Modeling Operational Risk With Bayesian Networks

Bayesian networks is an emerging tool for a wide range of risk management applications, one of which is the modeling of operational risk. This comes at a time when changes in the supervision of financial institutions have resulted in increased scrutiny on the risk management of banks and insurance companies, thus giving the industry an impetus to measure and manage operational risk. The more established methods for risk quantification are linear models such as time series models, econometric models, empirical actuarial models, and extreme value theory. Due to data limitations and complex interaction between operational risk variables, various nonlinear methods have been proposed, one of which is the focus of this article: Bayesian networks. Using an idealized example of a fictitious on line business, we construct a Bayesian network that models various risk factors and their combination into an overall loss distribution. Using this model, we show how established Bayesian network methodology can be applied to: (1) form posterior marginal distributions of variables based on evidence, (2) simulate scenarios, (3) update the parameters of the model using data, and (4) quantify in real‐time how well the model predictions compare to actual data. A specific example of Bayesian networks application to operational risk in an insurance setting is then suggested.     

Value at risk and efficiency under dependence and heavy-tailedness: models with common shocks

This paper presents an analysis of diversification and portfolio value at risk for heavy-tailed dependent risks in models with multiple common shocks. We show that, in the framework of value at risk comparisons, diversification is optimal for moderately heavy-tailed dependent risks with common shocks and finite first moments, provided that the model is balanced, i.e., that all the risks are available for portfolio formation. However, diversification is inferior in balanced extremely heavy-tailed risk models with common factors. Finally, in several unbalanced dependent models, diversification is optimal, even though there is extreme heavy-tailedness in common shocks or in idiosyncratic parts of the risks. Analogues of the obtained results further hold for efficiency comparisons of linear estimators in random effects models with dependent and heavy-tailed observations.     

An Extreme Value Analysis Of Advanced Age Mortality Data

Abstract

Extreme value theory describes the behavior of random variables at extremely high or low levels. The application of extreme value theory to statistics allows us to fit models to data from the upper tail of a distribution. This paper presents a statistical analysis of advanced age mortality data, using extreme value models to quantify the upper tail of the distribution of human life spans.

Our analysis focuses on mortality data from two sources. Statistics Canada publishes the annual number of deaths in Canada, broken down by angender and age. We use the deaths data from 1949 to 1997 in our analysis. The Japanese Ministry of Health, Labor, and Welfare also publishes detailed annual mortality data, including the 10 oldest reported ages at death in each year. We analyze the Japanese data over the period from 1980 to 2000.

Using the r-largest and peaks-over-threshold approaches to extreme value modeling, we fit generalized extreme value and generalized Pareto distributions to the life span data. Changes in distribution by birth cohort or over time are modeled through the use of covariates. We then evaluate the appropriateness of the fitted models and discuss reasons for their shortcomings. Finally, we use our findings to address the existence of a finite upper bound on the life span distribution and the behavior of the force of mortality at advanced ages.     

The limits of diversification when losses may be large

Recent results in value at risk analysis show that, for extremely heavy-tailed risks with unbounded distribution support, diversification may increase value at risk, and that generally it is difficult to construct an appropriate risk measure for such distributions. We further analyze the limitations of diversification for heavy-tailed risks. We provide additional insight in two ways. First, we show that similar non-diversification results are valid for a large class of risks with bounded support, as long as the risks are concentrated on a sufficiently large interval. The required length of the support depends on the number of risks available and on the degree of heavy-tailedness. Second, we relate the value at risk approach to more general risk frameworks. We argue that in markets for risky assets where the number of assets is limited compared with the (bounded) distribution support of the risks, unbounded heavy-tailed risks may provide a reasonable approximation. We suggest that this type of analysis may have a role in explaining various types of market failures in markets for assets with possibly large negative outcomes.     

