On an Optimal Stopping Problem with a Discontinuous Reward

This paper analyzes an optimal stopping problem with a discontinuous reward function motivated by variable annuity pricing, establishing conditions under which surrender occurs only at maturity and deriving new analytical representations for the value function and surrender region based on fee and surrender charge structures.

The Gist

Imagine you have bought a special "magic savings account" called a Variable Annuity. It's like a piggy bank that is invested in the stock market.

Here's the deal:

1. The Guarantee: When you decide to cash out (at a specific future date, say 15 years from now), the insurance company promises: "You will get at least $100, even if the stock market crashed and your account is worth $0." If the market did well and your account is worth $200, you get the $200.
2. The Catch (The Fee): To keep this safety net, you pay a small daily fee (like a subscription cost) taken directly out of your account balance.
3. The Escape Hatch (Surrender): You are allowed to cash out early. However, if you leave before the 15 years are up, the insurance company charges you a penalty (a surrender charge). The longer you stay, the smaller this penalty gets.

The Big Question: When should you leave?

This paper asks a very tricky question: When is it smartest for you to cash out?

You want to maximize your money. You have two choices at any moment:

• Option A: Cash out now. You get your current balance minus the penalty.
• Option B: Stay in the account. You keep hoping the market goes up, but you keep paying the daily fee.

Usually, in finance, if you have a "free option" to leave early (like an American stock option), the math is smooth and predictable. But this paper says: "Not so fast!"

The "Cliff" Problem (The Discontinuity)

The authors point out a weird, jagged edge in this problem.

• Day 14 years, 11 months, 29 days: If you cash out, you get your balance minus a penalty.
• Day 15 years (Maturity): If you cash out, you get your balance OR the $100 guarantee, whichever is higher. No penalty.

Imagine walking toward a cliff. For 99% of the journey, the ground is flat. But right at the finish line, there is a sudden, magical jump up in value. This "jump" makes the math extremely difficult. Standard financial formulas break because they assume the ground is smooth.

The Authors' Solution: The "Smooth Mirror"

The authors (Anne and Marie-Claude) found a clever trick. They realized that even though the "real" problem has a jagged cliff at the end, you can look at it through a "Smooth Mirror."

They created a fake, smooth version of the problem where the value changes gradually.

• In this smooth version, the math is easy.
• They proved that the "best time to leave" in the smooth version is almost exactly the same as the "best time to leave" in the real, jagged version.
• This allowed them to use standard, reliable math tools to solve a problem that looked unsolvable.

The "Fee vs. Penalty" Dance

The paper discovers a fascinating dance between the Fee (the cost to stay) and the Penalty (the cost to leave).

• The "Stay" Zone: If the penalty for leaving is high enough compared to the fee you pay to stay, it's never smart to leave early. You just ride it out until the end.
• The "Leave" Zone: If the fee is too high or the penalty is too low, you might want to leave early.

The Big Surprise:
In most financial problems, the "Leave Zone" is a simple, solid block. For example, "If your account drops below $50, leave."

But this paper shows that with Variable Annuities, the "Leave Zone" can be weird and disconnected.

• Imagine a map where you should leave if your account is between $50 and $100.
• But then, if your account goes up to $150, you should stay.
• But if it goes up to $200, you should leave again.

It's like a game where the rules change depending on how high you've climbed. The "Leave Zone" can have holes in it. This happens because the fee structure and the penalty structure interact in complex ways.

The "Continuation Premium" (The Value of Waiting)

The authors invented a new way to think about the value of your contract. They call it the "Continuation Premium."

Think of it like this:

• Surrender Value: What you get if you walk away today.
• Continuation Premium: The "extra value" you get just for keeping the contract alive.

This extra value comes from two things:

1. The Safety Net: The chance that the market crashes, but the $100 guarantee saves you.
2. The Option to Wait: The chance that the market will boom later, and you can cash out then with a smaller penalty.

The paper shows that if the "Continuation Premium" is positive, you should stay. If it drops to zero, you should leave.

