We propose a new class of observation-driven models, the Generalized Autoregressive Conditional Betas, which describe the joint dynamics of the time-varying slopes in a system of conditionally heteroskedastic simultaneous multiple regressions. The model accommodates large dimensions, parametric longitudinal restrictions, exogenous variables, and the coexistence of constant and time-varying slopes. It also introduces new mechanisms for the transmission of shocks, namely beta spillovers. We derive stationarity and uniform invertibility conditions and present beta and covariance tracking constraints. We show consistency and asymptotic normality of the Gaussian quasi-maximum likelihood estimator, and propose several computationally convenient quasi-maximum likelihood estimators, both parallel and sequential. Their finite sample properties are evaluated by Monte Carlo experiments. Finally, we illustrate the usefulness of modeling beta spillovers in the Fama-French three-factor asset pricing model. The results show that our model is useful for describing the transmission of shocks in financial markets.