A parametric alternative to the Hill estimator for heavy-tailed distributions

Despite its wide use, the Hill estimator and its plot remain to be difficult to use in Extreme Value Theory (EVT) due to substantial sampling variations in extreme sample quantiles. In this paper, we propose a new plot we call the eigenvalue plot which can be seen as a generalization of the Hill plot. The theory behind the plot is based on a heavy-tailed parametric distribution class called the scaled Log phase-type (LogPH) distributions, a generalization of the ordinary LogPH distribution class which was previously used to model insurance claims data. We show that its tail property and moment condition are well aligned with EVT. Based on our findings, we construct the eigenvalue plot from fitting a shifted PH distribution to the excess log data with a minimal phase size. Through various numerical examples we illustrate and compare our method against the Hill plot.     

中国商业银行操作风险损失分布甄别与分析：基于贝叶斯MCMC频率方法

确切的操作风险损失分布保障了风险度量的准确性。对银行操作风险损失数据的分析,国外学者一致认为操作风险分布近似泊松分布或负的贝奴里分布。基于中国商业银行1994～2008年的操作风险损失数据,通过对操作风险损失分布的检验、贝叶斯马尔科夫蒙特卡洛频率分析,发现中国商业银行操作风险损失分布近似服从广义极值分布(Generalized Extreme Value)。     

On multivariate tests of the CAPM

This paper evaluates the power of multivariate tests of the Capital Asset Pricing Model. The results indicate that when employing an unspecified alternative hypothesis, the ability of the tests to distinguish between the CAPM and other pricing models is poor. An upper bound is derived for the distance the alternative distribution of the test statistic can be from the null distribution when the deviations from the CAPM are due to missing factors. This upper bound explains the low power of the tests.     

On the price of risk in a mean-risk optimization model

We investigate a mean-risk model for portfolio optimization where the risk quantifier is selected as a semi-deviation or as a standard deviation of the portfolio return. We analyse the existence of solutions to the problem under general assumptions. When the short positions are not constrained, we establish a lower bound on the cost of risk associated with optimizing the mean–standard deviation model and show that optimal solutions do not exist for any positive price of risk which is smaller than that bound. If the investment allocations are constrained, then we obtain a lower bound on the price of risk in terms of the shadow prices of said constraints and the data of the problem. A Value-at-Risk constraint in the model implies an upper bound on the price of risk for all feasible portfolios. Furthermore, we provide conditions under which using this upper bound as the cost of risk parameter in the model provides a non-dominated optimal portfolio with respect to the second-order stochastic dominance. Additionally, we study the relationship between minimizing the mean–standard deviation objective and maximizing the coefficient of variation and show that both problems are equivalent when the upper bound is used as the cost of risk. Additional relations between the Value-at-Risk constraint and the coefficient of variation are discussed as well. We illustrate the results numerically.     

Oversigt over de udenlandske Aktuartidsskrifter

Abstract

In the classical compound Poisson risk model, Lundberg's inequality provides both an upper bound for, and an approximation to, the probability of ultimate ruin. The result can be applied only when the moment generating function of the individual claim amount distribution exists. In this paper we derive an upper bound for the probability of ultimate ruin when the moment generating function of the individual claim amount distribution does not exist.     

On a Sparre Andersen risk model perturbed by a spectrally negative Lévy process

In this paper, we consider a Sparre Andersen risk model perturbed by a spectrally negative Lévy process (SNLP). Assuming that the interclaim times follow a Coxian distribution, we show that the Laplace transforms and defective renewal equations for the Gerber–Shiu functions can be obtained by employing the roots of a generalized Lundberg equation. When the SNLP is a combination of a Brownian motion and a compound Poisson process with exponential jumps, explicit expressions and asymptotic formulas for the Gerber–Shiu functions are obtained for exponential claim size distribution and heavy-tailed claim size distribution, respectively.     

