In order to test for the validity of your analysis when using GARCH models, you should make sure that the model adequately captures the dynamics of the data. In other words, make sure that standardised residuals and squared standardised residuals are free from serial autocorrelation (you can employ the Box-Pierce portmanteau statistic).
In addition, you have to make sure that your model capture all ARCH effects (employ Lagrange multiplier test)
The following three aspects of the residuals from fitted GARCH model should be tested:
1. The standardized residuals from the GARCH model should approach normal distribution. One can use Shapiro-Wilk test and Jarque-Bera normality test. Histogram of the residuals is also a good visual tool to check normality.
2. The standardized squared residuals from the GARCH model should not be autocorrelated. One can use Ljung-box Q-statistic for this purpose.
3. ARCH LM test on the residuals can also be conducted to check for remaining ARCH effects in the residuals.
"The standardized residuals from the GARCH model should approach normal distribution. One can use Shapiro-Wilk test and Jarque-Bera normality test. Histogram of the residuals is also a good visual tool to check normality"
This is right only if conditional normality of residuals is assumed. However, a researcher may use other distribution (e.g. student t, GED, skewed t, and skewed GED).
If the number of observations used is large enough, t-distribution may also approach normal distribution. Thus, the residuals from a GARCH model with t-distribution may also follow normal distribution.
However, the distribution of many economic series is flat-tailed, which require a GARCH with t-distribution. One should check the distribution of the variable before choosing the type of distribution to be used with GARCH estimation.
how to examine the distribution of the variable before choosing the type of distribution to be used with GARCH estimation.(like student t, GED, skewed t, and skewed GED).
To test the validity of GARCH model, after the estimation of volatility we need to check whether the model has adequatley captured the voltility of data or not, we need to run Residual Diagnostic test to check whether the data series is free from Serial correlation and ARCH effect or not (This includes 3 tests).
To test the validity of GARCH model, after the estimation of volatility we need to check whether the model has adequatley captured the voltility of data or not, we need to run Residual Diagnostic test to check whether the data series is free from Serial correlation and ARCH effect or not (This includes 3 tests).