Article from: The Actuary Magazine

February/March 2015 – Volume 12, Issue 1

Controlling Our Behavior

20 | THE ACTUARY | FEBRUARY/MARCH 2015 Controlling Our Behavior

FEBRUARY/MARCH 2015 | THE ACTUARY | 21 MANY MANAGERS ARE STRUGGLING WITH THE DAUNTING TASK OF DESIGNING A CONTROL ENVIRONMENT FOR THE ACTUARIAL ASPECTS OF AN INSURANCE COMPANY. BY MARK GRIFFIN

ehavioral finance (sometimes Prize for his work in behavioral finance. Other more overconfident one becomes. referred to as behavioral ) authors in this field include Robert Shiller, – We draw unfounded conclusions B represents a synthesis between Hersh Shefrin and . In 1995, from small samples. psychology, neurology and anthropology. the Association for Investment Management – Experts see things in a much more It demonstrates consistent economic biases and Research (AIMR) published “Behavioral differentiated form, and they may and blind spots in the behavior of individuals Finance and Decision Theory in Investment overlook other perspectives. and groups. Management.” • Loss aversion: Investors are overly Insurance company managements are Their work and that of others has shown that reluctant to sell their loss-making in the process of building the “control behavioral finance demonstrates a number of positions because that would force environment” for financial reporting decision-making traps: them to admit they had made a envisioned by the Sarbanes-Oxley Act mistake. Instead, they hold on to these of 2002 and the 2013 COSO “Internal • Endowment effect: We value the things positions longer than positions with Control—Integrated Framework.” The risk we have, and the things we have gains. management, actuarial, compliance and invested time in, more than we should. internal audit areas within an insurance ASSUMPTION RESETTING EXAMPLE The confirmation bias suggests that those who Consider the following example. An have the initial assumption as their “anchor” may insurance company enters the simplified issue term insurance market. A mortality table is not take the new information seriously or apply a carefully chosen. A couple of years into the disciplined process to its arrival. new venture, the quarterly financials of the company contain a poor mortality result for company (sometimes referred to as • Anchoring: People don’t make this line of business. In preparing to describe the control functions), together with sufficient adjustments from an initial the results to the board of directors and the finance function, will be critical in “anchor,” and give disproportionate outside investors, the finance department building this environment. This article will weight to the first information they “asks” the actuarial department if this result help the reader understand behavioral receive. In Kahneman’s words, this is is merely noise or evidence that the mortality finance and its relevance in identifying “one of the most reliable and robust assumption is too aggressive. analytics and processes suitable for a results of experimental psychology.” control environment. Three examples In this case, for the actuaries involved in are provided. More broadly, the article • Confirmation bias: We tend to look for setting the assumption, the “endowment” is will help any actuary involved in making evidence that confirms our existing their investment in the actuarial designation, and resetting assumptions to understand view and to disregard findings that their stature within the organization and the behavioral finance and thereby avoid contradict it. This is sometimes referred assumption itself. For everyone within the some common biases. to as the “status quo bias.” organization who was not involved in setting the assumption, the assumption is the first THE PITFALLS • Overconfidence: information they receive and becomes the In 2002, won a Nobel – The more expertise one has, the anchor. The confirmation bias suggests that

22 | THE ACTUARY | FEBRUARY/MARCH 2015 those who have the initial assumption as their behavioral finance, we “anchor” may not take the new information should also recognize seriously or apply a disciplined process to that, as soon as we its arrival. With respect to overconfidence, devote time and attention the simplified issue mortality assumption to making an assumption, was probably either made by, or approved we lose our objectivity with by, the ranking mortality “expert” within the respect to the possible future company. In this situation, the expert status need to reset the assumption. of the decision-maker(s) may cause the Therefore, the company’s company to be too reliant on the initial data process should not rely on the and on the initial assumption drawn from that assumption-making body or data. The prospect of changing the mortality person to raise its hand proactively assumption may trigger loss aversion within and identify an issue. the company, as the resulting reserve increase might be viewed as admitting a mistake. The NERVOUS SYSTEM temptation to hold off on the assumption In the December 2014/January 2015 change and pray silently for future reversion edition of The Actuary, a “nervous may be too strong. system” for managing insurance product assumptions was proposed. Standard In the example, there is a possibility that deviations are calculated to differentiate the company will not react promptly to between variation in insurance product experience that is more than noise. Delayed experience that represents noise and applied, not just recognition increases the magnitude of the variation that indeed is an early warning on to assumptions eventual corrective action and may result in the need to change product assumptions. As such as mortality, the capital markets losing overall confidence an example, variation in experience beyond withdrawals and morbidity, in the company’s financials. one standard deviation would automatically but also to other profitability mandate an assumption review, and drivers such as the distribution of In his best-seller, Emotional Intelligence: Why experience beyond two standard deviations business by age, policy size, etc. The It Can Matter More Than IQ, Daniel Goleman would mandate a revision. nervous system can form an important part describes how initial reactions to unexpected of a model governance policy within a events may be emotional rather than rational. The nervous system can be applied control environment, as described earlier. The emergence of negative experience with consistently within an organization as well The analytics can be made available to, respect to an insurance product may constitute as across life, health, property and casualty, and are easily understood by, the actuarial, such an unexpected event. Even apart from and pension assumptions. It should be risk, finance and internal audit teams. While “control environment” considerations, the opportunity to replace what may be a rather awkward, emotional situation with a consistent Reinsurance process should be welcomed. THE RENEWAL PROCESS WITHIN A REINSURER is subject to the same All actuaries must realize that, having made challenges listed in the direct simplified issue example. Applying consistent a large personal investment in actuarial “review and revise” triggers based on the original assumptions at the point of training, we consciously or subconsciously renewal will ensure a disciplined approach and represent an effective control. feel empowered and entitled to make assumptions. Based on the findings of

