GBP_RBV`+data.estimate$`EUR/GBP_RBV_5`+data.estimate$`EUR/GBP_RBV_22`)
  EURGBP_HAR_RVBt5_NT_train <- lm(data.estimate$`EUR/GBP_RBV t+1,t+5`~data.estimate$`EUR/GBP_RBV`+data.estimate$`EUR/GBP_RBV_5`+data.estimate$`EUR/GBP_RBV_22`+data.estimate$`EUR/GBP_TotalCount`)
  EURGBP_HAR_RVGt5_train <- lm(data.estimate$`EUR/GBP_RGV t+1,t+5`~data.estimate$`EUR/GBP_RGV`+data.estimate$`EUR/GBP_RGV_5`+data.estimate$`EUR/GBP_RGV_22`)
  EURGBP_HAR_RVGt5_NT_train <- lm(data.estimate$`EUR/GBP_RGV t+1,t+5`~data.estimate$`EUR/GBP_RGV`+data.estimate$`EUR/GBP_RGV_5`+data.estimate$`EUR/GBP_RGV_22`+data.estimate$`EUR/GBP_TotalCount`)
  EURGBP_HAR_RVt5_Bloomberg_train <- lm(data.estimate$`EUR/GBP_RV t+1,t+5`~data.estimate$`EUR/GBP_RV`+data.estimate$`EUR/GBP_RV_5`+data.estimate$`EUR/GBP_RV_22`+data.estimate$`EUR/GBP_BloombergCount`)
  EURGBP_HAR_RVt5_SocialMedia_train <- lm(data.estimate$`EUR/GBP_RV t+1,t+5`~data.estimate$`EUR/GBP_RV`+data.estimate$`EUR/GBP_RV_5`+data.estimate$`EUR/GBP_RV_22`+data.estimate$`EUR/GBP_SocialMediaCount`)
  EURGBP_HAR_RVt5_News_train <- lm(data.estimate$`EUR/GBP_RV t+1,t+5`~data.estimate$`EUR/GBP_RV`+data.estimate$`EUR/GBP_RV_5`+data.estimate$`EUR/GBP_RV_22`+data.estimate$`EUR/GBP_NewsCount`)
  
  EURGBP_forecast.5[, 1 ] <- predict( EURGBP_HAR_RVt5_train, newdata=data.forecast )            
  EURGBP_forecast.5[, 2 ] <- predict( EURGBP_HAR_RVt5_NT_train, newdata=data.forecast )         
  EURGBP_forecast.5[, 3 ] <- predict( EURGBP_HAR_RVBt5_train, newdata=data.forecast )           
  EURGBP_forecast.5[, 4 ] <- predict( EURGBP_HAR_RVBt5_NT_train, newdata=data.forecast )            
  EURGBP_forecast.5[, 5 ] <- predict( EURGBP_HAR_RVGt5_train, newdata=data.forecast )           
  EURGBP_forecast.5[, 6 ] <- predict( EURGBP_HAR_RVGt5_NT_train, newdata=data.forecast )            
  EURGBP_forecast.5[, 7 ] <- predict( EURGBP_HAR_RVt5_Bloomberg_train, newdata=data.forecast )
  EURGBP_forecast.5[, 8 ] <- predict( EURGBP_HAR_RVt5_SocialMedia_train, newdata=data.forecast )
  EURGBP_forecast.5[, 9 ] <- predict( EURGBP_HAR_RVt5_News_train, newdata=data.forecast )
  
