2.1 Fundamentación Teórica
2.1.2 Plan de negocios
550. Annex X, Part 3, Paragraph 12 of the CRD requires, among others things, that the risk measurement system be internally consistent.
551. Institutions should adopt appropriate methods and procedures to guarantee the consistency and quality of the input, execution, and output phases of the model.
Model input, execution and output
552. Institutions should seek to identify operational risk classes within which loss amounts are independent and identically distributed. Alternatively, institutions may wish to adjust their data for known drivers in order to simplify the modeling process, which needs to be justified.
553. Where modelling requires independence of loss events, the following tools and techniques may be used.
· Time plots and appropriate functions, which serve, respectively, to check the level of stationarity of the data over time and to address the detected nonstationarity of the frequency and severity components.
· Techniques that reduce the effects of data clustering and seasonality and that adjust the data for inflation.
· Qualitative criteria (based, for example, on the qualitative information relating to the data) and/or quantitative tests to detect outliers and techniques to eliminate or mitigate their influence.
554. The execution of models for generating regulatory operational risk figures should be supported by a transparent and consistent process.
Depending on the modelling methodology chosen, institutions could take the following steps which are to be intended neither exhaustive nor binding:
1. Preliminary identification of a set of probability distributions to be fitted to the data. 2. Appropriate techniques for estimating parameters.
3. Appropriate diagnostic tools for evaluating how well the distributions fit the data. 4. Sound methods for selecting distributions, when the results from point 3 do not
indicate a clear choice.
555. In order to determine an overall operational risk capital figure that is credible and justifiable, the model should be built in a way that ensures the production of results that are as stable as possible. In addition, institutions should be able to evaluate the accuracy of the operational risk capital figures.
556. Some of the elements that institutions could take into consideration when calculating operational risk capital figure are presented in Annex VII of these guidelines.
Observation period
557. Annex X, Part 3, Paragraph 13 of the CRD requires institutions to base their internally generated operational risk measures on a minimum historical observation period of five years (three years when an institution first moves to an AMA).
558. A lowfrequency operational risk class may need a historical observation period longer than five years in order to collect sufficient data to generate reliable operational risk measures.
General examples of methods that institutions could use to obtain a sufficient amount of data reflecting the current operational risk profile include the following:
· Reducing the impact of oldest, least relevant internal data by appropriate weighting techniques (e.g. moving average); using quantitative indicators (e.g. inflation or business/structure risk drivers) or qualitative factors that reflect changes in the institution’s internal/external environment.
· Supplementing the most recent years of internal data with the corresponding years of external data from similar institutions/peer groups, after appropriate adjustments to the external data.
· Constructing data for operational losses in past years by means of scenario generated data or by scaling back more recent years of internal/external observations through appropriate techniques or indicators.
559. In the absence of sufficient data, institutions should make conservative risk estimates.
Confidence level
560. Annex X, Part 3, Paragraph 8 of the CRD states that the regulatory operational risk measure must capture potentially severe tail events, achieving a soundness standard comparable to a 99.9 percent confidence level.
561. In order to generate a regulatory operational risk measure at a soundness standard comparable to a 99.9 percent confidence level, institutions can perform a direct calculation at the 99.9 percent confidence level, or they can calculate an initial measure at a lower confidence level, located in the right end of the loss distribution, and then scale it up to the 99.9 percent confidence level using appropriate methods.
562. The institution should be able to demonstrate that the scaling method yields an output that is plausible and reliable. The confidence level used should not necessarily be interpreted as a boundary between the body and the tail of the distribution.
563. The confidence level at which the initial operational risk measure can be computed should be located in the rightend of the distribution of the losses. The level should be appropriate.
564. If scaling is used, institutions should be able to demonstrate the soundness, appropriateness, and reliability of the scaling technique and to analyse the overall accuracy of the scaling mechanism.
The following are examples of techniques that can be used to scale an initial calculation at a confidence level below 99.9 percent to a level comparable to 99.9 percent:
· Imposing a reasonable distribution for the highest percentiles of the data, based, for example, on information collected by peer groups or scenario analysis.
· If an Extreme Value Theory approach is used, resorting to the stability property of the model (the socalled 'Peaks over Threshold' stability property) that makes it possible to compute the highest percentiles of the frequency and severity distributions from figures estimated at a lower level (usually at the threshold level).
4.3.4.3. Expected losses, correlation, Insurance and other Risk transfer