The GARCH models and heterogeneous autoregressive (HAR) model are used to evaluate the information about the continuous/jump compositions of variances. The continuous/jump components of the implied variance are extracted from FTSE 100 data. We compare the performance of the GARCH models in terms of sample fit and out-of-sample performance measures. Our results suggest that both the realized variance and model-free implied volatility are important information resources in variance forecasting. Using the HAR model, we find that the realized measurements contain incremental information relative to implied variances. Implied volatilities are improved as the GB2 distribution fitting is applied.