我正在尝试计算由 5 个季度时间序列变量组成的结构 VAR 系统的脉冲响应函数：
Uncertainty = 0.01516483 0.01576851 0.01514961 0.01505102 0.01512489 0.01507230 0.01527489 0.01542124 0.01542613 0.01588306 0.01535583 0.01521866 0.01532704 0.01561553 0.01533545 0.01546499 0.01541305 0.01521282 0.01519695 0.01518378 0.01549650 0.01567024 0.01525345 0.01524214 0.01528679 0.01551463 0.01523704 0.01593742 0.01612399 0.01616118 0.01560939 0.01596426 0.01619084 0.01638339 0.01629785 0.01648435 0.01814780 0.01744788 0.01975361 0.01567541 0.01547753 0.01641009 0.01568128 0.01552354
Bank Rates = 4.750000 4.750000 4.583333 4.500000 4.500000 4.500000 4.666667 4.916667 5.250000 5.416667 5.750000 5.666667 5.333333 5.000000 5.000000 3.166667 1.000000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000 0.500000
Net Trade = -2232 -2355 -2480 -2423 -2411 -2361 -2739 -2655 -3026 -2974 -3227 -2487 -3333 -2631 -2291 -1850 -2042 -2122 -2078 -2246 -2468 -2196 -2481 -2606 -2276 -1876 -1897 -1881 -1855 -2126 -2149 -3058 -2169 -2338 -2353 -2807 -2805 -2912 -3104 -2799 -3444 -3654 -4052 -3769
Employment = 73.3 72.9 73.1 73.1 72.9 72.7 73.5 74.2 74.5 74.9 74.2 74.3 74.3 73.9 73.3 73.1 72.0 71.8 71.5 69.9 70.1 70.7 71.2 71.0 71.6 71.1 70.9 71.1 71.3 70.5 70.8 71.6 72.2 72.7 72.5 73.2 73.5 73.9 74.2 74.4 74.1 74.1 74.6
GDP = 81.7 83.3 83.3 84.4 84.4 84.7 85.6 86.2 86.4 87.1 88.0 89.7 89.2 91.7 93.3 94.2 93.2 93.4 94.5 94.3 95.0 96.0 96.1 96.5 98.3 99.2 99.5 99.6 99.8 99.9 100.3 100.0 101.8 101.7 102.8 102.8 103.2 103.4 103.5 103.8 103.1 103.2 103.9 104.0
根据 Augmented Dickey Fuller 测试，它们都是静止的。此外，在检查时间序列对之间的协整时，没有发现（Engle-Granger 检验）。
将所有变量分组后：
Sample = cbind(Uncertainty Bank Rates, Net Trade, Employment, GDP )
我尝试研究用于 var 的最佳滞后数：
VARselect(Sample, lag.max = 20, type="const")
这告诉我 6.... 这是第一个问题：季度数据的滞后数不切实际....此外，当将测试应用于每个单独的系列时，其中大多数显示 2 作为最佳滞后数。考虑到这一点，我尝试使用误差方差-协方差矩阵的 Cholesky 分解来计算 SVAR，并描绘脉冲响应函数：
var <- VAR(Sample, p=2, type = "both")
var
amat <- diag(5)
diag(amat) <- NA
svar.a <- SVAR(var, estmethod = "direct", Amat = amat)
plot(irf(svar.a, impulse = "Topic.15",
response = c("BankRates", "NetTrade", "Employ.per", "GDP"),
n.ahead = 20, ortho = TRUE,
cumulative = FALSE,
boot = TRUE, ci = 0.95,
runs = 100))
我得到：
