💰 quantitative finance

An Analytic COS Method for Compound Option Valuation

This paper introduces an analytic Fourier cosine (COS) method for valuing compound options that derives closed-form expressions for cosine coefficients to eliminate numerical quadrature in intermediate stages, thereby achieving superior computational efficiency and accuracy across various stochastic models and multi-stage structures.

Original authors: Zhipeng Huang, Cornelis W. Oosterlee

Published 2026-07-29

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Zhipeng Huang, Cornelis W. Oosterlee

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are a detective trying to solve a mystery that hasn't happened yet. In the world of finance, this is what happens when companies try to figure out the value of big, complicated bets called "options." Think of an option like a coupon that lets you buy something later at a fixed price. But what if that coupon itself is a coupon? That's a "compound option"—a bet on a bet. It's like buying a ticket to a raffle, where the prize is another ticket to a bigger raffle. To figure out how much that first ticket is worth, you have to guess what the second ticket will be worth, which depends on what happens in the future.

The tricky part is that the future is messy. Prices don't just move in smooth, predictable lines like a car on a highway; sometimes they jump, stutter, or get stuck in traffic jams of sudden news. Mathematicians use special tools called "characteristic functions" to describe these messy movements, kind of like a weather forecast that predicts not just rain, but also the chance of a sudden hailstorm. For a long time, calculating the value of these compound options was like trying to count every single grain of sand on a beach by hand—slow, tedious, and prone to mistakes. You had to take tiny steps, measure, add, measure, and add again, over and over.

This paper introduces a clever shortcut. The authors, Zhipeng Huang and Cornelis W. Oosterlee, have developed a new way to do these calculations using a method called the "COS method." Instead of counting every grain of sand, they found a way to write a single, perfect formula that tells you exactly how many grains there are, skipping the tedious counting entirely. They show that by using the mathematical "shape" of the future possibilities, they can calculate the value of these complex, multi-stage bets instantly and with incredible precision. This matters because it helps companies make smarter decisions about huge investments, like developing new medicines or building infrastructure, where the cost of being wrong is very high.

The Magic of the "Cosine" Shortcut

So, how does this new method actually work? Let's use an analogy. Imagine you are trying to predict the final score of a soccer tournament that has several rounds. In the old way (the "quadrature" method), to figure out the value of the team in Round 1, you would have to simulate thousands of possible games for Round 2, then thousands more for Round 3, and so on, averaging them all out. It's like watching every possible game in the universe to see who wins. It takes forever.

The authors' new method is like having a magic crystal ball that doesn't need to watch the games. Instead, it looks at the pattern of how teams usually play. The "COS method" (which stands for Fourier Cosine) is a way of describing any complex pattern using a stack of simple, wavy lines called "cosines." Think of it like describing a complex song by stacking simple musical notes on top of each other.

The big breakthrough in this paper is that the authors figured out how to stack these notes without doing the heavy lifting. Usually, when you have a "bet on a bet," you have to stop and do a lot of messy math (numerical quadrature) to figure out the next layer of the stack. The authors discovered that because the math of these waves is so regular, you can actually write down the answer for the next layer using a neat, closed-form formula. It's like realizing that instead of counting the bricks in a wall one by one, you can just multiply the height by the width because you know the pattern is perfect.

What They Found and Why It Matters

The paper demonstrates that this new "analytic" approach is not only faster but also more accurate. In their tests, they compared their method to the old, slow way of counting.

Speed: When they tested it on a standard financial model (the Black-Scholes model), their method was roughly 10 times faster than the old way while giving the same level of accuracy. In one test, the old method took about 3.5 milliseconds to get a very precise answer, while the new method did it in just 0.3 milliseconds.
Accuracy: They showed that as they added more "cosine waves" to their stack, the answer got closer and closer to the true value incredibly fast. With just 128 waves, their error was smaller than 1 in 10 billion.