  EURGBP_actuals.5[, 1] <- data.forecast$`EUR/GBP_RV t+1,t+5`
  
  # --------------------------------------
  # Forecast horizon 22
  # --------------------------------------
  
  EURGBP_HAR_RVt22_train <- lm(data.estimate$`EUR/GBP_RV t+1,t+22`~data.estimate$`EUR/GBP_RV`+data.estimate$`EUR/GBP_RV_5`+data.estimate$`EUR/GBP_RV_22`)
  EURGBP_HAR_RVt22_NT_train <- lm(data.estimate$`EUR/GBP_RV t+1,t+22`~data.estimate$`EUR/GBP_RV`+data.estimate$`EUR/GBP_RV_5`+data.estimate$`EUR/GBP_RV_22`+data.estimate$`EUR/GBP_TotalCount`)
  EURGBP_HAR_RVBt22_train <- lm(data.estimate$`EUR/GBP_RBV t+1,t+22`~data.estimate$`EUR/GBP_RBV`+data.estimate$`EUR/GBP_RBV_5`+data.estimate$`EUR/GBP_RBV_22`)
  EURGBP_HAR_RVBt22_NT_train <- lm(data.estimate$`EUR/GBP_RBV t+1,t+22`~data.estimate$`EUR/GBP_RBV`+data.estimate$`EUR/GBP_RBV_5`+data.estimate$`EUR/GBP_RBV_22`+data.estimate$`EUR/GBP_TotalCount`)
  EURGBP_HAR_RVGt22_train <- lm(data.estimate$`EUR/GBP_RGV t+1,t+22`~data.estimate$`EUR/GBP_RGV`+data.estimate$`EUR/GBP_RGV_5`+data.estimate$`EUR/GBP_RGV_22`)
  EURGBP_HAR_RVGt22_NT_train <- lm(data.estimate$`EUR/GBP_RGV t+1,t+22`~data.estimate$`EUR/GBP_RGV`+data.estimate$`EUR/GBP_RGV_5`+data.estimate$`EUR/GBP_RGV_22`+data.estimate$`EUR/GBP_TotalCount`)
  EURGBP_HAR_RVt22_Bloomberg_train <- lm(data.estimate$`EUR/GBP_RV t+1,t+22`~data.estimate$`EUR/GBP_RV`+data.estimate$`EUR/GBP_RV_5`+data.estimate$`EUR/GBP_RV_22`+data.estimate$`EUR/GBP_BloombergCount`)
  EURGBP_HAR_RVt22_SocialMedia_train <- lm(data.estimate$`EUR/GBP_RV t+1,t+22`~data.estimate$`EUR/GBP_RV`+data.estimate$`EUR/GBP_RV_5`+data.estimate$`EUR/GBP_RV_22`+data.estimate$`EUR/GBP_SocialMediaCount`)
  EURGBP_HAR_RVt22_News_train <- lm(data.estimate$`EUR/GBP_RV t+1,t+22`~data.estimate$`EUR/GBP_RV`+data.estimate$`EUR/GBP_RV_5`+data.estimate$`EUR/GBP_RV_22`+data.estimate$`EUR/GBP_NewsCount`)
  
  EURGBP_forecast.22[, 1 ] <- predict( EURGBP_HAR_RVt22_train, newdata=data.forecast )          
  EURGBP_forecast.22[, 2 ] <- predict( EURGBP_HAR_RVt22_NT_train, newdata=data.forecast )           
  EURGBP_forecast.22[, 3 ] <- predict( EURGBP_HAR_RVBt22_train, newdata=data.forecast )         
  EURGBP_forecast.22[, 4 ] <- predict( EURGBP_HAR_RVBt22_NT_train, newdata=data.forecast )          
  EURGBP_forecast.22[, 5 ] <- predict( EURGBP_HAR_RVGt22_train, newdata=data.forecast )         
  EURGBP_forecast.22[, 6 ] <- predict( EURGBP_HAR_RVGt22_NT_train, newdata=data.forecast )          
  EURGBP_forecast.22[, 7 ] <- predict( EURGBP_HAR_RVt22_Bloomberg_train, newdata=data.forecast )
  EURGBP_forecast.22[, 8 ] <- predict( EURGBP_HAR_RVt22_SocialMedia_train, newdata=data.forecast )
  EURGBP_forecast.22[, 9 ] <- predict( EURGBP_HAR_RVt22_News_train, newdata=data.forecast )
  
