Developing Risk-Modeling Knowledge: The Analytics Approach

This section presents the components of a method that can be used to support risk-modeling knowledge. The proposed method requires the implementation of an ERKMAS. This system supports the interdependent dynamic among the movements between tacit and explicit risk-modeling knowledge in the development of a RISKMAN:

  • ? Socialization: social interaction among the RM employees and shared riskmodeling experience
  • ? Combination: merging, categorizing, reclassifying, and synthesizing the riskmodeling process
  • ? Externalization: articulation of best practices and lessons learned in the riskmodeling process
  • ? Internalization: learning and understanding from discussions and mathematical modeling review

Figure 2.1 shows risk-modeling knowledge as a common area among RM processes in the organization. All these RM processes create models to describe the phenomenon. There is the possibility of using similar knowledge, techniques, and tools to solve different risk problems. Figure 2.2 shows that in three main components of ERM there are links among data, the search for problem solutions, and

Different ERM processes have risk-modeling knowledge as a common process

Figure 2.1 Different ERM processes have risk-modeling knowledge as a common process.

KM acts through risk modeling in different components of ERM processes

Figure 2.2 KM acts through risk modeling in different components of ERM processes.

policies, and the organization of outcomes such as risk-modeling knowledge has been conceived. Based on these previous points, the proposed method looks to use the context and experience to improve the risk-modeling process. Thus, the methodology proposed for developing risk-modeling knowledge is composed of the following seven steps:

  • 1. Answering questions related to strategy and strategic planning in the organization
  • 2. Identifying the enablers for the transfer of risk knowledge from tacit to explicit knowledge and vice versa
  • 3. Understanding flows of information to produce knowledge
  • 4. Understanding risk knowledge organization
  • 5. Searching for KM technologies and techniques
  • 6. Designing the ERKMAS to support risk modeling
  • 7. Connecting organizational performance metrics and the risk-modeling process

The first step in the method is to get answers to questions such as what resources does the organization need? What does the organization want? What does the organization measure? What is the impact? What has been the experience? What are the errors? Where is the failure based on lack of knowledge management?

This step is based on the needs of the stakeholders, their value definition, and the strategy planning process. The risk-modeling process is in agreement with the “design approach to planning” introduced by Ackoff (1981) following his five proposed phases: formulating the systems of problems; identifying ideals, objectives, and goals; means planning; resource planning; and design of implementation and control. In summary, the risk-modeling process starts with the recognition of the strategic context and the contribution that it will make to the strategic process.

Based on this strategic orientation, the second, third, and fourth steps are associated with the understanding that individual minds and knowledge creation require three elements in order to discover KM and ERM process relationships: identification of the ways to transfer tacit to explicit knowledge and vice versa (Nonaka and Takeuchi 1995); clarity about the flows of information and how they produce knowledge (Choo 1998; Weick 2001); and understanding of the way that the risk knowledge is organized (Wiig 1993).

The second step refers to enabler analysis of risk knowledge transfer, studying traps, errors, and constraints of the process. Transferring risk knowledge in both directions, tacit to explicit and vice versa, starts with the identification of traps in the decision-making process (Hammond et al. 2006) affected by the modeling process. The risk-modeling process needs the understanding of risk knowledge transfer to tune up people’s efforts and to reduce the wrong application and interpretation of concepts, relationships, and results. These traps are the following (Hammond et al. 2006):

  • ? “The mind gives disproportionate weight to the first information it receives.”
  • ? “Decision makers display, for example, a strong bias toward alternatives that perpetuate the status quo.”
  • ? “ ... is to make choices in a way that justifies past choices ...”
  • ? “The bias leads us to seek out information that supports our existing instinct or point of view while avoiding information that contradicts it.”
  • ? “The way a problem is framed can profoundly influence the choices you make.”
  • ? “While managers continually make such estimates and forecasts, they rarely get clear feedback about their accuracy.”

At the same time, it is necessary to clarify whether the models will be used for decision automation or for getting insight to problems only. For example, the automation of quantitative solutions in a trading operation can produce issues in a market if everyone is using the same strategy. For instance, “the report suggests that many of the quantitative portfolio construction techniques are based on the same historical data, such as value premium, size premium, earnings surprise, etc., and that there is a widespread use of standardized factor risk models that would explain why quant funds act in unison” (Avery 2007).

