面板数据上的HoltWinters

Question

我试图在面板数据上运行HoltWinters过程以获得公司列表的销售预测。我的数据框包含“公司”，“年”和“销售”字段。我有兴趣在此数据框中添加一个额外的列，以显示使用HoltWinters获得的合适销售额。

对于单个公司来说，这个练习是微不足道的，因为我可以运行HoltWinters。我可以通过跨公司循环来扩展这种方法，但我正在寻找避免循环的解决方案。

有什么想法？

Answer 1

阅读帮助文件HoltWinters，它看起来不像是矢量化的，所以我不认为你可以避免循环。我能做的最好的事情就是让lapply的循环更漂亮（也可能有小的速度增益）。假设你已经将个人作为单独的时间序列（或时间序列工会）：

set.seed(0) 
a <- ts(rnorm(10), start=c(2001, 5), freq=12) 
b <- ts(rnorm(10), start=c(2001, 5), freq=12) 
c <- ts(rnorm(10), start=c(2001, 5), freq=12) 
d <- ts(rnorm(10), start=c(2001, 5), freq=12) 
e <- ts(rnorm(10), start=c(2001, 5), freq=12) 
f <- ts(rnorm(10), start=c(2001, 5), freq=12) 
individual.one <- ts.union(a, b, c) 
individual.two <- ts.union(d, e, f) 
panel <- list(individual.one, individual.two) 

ans <- lapply(panel, HoltWinters) 

产量：

> ans 
Holt-Winters exponential smoothing with trend and additive seasonal component. 

Call: 
FUN(x = X[[1L]]) 

Smoothing parameters: 
alpha: 0 
beta : 0 
gamma: 0 

Coefficients: 
      [,1] 
a -0.504984544 
b -0.035336155 
s1 -1.085615710 
s2 -0.352859142 
s3 0.002437573 
s4 2.550897907 
s5 0.986769568 
s6 -0.596066930 
s7 -1.028559659 
s8 -0.187923905 
s9 -0.149289171 
s10 -0.168011617 
s11 0.606452449 
s12 -0.578231362 

[[2]] 
Holt-Winters exponential smoothing with trend and additive seasonal component. 

Call: 
FUN(x = X[[2L]]) 

Smoothing parameters: 
alpha: 0 
beta : 0 
gamma: 0 

Coefficients: 
      [,1] 
a -0.39575955 
b -0.04035375 
s1 0.75505039 
s2 -0.64553006 
s3 1.06488778 
s4 -0.40487180 
s5 1.74515472 
s6 0.64324387 
s7 -0.36380752 
s8 -0.74481981 
s9 -1.04726447 
s10 -0.90172103 
s11 -1.42433355 
s12 1.32401148 

其中：

> panel 
[[1]] 
        a   b   c 
May 2001 1.262954285 0.7635935 -0.22426789 
Jun 2001 -0.326233361 -0.7990092 0.37739565 
Jul 2001 1.329799263 -1.1476570 0.13333636 
Aug 2001 1.272429321 -0.2894616 0.80418951 
Sep 2001 0.414641434 -0.2992151 -0.05710677 
Oct 2001 -1.539950042 -0.4115108 0.50360797 
Nov 2001 -0.928567035 0.2522234 1.08576936 
Dec 2001 -0.294720447 -0.8919211 -0.69095384 
Jan 2002 -0.005767173 0.4356833 -1.28459935 
Feb 2002 2.404653389 -1.2375384 0.04672617 

[[2]] 
        d   e   f 
May 2001 -0.2357066 1.7579031 0.26613736 
Jun 2001 -0.5428883 0.5607461 -0.37670272 
Jul 2001 -0.4333103 -0.4527840 2.44136463 
Aug 2001 -0.6494716 -0.8320433 -0.79533912 
Sep 2001 0.7267507 -1.1665705 -0.05487747 
Oct 2001 1.1519118 -1.0655906 0.25014132 
Nov 2001 0.9921604 -1.5637821 0.61824329 
Dec 2001 -0.4295131 1.1565370 -0.17262350 
Jan 2002 1.2383041 0.8320471 -2.22390027 
Feb 2002 -0.2793463 -0.2273287 -1.26361438