The Use of Predictive Analytics to Set Valuation Assumptions

I will be one of the speakers at session 39 of the SOA Valuation Actuary Symposium, which will cover this hot topic for annuities and other product lines.  Hope to see you there.


Super Models - SOA webcast

August 2, 2019

12:00 - 1:30pm Eastern

Register

I will be the presenter at this SOA webcast.  If you like me, Super Models, or continuing education credit, this is the webcast for you!

What is a "super model"? In this context, super models are developed based on rigorous data analytics techniques, and they provide a range of potential outcomes and financial metrics that can be used to evaluate when material changes are necessary. “Assumptions” can be extracted from super models for various applications, but the super model itself is more robust than that. It is a framework for analysis and risk management, not a point-in-time set of numbers.

This session will emphasize data visualization and communication of highly technical concepts to colleagues and non-actuarial stakeholders. There will be examples based on industry-level policyholder behavior data from the U.S. annuity market, but the concepts transcend product lines.


Ruark Releases 2019 Variable Annuity Study Results

Increasing data exposure in key areas

Ruark Consulting, LLC today released the results of its 2019 industry studies of variable annuity (VA) policyholder behavior, which include surrenders, income utilization and partial withdrawals, and annuitizations.

“Data exposures in key areas have increased considerably since last year’s studies, allowing for more detailed analysis and higher credibility of results,” said Timothy Paris, Ruark’s CEO.

Among the notable increases in exposure:

  • Nearly double the exposure in years 11 and later, including income commencement behavior after common 10-year deferral incentives for guaranteed lifetime withdrawal benefits (GLWB).
  • 12% increase in exposure for in-the-money GLWBs, following equity market declines during the fourth quarter of 2018.
  • 29% increase in exposure for guaranteed minimum income benefits (GMIB) past their waiting period.

Total data comprises 85 million years of exposure and 14 million policyholders from 24 participating companies spanning the 11-year period from 2008-2018, with $795 billion in account value as of the end of the study period.

Highlights include:

  • GLWB deferral incentives appear to be effective. Income commencement rates are low overall; about 12% in the first year and falling to about half of that in years 2-10. However, commencement rates more than double in year 11 with the expiration of common 10-year bonuses for deferring income, before falling to an ultimate rate.  The pattern for GMIB is similar, although somewhat muted. After commencement, continuation rates are over 80%.

  • Annual withdrawal frequency rates for GLWB and GMIB have continued to increase and have become more efficient with approximately 60% of recent experience at the full guaranteed income amount.

  • The effects of "moneyness" (account value relative to the guarantee base) on partial withdrawal behavior differ depending on circumstances. Income commencement rates increase when GLWBs are more in-the-money. This effect is quite pronounced after the expiration of common 10-year deferral incentives, with commencement rates ranging from low single digits to nearly 40% depending on moneyness. At all durations, when guarantees move out-of-the-money, withdrawals in excess of the maximum amount are more common, which is suggestive of policyholders taking investment gains out of the contract.
  • On contracts without GLWB or GMIB, free partial withdrawal amounts increase after the end of the surrender charge period, similar to the familiar “shock” in surrender rates.

  • Surrender rates have not returned to 2008 levels, even as strong equity market performance has boosted account values in recent years. Newer sales include more GLWBs which have strong incentives for persistency. Also, VA writers de-risked their product offerings in the wake of the 2008 financial crisis and the low interest rate environment has reduced the attractiveness of non-VA investment alternatives.

  • Three surrender regimes are evident during the study period: surrenders at the shock duration were nearly 30% at the onset of the 2008 economic crisis, and in a range of 12-16% subsequently except for 2016 when they reached their nadir below 10%. The 2016 dip is believed to be the result of uncertainty surrounding the DOL’s proposed Fiduciary Rule and other political factors.
  • Contracts with GLWB and GMIB have much lower surrender rates, and this effect is even more pronounced for those limiting their partial withdrawals to the guaranteed income amount.

  • Policyholders that take systematic withdrawals on GLWB and GMIB exhibit a select-and-ultimate effect, with very low surrenders in the first systematic withdrawal year and increasing thereafter. In the fourth systematic withdrawal year and beyond, surrender rates are comparable to those of contracts that have not taken any withdrawals.
  • When calculating relative value for GLWBs, use of a nominal moneyness basis (account value relative to the GLWB benefit base) can be deceiving, since it fails to reflect important aspects of the product’s economics. Therefore, it may be preferable in many cases to use an actuarial basis that incorporates interest and mortality rates. Surrenders exhibit a dynamic relationship to moneyness, whether measured on a nominal or actuarial basis. On a nominal basis 81% of GLWB exposure is in-the-money, whereas on an actuarial basis only 11% is in-the-money.

