Taking the bank and non-bank variables as the research objects, the threshold Multivariate Multiquantile Conditional Autoregressive Value at Risk model(threshold MVMQ-CAViaR model)is constructed from the trend dimension, volatility dimension, and stability dimension to study the level of tail risk spillover between banks and non-bank financial institutions, and depict the dynamic risk and contagion relationships of the trend dimension at the 1%, 5%, and 10% quantile levels. The results show that the threshold MVMQ-CAViaR model can effectively identify and predict the tail risk aggregation and contagion in all three dimensions, and the tail risk spillover between the bank and nonbank variables is bidirectional and asymmetric. Non-bank financial institutions exert greater risk spillover on banks during prosperous periods, while banks exert greater risk spillover on non-bank financial institutions during downturns. Therefore, it is necessary to strengthen regulatory supervision on non-bank's risk transmission during downturns.