Additionally, the risk-modeling process can be affected by the process of transferring knowledge that produces risky exposure. The reason is that a lack of coordination of knowledge processes, according to some examples from financial practice, can influence losses:

  • ? Expansion: Growth affected the operations at American Express. Expansion ran faster than growth of capacity. The knowledge support was minimal (Simmons 1999).
  • ? Culture: The Banker Trust expansion reduced the quality of the product presentation to the clients. The reason was cultural pressures. There was a lack of information flow, and the products were not well understood. The culture of avoiding bad news reduced the possibility of finding solutions to errors (Simmons 1999).
  • ? Controls: Barings Bank’s failure is related to the creation of early warning systems and the relationship to a work environment of rewards and recognition. A short-term performance view and internal competition contributed to the bad results (Simmons 1999).
  • ? Lack of understanding: What is happening, the complexity increment, transaction creation, lack of control, information management, and cost as the only important factors to manage, reducing the ability to react in difficult or good opportunity times. This complexity and the cost of knowledge show the need for managing the understanding and use of information rather than the information itself (Sutcliffe and Weber 2003).
  • ? Lack of communication of business values in an understandable way, which people can embrace. Possibly the identification of off-limits actions was not clear (Simmons 1999).
  • ? Reduced stimulation of a learning system in order to review processes and to discuss the results and adequate diagnostic control systems (Simmons 1999).

Finally, it is important to take into consideration that there are additional factors to add to the above list affecting the coordination of knowledge: new and different worker mentalities open to new technology and with different means of communication, the existence of a culture of knowledge silos, and a greater desire for understanding and resolving doubts. There are new problems with a higher degree of complexity and demanding the transformation of organizations in order to achieve solutions that require enterprise-wide answers with the appropriate technological support.

The third step relates to the understanding of flows of information to produce knowledge and how to use these flows in risk modeling. This means analyzing experiences of KM processes, methods, and technologies used in RM problems in order to develop risk knowledge management capacity. Some examples of the search for KM support in order to improve risk-modeling knowledge are the following:

? Application of prediction and classification models (Burstein et al. 2002), such as financial service technology and knowledge development of the organization

  • ? Data mining practice as a means to support the customer focus; risk classification and loss estimation (Hormozi and Giles 2004; Dzinkowski 2002); the emphasis put on cost of integrating risk analyses, control, and risk policy creation, deployment, and application (Cumming and Hirtle 2001)
  • ? The emphasis on acquiring knowledge and problem solving or on increasing the orientation to people and processes (Edwards et al. 2003)
  • ? The search for a solution of sliced risk management data (McKibben 2004), the development of solutions to control risk exposure, and data structures to share them with different areas in the problem-solving process
  • ? Orientation to new technology for data and information management and for the modeling process (Shaw 2005)

The assumptions behind decisions in hedging or investment can be different; the lack of sharing can create issues in RM processes, and the controls may not be enough. It is important to search for the truth outside the silo generated by risk analysis in order to get better answers. Lack of knowledge access can create failures. Weak means for transferring knowledge can provide insufficient knowledge of the operation, poor assessments of the lessons learned, and poor understanding of the present and the forecasts made through risk knowledge.

This lack of knowledge can happen because of an interruption in the flow of information, which is a component of the modeling work that is complemented and used properly by the expert. Goovaerts et al. (1984) wrote that only incomplete information is available, and it is the actuary who decides the principles and distributions that are to be used. Information use, with interpretation and context content or, in other words, knowledge is part of the risk-modeling process as a common area of the RM processes for the analysis of market risk, operational risk, strategic risk, credit risk, and actions of risk mitigation, risk transfer, and risk capacity evaluation.

On the whole, the flow of information for risk-modeling knowledge in ERM processes is related to KM processes associated with risk assessment and risk knowledge creation. This flow of information uses data, and information follows as a method for storing and retrieving raw and created data and transferring results for knowledge applications. Figure 2.3 shows some of the examples of RM activities classified by KM processes.

