R语言中MSGARCH包如何像rugarch设定均值方程外部回归变量？

Great question! Moving from rugarch to MSGARCH (for regime-switching or standard GARCH models) means adjusting to a slightly different specification structure, but adding external regressors to the mean equation is straightforward once you know where to look.

Key Difference from rugarch #

In rugarch, you specify mean equation regressors inside the mean.model$external.regressors argument of ugarchspec(). In MSGARCH, the equivalent setting lives within the mean.spec list passed to the CreateSpec() function—this is the core spot to target.

Step-by-Step Implementation #

Here’s a concrete, reproducible example to walk you through the process:

1. Load the package and prepare your data
   First, load MSGARCH and create your time series returns + external regressor matrix (each column is a separate regressor, and rows must match the length of your returns data):

   library(MSGARCH)
   
   # Replace with your actual returns data
   returns <- rnorm(100)
   
   # Create external regressors (2 regressors for 100 observations)
   external_regs <- matrix(rnorm(100 * 2), ncol = 2, dimnames = list(NULL, c("macro_reg", "firm_reg")))
   

2. Define the MSGARCH specification
   Use CreateSpec() to build your model, and plug in the external regressors directly in the mean.spec list:

   spec <- CreateSpec(
     # Match your sGARCH(1,1) variance setup from rugarch
     variance.spec = list(model = "sGARCH", garchOrder = c(1, 1)),
     # Mean model with ARMA(1,1) + external regressors
     mean.spec = list(
       model = "ARMA",
       armaOrder = c(1, 1),
       include.mean = TRUE,  # Keeps the intercept in the mean equation
       external.regressors = external_regs  # Your regressors go here
     ),
     # Match your normal distribution setting
     distribution.spec = list(distribution = "norm"),
     # Disable regime switching if you don't need it (set to TRUE for multi-regime models)
     switch.spec = list(do.mix = FALSE)
   )
   

3. Fit the model and inspect results
   Estimate the model with FitML() and check the summary to confirm your regressor coefficients are included in the mean equation output:

   fit <- FitML(spec = spec, data = returns)
   summary(fit)
   

Critical Details to Keep in Mind #

• Regressor Format: external.regressors requires a matrix (not a data frame or vector). Double-check that the number of rows exactly matches your returns data length to avoid errors.
• Regime Switching Support: If you’re building a multi-regime model (switch.spec = list(do.mix = TRUE)), use the switch.mean parameter in switch.spec to control whether regressor coefficients vary across regimes: set switch.mean = TRUE for regime-specific coefficients, or FALSE to share coefficients across regimes.
• Mean Equation Flexibility: If you set include.mean = FALSE, the mean equation will omit the intercept and only include ARMA terms + external regressors.

内容的提问来源于stack exchange，提问作者Markoff Chainz