Theoretical and practical issues in Solvency II

Forecasting the future is a very important step for getting knowledge of various risks to which a company may be exposed.
There are different methods for doing this. For example, we can make a forecast by stochastic simulations. Those have their foundations in classical risk theory and cash flow modeling. The method strives for statistical precision such as the probability of ruin.
Another approach is stress testing procedures. This approach do not strives for statistical precision and has been adopted also in the banking system as the generic term that describes the potential effects of exceptional events using different techniques. These procedures were chosen because they are more understandable to management, regulators, investors and non-technical persons.
Moreover, the stress testing can be made using three different approaches, indeed it can be made: by a sensitivity test,by a scenario test, or by themaximum loss approach.

The sensitivity test isolates the short-term impact on assets or liabilities of the insurance undertaking of a series of predefined changes in particular risk factors, such as a shock change in interest rate of ±2%, 3%,etc.
The scenario test specifies the shocks that could affect a number of risk factors simultaneously, when extreme events occur. The scenarios could be based on either historical events or hypothetical events.
The maximum loss approach is usually used for forecast catastrophic events. It is a test for identifying the most damaging combinations of catastrophic events and their impact on the business.

In 2003 the IAIS published a guidance paper on stress testing in which it is understood that ”insurers would undertake stress testing as a matter of good corporate governance” and that ”the supervisor may develop standard stress tests and require insurers to perform such tests.”The guidance paper also includes discussion on designing the test.
Internal models can be seen also as early warning systems. Since supervision and modeling may be forward looking, and in order to properly analyze the data and to construct a satisfactory model, it is essential to understand the environment in which the data has been collected, and, in particular, how the environment has changed in the past.

Indeed, a good model may lose validity if some condition change trought the time, thus there will be an inaccurate representation of reality and the ability to form a good base for decision-making can be affected.