Modelling Operational Risk by AMA - Essay Example

Modelling Operational Risk by AMA The overall aim of the Basel II Advanced Measurement Approach (AMA) is to have a more quantitative approach to modelling operational risk whereby it will be able to have a statistical quantification of the level of operational risk losses banks are exposed to (Shevchenko and Temnov, 2009). This is exactly where the Bayesian interference comes in as an important form of Loss Distribution Approach (LDA) used from a statistical perspective in quantifying the frequency and severity distributions for bank operational risks (Akaike, 1983). On the whole, the Bayesian inference works by quantifying distributions for both frequency and severity of operational losses by the use of statistical procedures (Neil, Häger and Andersen, 2009). Lavin and Scherrish (1999) stressed that these statistical procedures are always expressed as random vector of data based on risk cells that have specified density for a given vector of parameter. Shevchenko (2011) associated that Bayesian inference to a number of advantages, for which they are used to model operational risk. A typical example of this is what Embrechts and Puccetti (2008) noted to be the consistency and convenience associated with the statistical framework used in quantifying uncertainties. As a quantitative approach, the outcomes with Bayesian inference are always guaranteed to be the same whenever the similar variables are used. This makes the outcomes with Bayesian inference highly reliable and consistent among similar set of operational variables within a bank (Lambrigger, Shevchenko and Wüthrich, 2007 and Neil, Fenton and Tailor, 2005). What is more, Shevchenko (2011) acknowledged the fact that the Bayesian inference is highly accommodating and versatile as it incorporates expert opinions with historical internal and external data used in various operational risk estimations (Burnecki, Kukla and Taylor, 2005). Even though the Bayesian inference has several strengths and advantages for usage, Shevchenko (2010) lamented that the approach’s over reliance on scenario analysis and expert judgement acts as a major setback for usage within a good number of firms. Adding to this, Wasserman (1997) and Alderweireld, Garcia and Léonard (2006) agreed that even though both scenario analysis and expert judgement provide important information for forecasting and decision making, banks with relatively limited dataset, and those that only started business may not have enough of these to use the Bayesian inference. Further improvement for using the method in modelling operational risk should therefore focus on ways in which banks can focus more on scenario analysis from an external perspective rather than focusing more internally. As far as internal scenario analysis is concerned, much emphasis ought to be placed on current and planned controls as these are common for both older and newer banks (Shevchenko, 2011). Apart from the use of Bayesian inference as a component of LDA, LDA has also been used as an independent method for modelling operational risk (Frachot, Georges and Roncalli, 2001). Chater and Oaksford (2008) stressed that the LDA is also a statistical or actuarial method for modelling loss distribution and thereby giving room to compute the capital charge of a bank’s operational risk. As the name implies, the LDA helps in realising operational risk by generally computing for aggregate loss distributions based on capital charge of value-at-risk measurement of risk (Carter andVan Brunt, 2000). In practice, banks implement the LDA by putting together three major forms of estimates which are risk type cell, probability distribution functions of single event impact, and event frequency known for the next year (Lee and Wagenmakers, 2005 and Frachot, Georges and Roncalli, 2001). Edwards et al. (1963) noted that the estimates give rise to the cumulative operational loss which are computed from the probability distribution function. A major merit associated with the use of the LDA method is the way and manner in which is embraces variables available to banks of all forms and sizes. This is because the key variables used in the computation of probability distribution function of cumulative operational loss are available to virtually any kind of bank (Kruschke, 2008). Again, the LDA has higher reliability because of the statistical models that are employed in its computations (Doya et al. (2007). Many have however