Two-Sided Bounds for Tails of Compound Negative Binomial Distributions in the Exponential and Heavy-Tailed Cases

This paper derives two-sided bounds for tails of compound negative binomial distributions, both in the exponential and heavy-tailed cases. Two approaches are employed to derive the two-sided bounds in the case of exponential tails. One is the convolution technique, as in Willmot & Lin (1997). The other is based on an identity of compound negative binomial distributions; they can be represented as a compound Poisson distribution with a compound logarithmic distribution as the underlying claims distribution. This connection between the compound negative binomial, Poisson and logarithmic distributions results in two-sided bounds for the tails of the compound negative binomial distribution, which also generalize and improve a result of Willmot & Lin (1997). For the heavy-tailed case, we use the method developed by Cai & Garrido (1999b). In addition, we give two-sided bounds for stop-loss premiums of compound negative binomial distributions. Furthermore, we derive bounds for the stop-loss premiums of general compound distributions among the classes of HNBUE and HNWUE.     

收益率曲线动态与银行经济价值的利率风险

本文在收益率曲线动态的主成分分析基础上,运用MonteCarlo模拟的主成分VaR方法,以我国五家商业银行为样本研究银行账户经济价值利率风险的计量方法,并与巴塞尔委员会标准久期法的结果进行比较。同时,对VaR模型的有效性进行了样本外的返回检验。研究发现,五家银行的经济价值面临的是利率上升的风险;非正态主成分VaR模型估计的经济价值利率风险,都要大于正态主成分VaR模型的结果,这反映了利率波动的厚尾特征,正态假设有可能低估风险。     

Extreme CEO pay cuts and audit fees

This study investigates whether sudden and severe reductions in total CEO compensation affect auditor perceptions of risk. We argue that extreme CEO pay cuts can incentivize the CEO to manipulate the financial reports or make risky operational decisions in a desperate attempt to improve firm performance. This incentive, in turn, is likely to impact auditor assessments of audit risk and auditor business risk, leading to higher audit fees. Consistent with our hypothesis, we find evidence of a positive and highly significant association between extreme CEO pay cuts and audit fees. The results suggest that audit fees are 4.6% higher when there is an extreme CEO pay cut, which corresponds to an audit fee that is $111,458 higher for the average firm-year observation in our sample.     

Lifetime dependence models generated by multiply monotone functions

We study a family of distributions generated from multiply monotone functions that includes a multivariate Pareto and, previously unidentified, exponential-Pareto distribution. We utilize an established link with Archimedean survival copulas to provide further examples, including a multivariate Weibull distribution, that may be used to fit light, or heavy-tailed phenomena, and which exhibit various forms of dependence, ranging from positive to negative. Because the model is intended for the study of joint lifetimes, we consider the effect of truncation and formulate properties required for a number of parameter estimation procedures based on moments and quantiles. For the quantile-based estimation procedure applied to the multivariate Weibull distribution, we also address the problem of optimal quantile selection.     

Portfolio diversification and value at risk under thick-tailedness†

This paper focuses on the study of portfolio diversification and value at risk analysis under heavy-tailedness. We use a notion of diversification based on majorization theory that will be explained in the text. The paper shows that the stylized fact that portfolio diversification is preferable is reversed for extremely heavy-tailed risks or returns. However, the stylized facts on diversification are robust to heavy-tailedness of risks or returns as long as their distributions are moderately heavy-tailed. Extensions of the results to the case of dependence, including convolutions of α-symmetric distributions and models with common shocks are provided.     

Risk assessment based on the analysis of the impact of contagion flow

This paper presents a new framework to model and calibrate the process of firm value evolution when an unanticipated exogenous event impacting one firm can contagiously affect other firms. The nature of propagation of such contagion is determined by the underlying connections between firms, which can adversely affect the tail risks of firm value, hence the securities issued by the firm. This paper combines the insights gained from the existing firm-value models and historical events into a structural model for flow of contagion among firms using a network-based approach. Rather than using stylized networks, we develop a data-driven approach for network construction where we define and calibrate several contagion variables to model the spread of contagion. This framework is applied for assessing firm-level risk under downside risk measures. Using actual data, our model illustrates how connections between firms can lead to heavy-tailed default distributions and default clustering observed in practice.     