  • Annuitization rates for GMIBs are in the low single digits and continue to decline. “Hybrid” versions that allow partial dollar-for-dollar withdrawals have much lower rates than traditional versions which reduce the guarantee in a pro-rata fashion, especially in the first year of eligibility. Factors influencing annuitization rates include age, duration, last year of eligibility, death benefit type, contract size, and moneyness.

Detailed study results, including company-level analytics and customized behavioral assumption models calibrated to the study data, are available for purchase by participating companies.


2019 VA industry studies are coming

This year's VA industry studies of policyholder behavior (surrenders, partial withdrawals and income utilization, and GMIB annuitization) should be complete and delivered to purchasers in mid-June.

If you have not yet decided, we continue to see significant increases in exposures and credibility in key areas that are critical for assumption models:

Total exposure years: 85 million (+5.1% since last year)
GMIB (10yr) exposure years: +28.7%
GLWB in-the-money by 25% or more: +12.3%
GLWB income commencement exposure in years 11 or later: +94.5%

With all of this data in our studies, we can identify key factors of influence and interactions that are typically not evident at the individual company level, even for large blocks. Moreover, we can use this data to develop and calibrate customized behavioral models that are readily implementable with sensible factors of influence, strong goodness-of-fit, and dramatically improved predictive power in terms of A/E ratios. This is not a budget question, it is an investment in risk management with quantifiable benefits. You can check this out as part of our Premium service option.

Also, data gathering is in progress for new industry studies for payout annuities (incl. GMIB/GLWB post-annuitization, SPIA, and structured settlements, respectively), fixed deferred annuities, and the growing market of structured/indexed variable annuities.

Details and pricing are available upon request, including all of the above along with our 2019 FIA studies and triennial (2018) VA and FIA mortality studies. Please contact me with your purchase decisions or if a discussion would be helpful. We will be back in touch in a few weeks when the VA study results are available.

On behalf of our team, thanks for your ongoing support! Talk to you soon.

Tim

timothyparis@ruark.co

860.866.7786


Pictures of Super Models

Delighted to contribute this cover article to The Modeling Platform newsletter of the Society of Actuaries Modeling section.

And thanks to those of you who attended my presentation at session 08 of the Life & Annuity Symposium this week, which was in a similar vein.

Please contact me at timothyparis@ruark.co if you would like to learn more about how we use industry-level data and techniques like these to develop a cost-benefit equation for data, and quantifiably improve actuarial models and risk management.


See you at the SOA Life & Annuity Symposium!

I will be speaking at session 08 about policyholder behavior modeling and the risks of over-fitting data.  Hope to see you in sunny Florida!


Ruark releases 2019 Fixed Indexed Annuity study

Surrender rates climbing steadily

Ruark Consulting, LLC today released the results of its 2019 studies of fixed indexed annuity (FIA) policyholder behavior. The studies, which examine the factors driving surrender behavior and income utilization, were based on experience from 3.5 million policyholders spanning the period January, 2007 through September, 2018. A total of 15 FIA writers participated, comprising $240 billion in account value as of September, 2018.

“By aggregating across the industry, our studies offer even our
largest clients a way to achieve greater precision than they could by relying
only on their own data,” said Timothy Paris, Ruark’s CEO. “And as industry
experience develops, the underlying trends are becoming even clearer.”

Ruark’s FIA studies cover products both with and without a guaranteed lifetime income benefit (GLIB). GLIB exposure outside the surrender charge period increased 34% in this edition over 2018.

Study highlights include:

  • After a
    slow decline from mid-2010 through 2016, surrender rates have climbed in
    2017-18.
    Shock duration surrender rates in the most recent quarter were
    30%, a level last seen in 2009. Rates following the end of the surrender charge
    period have risen as well. The increase in surrenders correlates with market
    interest rate increases. In contrast to variable annuities, which exhibited a
    dip in 2015-16 corresponding to uncertainty surrounding the DOL’s “fiduciary
    rule”, the effect on FIA surrenders appears to have been more muted.

