RStudio中VAR/SVAR模型FEVD图表重叠问题的优化咨询

问题描述 #

对14个变量拟合VAR/SVAR模型后绘制FEVD（预测误差方差分解）图表时，出现以下问题：

• 图表元素严重重叠
• 变量颜色图例显示不全，且部分区域与图表重叠
• 调整par("mar")参数无效果

附示例数据及原绘图代码：

# 示例数据
data <- structure(c(-15.128286545, 9.54362378, -3.183102118, 2.455920412, 2.291578693, -2.375566455, -1.006170511, -0.3, 0.2, 0.07, -1.48, 1.66, 0.23, 2.69, 0.139816501, -0.067264912, 0.00820542499999988, 0.601791352, -0.534672832, 0.448348103, 0.201531496, 1.912707215, -2.298881719, 2.250845193, -100.270023362, 168.06388869, -29.13957144, -5.96895973, -65.094658064, 69.084085184, 77.852479576, -115.75847262, 1.07846378, 124.53936952, -136.67294405, 0.763694729, -3.181436503, 0.357452785, 1.419643891, -1.352477057, -0.352129671, -0.518658596,-40.982268397, 15.10208529, 8.81612737, 14.742141917, 0.331767914, -3.388439373, -1.705233303, 9.716651558, 17.163282893, -26.403831865, 5.522207566, -6.267360551, 8.793307298, 0.68190014, 6.571072841, 10.734436876, -15.768737701, 3.732193641, -3.013388691, 2.257042669, -3.092580854, -7.590788096, -0.377256109, -1.954321683, -0.655236714, 8.787007941, -11.048392421, -11.771665078, -9.286360071, 0.672766377, 4.540993395, -22.771472581, 33.33956332, -21.892993669, 3.082682577, -18.404892011, 6.039591015, -8.744388675, 4.5514676, 3.81637218,-8.616022054, 10.73600499, -27.792088246, 27.763201157, -20.582295417, 36.58994891, -33.462654905, 8.941944848, 4.666353971, -0.492222954, 0.443107902, -0.444468843, 0.378535534, 0.363633632, -0.396522691, 0.164774515), dim = c(7L, 14L), dimnames = list(NULL, c("QIM", "ML", "WPI", "BOT", "FDI", "TX", "FBT", "CP", "PHA", "NM", "AUT", "ISP", "FER", "ER")))

# 原绘图代码（假设estim已完成VAR拟合）
vd <- fevd(estim, n.ahead=36)
par(mar=c(1.2,5,1.5,4.7))
plot(vd)

解决方案 #

1. 调整绘图窗口尺寸与基础布局 #

vars包默认的FEVD绘图布局在变量较多时容易拥挤，可通过以下方式优化：

• 手动拖大RStudio绘图窗口，或用dev.new()指定更大的画布尺寸：

dev.new(width = 12, height = 8)  # 可根据变量数量灵活调整宽高
plot(vd)

• 使用gridExtra包自定义子图布局，分散14个变量的FEVD子图：

library(gridExtra)
# 将14个FEVD子图拆分为3行5列（可按需调整行列数）
fevd_plots <- lapply(names(vd), function(var) {
  plot(vd[[var]], main = var, plot.type = "single", mar = c(3,3,2,1))
})
grid.arrange(grobs = fevd_plots, nrow = 3, ncol = 5)

2. 直接修改plot.fevd的参数 #

给plot()函数传递专属参数控制图例和间距：

• 调整图例位置与大小：用legend.pos指定图例位置，legend.cex缩小图例文字：

plot(vd, legend.pos = "bottomright", legend.cex = 0.7)

• 手动设置子图边距与外间距：直接在plot()中传递mar和oma参数（vars的plot.fevd支持该参数）：

plot(vd, mar = c(4, 4, 2, 1), oma = c(0, 0, 2, 0))  # oma控制整体外边框距

3. 用ggplot2重绘FEVD（完全自定义样式） #

将FEVD数据转换为长格式，用ggplot2实现精细化控制：

library(ggplot2)
library(tidyr)

# 将FEVD对象转换为可用于ggplot的数据框
fevd_df <- do.call(rbind, lapply(names(vd), function(response_var) {
  df <- as.data.frame(vd[[response_var]])
  df$horizon <- 1:nrow(df)
  df$response_var <- response_var
  pivot_longer(df, cols = -c(horizon, response_var), 
               names_to = "shock_var", values_to = "contribution")
}))

# 绘制分面FEVD图
ggplot(fevd_df, aes(x = horizon, y = contribution, color = shock_var)) +
  geom_line(linewidth = 0.8) +
  facet_wrap(~response_var, nrow = 3, ncol = 5) +  # 按需调整行列数
  labs(x = "预测期", y = "方差贡献占比", color = "冲击变量") +
  theme_minimal() +
  theme(legend.position = "bottom",
        legend.text = element_text(size = 8),
        axis.text = element_text(size = 7),
        strip.text = element_text(size = 9))

4. 精简变量（可选） #

如果14个变量中有高度相关或逻辑上可合并的变量，可先合并变量后再绘制FEVD，从根源减少子图数量，避免拥挤。

内容的提问来源于stack exchange，提问作者Adeel Ahmed