A generalization of Panjer’s recursion and numerically stable risk aggregation

Portfolio credit risk models as well as models for operational risk can often be treated analogously to the collective risk model coming from insurance. Applying the classical Panjer recursion in the collective risk model can lead to numerical instabilities, for instance if the claim number distribution is extended negative binomial or extended logarithmic. We present a generalization of Panjer’s recursion that leads to numerically stable algorithms. The algorithm can be applied to the collective risk model, where the claim number follows, for example, a Poisson distribution mixed over a generalized tempered stable distribution with exponent in (0,1). De Pril’s recursion can be generalized in the same vein. We also present an analogue of our method for the collective model with a severity distribution having mixed support.     

Box-Cox stochastic volatility models with heavy-tails and correlated errors

This paper presents a Markov chain Monte Carlo (MCMC) algorithm to estimate parameters and latent stochastic processes in the asymmetric stochastic volatility (SV) model, in which the Box-Cox transformation of the squared volatility follows an autoregressive Gaussian distribution and the marginal density of asset returns has heavy-tails. We employed the Bayes factor and the Bayesian information criterion (BIC) to examine whether the Box-Cox transformation of squared volatility is favored against the log-transformation. When applying the heavy-tailed asymmetric Box-Cox transformed SV model, three competing SV models and the t-GARCH(1,1) model to continuously compounded daily returns of the Australian stock index, we find that the Box-Cox transformation of squared volatility is strongly favored by Bayes factors and BIC against the log-transformation. While both criteria strongly favor the t-GARCH(1,1) model against the heavy-tailed asymmetric Box-Cox transformed SV model and the other three competing SV models, we find that SV models fit the data better than the t-GARCH(1,1) model based on a measure of closeness between the distribution of the fitted residuals and the distribution of the model disturbance. When our model and its competing models are applied to daily returns of another five stock indices, we find that in terms of SV models, the Box-Cox transformation of squared volatility is strongly favored against the log-transformation for the five data sets.     

Dirichlet process mixture models for insurance loss data

In the recent insurance literature, a variety of finite-dimensional parametric models have been proposed for analyzing the hump-shaped, heavy-tailed, and highly skewed loss data often encountered in applications. These parametric models are relatively simple, but they lack flexibility in the sense that an actuary analyzing a new data-set cannot be sure that any one of these parametric models will be appropriate. As a consequence, the actuary must make a non-trivial choice among a collection of candidate models, putting him/herself at risk for various model misspecification biases. In this paper, we argue that, at least in cases where prediction of future insurance losses is the ultimate goal, there is reason to consider a single but more flexible nonparametric model. We focus here on Dirichlet process mixture models, and we reanalyze several of the standard insurance data-sets to support our claim that model misspecification biases can be avoided by taking a nonparametric approach, with little to no cost, compared to existing parametric approaches.     

Who is Not Poor? Dreaming of a World Truly Free of Poverty

When the World Bank dreams of "a world free of poverty," whatshould it be dreaming? In measuring global income or consumptionexpenditure poverty, the World Bank has widely adopted the $1a day standard as a lower bound. Because this standard is basedon poverty lines in the poorest countries, anyone with incomeor expenditures below this line will truly be poor. But thereis no consensus standard for the upper bound of the global povertyline: above what level of income or expenditures is someonetruly not poor? This article proposes that the World Bank computeits lower and upper bounds in a methodologically equivalentway, using the poverty lines of the poorest countries for thelower bound and the poverty lines of the richest countries forthe upper bound. The resulting upper bound global poverty linewould be 10 times higher than the current lower bound and atleast 5 times higher than the currently used alternative lowerbound of $2 a day. And in tracking progress toward a world freeof poverty, the World Bank should compute measures of globalpoverty using a variety of weights on the depth and intensityof poverty for a range of poverty lines between the global lowerand upper bounds. For instance, rather than trying to artificiallyforce the global population of 6.2 billion (a billion is 1,000million) into just two categories "poor" and "not poor," withthe new range of poverty lines the estimates would be that 1.3billion people are "destitute" (below $1 a day), another 1.6billion are in "extreme poverty" (above $1 a day but below $2dollar a day), and another 2.5 billion are in "global poverty"(above extreme poverty but below the upper bound poverty line).