如何在 R 中进行 Ljung-Box 测试？答案

【问题标题】：How to do the Ljung-Box test in R?如何在 R 中进行 Ljung-Box 测试？
【发布时间】：2019-12-05 14:34:18
【问题描述】：

我正在尝试在 R 中进行 Ljung-Box 测试，但出现错误，我不明白问题出在哪里。

让我们使用 "Forecasting: Principles and Practice" 中示例中的代码来获取 Google 的每日收盘价

library(fpp3)
google_stock <- gafa_stock  %>%
  filter(Symbol == "GOOG")  %>%
  mutate(day = row_number()) %>%
  update_tsibble(index = day, regular = TRUE)

google_2015 <- google_stock %>% filter(year(Date) == 2015)

我使用朴素方法从预测序列中获得残差

aug <- google_2015 %>% model(NAIVE(Close)) %>% augment()

现在我想使用 Ljung-Box 测试对残差中的自相关进行正式测试

aug %>% features(.resid, ljung_box, lag=10, dof=0)

但我在输出中收到此错误：

# A tibble: 1 x 2
  Symbol .model      
  <chr>  <chr>       
1 GOOG   NAIVE(Close)
Warning message:
1 error encountered for feature 1
[1] 'ts' object must have one or more observations

我做错了什么？

这是数据集的一个样本（使用 dput() ）：

    > dput(head(google_stock, 10))
structure(list(Symbol = c("GOOG", "GOOG", "GOOG", "GOOG", "GOOG", 
"GOOG", "GOOG", "GOOG", "GOOG", "GOOG"), Date = structure(c(16072, 
16073, 16076, 16077, 16078, 16079, 16080, 16083, 16084, 16085
), class = "Date"), Open = c(554.125916, 553.897461, 552.908875, 
558.865112, 569.297241, 568.025513, 565.859619, 559.595398, 565.298279, 
572.769714), High = c(555.26355, 554.856201, 555.814941, 566.162659, 
569.953003, 568.413025, 565.859619, 569.749329, 571.781128, 573.768188
), Low = c(550.549194, 548.894958, 549.645081, 556.95752, 562.983337, 
559.143311, 557.499023, 554.975403, 560.400146, 568.199402), 
    Close = c(552.963501, 548.929749, 555.049927, 565.750366, 
    566.927673, 561.468201, 561.438354, 557.861633, 570.986267, 
    570.598816), Adj_Close = c(552.963501, 548.929749, 555.049927, 
    565.750366, 566.927673, 561.468201, 561.438354, 557.861633, 
    570.986267, 570.598816), Volume = c(3666400, 3355000, 3561600, 
    5138400, 4514100, 4196000, 4314700, 4869100, 4997400, 3925700
    ), day = 1:10), row.names = c(NA, -10L), key = structure(list(
    Symbol = "GOOG", .rows = list(1:10)), row.names = c(NA, -1L
), class = c("tbl_df", "tbl", "data.frame"), .drop = TRUE), index = structure("day", ordered = TRUE), index2 = "day", interval = structure(list(
    year = 0, quarter = 0, month = 0, week = 0, day = 0, hour = 0, 
    minute = 0, second = 0, millisecond = 0, microsecond = 0, 
    nanosecond = 0, unit = 1), class = "interval"), class = c("tbl_ts", 
"tbl_df", "tbl", "data.frame"))

【问题讨论】：

请提供您的数据集样本dput(head(df, 10))
我使用 dput() 编辑了带有示例的帖子
ljung_box(aug[,'.resid'], lag=10,dof=0) 返回值没有错误，因此这表明处理 %>% 存在一些问题？
这是在 github 上的 fabletools 包中修复的错误。见github.com/tidyverts/feasts/issues/79

标签： r time-series forecasting autocorrelation

【解决方案1】：

看起来您需要先选择感兴趣的列 (.resid)，然后再传递给 features() 函数：

aug %>% 
select(.resid) %>% 
features(.resid, ljung_box, lag = 10, dof = 0)
# Output
# Selecting index: "day"
# A tibble: 1 x 2
  lb_stat lb_pvalue
    <dbl>     <dbl>
1    7.91     0.637

如果您没有在features() 函数中指定.resid，它仍然可以通过自动选择该列（因为它是除索引列之外的唯一列）来工作：

aug %>% 
select(.resid) %>% 
features(features = ljung_box, lag = 10, dof = 0)
# Selecting index: "day"
# Feature variable not specified, automatically selected `.var = .resid`
# A tibble: 1 x 2
  lb_stat lb_pvalue
    <dbl>     <dbl>
1    7.91     0.637

【讨论】：