Actuaries Abroad

DFA or DST or ALM?
(Or Too Many Silly Acronyms)

by Kendra M. Felisky-Watson

On a last day of summer (okay, it was only June 19 and summer hadn't actually officially started, but here in England we'd already had our allotted two days of sun and didn't expect any more), 90 people attended the second Joint Seminar of the CAS and the Institute/Faculty of Actuaries. Interestingly enough, the people attending came not only from the U.K. and the U.S., but also Albania, Bermuda, Cyprus, Denmark, Germany, India, Norway, and Switzerland.

I think the seminar was very successful in that, not only did we learn something about DFA, but some interesting debate was generated throughout the day. Quite a lot of the discussion centered on the different approaches to DFA taken by U.S. and U.K. theoreticians and practitioners, which seemed to me to be the whole point of the seminar.

After a brilliant introduction by the day's chairperson, Julian Lowe, the two presidents gave quick presentations on their respective organizations. Our illustrious president, Alice Gannon, discussed the mutual recognition issue and also talked about the CAS Survey on Nontraditional Practice Areas (a.k.a., Wider Fields in U.K.-speak). Then Paul Thornton, president of the Institute of Actuaries, discussed the incredible growth in general insurance/casualty actuaries over the last twenty years and how a statutory role is being discussed with the regulators. He also touched on mutual recognition, exam structure, and the role of the Institute in India and Southeast Asia. Alice appeared to be a bit jealous of Paul's ornate chain of office draped around his neck.

Then it was on to the real topic of the day, dynamic financial analysis or dynamic solvency testing as most people in the U.K. call it. Bryan Joseph kicked off with an introduction to DFA and the regulatory and business planning uses of DFA. He discussed what an ideal solvency test required by regulators should incorporate, and how well the tests of four different regulatory environments (E.U. Capital Requirement, Lloyd's Risk-Based Capital Approach, U.S. Risk-Based Capital Requirement, and the Canadian Dynamic Capital Adequacy Testing) met his criteria. According to Bryan, these criteria should be underwriting risk, asset risk, credit risk, and operational risk, as well as actual mix of business, financial strength of reinsurers, and no double gearing (that is, of capital).

Bryan then went on to discuss how DFA can be a solvency test as well as a business-planning tool for the company. He also talked about model construction and output. Interestingly, Bryan used the banking industry's adoption of capital adequacy models as a prediction of the insurance industry's future use of DFA. Finally, he discussed the hurdles for the acceptance of DFA, primarily the "black box" problem. In discussion, Chris Daykin mentioned that the Netherlands, Finland, and some other parts of the E.U. are investigating DFA but that there were no plans for an E.U.-wide risk-based capital model.

Sholom Feldblum discussed the thought processes and decisions that were taken when his company implemented its DFA model a year and a half ago. He described the procedure for using the targeted return on capital to derive a target loss ratio by line of business for pricing purposes. Sholom then worked through his company's model for capital allocation using an option-pricing analysis to derive the expected policyholder deficit as a risk measure. He paid particular attention to describing a stochastic reserving model. Some pros of his DFA model are that it is theoretically sound and has consistent capital determination across lines of business; drawbacks are that some risk distributions are not known, along with the usual bugbear of parameter risk.

After lunch, our local "mega-economic wizard," Andrew Smith woke everybody up with an intriguing discussion of the most important step after construction of a DFA model: how to use the outputs. He started by debunking traditional methods of deciding between different strategies based on the outputs from a DFA model. He described how DFA models give distributions of the outcomes but no guidance as to how to select the best scenario. There is an overload of data, but little actual information! To illustrate his point, Andrew constructed a simple model with five scenarios leading to five different results, each with its own distribution. He then set about comparing and contrasting the different measures of risk and return and how some traditional methods would pick inappropriate strategies if there were incomplete understanding of the underlying scenarios. Andrew then introduced the concept of state price deflators (conventionally called "deflators") acting as a stochastic discount factor. Andrew concluded with a warning that you must get an answer from your DFA model that you believe to be robust and then communicate it to management.

A jet-lagged Stephen Lowe stepped up to describe his DFA model and he definitely deserves some sort of award for managing to link DFA models with cans of beer, thermodynamics, and fractals. Stephen described how there are three markets (capital, factor, and product) that are all sources of systematic risk. He described the external risks (systematic and event) and internal risks (operational). He then went on to discuss the importance of calibration of the DFA model to ensure that the results are realistic. Finally, Stephen spent some time describing how to take the results of a DFA model and translate them into information for management. Specifically, he has used his model for asset allocation, determining the debt/equity mix of capital, comparing different strategies in an underwriting cycle, determining the optimal mix of business, and comparing alternative reinsurance structures to maximize retention levels.

The final presenter of the day was Stavros Christofides with a ramble through the history of DFA, or asset liability modelling as it used to be called. It was an interesting walk from the 1960s through risk theory models in the 1970s and ALM and solvency models in the 1980s to the DFA developments in the 1990s. The early models were focused on ruin probabilities and solvency whereas the latest models are stochastic planning systems. Stavros described how the DFA process could be used to test different strategies. He also discussed the various issues with model specification and model calibration and how we must understand what drives the results. It is important to comprehend the exchange of risk for reward. Stavros concluded by warning that there remains plenty of scope for expensive failure.

The next Joint Seminar will be held in October 2001 where the topic will be International Insurance Issues. I am sure that an interesting debate will be stimulated again.