Professional Insight

Actuaries Have the Skills to Take It to the Bank(ers)

It is easy to think that what bankers and currency traders know, we actuaries cannot.

But their worlds aren’t mysterious, and basic actuarial tools can help manage financial risks, two actuaries told attendees in the session “Actuarial Approaches for Measuring and Managing Nontraditional Risks” at the CAS Annual Meeting in Honolulu in November 2019.

James McNichols, ACAS, consulting actuary at Huggins Actuarial Services, Inc., and Michael Schmitz, FCAS, principal and consulting actuary at Milliman, Inc., urged actuaries to boldly venture into banking, economic forecasting and currency risk management.

Both actuaries said that the actuarial toolkit, tweaked slightly, offers lots of opportunities for property-casualty actuaries to expand into other financial services fields.

McNichols gave two examples: a reinsurance pool that protects several firms’ currency risks and the risk that the economy will fall into a recession. Schmitz outlined how actuaries can model various aspects of mortgage risk.

To show that banking-style risks such as foreign exchange trading and economic forecasting aren’t radically different from traditional insurance risks such as personal or commercial auto, McNichols assessed them side by side in an “actuarial risk ranking.”

It can be a little hard to follow, but here is what McNichols did.

He realized, as many have, that all of these financial risks can be modeled across the three dimensions of pure risk:

  • Likelihood of an event (frequency).
  • Severity of the event once it has occurred.
  • Predictability of overall outcomes (in insurance this is aggregate claims).

He then ranked financial risks by assessing how susceptible each one is to the three types of measurement risk that challenge any financial model:

  • Model risk (that the model you have chosen is inappropriate).
  • Parameter risk (that the model could be correct but the parameters you’ve chosen such as mean and variance are incorrect).
  • Process risk (that the results are random even when you have the correct model and parameters).

By mapping this basic risk geometry, he had two key insights: Some insurance lines, like personal auto and workers’ compensation, present less actuarial risk modeling uncertainty than is exhibited by financial risks, but others, like cyberrisk and asbestos reserves, have a lot of modeling uncertainty — as much or more than banking-style risks such as mortgages and foreign exchange.

Foreign exchange risk, in particular, is relatively high in measurement risk, but this risk is concentrated in the process risk of the frequency distribution. Other dimensions of the risk are relatively easy to ascertain.

Currency risk is tricky because the vast majority of rate changes occur in a very small range. But when crazy things happen, they are really crazy, fat-tailed events. A normal distribution does OK explaining typical events but breaks down with crazy ones. Those are the events that multinational corporations fear the most because an unfortunate turn for them in the foreign exchange markets can severely damage revenue and profits.

Actuaries are experts at understanding the skewness of a distribution — and fitting a distribution with lots of small losses and fewer large losses. (An obvious example is the lognormal distribution.)

“We are experts at skew,” he said.

But currency risk needs an understanding of kurtosis — the spikiness of the distribution. McNichols considers expertise in kurtosis an easy addition to the actuarial toolkit. Why does it matter? Because a particularly spiky (leptokurtic) curve has a much fatter tail than a lognormal distribution.

In a really crazy year, the hefty premium is designed to be enough to cover any losses. Most years, a lot of the premium — say, 70% — is refunded. And the overall cost is competitive with traditional currency hedges such as options and futures, McNichols said. That insight lets him pool several different companies’ foreign exchange risk into a retrospectively rated reinsurance pool.

Next, McNichols focused on economics. In his actuarial risk rating, the forecasting of recessions was roughly on a par with modeling excess umbrella losses and could in theory be modeled by looking at frequency and severity. Frequency measures how often we have recession. Severity measures how long the recession lasts.

In most traditional actuarial analyses, frequency and severity are assumed to be independent. But we don’t know if that is a valid assumption when modeling a recession.

If both the frequency and severity can be either independent or dependent, then there are four scenarios:

  1. Both frequency and severity are independent of each other.
  2. Both are dependent on each other.
  3. One is dependent and the other independent.
  4. The other way around.

Modeling each of the scenarios, McNichols said, would provide insights to clients. He recommended a Bayesian approach — creating an initial estimate and refining it as more information becomes available — and proceeding with caution. Clearly, the model risk associated with predictions of economic activity is relatively high and thus a structured reinsurance solution can help address this unique enterprise risk management problem.

Currently, not many actuaries are modeling currency risk or predicting recessions, but a number of banking-related products use actuarial techniques. The second session speaker, Schmitz, described those products and showed how traditional techniques can model their risks, particularly if those techniques get a couple of tweaks. He gave similar information, and a bit more detail, at a second Annual Meeting talk, “Banking on Actuarial Talent: Why Actuaries Are Well-Positioned to Boost the Banking Sector.”

