In this paper, we estimate the effects of the COVID-19 pandemic on the banking system and the real economy and simulate potential policy responses. We combine machine learning algorithms, namely a Random Regression Forest and a Long Short Term Memory neural network, with an agent-based framework to calculate the expected results of the pandemic, according to different scenarios regarding financial stability. We then simulate government responses to this crisis and find that traditional demand and supply stimuli are outperformed by our suggestion of relaxing bank regulation. We examine two alternatives of our suggested policy and find that they result in optimised outcomes for most variables examined. Our findings have important policy implications as authorities are formulating post-crisis recovery plans amidst budgetary constraints.