GSIB (global systematically important banks) balance sheet simulations are part of the CCAR program data presented to the FRB (Federal Reserve Bank), I think semiannually. The Banks use the balance sheet simulations to demonstrate their current and projected capabilities in liquidity management to their senior management and the Federal Reserve Bank. Prior to the CCAR submission there is substantial effort at each of the GSIBs to check the balance sheet simulations for completeness and accuracy. After all the GSIB CCAR submissions, the FRB presumably expends substantial effort to verify that the aggregated balance sheet simulation data is complete and accurate.
The Fed and the banks all agree on a set of market environment data including econometric variables in addition to yield curves, credit spreads and FX exchange rates. They agree on the specific variables and the levels they assume on a given reporting as_of date. This baseline market environment represents a snap shot of the current global market prices. The Fed and the banks further agree on a set of future market environments (differing in various ways than the implied future market environments) denoted with names like Adverse or Severely Adverse. All these market environments are standard inputs to the balance sheet simulator along with the current balances and rates. The balances and rates are in a matrix of dimension account by simulated forward months. Assume there are no more than a million accounts at any one GSIB, practically there are about 10K to 100K accounts, but let’s go with a million for estimates. There are 60 months to simulate forward. The output of the simulation is a matrix with 60 million elements representing the balance and rate movment over the 5 year simulation horizon for each account in the bank balance sheet. There is one 60mm element matrix output for each of the predetermined input scenarios.
The idea is that we can resimulate the balance sheet entirely to check the CCAR submissions. How many resimulations can we do? Given a 10 hour day to check we can resimulate about 10mm times for free (say on your desktop) or a billion times if you use a modest multiprocessor. What can you check and discover about a GSIB CCAR submission if you can ask a billion automated questions before submitting it to the FRB? What can the FRB find out about the aggregated CCAR balance sheet submissions in one day by asking a billion automated follow up questions? The time and money cost is immaterial, who says no? Banks may not be excited to share their proprietary quantitative balance and rate models, since their revenue depends on them. The FRB can develop their own standard product balance and rate models, give them to the banks and have the banks report both native and standard model balance sheet simulation results. FRB then has an easier job and a standard way to rerun the bank’s simulation.