The end of a surrender charge is a maturity date

A fixed annuity is usually described as a long liability. A policyholder deposits money, we credit interest, and the money stays with us for years, often until it becomes retirement income. Much of our asset strategy rests on that description. Because the liabilities are long, we can buy long and less liquid assets, and the extra yield on those assets pays for a good part of the spread we earn.

The description is incomplete. Most fixed and indexed annuities carry a surrender charge that declines over a set period, commonly five to ten years, and then disappears. Before that date, a policyholder who leaves pays a penalty and, in many contracts, a market value adjustment that passes on part of the cost of higher rates. After it, the policyholder can leave freely. For a company with a large block sold in one year, the end of the surrender period is not a soft behavioural assumption. It is a date written into thousands of contracts at once.

The cliff we built ourselves

Between 2017 and 2019 Wrenfield sold heavily, as did most of the industry. Interest rates were low, five-year fixed annuities compared well with bank deposits, and marketing organisations steered business to whichever carrier offered the best five-year rate. We credited between 2.6 and 3.4 per cent. The assets we bought to back that business yielded about 3.8 per cent, and many would not mature for a decade or more.

When rates rose in 2022, the arithmetic turned. By 2023 new five-year contracts were crediting above 5 per cent. Policyholders whose charges were ending could move to a competitor at a much higher rate with no penalty, and their advisers had every reason to suggest it. We had $3.7 billion of policies reaching the end of their surrender periods across 2023 and 2024, and the assets behind them were worth well below book value.

None of this was unforeseeable. The dates were in our own policy files. What we lacked was the habit of treating them as maturities. Our liquidity stress test modelled surrenders as a percentage of the whole block, adjusted for interest rates, rather than as a schedule of contractual dates on which the penalty simply ended. The averages hid the cliff.

Renewal rates are a pricing decision, not a retention budget

When a large cohort comes free, a distribution team naturally wants to keep it, and it usually proposes a renewal rate close to the market's new-money rate. Our team proposed 4.9 per cent. The argument was sound on its own terms. Keeping the business avoids selling assets at a loss, and marketing organisations remember which carriers treated their clients well.

The difficulty is that a renewal rate must be paid for by the assets the company actually holds, or by the assets it buys to replace them. Ours could support about 4.1 per cent. Paying 4.9 meant one of three things: taking more credit risk on new purchases than our guidelines allowed, crediting more than the old assets earned and calling the difference an investment in retention, or charging new policyholders more so that old ones could be paid more. Each of those has been done somewhere in the industry. None of them is honest pricing.

We renewed at 4.1 per cent and planned for most of the money to leave. In the event $2.3 billion went over eighteen months. Because we had prepared a year in advance, with secured borrowing capacity and new purchases timed to mature in the outflow quarters, we met the surrenders while selling only $1.1 billion of bonds. Realised losses were $160 million, mostly absorbed in the interest maintenance reserve, which spreads interest-related gains and losses over the remaining life of the assets sold. Statutory rules allowed some temporary relief for negative balances in that reserve during this period. We used it, but I would not build a company's liquidity plan on relief that regulators granted as an exception.

The commercial price was real. Two large marketing organisations moved us down their recommended carrier lists, and our 2024 sales fell 12 per cent. Our distribution head had predicted exactly that, and he was right to say so loudly.

Ladder the cliffs when the product is designed

The lasting lesson concerns design rather than crisis management. A company that sells most of a five-year product in a single year has created a concentrated maturity, whether or not its accounts describe it that way. The cheapest moment to spread that maturity is when the product is sold, not when the charge runs out.

Since 2024 our product committee has applied three rules. No single quarter may hold more than a set share of any year's surrender-charge expiries; when sales of one duration run ahead of plan, we adjust rates on other durations to spread the dates. Our liquidity stress test treats every expiry as a contractual maturity, with a range of assumed take-up rates, and the board sees the schedule two years forward. And the assets bought for a cohort are chosen with its expiry date in view, so that some of them mature close to the day the penalty ends.

These rules cost us something. They sometimes mean pricing a popular duration below the market leaders, which loses sales in the year we do it. They also make us hold more short assets than a pure yield calculation would choose. I think those costs are the price of an annuity company knowing its own maturity profile.

There is a fairness point as well. A policyholder whose charge has ended is entitled to leave, and a company that has planned for the departure can let the money go without resentment and without asking new customers to fund old ones. The renewal rate we offered in 2023 was the rate we could honestly pay. Those who stayed were not subsidised and those who left were not penalised, which is how an annuity ought to behave.

For directors the question is easy to ask. Show me, quarter by quarter, how much of our in-force business loses its surrender protection over the next three years, and what we would have to sell if half of it left. A management team that answers quickly has made those dates part of its balance sheet. One that answers with an average lapse rate has not yet found its cliffs.