The fourth step consists of understanding the organization of risk-modeling knowledge. Risk modeling requires following the mathematical modeling process. Knowledge within the risk-modeling process can be organized as a collaborative effort and as the application of knowledge from different sources and disciplines.

The organization of risk-modeling knowledge means the identification of KM processes of the mathematical modeling experience. This refers to getting clarity about what to do and what to know and what the process to build a mathematical model is. Mathematics (Aleksandrov et al. 1969) has as its characteristics abstraction, demonstration, and applications under precision and logical rigor. Abstraction

Some information flows and relationships that KM processes have with ERM processes

Figure 2.3 Some information flows and relationships that KM processes have with ERM processes.

is the search for quantitative relationships; demonstration is part of the human knowledge needed in order to get generalizations about the quantitative relationships of the members of a group.

The organization needs to develop the capacity to solve problems and to provide support for the modeling process under the following premises (Mladenic et al. 2003):

  • ? Decisions come from humans and machines. Machines use predefined decision systems and humans use some theory frameworks and decision support systems.
  • ? People as the core of all processes: Management is about knowledge, and KM is about people, processes, technology, and organizational structure.
  • ? Old computing is about what computers can do, and new computing is about what people can do. New computing is about reliability, comprehensibility, universality, and harmony with human needs.
  • ? Power comes not from having knowledge. Power comes from sharing knowledge and using it to add value and to create a sustainable competitive advantage, which is similar to thinking in KM and business processes instead of business processes and information systems.
  • ? Technology is more than software or hardware. It is the answer to the “how” for finding a solution. For example, modeling process capacity is competitive capacity (Davenport 2006).

The organization of risk-modeling knowledge can use experiences in KM practice in mathematical development, under the previous premises, with the development of communities of practice (Wenger 2000), such as the Bourbaki Group and the Cambridge Club of 1930. In both cases, the scientific work was based on common interest, and the organization was formed by volunteer members working for better development of the science. The Bourbaki Group was composed originally of Henri Cartan, Claude Chevalley, Jean Coulomb, Jean Delsarte, Jean Dieudonne, Charles Ehresmann, Rene de Possel, Szolem Mandelbrojt, and Andre Weil. The group was created in 1935 with the purpose of writing a comprehensive text on mathematics based on set theory and axiomatic foundation. There were meetings to review the content, to identify the production logic, and to decide the structures and mathematical development. The main point was to create forums for discussion and idea development based on formalism and axioms as proposed by Hilbert. The experience of communities of practice enhance RM process and, in particular, risk modeling because of the development and testing of modeling steps and result validations.

Another means of knowledge collaboration was the one described by Foster (1985), who wrote about what he called the Cambridge Club 1930: “These four men did not comprise a school since they did not appear to have a common objective, but they knew each other and influenced each other and so might be called a club.” Foster referred to Sir Arthur Eddington, Sir James Jeans, Bertrand Russell, and A. N. Whitehead, all of whom were working at Cambridge at the end of 1930. The difference with the Bourbaki Group was the regularity of meetings and the specific objective. In both cases, the knowledge transfer was fundamental to contributing to the formalization of mathematics and to the new mathematical physics that emerged with the theory of relativity, quantum theory, and the uncertainty principle. These examples of the Bourbaki Group and Cambridge Club possess the attributes of the communities of practice presented by Wenger (2000) and are clear samples of collaborative work in knowledge creation.

Finally, a powerful example of modeling collaboration has been the development of open source in our society. Access to multiple tools is an example of what can be done with high levels of quality because of knowledge sharing for knowledge creation and knowledge application. Some examples of this modeling collaboration are R, Octave, Content Management Systems, and many other software solutions that are available for several problems.

A second point of view for using KM experience is to recognize learning processes in organizations (Senge 1990). Knowledge discovery and knowledge transfer from other disciplines have been used to provide solutions for risk-modeling problems. There are examples of theories that come from general stochastic process analysis (Compound Poisson Process, Brownian Motion; Karlin and Taylor 1998) or from other observations, abstraction, and reference theories, such as fractal geometry, or symmetry analysis of nature, which represent knowledge transfer from other disciplines into risk management. The Brownian theory of motion that comes from physics is one of the bases for financial mathematics or as the application of the general stochastic process, martingales, and compound Poisson processes are also the basis for the financial models and loss distribution modeling. The symmetry study through group theory is an example of starting from the observation of geometric figures to apply concepts to many different mathematical branches and solutions to practical problems in many disciplines. In addition, discovery of risk-modeling knowledge can develop innovation and application of methods and outcomes to problem solving in other disciplines.

