Once a set of adjustment scenarios has been produced in a consistent macroframework, the next step is to translate the various outputs into the balance sheets and income statements of financial institutions. There are two main approaches to translating or “mapping” scenarios into balance sheets: the “bottom-up” approach, in which estimates are based on data for individual portfolios, which can then be aggregated, and the “top-down” approach, which uses aggregated or macrolevel data to estimate the effect.4
Under the bottom-up approach, the response to various shocks in a scenario is estimated at the portfolio level while using highly disaggregated data from individual financial institutions at a point in time. The results of the bottom-up approach can then be aggregated or compared to analyze the sensitivity of the entire sector or group of institutions. The bottom-up approach has the advantage of making better use of individual portfolio data; however, if individual institutions provide their own estimates, then the approach may introduce some inconsistencies about how each institution applies the scenario and produces its numerical estimates. The bottom-up approach also provides information on how the effect of shocks varies across institutions and on the variance or dispersion of this effect, which is an important statistic on financial stability of the system insofar as large losses in one institution can trigger contagion.
The top-down approach is used to estimate the responsiveness of a group of institutions to a particular scenario. Under this approach, a common parameter or estimate is applied to all institutions in the data set, (e. g., using a panel regression or a regression of aggregated information) to arrive at an estimate of the aggregate effect. The top-down approach is often easier to implement, because it requires only time series of aggregated
data and is a consistent and uniform method that implicitly takes into account the responses of banks to shocks over time. However, aggregate historical relationships may not hold in the future. Ideally, both methods should be applied, but data limitations may preclude the application of both methods in many countries.
The remainder of this section discusses the various steps involved in implementing a system-focused stress test by addressing a series of key questions. The questions include the following: Who should perform the empirical analysis? Which institutions should be included? What are the data constraints? How large should the shocks be? How do we link the macroadjustment scenarios to individual balance sheets and income statements?