Optimal Investment for an Insurer in Two Currency Markets: A Stochastic Control Analysis

Table of Contents

1. Introduction

This paper addresses a critical gap in actuarial risk management literature: the optimal investment strategy for an insurance company operating across multiple currency markets. Traditional models often confine insurers to a single currency domain, ignoring the realities of globalized finance. The authors, Zhou and Guo, extend the classical Cramér-Lundberg surplus model into a two-currency setting, incorporating stochastic foreign exchange (FX) rate dynamics modeled by an Ornstein-Uhlenbeck (OU) process. The primary objective is to maximize the expected exponential utility of the insurer's terminal wealth, a common risk-averse criterion in finance.

2. Model Framework

2.1 Surplus Process

The insurer's surplus process $R(t)$ is modeled using the diffusion approximation of the classical Cramér-Lundberg model: $$dR(t) = c dt - d\left(\sum_{i=1}^{N(t)} Y_i\right) \approx \mu dt + \sigma_R dW_R(t)$$ where $c$ is the premium rate, $\mu$ is the drift, and $\sigma_R$ represents the volatility from the claims process, approximated by a Brownian motion $W_R(t)$.

2.2 Investment Assets

The insurer allocates its wealth between:

1. A domestic risk-free asset (e.g., government bonds) with a constant interest rate $r_d$.
2. A foreign risky asset (e.g., a foreign stock index) with a stochastic return process. The return in foreign currency is modeled as a geometric Brownian motion.

The key variable is the proportion of wealth $\pi(t)$ invested in the foreign risky asset.

2.3 Foreign Exchange Rate Dynamics

A central innovation is modeling the FX rate $S(t)$ (domestic currency per unit of foreign currency). Its instantaneous mean growth rate $\theta(t)$ follows an Ornstein-Uhlenbeck process: $$d\theta(t) = \kappa(\bar{\theta} - \theta(t))dt + \sigma_\theta dW_\theta(t)$$ $$dS(t) = S(t)[\theta(t)dt + \sigma_S dW_S(t)]$$ where $\kappa$ is the mean-reversion speed, $\bar{\theta}$ is the long-term mean, and $W_\theta(t)$, $W_S(t)$ are correlated Brownian motions. This captures the stylized fact that FX rates exhibit mean reversion and stochastic drift, influenced by factors like inflation differentials and interest rate spreads.

3. Optimization Problem

3.1 Objective Function

The insurer aims to maximize the expected exponential utility of terminal wealth $X(T)$ at time $T$: $$\sup_{\pi(\cdot)} \mathbb{E}\left[ -\frac{1}{\gamma} e^{-\gamma X(T)} \right]$$ where $\gamma > 0$ is the constant absolute risk aversion coefficient. The wealth process $X(t)$ evolves based on the surplus, investment returns, and FX conversions.

3.2 Hamilton-Jacobi-Bellman Equation

Using dynamic programming, the value function $V(t, x, \theta)$ is defined as the supremum of the expected utility from time $t$ with wealth $x$ and FX drift $\theta$. The associated HJB equation is a nonlinear partial differential equation (PDE): $$\sup_{\pi} \left\{ V_t + \mathcal{L}^{\pi} V \right\} = 0$$ with the terminal condition $V(T, x, \theta) = -\frac{1}{\gamma}e^{-\gamma x}$. Here, $\mathcal{L}^{\pi}$ is the infinitesimal generator of the controlled wealth process, incorporating terms from the surplus, asset returns, and FX dynamics.

4. Analytical Solution

4.1 Optimal Investment Strategies

The authors derive the optimal investment strategy $\pi^*(t)$ in feedback form. It is a function of the current state variables, particularly the stochastic FX drift $\theta(t)$ and the risk aversion $\gamma$. $$\pi^*(t) = \frac{1}{\gamma \sigma_S^2} \left( \theta(t) - r_d + r_f + \rho_{S\theta}\sigma_S\sigma_\theta \frac{V_\theta}{V_x} \right)$$ where $r_f$ is the foreign risk-free rate, $\rho_{S\theta}$ is the correlation between FX price and its drift, and $V_x$, $V_\theta$ are partial derivatives of the value function. The strategy consists of a myopic component (first term) and a hedging component (second term) against fluctuations in the FX drift.

