As seen in the previous section, the emphasis between capital and controls varies according to the nature of the risk. However, securities firms have put forward a number of arguments against the way existing capital requirements are structured. These can be summarised as follows:
(i) they are uneconomic, and drive business into unregulated or less regulated entities;
(ii) they have little relevance to how firms internally manage their business and hence create unnecessary costs;
(iii) they do not reflect the real risk at an individual firm, specifically because they offer only limited recognition to hedges, correlations and offsets.
The Technical Committee agrees that capital requirements should be based on risk (insofar as this can be measured), not least in order to give firms the right incentives to reduce risk. The increased use and experience of VaR models by securities firms for internal risk management purposes has caused the Technical Committee to look again at the case for basing some part of regulatory capital requirements on the output of VaR models. In the last four to five years, securities firms and supervisors have gained practical experience in modelling, particularly of market risk. This has been assisted by a number of studies (see the appendix for references).
The above work has established that VaR models are more readily applicable to environments with the following characteristics:
- Where there is a long run of data for the risk factors being modelled, with a very high frequency of observations. The more information there is to calibrate the model, and then test it, the more confidence firms (and supervisors) can have in the appropriateness of the output.
- Where the distributional assumptions made by the model reflect the essential features of the market, e.g. if the model assumes a normal distribution, the actual returns exhibit a low degree of skew or fat tailedness. 3
- Where underlying exposures lend themselves to aggregation without undue loss of risk sensitivity.
From this combination of practical experience and research, supervisors have been in a position to develop a framework to approach the use of models for regulatory capital purposes. It is recognised that there can be no certainty that the future will resemble the past. In addition, assumptions that seemed reasonable at the time of developing the model could well, at some point in the future, lose their validity. Considering these points, the recognition of models requires supervisory involvement in the following areas:
(i) verifying that VaR models are fully integrated into the day to day risk management processes of the firm with an appropriate level of independent verification;
(ii) approving the parameters of the models such as minimum holding period, confidence level and historical observation period;
(iii) assessing the reliability of the models including the use of correlations (e.g. through establishing a framework for backtesting);
(iv) ensuring that firms regularly review the performance of their models and take action on the outcomes of their internal backtesting;
(v) ensuring that firms supplement their modelling approach by a programme of stress testing that is appropriate to their particular risk profile and concentration of exposures;
(vi) obtaining appropriate and timely information on the outcomes of backtesting and stress testing;
satisfying themselves that firms have adequate financial and human resources to adopt and operate a VaR model;
(viii) determining an additional amount of capital to be added to the value at risk number as a safeguard against unavoidable shortcomings of the VAR approach;
There are some qualifications in a move to VaR modelling that securities supervisors should be mindful of. In particular the following should be noted: