我有这样一个数据集:
data <- as.zoo(ts.union(a=arima.sim(model=list(ar=c(.9,-.2)), n=144),
b=arima.sim(model=list(ar=c(.6, -.3)), n=144),
c=arima.sim(model=list(ar=c(-.2,-.6)), n=144)))
我为 a 做了一个滚动窗口预测，它为每一步提供了 6 步提前预测。:
rolling.window <- rollapply(data, width = 132,
FUN = function(x) predict(VAR(x, type="const", ic="FPE"),
n.ahead=6, ci=0.95)$fcst$a[,1],
by.column = F, align = "right")
head(rolling.window)
132 0.086474 0.031416 0.00071186 -0.016284 -0.025615 -0.030692
133 1.289223 0.762734 0.46166288 0.284157 0.180816 0.120837
134 0.307354 0.332732 0.28306490 0.223481 0.171789 0.132596
135 0.105074 0.148357 0.14704495 0.128852 0.109577 0.093722
136 -0.469992 -0.496095 -0.39268676 -0.263921 -0.155009 -0.074600
137 -1.047158 -0.720692 -0.45201041 -0.251064 -0.115632 -0.029640
现在，我想像这样自动将这些预测存储在矩阵(或多个时间序列对象)中:
w132 w133 w134 w135 w136 w137
133 0.08647370 NA NA NA NA NA
134 0.03141553 1.28922 NA NA NA NA
135 0.00071186 0.76273 0.30735 NA NA NA
136 -0.01628371 0,46166 0.33273 0.105074 NA NA
137 -0.02561482 0.28416 0.28306 0.148357 -0.46999 NA
138 -0.03069235 0.18082 0.22348 0.147045 -0.49610 -1.04716
139 NA 0.12084 0.17179 0.128852 -0.39269 -0.72069
140 NA NA 0.13260 0.109577 -0.26392 -0.45201
141 NA NA NA 0.093722 -0.15501 -0.25106
142 NA NA NA NA -0.07460 -0.11563
143 NA NA NA NA NA -0.02964
等等。我希望每个滚动窗口步骤都在相应时间提前预测 6 步。不幸的是，我完全不知道应该从哪里开始。我尝试使用 lag()，但这一次只能用于一个系列。我也不知道如何在 rollapply() 函数中做到这一点。你能给我一个提示吗？
最佳答案
你可以这样做:
do.call(cbind, lapply(1:nrow(df), function(i) c(rep(NA,i-1), df[i,], rep(NA, nrow(df)-i))))
[,1] [,2] [,3] [,4] [,5] [,6]
V2 0.086474 NA NA NA NA NA
V3 0.031416 1.289223 NA NA NA NA
V4 0.00071186 0.762734 0.307354 NA NA NA
V5 -0.016284 0.4616629 0.332732 0.105074 NA NA
V6 -0.025615 0.284157 0.2830649 0.148357 -0.469992 NA
V7 -0.030692 0.180816 0.223481 0.147045 -0.496095 -1.047158
NA 0.120837 0.171789 0.128852 -0.3926868 -0.720692
NA NA 0.132596 0.109577 -0.263921 -0.4520104
NA NA NA 0.093722 -0.155009 -0.251064
NA NA NA NA -0.0746 -0.115632
NA NA NA NA NA -0.02964
数据:
df = structure(list(V2 = c(0.086474, 1.289223, 0.307354, 0.105074,
-0.469992, -1.047158), V3 = c(0.031416, 0.762734, 0.332732, 0.148357,
-0.496095, -0.720692), V4 = c(0.00071186, 0.46166288, 0.2830649,
0.14704495, -0.39268676, -0.45201041), V5 = c(-0.016284, 0.284157,
0.223481, 0.128852, -0.263921, -0.251064), V6 = c(-0.025615,
0.180816, 0.171789, 0.109577, -0.155009, -0.115632), V7 = c(-0.030692,
0.120837, 0.132596, 0.093722, -0.0746, -0.02964)), .Names = c("V2",
"V3", "V4", "V5", "V6", "V7"), row.names = 132:137, class = "data.frame")
关于r - 滚动窗口后创建预测矩阵，我们在Stack Overflow上找到一个类似的问题： https://stackoverflow.com/questions/32454061/