But the real magic happens when things get messy. The authors tested their method on scenarios where the "future" isn't smooth. They looked at situations where "jumps" happen—sudden, big changes in value, like a drug trial suddenly succeeding or failing. They even modeled "clustered" jumps, where one piece of bad news triggers a wave of other bad news (like a domino effect).

In a simulation of a pharmaceutical company's research and development (R&D), they found that the timing of these jumps matters a lot.

If jumps are random and independent (like the old "Merton" model), the value of the project is one thing.
But if jumps are "clustered" (like their new "Q-Hawkes" model), the value changes slightly but significantly. For a two-stage drug investment, the clustered model suggested the project was worth about 47.30, while the random model said 45.57.
For a multi-stage project with five steps, the difference was even more telling. The clustered model valued the project at 6.937, compared to 6.751 for the random model.

This suggests that the structure of uncertainty—whether bad news comes in isolated bursts or in a chaotic storm—actually changes how much an investment is worth. The old methods struggled to handle these complex, jumping scenarios without getting bogged down in slow calculations. The new COS method handles them with ease, allowing companies to see how different types of risk affect their decisions.

The Bottom Line

The authors didn't just invent a faster calculator; they built a new way of thinking about nested decisions. They proved that you don't need to approximate your way through every step of a complex, multi-stage problem. By using the mathematical symmetry of the waves, they turned a slow, step-by-step grind into a single, elegant leap.

They showed that this works for everything from simple two-step bets to complex, multi-stage real-world projects like building a new drug pipeline. Whether the future is smooth or full of sudden jumps, this method provides a clear, fast, and accurate way to see what the future might be worth. It's a tool that turns the fog of uncertainty into a clear map, helping investors and companies navigate the tricky waters of compound decisions with confidence.

Technical Summary: An Analytic COS Method for Compound Option Valuation

Problem Statement

Compound options are derivatives where the underlying asset is itself an option. While Geske [Ges79] derived a closed-form solution for compound options under the Black–Scholes model, such analytical solutions are generally unavailable when the underlying asset follows more complex stochastic dynamics (e.g., jump-diffusion or Lévy processes). These complex dynamics are crucial in real-world applications like staged R&D investments, where uncertainty characteristics (such as jump risk and clustering) may vary between phases.

Mathematically, valuing compound options involves nested conditional expectations. In a multi-stage setting, the value at an intermediate decision date depends on the continuation value of the subsequent stages. Traditional numerical approaches to solve these nested expectations often rely on numerical quadrature to compute the outer cosine coefficients in Fourier-based methods. This introduces an additional layer of approximation error and computational cost, potentially obscuring the spectral convergence properties of the underlying method and limiting scalability in multi-stage problems.

Methodology

The paper proposes a fully analytic Fourier cosine (COS) method for valuing compound options. The COS method, originally introduced by Fang and Oosterlee [FO08], approximates the transition density of the log-asset price using a Fourier-cosine series, with coefficients derived directly from the characteristic function of the process.

The core innovation of this work is the derivation of closed-form expressions for the outer cosine coefficients in the compound option recursion. The methodology proceeds as follows:

Inner Stage Approximation: The value of the inner option (the underlying option at the intermediate decision time $T_1$ ) is approximated using the standard COS expansion. Due to the independent-increment assumption of the underlying log-process, this inner value $V_{inner}(x)$ can be expressed as a finite trigonometric polynomial (a sum of sines and cosines) with analytically computable coefficients.
Analytic Outer Coefficients: Instead of using numerical quadrature to compute the coefficients for the outer option (which has a payoff of $(V_{inner}(x) - K_1)^+$ ), the authors substitute the trigonometric representation of $V_{inner}(x)$ directly into the integral definition of the outer coefficients.
Closed-Form Integration: By exploiting trigonometric product identities, the resulting integrals are evaluated in closed form. This eliminates the need for numerical integration at the outer stage.
Exercise Boundary: The method requires solving a single scalar nonlinear equation to find the numerical exercise boundary $x^*$ where $V_{inner}(x^*) = K_1$ . This is efficiently solved using a safeguarded Newton method or Brent's method, leveraging the analytic derivative of the trigonometric expansion.
Extension to Multi-Stage: The framework is extended recursively to $m$ compound decision dates. At each step backward from maturity, the continuation value is represented as a trigonometric polynomial, allowing the coefficients for the previous stage to be computed analytically without quadrature.
Model Flexibility: The method applies to any model where the conditional characteristic function of the log-price increments is known in closed form, including Geometric Brownian Motion (GBM), Merton jump-diffusion, and the Queue–Hawkes (Q-Hawkes) process for clustered jumps.