criticised this method as only being an old grade version of the Bayesian inference since it continues to overly rely on expert judgements when Bayesian inference has actually gone ahead to incorporate other modalities such as the scenario analysis in its computations. There are therefore those who believe that the computational outcomes of LDA cannot be as evidence based as that of the Bayesian inference (Griffiths et al., 2008). Apart from the two methods discussed above, the internal measure approach (IMA) is also used very extensively in practice in modelling operational risk (Wagenmakers, 2007). But as reflected in its name, the IMA focuses on a bank’s underlying risk by making use of the internal loss data available to the bank. These internal loss data are this used as the key input for capital computations (Kruschke, 2010). The IMA has been praised by many practitioners for its simplicity and usability (Frachot, Georges and Roncalli, 2001). Further improvements are however needed for this method given the fact that it hardly caters for issues of overestimation and underestimation. This is because unlike the previous methods which have some forms of external variables, this one will particularly emphasise on internal variable which are difficult to verify (Frachot, Georges and Roncalli, 2001). This means that there could be room for accounting inaccuracies under this method. References Akaike, H. (1983). “Information measure and model selection”. Bulletin of the International Statistical Institute 50: pp. 277–290 Alderweireld, T., Garcia, J. and Léonard, L. (2006). “A practical operational risk scenario analysis quantification”. Risk Magazine. Vol. 19 No. 2: pp. 93–95 Berger, J.O. and Berry, D.A. (1988) “Statistical analysis and the illusion of objectivity”. Am. Sci. 76: pp. 159–165. Burnecki, K., Kukla, G. and Taylor, D. (2005). Pricing of catastrophic bonds. New York, NY: Springer Carter, M. andVan Brunt, B. (2000). The Lebesgue-Stieltjes Integral. A Practical Introduction. New York, NY: Springer. Chater, N. and Oaksford, M., eds (2008). The Probabilistic Mind. Oxford: Oxford University Press Doya, K. et al., eds (2007). Bayesian Brain: Probabilistic Approaches to Neural Coding, Massachusetts: MIT Press Edwards, W. et al. (1963). “Bayesian statistical inference for psychological research”. Psychol. Rev. 70: pp. 193–242 Embrechts, P. and Puccetti, G. (2008). “Aggregation operational risk across matrix structured loss data”. The Journal of Operational Risk. Vol. 3 N. 2: pp. 29–44. Frachot A., Georges P. and Roncalli T. (2001). Loss Distribution Approach for operational risk. [Online] Available at [November 4, 2014] Griffiths, T.L. et al. (2008). Bayesian models of cognition. Cambridge: Cambridge University Press Kruschke, J.K. (2008). “Bayesian approaches to associative learning: from passive to active learning”. Learn. Behav. 36: pp. 210–226 Kruschke, J.K. (2010). “Bayesian data analysis”. Rev. Cogn. Sci. 36: pp. 110–116 Lambrigger, D.D., Shevchenko, P.V. and Wüthrich, M.V. (2007). “The quantification of operational risk using internal data, relevant external data and expert opinions”. The Journal of Operational Risk Vol. 2 No. 3: pp. 3–27 Lavin, M. and Scherrish, M. (1999). “Bayes factors: what they are and what they are not”. The American Statistician 53: pp. 119–122 Lee, M.D. and Wagenmakers, E.J. (2005) “Bayesian statistical inference in psychology. Psychol. Rev. 112: pp. 662–668 Neil, M., Fenton, N.E. and Tailor, M. (2005). “Using Bayesian networks to model expected and unexpected operational losses”. Risk Analysis Vol. 25 No. 4: pp. 963–972 Neil, M., Häger, D. and Andersen, L.B. (2009). “Modeling operational risk in financial institutions using hybrid dynamic Bayesian networks”. Journal of Operational Risk Vol. 4 No. 1: 3–33 Shevchenko, P.V. (2010). “Implementing loss distribution approach for operational risk”. Applied Stochastic Models in Business and Industry Vol. 26 No. 3: pp. 277–307 Shevchenko, P.V. (2011). Modeling Operational Risk Using Bayesian Inference. New York: Springer Shevchenko, P.V. and Temnov, G. (2009). “Modeling operational risk data reported above a time-varying threshold”. The Journal of Operational Risk Vol. 4 No. 2: pp. 19–42 Wagenmakers, E.J. (2007). “A practical solution to the pervasive problems of p values”. Psychon. Bull. Rev. 14: pp. 779–804 Wasserman, L. (1997). Bayesian model selection and model averaging. Technical report, Statistics Department, Carnegie Mellon University . Read More