FEBRUARY/MARCH 2015 | THE ACTUARY | 23 process to review and revise assumptions is AN OUTSIDE VIEW echoed by the Federal Reserve’s “Supervisory The example of the nervous system is a Guidance on Model Risk Management,”1 specific combination of processes and analytics that helps address the decision- Validation also can reveal making biases listed earlier. The second deterioration in model performance example, the outside review, can be thought over time and can set thresholds of as a broad principle. It addresses the same for acceptable levels of error, set of decision-making tendencies as the through analysis of the distribution nervous system. of outcomes around expected or predicted values. … An outsider is not as “endowed with,” or “anchored to,” the original assumption, and The objective of the [back-testing] should be able to look more objectively analysis is to determine whether on the assumption and on new evidence. differences stem from the omission of The overconfidence finding suggests that material factors from the model … or the outside view is necessary and it may be whether they are purely random and best if it comes from someone who is less thus consistent with acceptable model of an expert on the topic, and therefore less performance. bound to traditional approaches.

The traps listed earlier obviously all involve Kahneman devotes a chapter to “The Outside behavior. The use of a transparent, consistent View” in his behavioral finance best-seller, the approach is not analytically elegant, it process such as the nervous system removes Thinking Fast and Slow. The value of an is understandable by those without deep the opportunities to fall into these traps. A outside perspective is echoed by others. statistical knowledge. Therefore, it will company using the nervous system doesn’t resonate with senior management, industry need to rely on the parties responsible for The Fed makes explicit reference to outside 2 analysts, investors, auditors, regulators, etc. setting assumptions to raise their hand and perspectives: identify the need for assumption changes, THE FED’S PERSPECTIVE and there is no need for one party within the A guiding principle for managing This same need for an objective, transparent insurance company to challenge another party. model risk is “effective challenge” of models, that is, critical analysis by objective, informed parties who can identify model limitations and Pensions assumptions and produce appropriate changes. … THE PROCESS OF SETTING PENSION PLAN ASSUMPTIONS is subject to the same behavioral tendencies as those described in the term insurance Generally, validation should be done example. The analytical and process discipline of the nervous system can by people who are not responsible be applied to assumptions such as early retirement incidence, employment for development or use and do not termination and mortality experience. It is important to discern what is have a stake in whether a model is noise and what is a genuine trend, and be prepared to change your initial determined to be valid. assumptions when warranted. One of the International Association of Insurance Supervisor’s Insurance Core