Additionally, learning in organizations emerges from the analysis of experience and the identification of subprocesses in a risk-modeling process; this means the recognition of steps in the building model process that facilitate the identification of tasks and subprocesses that are based on knowledge and can be oriented and used toward the solution of different problems. The identification of the subprocesses of three risk-modeling examples is based on the work of Carnap (1966), Raiffa (1968), and Leonard (1998). Carnap introduced the idea of a law in science as statements expressing regularities in the world. He identified that not all the laws are universal but are, as he called them, statistical laws. The risk-modeling process belongs to the search for statistical laws and, as Carnap said, the process starts with direct observations of facts that in RM are called claims, losses, and exposures. Additionally, these laws are used to “explain facts already known, and they are used to predict facts not yet known.”

Now, Raiffa (1968) introduced decision analysis, and he identified the value of the outcome of the models based on the relevance to a real-world problem. He said, “In these lectures I have indicated how a decision maker’s preferences for consequences, attitudes towards risk, and judgments about uncertain events can be scaled in terms of subjective utilities and probabilities and how these can be incorporated into formal analysis.” This reflection supports the review and identification of the subprocesses of risk modeling that include understanding, interpretation, and the possible application to other problems.

In order to complement Carnap’s (1966) and Raiffa’s (1968) views, Leonard’s model (1998) is used. This model, called “knowledge creating and diffusing activities,” considers a cycle in which core capabilities for shared problem solving in the present are connected to implementing and integrating, experimenting, prototyping, importing, and absorbing knowledge. The core capabilities have to reduce the core rigidity that comes from skills and knowledge, managerial systems, physical systems, and values. Subprocesses in risk-modeling processes are looking for a better knowledge development in order to use knowledge with different problems.

Based on the above ideas, the next examples (Tables 2.3 through 2.5) provide an identification of the attributes of mathematical modeling that are used for risk analysis and that share common knowledge following the steps proposed by Klugman et al. (1998). These examples are the use of the compound Poisson model for loss distribution modeling, a modeling process for risk classification, and a modeling Markov process for credit risk evaluation and behavioral prediction. From the review of these three examples, it is possible to identify four main components in each modeling process: information management, mathematical work, experimenting and prototyping, and communicating. These four components of the modeling process are presented in Tables 2.3 through 2.5 and divided into subprocesses identified as common. These subprocesses go from data gathering up to the application of the theoretical concepts to different kinds of problems.

Tables 2.3 through 2.5 show the common steps in the analytics process to solve the problem in three different applications: risk classification, estimation of the loss distribution, and the Markov process. These common steps are described as part of knowledge process development. This means the application of the knowledge gained to get new solutions, possibly with different data and relationships.

Many steps have common knowledge that needs to be aligned and to produce capacity for risk modeling. For instance, describing groups (groups can be defined as the sets of customers according to or as examples of the due date or delinquency level) clustering, selecting variables from a linear approach, and the classification process profile of payment quality. One of the steps that is central in risk analysis is using loss distribution for other applications, such as percentile analysis (VAR) or probable maximum loss. The groups of debt quality can be organized in a different way, producing a sequence of the credit process from selection to control of credit portfolio.

Table 2.3 Analytics Process and Risk Classification

Analytics Process

Risk Classification

Problem definition: understanding the phenomenon

Meaning of customer classification, attributes available, timing, groups, etc.