Key Contributions

Fully Analytic Formulation: The paper derives the first fully analytic compound COS formulation where outer coefficients are evaluated in closed form. This removes the numerical quadrature layer entirely from the compound recursion.
Computational Efficiency: By eliminating quadrature, the method reduces the computational complexity associated with the outer coefficients. The complexity shifts from $O(n_q N)$ (where $n_q$ is the number of quadrature points) to $O(N_{in} N_{out})$ , where $N$ represents the number of cosine terms.
Preservation of Spectral Convergence: The removal of the quadrature error layer ensures that the spectral convergence properties of the COS method are preserved throughout the recursive valuation, provided the underlying transition densities satisfy regularity conditions.
Handling Heterogeneous Dynamics: The framework naturally accommodates stage-dependent dynamics. Different characteristic functions (and thus different diffusion or jump distributions) can be used for different time intervals without altering the recursive structure.
Extension to Other Derivatives: The analytic structure is shown to extend beyond standard compound calls to other contracts with single-boundary continuation payoffs, such as chooser options and Bermudan put options.

Numerical Results

The authors validate the method through several numerical experiments:

Black–Scholes Validation: Under GBM dynamics, the method reproduces Geske's closed-form solution with high accuracy. With $N=128$ cosine terms, the absolute error is below $10^{-10}$ .
Efficiency Comparison: Compared to a quadrature-based COS implementation, the analytic method achieves comparable accuracy (error $\approx 10^{-5}$ ) significantly faster. For instance, to match the accuracy of the analytic method with $N=64$ , the quadrature method required 5,120 quadrature points and took approximately an order of magnitude longer in CPU time.
Real Options with Jump Risk: The method is applied to two-stage and multi-stage pharmaceutical R&D investment problems.
- Two-Stage: The study compares GBM, Merton jump-diffusion (independent jumps), and Q-Hawkes (self-exciting clustered jumps). Results indicate that jump risk significantly impacts valuation, and the self-exciting nature of Q-Hawkes (clustering) leads to slightly higher option values than independent jumps, even when the expected number of jumps and jump-size distributions are matched.
- Multi-Stage: A four-stage R&D project is modeled where jump risk is introduced only in the final stage. The results demonstrate that anticipated late-stage jump risk propagates backward, materially affecting early-stage investment thresholds and project values. The analytic method handles these stage-dependent dynamics seamlessly.

Significance and Claims

The paper claims that the proposed analytic compound COS method provides a robust, efficient, and accurate tool for pricing nested optionality under realistic stochastic dynamics.

Elimination of Quadrature Error: The primary significance is the removal of the numerical quadrature layer, which not only speeds up computation but also prevents the introduction of additional approximation errors that can degrade the spectral convergence of the method in multi-stage settings.
Modeling Richness: The method enables the valuation of complex real-option problems with heterogeneous dynamics (e.g., switching from diffusion to jump-diffusion between stages) without the need for regression-based continuation estimation or nested simulation, which are typical in Monte Carlo approaches.
Transparency: The analytic formulation provides a transparent mathematical structure for the compound recursion, facilitating rigorous error analysis and demonstrating that the spectral convergence of the individual COS steps is preserved in the recursive compound setting.

The authors conclude that this approach is particularly valuable for multi-stage decision environments where both computational efficiency and modeling richness are essential, such as in staged R&D and infrastructure projects.