Solvency evaluation of the guaranty fund at a large financial cooperative 297was reviewed: the three studies of the Federal Deposit Corporation by Sheehan [5],
Bennett [1] and Kuritzkes et al. [2] and the study of the Canada Deposit Insurance
Corporation by McIntyre [3]. Overall, the survey of this literature showed that the
dominant approach to estimate the solvency of a deposit insurer was the use of a credit
risk simulation model and it appeared natural to follow this practice. However, before
proceeding, it seemed appropriate to identify the methodological options and make a
deliberate choice.
3 Analysis and selection of methodologies
The process consisted in identifying the various possible approaches, evaluating them
and eventually selecting one or several for implementation. After some analysis, four
dimensions emerged to characterise the possible approaches, namely, the level of
aggregation of the data, the estimation technique, the depth of historical data and the
horizon considered for the future. The following sections will look at each in turn.
The demand for subsidies by the credit unions depends on the losses that these
incur, which depends in turn on the risks they bear. This observation leads to consider
the credit unions as a single aggregated entity or to proceed to a disaggregated analysis
of each credit union one by one. Similarly, the total risk can be analysed as an aggregate
of all risks or risks can be analysed individually (e.g., credit risk, market risk and
operational risk). Finally, credit risk itself can be analysed at the portfolio level or can
be analysed by segments according to the type, the size and the risk of loans.
To estimate the distribution of the demand for subsidies by credit unions, two
techniques appeared possible. If aggregatedata were used, it would be possible to fit
theoretical statistical distributions to the historical distribution. However if disaggre-
gated data were used, a Monte Carlo simulation model would be more appropriate to
estimate the distribution of the demand for subsidies.
If aggregated data were used, 25 years of data would be available. On the other
hand, if disaggregated data were used, only seven years of data would be available.
If theoretical statistical distribution were used, the model would be static and the
horizon of one year would logically follow from the yearly period of observation
of historical data. If a simulation model was used, the financial dynamics of credit
unions and of the guaranty fund itself could be modelled and trajectories over time
could be built. In this case, a horizon of 15 years was considered relevant.
As the analysis above has shown, even though the four dimensions were not
independent, several combinations of choices could be implemented and these could
be viewed as forming a spectrum mainly according to the level of aggregation of data,
which seemed to be the dimension that had the strongest impact on conditioning the
other choices. In this light, it was decided to move forward with the implementation of
two polar choices, namely a highly aggregated approach and a highly disaggregated
approach. Table 1 summarises the characteristics of each of these two approaches.
It was deemed interesting to implement two very different approaches and to ob-
serve whether they would converge or not to similar results. If similarity was obtained,
then a cross-validation effect would increase confidence in the results. If dissimilarity