Search for general models, theoretical support, and mathematics

New theory for parameter estimation, testing new model decision trees, regression trees, neural networks ... GLM

Reducing the core rigidity: People coordination and project development

Expert identification, blueprint, maps, plans. Blocks and steps definition, capacity, and roles identification

Data gathering

Data experience, profile variables, default definition, claims data, exposure set definition, clustering for outliers

Data storage

Data mart creation and access, record selection, variable field selection

Data selection and preparation

Learning set, out of time, out of sample

Data for control

Selection of the samples, out of time and out of sample

Programming and specialized software

Testing assumptions, normality ..., modeling, categorization, regression process, model identification, model preparation

Prototyping: Model/program Testing definition of structure and relationships/model structure selection

Input data to different models

Parameters estimation and solutions of relationships and equations

Testing different methods

Model performance evaluation

ROC, classification tests, Kolmogorov-Smirnov

Model improvements

Identification of a different set of variables, parsimonious metrics, testing more variables

Table 2.3 (Continued) Analytics Process and Risk Classification

Analytics Process

Risk Classification

Reporting

Problem solved, scope, model specifications, results, interpretation, new challenges, and priorities

Reducing the judgment gap: Results interpretation

Meaning of classification

Communication

Presentations to different groups, taking feedback

Search for a statistical law: New generalizations and weak assumptions

Look for panel data, time series indicators

New applications, input new models

Developing variance metrics, identification new significant variables, benchmarking, development of segmented models (economic sectors, by different clusters)

Generalization of a risk model can depend on the assumptions, theory, time, and data available. For example, time is a factor affecting whether the model is discrete or continuous in time. This has a big impact in RM modeling. In all these steps, knowledge is a component to be organized and promoted in order to achieve answers and to identify how to improve assumptions and methods.

The capacity for risk-modeling development is grounded in how people learn to work both independently and simultaneously with others, coordinately and looking at the forest and not just the trees. The challenge is that organizations need to coordinate resources under the understanding that they have knowledge workers and problem solvers and not just workers and jobs. Furthermore, organizations have problems to solve and not just tasks to do, which implies that organizations have managers and analytics professionals coexisting, which requires the organization of risk-modeling knowledge and the ability to use common knowledge in mathematical modeling.

The fifth step refers to identifying the KM processes and how to support the use of common knowledge of RM processes in mathematical modeling. The riskmodeling process needs to be supported in order to mobilize information flows to produce risk knowledge through a clear role of KM processes as a means to consolidate, integrate, and organize the risk knowledge. This can happen in the following way: knowledge creation with which knowledge is represented by risk assessment, knowledge storage and retrieval through the data support for external and internal users, knowledge transfer using the experience of many people in known cases, and

Table 2.4 Analytics Process and Loss Distribution Fitting

Analytics Process

Loss Distribution Fitting: Compound Poisson Process

Problem definition: Understanding the phenomenon

Concept of loss, claim process, cost associated, income associated, reinsurance, recoveries

Search for general models, theoretical support, mathematics

New approaches for numerical and analytical solutions of the stochastic processes

Reducing the core rigidity: People coordination and project development

Expert identification, blueprint, maps, plans. Blocks and steps definition, capacity and roles identification

Data gathering

Claims data, recoveries, reinsurance, investment, clustering outliers identification

Data store

Data mart creation and access, record selection, variable field selection

Data selection and preparation

Different periods of time, simulation points, empirical distribution, descriptive statistics

Data for control

Different periods of time, simulation points, filters

Programming and specialized software

Histograms, ways to estimate parameters, distributions simple and mixed. Tail analysis

Prototyping: Model and program, testing definition of structure and relationships, model structure selection

Input data to different models

Parameters estimation and solutions of relationships and equations

Testing different methods

Model performance evaluation

Fit tests, Kolmogorov-Smirnov, chi-square

Model improvements

Identification of special cases

Reporting

Problem solved, scope, model specifications, results, interpretation, new challenges, and priorities

Table 2.4 (Continued) Analytics Process and Loss Distribution Fitting

Analytics Process

Loss Distribution Fitting: Compound Poisson Process

Reducing the judgment gap: Results interpretation

Meaning of loss distribution and applications

Communication

Presentations to different groups, taking feedback, developing new options

Search for a statistical law: New generalizations and weak assumptions

Mixed distribution and special groups for managing claim. Relationships with marketing, pricing

New applications and input new models

Loss given default, behavioral models, preventive dashboards, risk indicators management

knowledge application to discover business process opportunities. These KM processes in a risk-modeling process are described in detail in the following sections.

 
Source
< Prev   CONTENTS   Source   Next >