Skip to main content
main content, press tab to continue
Article

Dynamic surrender assumptions

Calibration parameters tested for the first time

By Craig Michaud | May 25, 2023

The author explores the impacts of the increase in rates on the U.S. life insurance industry and reviews some of the key issues and risks faced.
Insurance Consulting and Technology
N/A

Overview

Economic rates throughout the world have risen sharply in recent years. When considering both the magnitude and the persistency of the higher rates, we find ourselves in an economic environment unlike any experienced in decades.

This article will explore the impacts of the increase in rates on the U.S. life insurance industry and review some of the key issues and risks faced. Additionally, this article will review some common actions many U.S. life insurers are performing to assess and manage these issues and risks.

Note: The article includes a short three-question survey in the right margin, and the results will be shared with all participants.

Historical and current rate movements

The 2008 mortgage-backed securities (MBS) crisis saw a coordinated response by central banks worldwide and led to historically low rates. Central bank rates and government lending rates worldwide reached 0% and, in some instances, even went negative. However, this long period of low rates was also a continuation of a decades-long trend of rates dropping. In the U.S., rates reached historical highs in the early 1980s. Since then, they have been on a long-term downward trend. After the long period of historically low rates precipitated by the MBS financial crisis, rates have recently risen significantly in the U.S. and across the world.

Figure 1 focuses on the 10-year U.S. Treasury par rate over this extended period. The graph demonstrates the long-term trend, with a maximum value of 15.32% in the early 1980s and a low value just above 0.5% (0.62%) in 2020, before the 10-year U.S. Treasury began rising.

Fixed-income portfolios

U.S. life insurance products are fundamentally based on a promise to pay future values using funds provided either all up front or over the life of the policy. The funds provided are largely invested in fixed-income portfolios.

Many products incorporate an accumulation component to the value of the insurance. Some common examples are many deferred annuities and some life products:

  • Traditional fixed deferred annuities, such as multi-year guaranteed annuities (MYGAs) and single or flexible premium deferred annuities
  • Hybrid equity and fixed-income deferred annuities, such as fixed indexed annuities (FIAs) and registered index linked annuities
  • Universal life and indexed universal life

In all cases, the foundation of the assets to support these products are fixed-income portfolios. A common practice for the investment portfolios is to use an asset liability management (ALM) approach, which aims to match durations.

Policyholder behavior

The rate at which policyholders elect to leave their policies (or surrender) has been shown to be influenced by the state of the economy. As such, most companies incorporate a dynamic surrender function into their actuarial projection assumptions.

The basic principle of a dynamic surrender involves comparing the value to a policyholder from staying in his or her current policy to the value he or she would be able to receive from surrendering the policy and purchasing a new policy. Numerous methods are available for estimating the value that can be obtained from a new policy. Most are fundamentally built around comparing a current key crediting rate in an existing policy, where the key rate depends on the product type (e.g., crediting rate for a MYGA, option budget for an FIA) to an estimated potential similar rate that could be received for a new policy. These are often referred to as the current rate (CR) and the market rate (MR). When the MR is lower than the policy’s CR, surrenders are reduced, and conversely when the MR is higher than the CR, surrenders are increased. Insurers have often used historical data to calibrate the parameters of their dynamic surrender functions to estimate their expectations of the amount of decrease or increase. From Figure 1, we see that historical data would overall be from a period where the trend for rates had been dropping.

When estimating the MR, while numerous practices are utilized, it is very common to base the estimate on a single tenor UST plus a long-term average assumption for credit spreads, which can be achieved with new money investments. This sum may then be decreased by an average for an assumed pricing spread that competitors will try to achieve. Said another way, a typical approach to estimating MR might be based on something like:

MR = n Year UST Rate + Assumed Credit Spread – Assumed Target Pricing Spread

Note: Some companies use just MR = n Year UST Rate + a spread, and the spread used reflects the net of the credit spread and the target pricing spread, so it is the same concept.

Of course, there is a wide variation in the products sold, with varying surrender charge periods and varying liability durations. The tenor of the UST used can and will vary depending on the product and the companies’ views of their competitors, as well as the actuarial model they use. That being said, U.S. life insurers almost always incorporate a large amount of their investment portfolios with assets with tenors of five, seven and 10 years. It is therefore common for U.S. insurers to use either a five-year or 10-year UST rate to represent the “n” in their MR calculation.

In this article, while we will continue to focus on the 10-year UST, the issues we highlight will apply to other tenor UST rates as well.

To further highlight how the 10-year UST rate has changed over time, Figure 2 shows the difference in the 10-year rate on a given date to what it was three years in the past and five years in the past. The difference in the key rates (five, seven or 10) in these time periods (three or five years ago) is of particular interest to U.S. insurers to show how policies sold three and five years ago are likely to behave. It is particularly relevant as products with significant fixed-income asset portfolios will often have surrender charge periods of five, seven or 10 years. Historical data have shown that policyholders tend to be more dynamic in their behavior when penalties for doing so are zero (or close to it).

Figure 2 shows the difference of the 10-year UST rates to values three or five years prior (the two dotted lines, with the difference value shown on the right-hand y-axis).

From this graph, we can see that compared with rates three and five years ago, the differences in 10-year UST rates are currently greater than they have been at any time in the past 40 years. This is important when considering the calibration of the dynamic surrenders using historical policyholder behavior. Prior to late 2022, policies have not been in the condition where market rates are as significantly above what they were when the policy was purchased. Figure 2 shows that for policies purchased three to five years ago, current market rates are approximately 1.5% to 3.5% greater than at policy issue. In fact, as shown by the arrow in the graph, the current 10-year UST is greater than it has been for well over 10 years.

