If you estimated a standard GARCH(s,r) model, the parameters were likely restricted to produce stationary conditional variance, which means it is mean reverting. This is what you see -- the variance gradually reverts towards its long-term mean.
An alternative GARCH-type of model that allows for non-mean-reverting volatility is integrated GARCH (IGARCH) that produces random-walk-type of volatility; or a GARCH model with exogenous variables (such as linear or nonlinear deterministic trend) in the conditional variance equation, which produces -- naturally -- a linear or nonlinear trend in volatility.
GARCH is more appropriate for forecasting volatility. Here, you can use $F$ test to compare the inflation volatility before and after the inflation targeting. It involves lesser assumptions and is more appropriate.
Let $X_1, \dotsc, X_n$ and $Y_1, \dotsc, Y_m$ be independent and identically distributed samples from two populations which each have a normal distribution. The expected values for the two populations can be different, and the hypothesis to be tested is that the variances are equal. Let
$$\overline{X} = \frac{1}{n}\sum_{i=1}^n X_i\text{ and }\overline{Y} = \frac{1}{m}\sum_{i=1}^m Y_i$$
be the sample means. Let
$$ S_X^2 = \frac{1}{n-1}\sum_{i=1}^n \left(X_i - \overline{X}\right)^2\text{ and }S_Y^2 = \frac{1}{m-1}\sum_{i=1}^m \left(Y_i - \overline{Y}\right)^2 $$
be the sample variances. Then the test statistic
$$ F = \frac{S_X^2}{S_Y^2} $$
has an $F$-distribution with $n − 1$ and $m − 1$ degrees of freedom if the null hypothesis of equality of variances is true. Otherwise it has a non-central $F$-distribution. The null hypothesis is rejected if $F$ is either too large or too small.
Best Answer
ARCH postestimation help file explains it all. You will most likely need
or
depending on what exactly you mean by ``volatility''. The former is the full prediction, the latter is the multiplier that goes in front of the $\hat\sigma^2$.