4.2 Value Function

Through an ansatz method common in exponential utility problems, the value function is conjectured to have a separable form: $$V(t, x, \theta) = -\frac{1}{\gamma}\exp\left\{-\gamma x e^{r_d (T-t)} + A(t) + B(t)\theta + \frac{1}{2}C(t)\theta^2 \right\}$$ Substituting this into the HJB equation reduces the PDE to a system of ordinary differential equations (ODEs) for the functions $A(t)$, $B(t)$, and $C(t)$, which can be solved numerically or, in special cases, analytically.

5. Numerical Analysis

The paper presents a numerical analysis to illustrate the properties of the optimal strategy. Key parameters are calibrated to realistic values: $\gamma=2$, $r_d=0.03$, $r_f=0.01$, $\kappa=0.5$, $\bar{\theta}=0.02$, $\sigma_S=0.15$, $\sigma_\theta=0.05$. The analysis likely demonstrates:

1. Sensitivity to FX Drift ($\theta$): As $\theta(t)$ increases (expected appreciation of foreign currency), the optimal allocation $\pi^*(t)$ to the foreign risky asset increases.
2. Impact of Risk Aversion ($\gamma$): Higher $\gamma$ leads to a more conservative strategy, reducing the magnitude of $\pi^*(t)$.
3. Effect of Mean-Reversion ($\kappa$): A higher $\kappa$ (faster mean reversion) reduces the hedging demand component, as deviations of $\theta(t)$ from its mean are expected to be short-lived.

6. Key Insights

• Two-Currency Hedging: The optimal strategy inherently hedges currency risk. It's not just about seeking higher returns abroad but dynamically managing the exposure to the stochastic FX drift.
• Role of Stochastic Drift: Modeling the FX drift as an OU process adds a state variable. The optimal policy depends not just on the current FX rate but on the estimated underlying trend ($\theta(t)$), which is more persistent.
• Separation of Concerns: The exponential utility leads to a separation where the optimal investment amount is independent of the insurer's current wealth level, a classic result for CARA utility.
• Practical Implementation Challenge: The strategy requires continuous estimation of the unobservable process $\theta(t)$, likely using filtering techniques (e.g., Kalman filter) on observed FX rates.

7. Core Analyst Insight

Core Insight: This paper isn't just a mathematical exercise; it's a formal rebuttal to the myopic, single-currency asset-liability management (ALM) still prevalent in many insurers. By rigorously integrating a mean-reverting stochastic FX drift, Zhou and Guo expose the significant model risk embedded in assuming constant or deterministic currency trends. Their work shows that ignoring the time-varying nature of FX fundamentals (like inflation differentials, which the paper rightly highlights) leads to suboptimal capital allocation and underestimated tail risk.

Logical Flow: The logic is elegant: (1) Start with a robust insurance surplus model (Cramér-Lundberg diffusion). (2) Acknowledge the global investment reality by adding a foreign asset. (3) Crucially, reject the simplistic Geometric Brownian Motion for FX, adopting a financially sensible OU process for its drift. (4) Apply stochastic control machinery (HJB) to derive the optimal feedback law. The chain is strong, but its weakest link is the diffusion approximation of claims, which smooths over jump risk—a core insurance risk.

Strengths & Flaws: Strengths: The model's main strength is its tractability leading to closed-form insights. The separation result is powerful for communication with non-quantitative executives. Incorporating a stochastic FX drift is a meaningful step beyond models like those in Browne (1995) or Wang (2007). The connection to economic fundamentals (inflation, balance of payments) in the introduction grounds the math in reality. Flaws: The elephant in the room is the assumption of a perfectly correlated diffusion approximation for insurance claims. This negates the very jump/ruin risk insurers exist to manage, as noted in foundational texts like Asmussen & Albrecher (2010). The model also assumes frictionless trading and no constraints (like short-selling limits common for insurers), limiting immediate practical application. Compared to the machine learning-driven approaches for FX forecasting seen in recent fintech literature (e.g., using LSTMs or Transformers), the OU process, while elegant, may be too simplistic to capture complex regime-switching behaviors.