Banking and insurance have overlapped for years in private mortgage insurance (PMI), which protects banks and mortgage investors when a home buyer finances more than 80% of the home loan. Actuaries have been writing and pricing it for decades.

But the PMI product has some key differences from a traditional insurance product. For decades, mortgage insurers only needed loss reserves for loans that were already delinquent. No loss reserve was needed for loans that would eventually become delinquent, the PMI equivalent of IBNR.

That left significant long-tail risk, Schmitz said, which was covered by contingency reserves — a separate liability on the insurer’s balance sheet intended to protect banks when the economy falters and lots of borrowers default.

The contingency reserve didn’t work terribly well in the financial crisis. Many of the main PMI writers lost billions and some exited the business.

The financial crisis highlighted some shortcomings of rigid adherence to traditional loss development methods. In a 2010 CAS E-Forum paper, Schmitz and co-author Kyle Mrotek, FCAS, noted that traditional triangular methods need to be adjusted for changes in underwriting standards.

In the first decade of the century, mortgage underwriters relaxed their standards.

One way to observe this relaxation is through the popularity of interest-only mortgages. A traditional mortgage requires the borrower’s payment to cover the interest that has accrued on the loan and a portion of the outstanding principal. For an interest-only loan, the payment only covers the accrued interest, meaning the borrower still owes the entire principal on the day the loan comes due.

In 2002, just 1% of the riskiest borrowers had interest-only loans. By 2005 one-third had them.

It was a significant change. As they pay down loans, borrowers have an ever-increasing stake in making sure they stay current on debts. And a borrower who can’t afford to pay a little extra to cover a sliver of principal is probably skating on thin ice financially.

Analysts who ignored that underwriting trend were likely to understate the frequency of defaults in the portfolio. And they did.

In his banking sector presentation, Schmitz noted that calendar year development patterns could also be affected by economic conditions and government programs. For example, missed payments — those harbingers of defaults — could indicate an economic recession.

An example of government’s role in changing patterns occurred in 2017, when missed payments spiked, but not because borrowers were facing financial ruin. The spike was caused by the extraordinary number of people affected by Hurricanes Harvey, Irma and Maria. In a federal disaster area, homeowners are allowed to skip mortgage payments without penalty, and many do — enough to distort development patterns.

Schmitz recommended focusing on Bornhuetter-Ferguson analysis, especially when data are volatile or immature. The a priori justification should consider modifications for economic patterns and account for standard underwriting characteristics like credit score and loan-to-value ratio. The analysis should be at the most granular level possible. Econometric forecasts of frequency and severity at the individual loan level are the standard.

The lessons of the PMI crisis are worth learning for actuaries. New mortgage-related products have emerged, some driven by regulatory changes and some by the dynamics of the marketplace.

One of these is credit risk transfer. The mortgage market is dominated by government-sponsored enterprises, or GSEs, namely Freddie Mac and Fannie Mae. These companies purchase and guarantee millions of individual mortgages from banks. They create securities whose principal and interest are paid with the funds the GSEs receive from the millions of people paying off the loans the GSEs bought. The GSEs guarantee the securities against default.

Of course, that guarantee leaves the GSEs with an enormous pool of risk. In the financial crisis, defaults triggered those guarantees and cost them hundreds of billions of dollars. The federal government bailed them out at a cost of $187 billion, according to the Shadow Open Market Committee. (The GSEs paid it back eventually and the government turned a profit.)

To avoid a repeat, the GSEs now create securities and credit insurance transactions that cede the pool of risk to investors and reinsurers. Other lenders do the same thing. The process is known as credit risk transfer (CRT). When the GSEs, who sit on the back end of the mortgage market, are the ones securing a CRT, it is called a back-end CRT. For other transactions, such as PMI, it is called a front-end CRT since the loans already have credit protection before they get to the GSEs.

The financial crisis also drove an accounting change that actuaries could address: current expected credit loss.

Beginning in 2020, banks over a certain size will have to post a larger reserve: not just for loans already impaired, but for the expected losses on loans when they are first written. In insurancespeak, loans have to be reserved at ultimate loss, not just reported. Small banks were granted an extension and have until 2023 to implement the new framework.

Schmitz said that the change “transforms how banks think about credit risk.” Just as they have done for decades with PMI, actuaries can estimate the reserve associated with credit risk. In South Africa, they already do, Schmitz said. And soon, more actuaries may be seen in banks and throughout the financial marketplace — not just insurance.

As Schmitz said: “Actuaries are well-positioned to do this.”

James P. Lynch, FCAS, is chief actuary and director of research for the Insurance Information Institute. He serves on the CAS Board of Directors.