  EURGBP_actuals.22[, 1] <- data.forecast$`EUR/GBP_RV t+1,t+22`
  
} 

这将返回以下错误：

EURGBP_forecast.1[, 1] <- 预测(EURGBP_HAR_RV_train, newdata = data.forecast) 中的错误： 要替换的项目数量不是替换长度的倍数
另外：警告消息： “newdata”有 1 行，但找到的变量有 500 行

我不知道我的错误是什么，我将不胜感激任何类型的帮助。

如果我需要提供更多信息，请告诉我。

Answer 1

我想我会设置几个这样的函数：

首先是一个生成模型的函数，给定一个 data.estimate 和一个horizo​​n。这将返回 lm 对象列表

gen_models <- function(data.estimate, horizon) {
  
  if(horizon == 1) {
    return(
      list(
        EURGBP_HAR_RV_train = lm(`EUR/GBP_RV t+1`~`EUR/GBP_RV`+`EUR/GBP_RV_5`+`EUR/GBP_RV_22`,data=data.estimate),
        EURGBP_HAR_RV_NT_train = lm(`EUR/GBP_RV t+1`~`EUR/GBP_RV`+`EUR/GBP_RV_5`+`EUR/GBP_RV_22`+`EUR/GBP_TotalCount`,data=data.estimate),
        EURGBP_HAR_RVB_train = lm(`EUR/GBP_RBV t+1`~`EUR/GBP_RBV`+`EUR/GBP_RBV_5`+`EUR/GBP_RBV_22`,data=data.estimate),
        EURGBP_HAR_RVB_NT_train = lm(`EUR/GBP_RBV t+1`~`EUR/GBP_RBV`+`EUR/GBP_RBV_5`+`EUR/GBP_RBV_22`+`EUR/GBP_TotalCount`,data=data.estimate),
        EURGBP_HAR_RVG_train = lm(`EUR/GBP_RGV t+1`~`EUR/GBP_RGV`+`EUR/GBP_RGV_5`+`EUR/GBP_RGV_22`,data=data.estimate),
        EURGBP_HAR_RVG_NT_train = lm(`EUR/GBP_RGV t+1`~`EUR/GBP_RGV`+`EUR/GBP_RGV_5`+`EUR/GBP_RGV_22`+`EUR/GBP_TotalCount`,data=data.estimate),
        EURGBP_HAR_RV_Bloomberg_train = lm(`EUR/GBP_RV t+1`~`EUR/GBP_RV`+`EUR/GBP_RV_5`+`EUR/GBP_RV_22`+`EUR/GBP_BloombergCount`,data=data.estimate),
        EURGBP_HAR_RV_SocialMedia_train = lm(`EUR/GBP_RV t+1`~`EUR/GBP_RV`+`EUR/GBP_RV_5`+`EUR/GBP_RV_22`+`EUR/GBP_SocialMediaCount`,data=data.estimate),
        EURGBP_HAR_RV_News_train = lm(`EUR/GBP_RV t+1`~`EUR/GBP_RV`+`EUR/GBP_RV_5`+`EUR/GBP_RV_22`+`EUR/GBP_NewsCount`,data=data.estimate)
      )
    )
  }
  if(horizon==5) {
    return(
      list(
        EURGBP_HAR_RVt5_train = lm(`EUR/GBP_RV t+1,t+5`~`EUR/GBP_RV`+`EUR/GBP_RV_5`+`EUR/GBP_RV_22`,data=data.estimate),
        EURGBP_HAR_RVt5_NT_train = lm(`EUR/GBP_RV t+1,t+5`~`EUR/GBP_RV`+`EUR/GBP_RV_5`+`EUR/GBP_RV_22`+`EUR/GBP_TotalCount`,data=data.estimate),
        EURGBP_HAR_RVBt5_train = lm(`EUR/GBP_RBV t+1,t+5`~`EUR/GBP_RBV`+`EUR/GBP_RBV_5`+`EUR/GBP_RBV_22`,data=data.estimate),
