如何检验GARCH过程是否为零均值？R语言对应的检验方法是什么？

Hey there! Let's tackle your two questions about verifying zero mean in GARCH processes step by step:

When we say a GARCH process has zero mean, we mean the constant term in its mean equation is statistically indistinguishable from zero. Here are the most reliable testing approaches:

• Robust sample mean t-test: Calculate the sample mean of the series, then run a t-test—but since GARCH processes have conditional heteroscedasticity, you must use robust standard errors (like Newey-West adjustments) to account for heteroscedasticity and potential autocorrelation. If the test returns a p-value above your chosen significance level (e.g., 0.05), you can't reject the zero-mean hypothesis.
• Test the constant term in a mean-inclusive GARCH model: Fit an ARMA-GARCH model where the mean equation includes a constant term (μ). Then check if μ is statistically significant. If the p-value for μ is > 0.05, there's no evidence against the zero-mean assumption.
• Non-parametric alternatives (for non-normal data): If your data violates normality assumptions, use tests like the sign test or Wilcoxon signed-rank test to check if the median (which equals the mean for symmetric distributions) is zero. These are less powerful than parametric tests but more robust to distributional violations.

R has several straightforward tools for this—here are the most common ones with code examples:

• Robust t-test for sample mean
  Use the sandwich and lmtest packages to compute Newey-West robust standard errors for a simple mean model:

  # Load required packages
  library(sandwich)
  library(lmtest)
  
  # Assume 'x' is your time series data
  mean_model <- lm(x ~ 1)  # Model with only an intercept (representing the mean)
  # Run t-test with Newey-West robust SEs (adjust lag based on your data's autocorrelation)
  robust_mean_test <- coeftest(mean_model, vcov = NeweyWest(mean_model, lag = 3))
  print(robust_mean_test)
  

  If the intercept's p-value is > 0.05, you can proceed with the zero-mean assumption.

• Test the constant term in a fitted GARCH model
  Two popular packages for GARCH modeling make this easy:

  1. Using rugarch:

  library(rugarch)
  
  # Specify a GARCH(1,1) model with a mean intercept
  garch_spec <- ugarchspec(
    variance.model = list(garchOrder = c(1, 1)),
    mean.model = list(armaOrder = c(0, 0), include.mean = TRUE)
  )
  
  # Fit the model to your data
  garch_fit <- ugarchfit(spec = garch_spec, data = x)
  
  # Check the coefficient table for the mean term (labeled 'mu')
  print(garch_fit@fit$matcoef)
  

  Look at the p-value for mu—if it's above 0.05, the zero-mean assumption holds.

  2. Using fGarch:

  library(fGarch)
  
  # Fit a GARCH(1,1) model with a mean intercept
  garch_fit <- garchFit(~ garch(1, 1), data = x, include.mean = TRUE)
  
  # Print the summary to view mean term significance
  summary(garch_fit)
  

  In the "Coefficient(s)" section, find the p-value for mu to assess significance.

内容的提问来源于stack exchange，提问作者Anna