Risk-sensitive portfolio optimization with transaction costs

Tomasz R. Bielecki, Jean‐Philippe Chancelier, Stanley R. Pliska, Agnès SulemThe Journal of Computational Finance 2004 · DOI

TL;DR

This paper tackles the complex problem of optimizing investment portfolios when facing transaction costs and a fluctuating interest rate. They introduce a novel computational framework based on risk-sensitive control theory and a sophisticated Howard algorithm to derive optimal trading strategies, offering valuable insights into investor behavior under realistic market conditions.

What Problem Does This Paper Address?

Managing an investment portfolio optimally is a long-standing challenge in finance. Traditional models often simplify reality by assuming no transaction costs or constant market conditions. This paper addresses a more realistic scenario: an investor trading two assets (a bank account and a risky security) with transaction costs (a fixed percentage of the transaction value) and a stochastic interest rate (modeled by a Vasicek process). The core difficulty lies in finding an optimal trading strategy that balances maximizing returns with managing risk, especially when the interest rate is unpredictable and trading incurs costs. Previous work has either simplified the problem by ignoring transaction costs, assuming constant interest rates, or focused on simpler risk measures. This paper aims to bridge this gap by providing a computational solution for a more complex, yet realistic, model.

Methods

The authors employ a sophisticated approach rooted in risk-sensitive impulsive control theory. They formulate the problem as finding an optimal trading strategy that maximizes a risk-sensitive performance criterion, which is essentially a trade-off between portfolio growth rate and volatility. This leads to a non-linear quasi-variational inequality (RSQVI). To tackle this, they transform the RSQVI into an equivalent ergodic Isaac-Hamilton-Jacobi equation associated with a min-max problem.

The key innovation lies in their computational algorithm. They develop an extended two-stage policy iteration algorithm, a variant of the Howard algorithm, specifically designed for these min-max problems. This algorithm involves:

  • Localisation: Discretizing the state space into a compact rectangle and using reflecting boundary conditions.
  • Finite-difference approximation: Approximating the continuous-time operators with discrete-time equivalents.
  • Linearization: Introducing an auxiliary control variable to linearize the RSQVI, making it amenable to numerical solution.
  • Howard Algorithm: A double-stage algorithm with major and minor iterations to iteratively find the optimal strategy and value function.

Datasets

Not discussed in the paper. The paper focuses on a theoretical framework and numerical simulations based on parameter choices.

Key Insights and Ideas

  • Risk-Sensitive Criterion as a Practical Objective: The paper highlights that the risk-sensitive criterion, while mathematically derived, offers a practical interpretation of balancing expected return and volatility, aligning with real-world investment objectives.
  • Transformation Simplifies Complexity: A crucial step is the change of variables that reduces the number of state variables from five to two, significantly simplifying the problem's dimensionality.
  • No-Trade Regions are Key: The optimal strategy is characterized by "no-trade regions" where the investor refrains from trading. Transactions only occur when the portfolio's risky fraction hits specific boundaries.
  • Interest Rate's Influence: The stochastic interest rate (factor) directly impacts the risky asset's appreciation rate, and its level influences the boundaries of the no-trade regions. Lower interest rates tend to make the risky asset more attractive.
  • Risk Aversion's Impact: As investors become more risk-averse (higher $\theta$), they tend to hold more of the less risky asset, shifting the no-trade region towards lower proportions of the risky asset.
  • Transaction Costs Shape Strategy: Higher transaction costs lead to larger no-trade regions, as the cost of rebalancing outweighs the potential gains from minor adjustments.

Key Results

The paper successfully implements a computational algorithm to solve the risk-sensitive portfolio optimization problem with transaction costs and a stochastic interest rate. They provide numerical results that:

  • Validate the Algorithm: The algorithm is shown to be efficient, converging in a few iterations.
  • Replicate Prior Work: They successfully reproduce results from a simpler, zero-factor case studied by Bielecki and Pliska [4], serving as a benchmark.
  • Illustrate the One-Factor Case: They present detailed numerical examples for the one-factor Vasicek interest rate model, showing typical optimal strategies with buy, sell, and no-transaction regions.
  • Demonstrate Sensitivity Analysis: They analyze how the optimal strategy changes with variations in key parameters:
  • Risk Aversion ($\theta$): Increased risk aversion leads to a larger no-transaction region.
  • Transaction Cost ($\alpha$): Higher transaction costs also expand the no-transaction region.
  • Vanishing Transaction Cost Limit: They show that as transaction costs approach zero, the optimal policy converges to the solution of the no-transaction cost problem.

Limitations and Future Work

  • Computational Challenges: The paper acknowledges that solving complex risk-sensitive quasi-variational inequalities presents significant computational challenges, which their algorithm aims to overcome.
  • Model Simplifications: While more realistic than previous models, the paper still relies on specific assumptions about the market dynamics (e.g., Gaussian factors, specific transaction cost structure).
  • Future Work: The authors suggest that their framework can be extended to more general models and that the computational algorithm can be further refined. The paper's focus on providing a computational tool for realistic problems implies a desire to enable more in-depth economic analysis and potentially inform practical investment decisions.

Summary updated

Abstract

We develop methods of risk-sensitive impulsive control theory in order to solve an optimal asset allocation problem with transaction costs and a stochastic interest rate. The optimal trading strategy and the risk-sensitized expected exponential growth rate of the investor’s portfolio are characterized in terms of a non-linear quasi-variational inequality. This problem can then be interpreted as the ergodic Isaac–Hamilton–Jacobi equation associated with a min–max problem. We use a numerical method based on an extended two-stage policy iteration algorithm for min–max problems and provide numerical results for the case of two assets and one factor that is a Vasicek interest rate.
BibTeX
@article{bielecki2004risk,
  author = {Tomasz R. Bielecki and Jean‐Philippe Chancelier and Stanley R. Pliska and Agnès Sulem},
  title = {{Risk-sensitive portfolio optimization with transaction costs}},
  journal = {The Journal of Computational Finance},
  year = {2004},
  doi = {10.21314/jcf.2004.115},
}
Figure