        EURGBP_HAR_RVBt5_NT_train = lm(`EUR/GBP_RBV t+1,t+5`~`EUR/GBP_RBV`+`EUR/GBP_RBV_5`+`EUR/GBP_RBV_22`+`EUR/GBP_TotalCount`,data=data.estimate),
        EURGBP_HAR_RVGt5_train = lm(`EUR/GBP_RGV t+1,t+5`~`EUR/GBP_RGV`+`EUR/GBP_RGV_5`+`EUR/GBP_RGV_22`,data=data.estimate),
        EURGBP_HAR_RVGt5_NT_train = lm(`EUR/GBP_RGV t+1,t+5`~`EUR/GBP_RGV`+`EUR/GBP_RGV_5`+`EUR/GBP_RGV_22`+`EUR/GBP_TotalCount`,data=data.estimate),
        EURGBP_HAR_RVt5_Bloomberg_train = lm(`EUR/GBP_RV t+1,t+5`~`EUR/GBP_RV`+`EUR/GBP_RV_5`+`EUR/GBP_RV_22`+`EUR/GBP_BloombergCount`,data=data.estimate),
        EURGBP_HAR_RVt5_SocialMedia_train = lm(`EUR/GBP_RV t+1,t+5`~`EUR/GBP_RV`+`EUR/GBP_RV_5`+`EUR/GBP_RV_22`+`EUR/GBP_SocialMediaCount`,data=data.estimate),
        EURGBP_HAR_RVt5_News_train = lm(`EUR/GBP_RV t+1,t+5`~`EUR/GBP_RV`+`EUR/GBP_RV_5`+`EUR/GBP_RV_22`+`EUR/GBP_NewsCount`,data=data.estimate)
      )
    )
  }
  if(horizon == 22) {
    return(
      list(
        EURGBP_HAR_RVt22_train = lm(`EUR/GBP_RV t+1,t+22`~`EUR/GBP_RV`+`EUR/GBP_RV_5`+`EUR/GBP_RV_22`,data=data.estimate),
        EURGBP_HAR_RVt22_NT_train = lm(`EUR/GBP_RV t+1,t+22`~`EUR/GBP_RV`+`EUR/GBP_RV_5`+`EUR/GBP_RV_22`+`EUR/GBP_TotalCount`,data=data.estimate),
        EURGBP_HAR_RVBt22_train = lm(`EUR/GBP_RBV t+1,t+22`~`EUR/GBP_RBV`+`EUR/GBP_RBV_5`+`EUR/GBP_RBV_22`,data=data.estimate),
        EURGBP_HAR_RVBt22_NT_train = lm(`EUR/GBP_RBV t+1,t+22`~`EUR/GBP_RBV`+`EUR/GBP_RBV_5`+`EUR/GBP_RBV_22`+`EUR/GBP_TotalCount`,data=data.estimate),
        EURGBP_HAR_RVGt22_train = lm(`EUR/GBP_RGV t+1,t+22`~`EUR/GBP_RGV`+`EUR/GBP_RGV_5`+`EUR/GBP_RGV_22`,data=data.estimate),
        EURGBP_HAR_RVGt22_NT_train = lm(`EUR/GBP_RGV t+1,t+22`~`EUR/GBP_RGV`+`EUR/GBP_RGV_5`+`EUR/GBP_RGV_22`+`EUR/GBP_TotalCount`,data=data.estimate),
        EURGBP_HAR_RVt22_Bloomberg_train = lm(`EUR/GBP_RV t+1,t+22`~`EUR/GBP_RV`+`EUR/GBP_RV_5`+`EUR/GBP_RV_22`+`EUR/GBP_BloombergCount`,data=data.estimate),
        EURGBP_HAR_RVt22_SocialMedia_train = lm(`EUR/GBP_RV t+1,t+22`~`EUR/GBP_RV`+`EUR/GBP_RV_5`+`EUR/GBP_RV_22`+`EUR/GBP_SocialMediaCount`,data=data.estimate),
        EURGBP_HAR_RVt22_News_train = lm(`EUR/GBP_RV t+1,t+22`~`EUR/GBP_RV`+`EUR/GBP_RV_5`+`EUR/GBP_RV_22`+`EUR/GBP_NewsCount`,data=data.estimate)
      )
    )
  }
}