Actionable Insights: 1. For Insurer CFOs & CROs: Demand that your ALM models incorporate stochastic currency risk premiums, not just volatile spot rates. This paper provides the blueprint. 2. For Quants: Use this framework as a benchmark. The next step is to embed the core idea—hedging stochastic FX drift—into more realistic settings: with jump-diffusion surplus (à la Yang & Zhang (2005)), under regulatory constraints (Solvency II / ICS), or with multiple correlated foreign currencies. 3. For Software Vendors: The need to estimate the latent state $\theta(t)$ in real-time is a direct business case for integrating Kalman filtering or particle filtering modules into treasury and risk management systems. In essence, this paper provides a crucial theoretical upgrade. The onus is now on the industry to implement its insights within more robust, computationally advanced, and regulated frameworks.

8. Technical Details & Mathematical Framework

The complete controlled wealth process dynamics are: $$dX(t) = [X(t)r_d + \pi(t)X(t)(\theta(t) + \alpha - r_d) + \mu]dt + \pi(t)X(t)\sigma_S dW_S(t) + \sigma_R dW_R(t)$$ where $\alpha$ is the excess return of the foreign risky asset in its local currency. The correlation structure between the Brownian motions $(W_R, W_S, W_\theta)$ is crucial. Typically, one might assume $W_R$ is independent of $(W_S, W_\theta)$, while $dW_S(t)dW_\theta(t) = \rho_{S\theta}dt$.

The HJB equation becomes: $$0 = \sup_{\pi} \{ V_t + [x r_d + \pi x (\theta + \alpha - r_d) + \mu]V_x + \kappa(\bar{\theta}-\theta)V_\theta + \frac{1}{2}(\pi^2 x^2 \sigma_S^2 + \sigma_R^2)V_{xx} + \frac{1}{2}\sigma_\theta^2 V_{\theta\theta} + \pi x \rho_{S\theta}\sigma_S\sigma_\theta V_{x\theta} \}$$ The first-order condition for the supremum yields the expression for $\pi^*$ provided in Section 4.1.

9. Experimental Results & Chart Description

While the provided PDF excerpt does not contain specific figures, a standard numerical analysis for this model would likely include the following charts:

1. Optimal Allocation vs. FX Drift ($\theta$): A positively sloped line or curve showing $\pi^*$ increasing with $\theta(t)$. Different lines would represent varying levels of risk aversion ($\gamma$), with steeper slopes for lower $\gamma$.
2. Dynamic Path Simulation: A multi-panel chart showing simulated paths over time for:
   • The OU process $\theta(t)$ mean-reverting around $\bar{\theta}$.
   • The corresponding optimal investment proportion $\pi^*(t)$ reacting to changes in $\theta(t)$.
   • The resulting insurer's wealth path $X(t)$ compared to a benchmark (e.g., invest-only-domestically strategy).
3. Sensitivity to Mean-Reversion Speed ($\kappa$): A chart showing the volatility or range of $\pi^*(t)$ decreasing as $\kappa$ increases, because the hedging motive against changes in $\theta$ diminishes.

The key takeaway from such charts would be the active, state-dependent nature of the strategy, as opposed to a static strategic asset allocation.

10. Analysis Framework: A Simplified Case Study

Scenario: A Japanese non-life insurer with a surplus drift ($\mu$) of JPY 5 billion per year and volatility ($\sigma_R$) of JPY 2 billion. It considers investing in US equity ETFs (risky foreign asset).

