Measurement And Empirical Research On Operational Risk Of Financial Institutions

With rapidly changing external environment and increasingly fierce industrial competition, financial institutions meet more and more complicated challenges. Because operational risk exists widely in the daily operation of financial institutions, and its characteristic and controlling method is significantly different from other risks. So facing over-growing operational risk, financial supervision department and financial institutions pay more and more attention on the measures against it.Nowadays, managing and controlling techniques of operational risk have been researched among the academia and practice circle. Although there was increasing focus on it for domestic financial institutions, preliminary results were only achieved at the following aspects:characteristic and formation mechanism. Large gap still exists in operational risk measurement and methods compared with foreign peers. Control and management of operational risk start from measurement. Therefore, it is urgent issue for discussion to solve how to measure operational risk for domestic financial industry.This thesis has the following three research significances:1. Operational risk of financial institutions is same in essence. This conclusion helps to measure operational risk for all types of financial institution.Nowadays, research of measurement on commercial bank’s operational risk has been relatively deep at both home and abroad. But researches on other types of financial institution are rare. Then, whether measurement technique adapted for commercial bank can also be applied for other financial institution? Only by revealing operational risk’s essential of three-type financial institutions can deepen knowledge of operational risk, avoid risk quantification blindly and find out common measurement technique of operational risk for all financial institutions.2. Measurement accuracy of operational risk is related to whether implement of management is effective or not. Risk measurement is the important point of risk management system. It is incorrect for the financial institutions to ignore risk measurement and discuss risk management directly. After all, the choice of measurement technique and method decide the effectiveness of risk internal control mechanism and risk management. But research on operational risk starts late compared with credit risk and market risk, unified understanding about operational risk measurement has’t been formed. Meanwhile, difference between domestic research and foreign research on operational risk is huge. Therefore, improved measurement technique of operational risk has vital significance to enhance competitiveness of financial institutions.3. Measurement accuracy of operational risk is related to whether economical capital can exert its effect or not.Economical capital is equal to unexpected loss amount of risk in quantity. It is used to measure and defend the part of loss exceeding expected loss. It is the core tool of optimizing resource allocation and raising risk adjusted return on capital (RAROC). After economical capitale is calculated, conomical capital can be allocated. From this perspective, measurement accuracy of operational risk is important for financial institutions.The main research content of this thesis is:1. The first two chapter introduce background, significance, innovation of this thesis. Then aimed at laying theoretical foundation for the following chapters, a comprehensive literature review is made on operational risk measurement method, particularly on three methods:Loss Distribution Approach, Extreme Theory Approach, Bayes Method and Credibility Model.2. If there is no profound understanding about relationship with concept, characteristic and type of operational risk, quantitative model is useless. So the third chapter firstly gives the unified definition of financial institution. Then classify operational risk according to cause and business department. Finally analyze what triggers operational risk and compare the operational risk’s feature of three-type financial institution. Through the above analysis, this thesis has one conclusion:operational risk of financial institution is same in essence. Thereby, operational risk’s measurement technique suited to one type’s financial institution can also be applied to the other type’s financial institution. This chapter establishes the solid foundation for the following chapter.3. In the fourth chapter, Loss Distribution Approach (LDA) is used to measure operational risk. BCBS proposes three types’operational risk measurement methods. Among the Advanced Measurement Approach (AMA), LDA is the most frequently used method. Because LDA is based on historical loss data, measurement result is relatively objective. Under the framework of LDA, Peaks over Threshold Approach (POT) is used to measure operational risk. The tail behavior of distribution is more important for the extreme huge loss data. Because Extreme Value Theory (EVT) is targeted on fitting tail of distribution, it is publicly accepted method. But how to select threshold precisely is still one key obstacle in the practical application.In the fourth chapter, to select threshold quantitatively and more accurately, a corrected method using change point theory to locate stable state of Hill curve is proposed. Empirical research with example of commercial bank’s operational risk loss data (totaled279pieces reported by domestic and overseas media) is done to test the effectiveness of this corrected method.4. In the fifth chapter, Bayes Method is used to measure operational risk. Although this method is widely accepted in the case of small sample, it still has one difficulty in the practical application.In the fifth chapter, Pareto Distribution and Negative Binominal Distribution are used to describe severity and frequency of operational risk respectively. Then difficulty is solved by adopting MCMC simulation method with using WinBUGS software. Because three-type financial institution face the exactly same operational risk, empirical research with example