Dynamic surrender calibration

As we have discussed, companies calibrate their dynamic surrender formulas based on historical behavior. Additionally, as we have seen in figures 1 and 2, the current macroeconomic environment has led to a condition that has not been seen at any point in recent history. Given this, any experience study data would be extremely limited in its ability to support calibrating a dynamic surrender formula. Parameters used to set the results of the dynamic surrender in such an increasing rate environment have been informed estimations (guessing) instead of based on actual results. Many companies are now questioning their estimates for policyholder behavior when the MR is significantly above the CR.

ALM implications: Bringing it together

The recent liquidity issues in the banking industry, highlighted by the collapse of Silicon Valley Bank, have served to further exacerbate the concern many in the U.S. life insurance industry have around their current projection methods and dynamic assumptions.

Of course, the liability impacts cannot be thought about in isolation, as ALM practices involve using estimated liability cash flows to develop and manage asset investments. Further, best practice risk management and inforce management are predicated on accurately estimating future liability cash flows. As such, best practices often incorporate monitoring of policyholder behavior as well as rigorous actual-to-expected analysis of the results. Based on this analysis, numerous U.S. life insurers have shared with WTW that they are seeing results for surrenders below what their current dynamic surrenders are projecting. In addition, many have shared that they are revisiting this assumption, and we expect that many will also be performing updated experience studies, perhaps timed to support the common assumption unlocking period of the third quarter in 2023.

When considering updated experience studies, there are some key considerations:

  • Should the most recent experience be given more weighting/credibility, since it is the first time rates have risen so much compared with rates three, five and 10 years ago?
  • Policyholder behavior is a complicated function to model. Should data analytics be used (e.g., generalized linear modeling)?
  • What resources are required to develop updated assumption parameters in a timely fashion? Getting an update that is close to accurate quickly may be more valuable than a more accurate result that takes longer to develop.
  • What is the best balance between the number of variables versus the practicality of modeling system capabilities and the resources required to develop the updated assumption parameters in a timely fashion? Some variables to consider are listed below:
    • Policy size due to the correlation with financial sophistication of the policyholder and the potential financial freedom to optimize behavior.
    • In or out of surrender charge period (even though it may be financially beneficial to leave, many policyholders have an aversion to losing money from a surrender charge; shock surrenders will be particularly important and may be better reflected with alternative dynamic behavior parameters for the shock surrender/lapse year)
    • Attained age, as policyholders’ need for funds changes as they enter and pass through their retirement
    • Qualified versus non-qualified policies, as qualified policies are affected by the IRS required minimum distribution rules
  • Policy year is closely related to the presence of the surrender charges, and most surrender charge schedules see decreasing surrender rates as the policy ages.
  • Policies that have been in force for a long time may have a form of policyholder behavior inertia and/or lack of awareness of the policy.
  • Policies with a higher guaranteed minimum interest rate (GMIR) will tend to have stronger persistency, and a 3% ultimate GMIR from an old policy still has a higher intrinsic value than currently lower ultimate GMIRs, even if the current credited rates are equal.
  • The presence of value trapped in a rider (an embedded derivative, such as a guaranteed lifetime withdrawal benefit [GLWB] or other excess value) compared with the value received upon surrender.
    • There is significant complexity in choosing how to represent the value of the riders commonly found in insurance products, such as a GLWB (e.g., economically versus a rider’s benefit base value).
    • Capturing the value of the rider at the time of surrender in historical experience studies can be complicated.
  • What is used to represent the MR (UST + fixed spread vs. UST + current market spreads)?
  • Numerous other exogenous macroeconomic factors may affect policyholders’ needs for their funds (e.g., a banking crisis, a pandemic or inflation).
  • Will policyholder behavior be influenced by secondary impacts similar to how prepayment speed estimates are represented in the mortgage industry? For example, mortgage policyholders tend to refinance after the first time rates decrease, but that tendency drops with subsequent rate decreases.

The only certainty is uncertainty

It is difficult to predict how market rates will behave in the future. UST rates may continue to stay elevated or potentially rise further to combat inflation, or they may drop back down to support an economy that enters a recession. Despite this, using the recently evolving unique experience, the insurance industry now has an opportunity to calibrate its dynamic behavior assumptions better. Many companies are revisiting the calibration of their dynamic behavior functions. Accurate calibration of this dynamic behavior function has some key challenges, such as:

  • There is a limited amount of surrender experience for how policy holders have behaved when rates have risen this much compared to rates a few years back.
  • The initial behavior in the limited experience that does exist may not accurately reflect behavior going forward for multiple reasons such as competitors are slowly raising their current new product rates.

It is worth considering if it may be best to limit any changes made to the current assumption parameters. It will be better to incorporate a partial move in the direction of recent behavior so they do not over-correct the dynamic parameters instead of changing them to fully match the recent behavior only to change them again in the opposite direction in the near future. In all instances, we expect companies will monitor this behavior very closely for the foreseeable future.

Author


Director, Insurance Consulting and Technology

Craig is a Director in WTW’s Insurance Consulting and Technology Life practice. His experience includes product pricing, valuation, managing ALM risk, generating financial plans, and financial risk management.

email Email

Contact us