24 | THE ACTUARY | FEBRUARY/MARCH 2015 Principles (ICPs)3 touches on this as well: • Avoid the problems we cannot easily TORNADO CHARTING The control functions (other than internal solve, often by substituting a problem A tool known as a tornado chart can be used audit) should be subject to periodic we can solve; this is often referred to as to rank and communicate a company’s top internal or external review by the insurer’s substitution. risks, mitigating these behavioral inclinations. internal auditor or an external reviewer. • “Choose not to choose.” The recent To prepare a tornado chart, all assumptions best-seller, Scarcity: Why Having Too are shocked in both directions by an arbitrary When a corporate credit analyst first Little Means so Much by percentage. For each shock, the severity is decides to approve a credit, it is similar and , describes calculated (leaving all other assumptions in many ways to the actuary’s choice of how we tend to “tunnel” on the things at the best estimate) in the risk “currency” a new assumption. For all of the same that will impact us most immediately at of the institution (such as embedded value, risk-based capital (RBC), GAAP earnings). A tool known as a tornado chart can be used to Most insurance companies have the ability to rank and communicate a company’s top risks, analyze such shocks, and this type of exercise will almost certainly be familiar for insurers mitigating these behavioral inclinations. subject to the European Union’s Solvency II Directive. These stress test results are ranked reasons, the original analyst will find the expense of other, more important in order of their negative impact and might it hard to be objective on the credit if risks that pose a less immediate threat. appear as shown on page 26. it subsequently starts to weaken. An insurance company’s process for In the January 2014 Health Section Given the behavioral observations listed identifying and monitoring deteriorating newsletter, an article by John Stark above, it is important that there be no “view” credits should not rely solely on the addressed insights from behavioral imparted on the shocks to be applied, just original decision-maker raising their hand finance with respect to the buying and a calculation of the severity of the arbitrary proactively and identifying an issue. selling of health insurance. The article shocks. Hence, a better name may be the Thankfully, views and metrics on corporate referred to substitution, as well as the Murphy’s law tornado chart. Most historical credits are available in the capital markets. endowment effect and loss aversion. Also financial surprises have resulted from a included is prospect theory, which is not misestimation of frequency, not an inability to FOCUSING OUR TIME AND TOOLS addressed here. calculate severity. While the first two examples were based on the behavioral finance findings around These behavioral finance observations If historic data is available for each major risk, decision-making, the third is based on on resource allocation strongly suggest the shocks can be calibrated to represent a behavioral finance findings regarding how we the need for a disciplined process to both specific frequency. As an example, the shocks allocate our time and attention and how we broadly set the agendas of the various could represent a historical one-in-every-10- as individuals, and groups, set priorities. control functions and more narrowly year “event.” In this “historic” tornado chart, A wide range of studies have shown that we: to ensure the modeling agenda is set as in the Murphy’s law version of the tornado • Choose projects for their ease, not their objectively. The 2014 publication “Model chart, it is critical to ensure there is no “view” importance. Validation for Insurance Enterprise Risk involved in the shocks that are used. • Spend too much time on small and Capital Models,” sponsored by the decisions and not enough on big ones. Society of Actuaries, Canadian Institute of The tornado chart can be held up against • Use the information that is close at Actuaries and Casualty Actuarial Society, the risk team’s agenda and resources to hand. suggests that the reader “check whether ensure that economic risks are the prime • Are more engaged by things we like a process is in place to determine which determinant, and that people’s expertise or than by things we dislike. risks need to be modeled.” interest, or the availability of analytic tools

FEBRUARY/MARCH 2015 | THE ACTUARY | 25 and resource allocation. Many managers are Stress Test Results struggling with the daunting task of designing a control environment for the actuarial -ve +ve aspects of an insurance company (or Credit risk pension valuations). Insights from behavioral finance can provide a helpful perspective on Equity market risk processes and analytics such as a nervous system, an outside view and tornado charting Interest rate risk that help form a control environment. A

Mortality END NOTES 1 Board of Governors of the Federal Reserve System, Longevity Office of the Comptroller of the Currency, “Supervisory Guidance on Model Risk Management,” OCC 2011-12 Withdrawals (April 4, 2011), www.occ.treas.gov/news-issuances/ bulletins/2011/bulletin-2011-12a.pdf. 2 Ibid.

3  or data, has not had an undue influence. trade-offs. If one can attribute sources of International Association of Insurance Supervisors, “Insurance Core Principles, Standards, Guidance and The same biases on resource allocation return to various risks and can measure the Assessment Methodology,” Oct. 1, 2011, ICP 8.2.6. listed earlier should be considered in setting costs of mitigating the risks symmetrically the agenda for the other control functions, or asymmetrically as well as measuring Mark Griffin, FSA, CERA, is executive vice president the capital (economic or regulatory) of actuarial, internal audit and compliance. and chief risk officer at The Phoenix Companies, Inc. As an example, the tornado chart might combinations (recognizing that the risks are in Hartford, Conn. He can be reached at Mark.Griffin@ phoenixwm.com. Behavioral finance reveals a number of consistent biases in both decision-making and resource allocation.

indicate that the risk team’s plan to develop not necessarily additive), the tool can be used a state-of-the-art elliptical copula formula for risk budgeting or strategizing. within the economic capital model should be reconsidered relative to a longevity risk As time passes, changes in the impact analysis and possible risk transfer. Perhaps of shocks are a gauge of the institution’s some other risk activity can be deferred changing exposure. It is also enlightening in favor of a corporate credit deep dive over time to compare the attribution between into a sector (for example, energy) that is actual and expected results of the institution undergoing structural changes with possible to the elements and magnitudes calculated in knock-on effects across other sectors. tornado charting.

RISK, RETURN AND CAPITAL SUMMARY The tornado chart can also be an effective Behavioral finance reveals a number of framework for presenting risk/return consistent biases in both decision-making

26 | THE ACTUARY | FEBRUARY/MARCH 2015