我正在估计带有回归变量的VAR模型，该回归变量是通过实现cheng y Liu方法的软件包tsoutliers获得的。 该模型的估计正确，但是当我尝试预测h向前的步长（我使用h = 2）时，我得到了NA值作为预测值。
因此，我的回归变量（矩阵）是哑变量，因为它们是AO类型的离群值。我还必须在参数dumvar中使用回归矩阵进行预测，因此我创建了行等于h（在示例中等于2）的矩阵。
我尝试了不同的时期进行预测，也更改了预测中包含的回归矩阵，使用随机统一值矩阵（我怀疑问题是零）。
# Usamos el método de Cheng y Liu
# Restringimos a encontrar outliers de tipo AO con valor de t mayor igual a 3.
agro18_out <- tsoutliers::tso(serie18_diff1[,"agro"], tsmethod = "auto.arima",
args.tsmethod = list(ic = "aicc", seasonal = FALSE, approximation = FALSE,
stepwise = FALSE), types = c("AO"), cval = 2.0,
discard.method = "en-masse", maxit.iloop = 100, maxit.oloop = 100)
industria18_out <- tsoutliers::tso(serie18_diff1[,"industria"], tsmethod = "auto.arima",
args.tsmethod = list(ic = "aicc", seasonal = FALSE, approximation = FALSE,
stepwise = FALSE), types = c("AO"), cval = 3.0,
discard.method = "en-masse", maxit.iloop = 100, maxit.oloop = 100)
comercio18_out <- tsoutliers::tso(serie18_diff1[,"comercio"], tsmethod = "auto.arima",
args.tsmethod = list(ic = "aicc", seasonal = FALSE, approximation = FALSE,
stepwise = FALSE), types = c("AO"), cval = 3.0,
discard.method = "en-masse", maxit.iloop = 100, maxit.oloop = 100)
# Estimación del VAR hasta 2018
# generamos los regresores hasta 2018-2
agro18_ao <- tsoutliers::outliers.effects(agro18_out$outliers, length(serie18_diff1[, "agro"]))
industria18_ao <- tsoutliers::outliers.effects(industria18_out$outliers, length(serie18_diff1[, "agro"]))
comercio18_ao <- tsoutliers::outliers.effects(comercio18_out$outliers, length(serie18_diff1[, "agro"]))
colnames(agro18_ao) <- paste("agr", colnames(agro18_ao), sep = "")
colnames(industria18_ao) <- paste("ind", colnames(industria18_ao), sep = "")
colnames(comercio18_ao) <- paste("com", colnames(comercio18_ao), sep = "")
xreg18 <- matrix(nrow = nrow(comercio18_ao), ncol = ncol(comercio18_ao) + ncol(industria18_ao) + ncol(agro18_ao))
xreg18 <- cbind(industria18_ao, comercio18_ao, agro18_ao)
xreg18
sum(apply(xreg18, 1, sum) > 1) # No tienen AO en los mismos periodos
h = 2
xreg18_pred <- matrix(nrow = 2, ncol = ncol(comercio18_ao) + ncol(industria18_ao) + ncol(agro18_ao))
agro18_ao <- tsoutliers::outliers.effects(agro18_out$outliers, length(serie18_diff1[, "agro"]) + h)
industria18_ao <- tsoutliers::outliers.effects(industria18_out$outliers, length(serie18_diff1[, "agro"]) + h)
comercio18_ao <- tsoutliers::outliers.effects(comercio18_out$outliers, length(serie18_diff1[, "agro"]) + h)
colnames(agro18_ao) <- paste("agr", colnames(agro18_ao), sep = "")
colnames(industria18_ao) <- paste("ind", colnames(industria18_ao), sep = "")
colnames(comercio18_ao) <- paste("com", colnames(comercio18_ao), sep = "")
xreg18_pred <- cbind(industria18_ao, comercio18_ao, agro18_ao)
xreg18_pred <- xreg18_pred[-seq_along(1:nrow(xreg18)),]
xreg18_pred
# Estimamos inicialmente un var
var18 <- vars::VAR(serie18_diff1, ic = "AIC", type = "none", lag.max = 10, exogen = xreg18)
# predicciones
# Se genera error al usar el dumvar...Por qué???
pred18 <- predict(object = var18, n.ahead = 2, ci = 0.95, dumvar = xreg18_pred)
这是我的实际输出 $农业 第一低下CI [1，] NA NA NA 0.06412661 [2，] NA NA NA 0.06449499
$ industria 第一低下CI [1，] NA NA NA 0.05815483 [2，] NA NA NA 0.05861298
$ comercio 第一低下CI [1，] NA NA NA 0.07034564 [2，] NA NA NA 0.07128827
这是我的预期结果
$ agro 第一低下CI [1，] 0.0007866399 -0.07264581 0.07421909 0.07343245 [2，] -0.0007444915 -0.07810270 0.07661372 0.07735821
$ industria 第一低下CI [1，] -0.0042103795 -0.07351935 0.06509859 0.06930897 [2，] -0.0003505954 -0.07081144 0.07011025 0.07046085
$ comercio 第一低下CI [1，] -0.0050694363 -0.08668936 0.07655049 0.08161992 [2，] -0.0002143067 -0.08376733 0.08333872 0.08355302