Why Does This Matter?

1. For Insurance Companies: They need to know exactly when people will quit so they can manage their risk. If they think people will stay, but the math says they will leave, the company could lose money.
2. For You (The Policyholder): It helps understand that sometimes, even if you have a lot of money in the account, it might be smart to leave because the fees are eating you alive. Conversely, sometimes it's smart to stay even if the account is low, because the guarantee is valuable.
3. For Mathematicians: They solved a puzzle that standard tools couldn't crack by realizing that a "jagged" problem could be solved by looking at a "smooth" reflection of it.

In a Nutshell

This paper is about figuring out the perfect time to quit a savings plan that has a safety net but charges a daily fee. The authors discovered that the answer isn't always a simple "stay" or "go." Sometimes, the best strategy is a zig-zag pattern that depends entirely on how the daily fees and the exit penalties are set up. They solved this by inventing a new mathematical lens that smooths out the jagged edges of the problem.

Technical Summary

Here is a detailed technical summary of the paper "On an Optimal Stopping Problem with a Discontinuous Reward" by Anne MacKay and Marie-Claude Vachon.

1. Problem Statement

The paper addresses the valuation of a Variable Annuity (VA) contract with a Guaranteed Minimum Maturity Benefit (GMMB). The core challenge is modeling the policyholder's optimal surrender strategy under the assumption that they act rationally to maximize the risk-neutral value of the contract.

• The Financial Setting: The underlying investment account $F_t$ follows a geometric Brownian motion under the risk-neutral measure $Q$, subject to a time- and state-dependent fee rate $c(t, F_t)$.
• The Reward Function: The contract offers a payoff at maturity $T$ of $\max(G, F_T)$, where $G$ is the guaranteed amount. If the policyholder surrenders at time $t < T$, they receive $g(t, F_t)F_t$, where $g$ is a surrender charge function (typically $1 - \text{penalty}$).
• The Discontinuity: The reward function $\phi(t, x)$ is defined as:
  $$\phi(t, x) = \begin{cases} g(t, x)x & \text{if } t < T \\ \max(G, x) & \text{if } t = T \end{cases}$$
  Crucially, if $x < G$, $\lim_{t \to T^-} \phi(t, x) = x < G = \phi(T, x)$. This creates a discontinuity at maturity. Furthermore, the reward is unbounded (as $x \to \infty$).
• The Objective: Solve the optimal stopping problem:
  $$v(t, x) = \sup_{\tau \in \mathcal{T}_{t,T}} \mathbb{E}\left[ e^{-r(\tau-t)} \phi(\tau, F_\tau^{t,x}) \right]$$
  Standard results from American option pricing (which assume continuous, bounded rewards) are not directly applicable due to the discontinuity and unboundedness.

2. Methodology

The authors employ a rigorous probabilistic and analytical approach to overcome the limitations of standard optimal stopping theory.

• Alternative Representation: To establish the continuity of the value function $v(t, x)$, the authors introduce an alternative reward function $\tilde{\phi}(t, x) = g(t, x)x \vee h(t, x)$, where $h(t, x)$ is the expected present value of the maturity benefit. They prove that the value function for the original discontinuous problem is identical to the value function of a problem with this continuous reward function.
• Regularity Analysis: Using the continuous representation, they prove that $v(t, x)$ is continuous on $[0, T] \times \mathbb{R}^+$, convex in $x$, and Lipschitz continuous in $x$. They also establish time-regularity (Hölder continuity).
• Free Boundary Formulation: They characterize the value function as the solution to a free boundary value problem and a variational inequality. This involves defining a surrender region $S$ (where $v = \phi$) and a continuation region $C$ (where $v > \phi$).
• Integral Representations: By applying a generalized Itô's formula to the value process, they derive two integral representations for $v(t, x)$:
  1. Early Surrender Premium: Analogous to the early exercise premium in American options.
  2. Continuation Premium: A novel decomposition expressing the value as the sum of the immediate surrender value and an integral term representing the value of holding the contract.
• Condition for Trivial Solution: They derive a partial differential inequality involving the fee $c$ and surrender charge $g$ (specifically the operator $L(t)$) that determines whether early surrender is ever optimal.