然后是一个返回预测的函数，给定模型列表和 data.forecast 框架

get_forecasts <- function(models, new_data) {
  forecasts = lapply(models, predict, newdata=new_data)
  names(forecasts) = names(models)
  forecasts
}

然后，您可以使用类似的方法循环遍历 number_of_forecasts，每次返回单行 tibbles 的列表。

window_size=500
last_window = nrow(EURGBP_final) - window_size
number_of_forecasts = dim( EURGBP_final)[1] - window_size

result = lapply(1:2, function(x) {
  
  data.estimate <- EURGBP_final %>% filter(between(row_number(),x,x+499))
  data.forecast <- EURGBP_final %>% filter(row_number() == 500+x)
  
  models1 = gen_models(data.estimate, horizon = 1)
  forecasts1 = get_forecasts(models1,data.forecast)
  
  models5 = gen_models(data.estimate, horizon = 5)
  forecasts5 = get_forecasts(models5,data.forecast)
  
  models22 = gen_models(data.estimate, horizon = 22)
  forecasts22 = get_forecasts(models22,data.forecast)
  
  list(
    f1 = as_tibble(forecasts1) %>% mutate(row=x+500) %>% relocate(row),
    f5 = as_tibble(forecasts5) %>% mutate(row=x+500) %>% relocate(row),
    f22 =  as_tibble(forecasts22) %>% mutate(row=x+500) %>% relocate(row)
  )
})

然后可以使用 bind_rows 将所得的 tibbles 绑定在一起

f1 = bind_rows(lapply(result, function(x) x[["f1"]]))
f5 = bind_rows(lapply(result, function(x) x[["f5"]]))
f22 = bind_rows(lapply(result, function(x) x[["f22"]]))

现在，f1、f5 和 f22 中的每一个都是包含预测的 tibbles从第 501 行到原始帧的末尾。如果您愿意，您可以轻松添加日期。 f1 如下所示：

# A tibble: 1,911 x 10
     row EURGBP_HAR_RV_train EURGBP_HAR_RV_NT_train EURGBP_HAR_RVB_train EURGBP_HAR_RVB_N~ EURGBP_HAR_RVG_t~ EURGBP_HAR_RVG_~ EURGBP_HAR_RV_B~ EURGBP_HAR_RV_S~
   <dbl>               <dbl>                  <dbl>                <dbl>             <dbl>             <dbl>            <dbl>            <dbl>            <dbl>
 1   501             0.00444                0.00446              0.00314           0.00314           0.00303          0.00302          0.00445          0.00446
 2   502             0.00445                0.00443              0.00315           0.00315           0.00309          0.00311          0.00445          0.00449
 3   503             0.00445                0.00441              0.00314           0.00315           0.00314          0.00318          0.00445          0.00442
 4   504             0.00444                0.00439              0.00318           0.00318           0.00313          0.00317          0.00444          0.00446
 5   505             0.00444                0.00445              0.00317           0.00316           0.00312          0.00312          0.00444          0.00447
 6   506             0.00444                0.00447              0.00317           0.00316           0.00310          0.00308          0.00445          0.00449
 7   507             0.00445                0.00443              0.00314           0.00315           0.00321          0.00322          0.00445          0.00449
 8   508             0.00444                0.00445              0.00313           0.00313           0.00312          0.00312          0.00444          0.00448
 9   509             0.00446                0.00443              0.00317           0.00317           0.00313          0.00314          0.00445          0.00440
10   510             0.00445                0.00447              0.00318           0.00317           0.00319          0.00317          0.00445          0.00442
# ... with 1,901 more rows, and 1 more variable: EURGBP_HAR_RV_News_train <dbl>