Parameter Assumptions (Illustrative):

• JPY Risk-free rate ($r_d$): 0.1%
• USD Risk-free rate ($r_f$): 2.5%
• USD Equity excess return ($\alpha$): 4%
• Current USD/JPY drift estimate ($\theta(t)$): -1% (expecting JPY strengthening)
• FX volatility ($\sigma_S$): 12%
• Insurer's risk aversion ($\gamma$): 1.5

Framework Application:

1. Estimate State: The insurer's treasury uses a Kalman filter on recent USD/JPY data to estimate the current $\theta(t)$ as -1%.
2. Compute Myopic Demand: $(\theta + \alpha - r_d) / (\gamma \sigma_S^2) = (-0.01 + 0.04 - 0.001) / (1.5 * 0.12^2) \approx 0.029 / 0.0216 \approx 1.34$. This suggests a 134% allocation based on immediate risk-return.
3. Adjust for Hedging Demand: The hedging component (involving $V_\theta/V_x$) would likely be negative when $\theta$ is below its long-term mean (if $\bar{\theta}$ is, say, 0%), reducing the final allocation. Assume it reduces the allocation by 0.5.
4. Final Strategy: $\pi^* \approx 1.34 - 0.5 = 0.84$. The model suggests investing 84% of the investable wealth in the US equity ETF, a significant but leveraged position that accounts for the expected JPY strength.

This case highlights how the model dynamically adjusts for currency views, unlike a static 60/40 portfolio.

11. Application Outlook & Future Directions

Immediate Applications:

• Strategic Asset Allocation (SAA) for Global Insurers: This model provides a quantitative foundation for dynamic SAA frameworks that explicitly model currency risk as a stochastic drift, improving upon constant-mix strategies.
• ALM System Enhancement: Risk technology providers (e.g., Moody's Analytics, Bloomberg) can integrate this type of stochastic control logic into their ALM simulation engines for insurers.

Future Research Directions:

1. Incorporating Jumps and Ruin Probability: The most critical extension is merging this framework with a jump-diffusion or pure jump surplus process to study the impact on optimal investment and minimizing the probability of ruin, a paramount insurer objective.
2. Regulatory Constraints: Imposing constraints like no short-selling ($0 \le \pi(t) \le 1$), leverage limits, or Solvency II capital charge constraints would make the model more practical. This leads to variational inequalities and free boundary problems.
3. Machine Learning for State Estimation: Replacing the OU process with a drift process learned via recurrent neural networks (RNNs) from high-frequency economic data could capture more complex dependencies.
4. Multiple Currencies and Assets: Extending the model to a basket of $n$ foreign currencies and $m$ risky assets, leading to a high-dimensional HJB equation solvable perhaps via deep reinforcement learning methods, as explored in recent literature for portfolio optimization.
5. Empirical Validation: A comprehensive back-testing study comparing the performance of this strategy against standard benchmarks for a panel of global insurers over the last 20 years.

12. References

1. Browne, S. (1995). Optimal Investment Policies for a Firm with a Random Risk Process: Exponential Utility and Minimizing the Probability of Ruin. Mathematics of Operations Research, 20(4), 937-958.
2. Yang, H., & Zhang, L. (2005). Optimal Investment for Insurer with Jump-Diffusion Risk Process. Insurance: Mathematics and Economics, 37(3), 615-634.
3. Schmidli, H. (2002). On Minimizing the Ruin Probability by Investment and Reinsurance. The Annals of Applied Probability, 12(3), 890-907.
4. Asmussen, S., & Albrecher, H. (2010). Ruin Probabilities (2nd ed.). World Scientific.
5. Wang, N. (2007). Optimal Investment for an Insurer with Exponential Utility Preference. Insurance: Mathematics and Economics, 40(1), 77-84.
6. Bai, L., & Guo, J. (2008). Optimal Proportional Reinsurance and Investment with Multiple Risky Assets. Insurance: Mathematics and Economics, 42(3), 968-975.
7. Goodfellow, I., et al. (2014). Generative Adversarial Nets. Advances in Neural Information Processing Systems (NeurIPS), 27. (As an example of advanced ML methodology applicable to future extensions).
8. Bank for International Settlements (BIS). (2023). Triennial Central Bank Survey of Foreign Exchange and Over-the-counter (OTC) Derivatives Markets. (Authoritative source on FX market structure).