3. Key Contributions

1. Handling Discontinuous Rewards: The paper provides a theoretical framework to handle optimal stopping problems with unbounded, time-dependent, and discontinuous rewards, extending the applicability of optimal stopping theory to VA pricing where standard American option results fail.
2. Continuity of the Value Function: By proving the equivalence to a continuous reward problem, the authors establish the continuity of $v(t, x)$, a prerequisite for applying free boundary techniques and generalized Itô's formulas.
3. The "Continuation Premium" Representation: The authors introduce a new decomposition of the VA value:
   $$v(t, x) = \text{Surrender Value} + \text{Continuation Premium}$$
   The continuation premium consists of the value of the maturity guarantee plus an integral term that accumulates only while the account is in the continuation region. This offers a new perspective for numerical pricing and risk analysis.
4. Characterization of the Surrender Region:
   • They identify a condition ($L(t) < 0$) under which the surrender region is non-empty and takes the form of a threshold type ($S_t = [b(t), \infty)$).
   • They prove that if $L(t) > 0$ over an interval, the surrender region is empty during that interval.
   • Novelty: They demonstrate that the surrender region can be disconnected (e.g., empty in the middle of the contract life and non-empty at the beginning and end), leading to a discontinuous optimal surrender boundary.
5. Equivalence Conditions: They provide conditions under which the original discontinuous problem is equivalent to the continuous reward problem, ensuring that numerical methods developed for continuous rewards are valid for VAs under specific fee structures.

4. Key Results

• Trivial Stopping Time: If the fee and surrender charge satisfy the inequality:
  $$g_t + (r - c + \sigma^2)x g_x + \frac{\sigma^2 x^2}{2} g_{xx} - c g \geq 0$$
  then the optimal strategy is to hold the contract until maturity ($\tau^* = T$). This effectively eliminates surrender risk.
• Structure of the Surrender Region:
  • If $L(t) < 0$ (where $L(t) = g_t - c g$ in the time-dependent case), the surrender region is non-empty and of threshold type ($x \geq b(t)$).
  • If $L(t) > 0$, the surrender region is empty ($S = \emptyset$).
  • Discontinuity: Numerical examples show that if $L(t)$ changes sign (e.g., positive in the middle of the contract due to a specific fee structure), the surrender region becomes disconnected, and the optimal surrender boundary $b(t)$ becomes discontinuous.
• Equivalence: Under the condition $L(t) < 0$, the optimal stopping times and surrender regions for the discontinuous problem and the continuous reward problem are identical.

5. Significance

• Theoretical Advancement: This work bridges a gap between standard American option theory and the specific complexities of Variable Annuity pricing. It validates the use of advanced numerical methods (like free boundary solvers) for VAs by proving the necessary regularity properties of the value function.
• Risk Management Insight: The derived condition ($L(t) < 0$ vs. $L(t) > 0$) provides insurers with a powerful analytical tool to design fee and surrender charge structures. It shows that simply having a high surrender charge is not enough; the interaction between the fee rate and the surrender charge over time determines whether early surrender incentives exist.
• Complexity of Behavior: The discovery that the surrender region can be disconnected challenges the common assumption in the literature that surrender regions are always connected intervals. This implies that policyholders might surrender early, then hold the contract, and surrender again (or vice versa) depending on the fee structure, a nuance critical for accurate liability modeling.
• New Decomposition: The "continuation premium" representation offers a fresh lens for understanding the value of the contract, separating the immediate liquidity value from the value of the guarantee and the option to wait.

In summary, the paper rigorously solves a complex optimal stopping problem arising in insurance, proving that despite the discontinuity in the reward, the value function is well-behaved, and providing new analytical tools to understand how fee structures dictate surrender behavior.