of commercial bank’s operational risk loss data (totaled279pieces reported by domestic and overseas media) can test the effectiveness of this corrected model.5. In the sixth chapter, Credibility Model is used to measure operational risk. To solving the problem of deficient loss data, majority of scholars combine the internal and external loss data directly. But due to the heterogeneity of risk, this data processing method is inappropriate. So how to integrate internal and external loss data, even how predict the operational risk loss of the next year just for one individual financial institution is still the difficult problem in the practical application. In the sixth chapter, Buhlmann-Straub credibility model is used to get risk exposure of operational risk by using MCMC algorithm with the implement of WinBUGS in the event of insufficient loss data. Empirical research with example of stated-own commercial bank’s operational risk loss data (totaled125pieces reported by domestic and overseas media) is done to test the effectiveness of this method.The main viewpoint of this thesis is:1. According to cause of formation, operational risk is classified into internal factor and external factor. Internal factor is also divided into four categories: system, skill, out of control as well as fraud. Here, system category refers to the operational risk caused by imperfect system and product. Skill category refers to the operational risk caused by employees’imperfect knowledge and incorrect understanding. Out of control category refers to the operational risk caused by employees’ subjective negligence without malice. While fraud category refers to the operational risk caused by employees’subjective negligence motivated by self-interest. External factor can be divided into three categories:Macro-policy, fraud and other. Among the above category, internal fraud risk has the biggest share in quantity and amount.2. Although POT can describe the characteristic of operational risk better, it has one key obstacle:how to determine threshold. If bigger threshold is selected, fewer samples. can be used. If smaller threshold is selected, more samples that don’t belong to tail of distribution will be used. The above two kinds of circumstances will bring biased result. So how to determine threshold more accurately is still worthy of researching.3. Although Bayes Method is has better prospective, it has one obstacle: high-dimensional value computation of posteriori distribution. This process is far more complicated. So how to put this method into practice is still worthy of researching.4. Due to insufficient loss data, most scholars solve this problem by mixing internal and external loss data directly. But every financial institution has different product line, operation procedure, risk preference and risk control system. In other words, risk heterogeny exists among financial institution. And the accuracy of measurement result can be doubted with application of this data process. So how to integrate internal and external loss data more efficiently, even how to predict the next year’s operational loss for one financial institution is still worthy of researching.The innovation of this thesis is:1. Operational risk of financial institution is same in essence.The premise of operational risk measurement is collecting loss data. Aimed at standardize data collection and build up solid foundation for the latter empirical research, definition of operational risk for financial institution is made. Although product line and risk control system of each financial institution have differences, the characteristic and cause of operational risk is basically same by analyzing sources of risk (organization structure, work flow, information system as well as employee). Meanwhile, through comparing loss data of three-type financial institution, we find that they face common main cause of formation:employee. So operational risk of three-type financial institution is same in essence.2. Threshold selection of POT model based on change point theory.Based on the former research, change point theory is introduced to solve the difficulty of selecting threshold quantitively and more accurately for Peak over Threshold Theory. Threshold is located at the stable state of Hill curve. It can be selected combined maximum value of first-order differential and maximum value of second-order differential. Empirical results show that a good evaluation of capital requirement for operational risk can be obtained.3. Applying Bayes Method based on MCMC simulation.Based on the former research, MCMC simulation by applying Bayesian approach is introduced to solve the difficulty of insufficient loss data. So this thesis discusses how to conduct a Markov Chain for Negative Binominal distribution and Pareto distribution with Gibbs sampling in order to get posterior distributions of loss severity and frequency dynamically as well as capital requirement for operational risk. Then put WINBUGS software to use to estimate parameters of distributions. Thus, better evaluation of capital requirement for operational risk can be obtained compared with the maximum likelihood estimation method.4. Mixing internal and external loss data and predicting loss for one financial institution by using credibility model.Based on the former research, credibility model is introduced to solve the difficulty of mixing internal and external loss data. Due to the deficient data and heterogeneity of risk, it’s hard for financial institution to get the unbiased estimated measurement by mixing the internal and external event of loss. Buhlmann-Straub credibility model is conducted to get risk exposure of operational risk by using MCMC algorithm with the implement of WinBUGS in the event of insufficient loss data. Then the amount of loss in the next year can be predicted for the individual financial institution. Empirical results show that measuring operational risk is inappropriate by mixing loss data of financial institution with different risk features. The stochastic simulation method applied by this thesis can improve the precision of estimators.