We propose an extension of standard asymmetric volatility models in the generalized autoregressive conditional heteroskedasticity (GARCH) class that admits conditional non-Gaussianities in a tractable fashion. Our ?bad environment-good environment” (BEGE) model utilizes two gamma-distributed shocks and generates a conditional shock distribution with time-varying heteroskedasticity, skewness, and kurtosis. The BEGE model features nontrivial news impact curves and closed-form solutions for higher-order moments. In an empirical application to stock returns, the BEGE model outperforms asymmetric GARCH and regime-switching models along several dimensions.