Christian Lochmüller

Fuzzy credibility for mixing different data sources in evaluating operational risk: Modelling operational risk

After the global financial crisis of 2008 the banking sector has shown a strong interest, internationally and in Colombia, in developing models to manage and measure the risks of business processes and in particular the risk associated with the operations of an organization (operational risk). This article proposes a model based on fuzzy credibility theory in order to mix different data sources for evaluating operational risk. The results of calculating the OpVaR are compared using both, credibility theory and fuzzy credibility. It can be concluded that these results differ, when the membership of the internal data to the external data set is low. In this case the fuzzy model gives more weight to external data compared to the model that applies credibility theory.

An Integrated Inverse Adaptive Neural Fuzzy System with Monte-Carlo Sampling Method for Operational Risk Management

Abstract Operational risk refers to deficiencies in processes, systems, people or external events, which may generate losses for an organization. The Basel Committee on Banking Supervision has defined different possibilities for the measurement of operational risk, although financial institutions are allowed to develop their own models to quantify operational risk. The advanced measurement approach, which is a risk-sensitive method for measuring operational risk, is the financial institutions preferred approach, among the available ones, in the expectation of having to hold less regulatory capital for covering operational risk with this approach than with alternative approaches. The advanced measurement approach includes the loss distribution approach as one way to assess operational risk. The loss distribution approach models loss distributions for business-line-risk combinations, with the regulatory capital being calculated as the 99,9% operational value at risk, a percentile of the distribution for the next year annual loss. One of the most important issues when estimating operational value at risk is related to the structure (type of distribution) and shape (long tail) of the loss distribution. The estimation of the loss distribution, in many cases, does not allow to integrate risk management and the evolution of risk; consequently, the assessment of the effects of risk impact management on loss distribution can take a long time. For this reason, this paper proposes a flexible integrated inverse adaptive fuzzy inference model, which is characterized by a Monte-Carlo behavior, that integrates the estimation of loss distribution and different risk profiles. This new model allows to see how the management of risk of an organization can evolve over time and it effects on the loss distribution used to estimate the operational value at risk. The experimental study results, reported in this paper, show the flexibility of the model in identifying (1) the structure and shape of the fuzzy input sets that represent the frequency and severity of risk; and (2) the risk profile of an organization. Therefore, the proposed model allows organizations or financial entities to assess the evolution of their risk impact management and its effect on loss distribution and operational value at risk in real time.

A fuzzy credibility model to estimate the operational value at risk using internal and external data of risk events

Abstract Operational Risk (OpR) refers to the possibility of suffering losses resulting from inadequate or failure of processes and/or technology, inadequate behaviour of people or external events. OpR was one of the main risks that led to the 2008 global financial crisis. Limitations of the analytical models that are applied in estimating this risk surface when qualitative information, frequently associated with OpR events, is used. To determine the magnitude of OpR in financial organisations, qualitative data and also historical data from risk events can be used. Current research trends that focus on the development of analytical models, by using different databases, to estimate the Operational Value at Risk (OpVaR) still lack models based on qualitative information, risk management profiles and the ability to integrate different databases of OpR events. In this paper we present a fuzzy model to estimate the OpVaR of an organisation by working with two different databases that contain internal available data and external or observed data. The proposed model considers: (1) the intrinsic properties of the data as fuzzy sets related to the linguistic variables of the observed data (external) and the data from available databases (internal), and (2) a series of management profiles to mitigate the effect that external data usually causes in estimating the OpVaR of an organisation. The results obtained with the proposed model allow an organisation to estimate and determine the behaviour of the OpVaR over time by using different risk profiles. The integration of qualitative information, different risk profiles (ranging from weak to strong risk management), and internal and external databases contributes to the advancement of estimating the OpVaR in risk management .

Flexible inverse adaptive fuzzy inference model to identify the evolution of Operational Value at Risk for improving operational risk management

Abstract Operational risk was one of the most important risks in the 2008 global financial crisis. This is due to limitations of the applied models in explaining and estimating this type of risk from highly qualitative information related to failures in the operations of financial organizations. A review of research literature on this area indicates an increase in the development of models for the estimation of the operational value at risk. However, there is a lack of models that use qualitative information for estimating this type of risk. Motivated by this finding, we propose a Flexible Inverse Adaptive Fuzzy Inference Model that integrates both a novel Montecarlo sampling method for the linguistic input variables of frequency and severity that allow the characterization of a risk event, the impact of risk management matrices to estimate the loss distribution and the associated operational value at risk. The methodology follows a loss distribution approach as defined by Basel II. A benefit of the proposed model is that it works with highly qualitative risk data and it also connects the risk measurement (operational value at risk) with risk management, based on risk management matrices. This way, we mitigate limitations related to a lack of available operational risk event data when assessing operational risk. We evaluate the experimental results obtained through the proposed model by using the Index of Agreement indicator. The results provide a flexible loss distribution under different risk profiles or risk management matrices that explain the evolution of operational risk in real time.

Fuzzy model Takagi Sugeno with structured evolution for determining consumer credit score

The rating (score) and knowledge of the payment behavior of a client to reduce the time of granting of consumer credit is one of the requirements that financial institutions have dedicated to providing these services. For qualifying customers, these entities are based on qualitative and quantitative information of a client, making it difficult a homogeneous rating. Because of the need to reduce response times regarding the approval or rejection of a credit application is important to use models that help to analyze a real-time credit. Hence, this paper develops and analyzes based on the principles of evolutionary computation model, and the principles of a fuzzy Takagi Sugeno type model, to estimate the score in the allocation of consumer loans. To optimize learning, the proposed model is subjected to a process of evolution, based on the EVOP model, which guides learning model, based on two parameters such as: the generation parameter and the parameter mutation thus generating a structured evolution that will take the model to different states in learning. The results obtained by the proposed model allows to decrease the time of granting of consumer credit, as allowed demonstrate the sensitivity of the model against the score, according to the variation of the amount requested by a customer.