shockvector

[JanFialkowski]

It should be possible to get the final shock values for each firm. I.e. both the h_d(T) and h_u(T). The R code does this with the option track_h. We should be able to do the same. Biggest hurdle: It's an enormous matrix that is not easy to put into a text file. Other solutions might be necessary.

[Devetak]

This cannot be done at scale i think.

Not even if storing in-betweens. 200_000 x 200_000 matrix of float8 is around 300GB.

Best to return vectors and let users handle intermediate computations and storage?

[JanFialkowski]

The output definitely needs to be some sparse format. What do you mean by in-betweens? In the most basic use-case: you calculate the ESRI cascade and get some shock value for each firm due to up- and downstream contagion. It should be possible to take that vector and somehow save it at the end of the cascade. It might be necessary to save it in a serialized format of julias sparse matrices.

The aspirational goal would be to track these shocks over time and see how it propagates through the network. But that might actually be impossible in the general case.