  • Contracts
    with a guaranteed living income benefit (GLIB) have much better persistency

    than those without. Surrender rates during the surrender charge period for
    contracts with GLIBs are less than half those of contracts without the
    guarantee. Among contracts that have begun taking income withdrawals,
    persistency is better still; shock duration rates are approximately 13%, as
    compared to 26% for contracts without GLIB.
  • Credited rates have a discernable effect on surrenders.
    We observe that contracts earning less
    than 2% exhibit sharply higher surrenders than those earning more
    . There is
    also differentiation among contracts with higher returns, although it is less
    pronounced after normalizing for living presence and utilization. As market interest rates increase, so do
    surrenders, and there is some indication that a higher credited rate tempers
    the increase.
    In contrast, equity returns are negatively correlated with
    surrenders. Taken together, these results suggest that policyholders consider their FIA contract’s performance in the context
    of fixed income investments
    rather than equity market investments.

  • The persistency of contracts with a GLIB rider
    appears insensitive to nominal moneyness, that is, the relationship of account
    value to the benefit base. However, when
    calculation of the guarantee’s relative value is performed by using an
    actuarial moneyness basis which discounts guaranteed income for interest and
    mortality rates, the picture changes.
    Using an actuarial measure, we see
    that (a) GLIB benefits are not generally as valuable to the policyholder as
    they might first appear; and (b) persistency is greater when the actuarial
    benefits are more valuable, as should be expected.

  • Annual
    GLIB benefit commencement rates are low, 7% overall in the first contract
    duration and then falling to the 2-3% range in years 3 and later.
    In contrast
    to variable annuities, which exhibit a spike following the expiration of
    benefit bonuses, FIA utilization remains consistent in later years.
    Benefit commencement rates appear largely insensitive to RMDs. Age, tax status, and contract size all
    influence commencement rates.
    When a living benefit contract does begin
    taking withdrawals, it is likely to continue in subsequent years; average
    continuation rates are 95%.

  • GLIB
    utilization increases when policies are in the money, that is, the benefit base
    exceeds the account value.
    After normalizing for age, tax, and contract
    duration, contracts that are 25% in the money or more exercise at a 10% rate.
    In contrast, when contracts with lifetime withdrawal benefits are out of the
    money, at the money, or modestly in the money, policyholders exercise at a
    base rate of about 2%
    .
  • FIA contracts typically offer the opportunity to
    take 10% of account value annually in penalty-free withdrawals, often following
    a 1-year waiting period. This is the case for contracts with and without a
    guaranteed living income benefit (GLIB) rider. Base contract withdrawals have
    been largely stable over the past decade. Behavior differs across four groups: Those
    taking the full penalty-free amount; those taking less; those taking excess;
    and those for which no penalty applies.

  • Free partial withdrawal activity on the base
    contract is influenced by age and required minimum distributions, as well as
    contract size and the presence of a waiting period for free partial withdrawals.
    Notably, withdrawal sizes spike in the year following the end of the surrender
    charge period, when all partial withdrawals become penalty-free. Average
    withdrawal sizes for contracts without GLIB jump 9 percentage points following
    the end of the surrender charge period.

Detailed study results, including company-level analytics, are available for purchase by participating companies. For further information, please contact Timothy Paris, CEO.


2019 *NEW* Ruark annuity industry studies

In response to client demand, we would like begin the process of gathering policyholder behavior and mortality data for products beyond VA and FIA, as follows:

  • Post-annuitization deferred annuity mortality, including GMIBs and GLWBs in the payout phase.
  • Immediate annuity mortality.
  • Structured settlement annuity mortality.
  • Fixed deferred annuity surrenders, income utilization, and mortality.
  • Structured/indexed variable annuity surrenders, income utilization, and mortality.

If you have not already received the data request templates, please contact us. We request your data submissions by March 31, 2019. Pending your data submissions and purchase commitments, we will circle back with you to let you know what studies we can proceed with and when.

We look forward to broadening our relationship with you through these additional studies and corresponding behavioral modeling. Please contact Tim (timothyparis@ruark.co) or any of us at Ruark with any questions.


So many presentations in the last few months...

Actuaries' Club of Hartford & Springfield meeting
Equity-Based Insurance Guarantees Conference
SOA Annual Meeting
Valuation Actuary Symposium

Thanks to all of these organizations, and I appreciated the attendee questions and discussions. Each presentation was at least a little different, but this one from the ACHS meeting is representative of the type of insights that our industry experience studies give to our clients, and how we demonstrate the value of our work + industry data in developing customized behavioral models that help insurance companies better manage policyholder behavior and mortality/longevity risks for annuities.