Is it correct that you use the RCP scenarios (as a general circulation model) and then you downscale them (regionalized) via the CMIP5?

We do not perform our own downscaling of Global Climate Models. Currently, the majority of our models use data from the coordinated regional downscaling experiment (CORDEX). This simulation data uses CMIP5 projection data, which is a global projection based on the RCP scenarios, and downscales them to individual regions such as North America, Europe or Australia. Newer hazard models use CMIP6 data, which is the updated version of CMIP5 that uses shared socio-economic pathways (SSPs) as scenarios, which are similar to the RCPs for physical risk (though some additional SSPs exist). 

For further details please see XDI Climate and Hazard Modelling Methodology documentation.