Source-linked AI summary
Time-Inconsistent Stochastic Linear--Quadratic Control
Ying Hu, Hanqing Jin, Xun Yu Zhou
TL;DR
The paper studies time-inconsistent stochastic LQ control caused by expected-state and state-dependent objective terms. It defines open-loop equilibrium controls, derives FBSDE-based sufficient conditions, and applies them to obtain explicit controls and portfolio strategies under tractable settings.
Problem
Time-inconsistent stochastic LQ problems lack a dynamic notion of solution when objective terms depend quadratically on expected state or on the current state.
Method
The paper defines equilibrium through infinitesimal open-loop spike deviations and derives a sufficient condition using a flow of forward–backward stochastic differential equations.
Results
The paper obtains explicit equilibrium controls for scalar states with deterministic coefficients and explicit mean–variance strategies when the risk premium is deterministic or stochastic.
Takeaways & Limitations
Equilibrium definitions matter: the paper’s open-loop strategies are generally different from feedback-control strategies in the deterministic mean–variance case.
Takeaways & Limitations
General solvability of the FBSDE flow and its constraints remains an open problem, with the paper solving the problem thoroughly only under deterministic scalar-case coefficients.
Abstract
from arXiv · showhide
In this paper, we formulate a general time-inconsistent stochastic linear--quadratic (LQ) control problem. The time-inconsistency arises from the presence of a quadratic term of the expected state as well as a state-dependent term in the objective functional. We define an equilibrium, instead of optimal, solution within the class of open-loop controls, and derive a sufficient condition for equilibrium controls via a flow of forward--backward stochastic differential equations. When the state is one dimensional and the coefficients in the problem are all deterministic, we find an explicit equilibrium control. As an application, we then consider a mean-variance portfolio selection model in a complete financial market where the risk-free rate is a deterministic function of time but all the other market parameters are possibly stochastic processes. Applying the general sufficient condition, we obtain explicit equilibrium strategies when the risk premium is both deterministic and stochastic.
1. Introduction.
The paper addresses time-inconsistent stochastic control by defining open-loop equilibrium controls and deriving FBSDE-based sufficient conditions, then applying them to mean–variance portfolio selection.
- Time inconsistency arises in problems with hyperbolic discounting, mean–variance portfolio selection, or probability distortion.
- Precommitted controls are optimal only from the initial time and do not provide a dynamic solution to time inconsistency.
- The paper formulates a stochastic LQ problem with quadratic expected-state and state-dependent objective terms, defining equilibrium within open-loop controls.
- A sufficient equilibrium condition is derived through a time-parameterized flow of forward–backward stochastic differential equations.
- For mean–variance portfolio selection with potentially random market parameters, the paper obtains explicit equilibrium strategies and finds them generally different from feedback-control strategies.
2. Problem Setting.
The paper sets up a non-homogeneous stochastic linear system and an objective whose expected-state and state-dependent terms create time inconsistency, motivating open-loop equilibrium controls.
- Problem Setting: The control problem uses a continuous-time, n-dimensional non-homogeneous linear controlled system with an adapted control and state process.
- Problem Setting: The system permits deterministic A and essentially bounded adapted coefficients B, Cj, and Dj, alongside stochastic drift and diffusion terms.
- Problem Setting: The objective combines standard LQ state and control costs with a quadratic expected-terminal-state term and a state-dependent expected-state term.
- Problem Setting: The quadratic expected-state term is motivated by variance, while the state-dependent term stems from state-dependent utility.
- Problem Setting: An equilibrium is required to withstand infinitesimal spike deviations at every time, rather than being globally optimal from the initial time.
- Problem Setting: The paper’s objective is to characterize equilibrium controls generally and identify them in special cases including mean–variance portfolio selection.
3. Sufficient Condition of Equilibrium Controls.
The paper derives a sufficient equilibrium condition from spike-variation analysis as a flow of FBSDEs, but general solvability remains open; tractable results require further restrictions.
- Sufficient Condition of Equilibrium Controls: For each fixed initial time, the adjoint equations are BSDEs, but collectively they form a flow of BSDEs with a time-indexed family of unknowns.
- Sufficient Condition of Equilibrium Controls: The condition requires a solution to a system involving equilibrium control, state, and adjoint processes satisfying additional constraints.
- Sufficient Condition of Equilibrium Controls: General existence and unique solvability of the flow equations remain challenging open problems, even for scalar states.
- Sufficient Condition of Equilibrium Controls: The scalar-state version provides an equilibrium-control criterion through the corresponding FBSDE system and constraint.
- Sufficient Condition of Equilibrium Controls: When all scalar-case coefficients are deterministic, the paper can solve the problem more thoroughly than in the general setting.
4. Equilibrium When Coefficients Are Deterministic.
With deterministic coefficients, the equilibrium-control construction reduces the problem to coupled ODE/Riccati systems, whose positive solutions yield an explicit equilibrium control under stated conditions.
- Deterministic-coefficient reduction: Under deterministic coefficients, the BSDE becomes an ODE, with K ≡ 0 and a deterministic representation for P(s; t).The analysis then seeks a linear feedback equilibrium through an Ansatz.
- Positive-solution analysis: If system (4.10) admits a positive solution pair (M, J), then system (4.9) admits a positive solution pair (M, N).This provides a route from the auxiliary system to the original Riccati system.
- Positive-solution analysis: Theorem 4.2 gives unique positive solution pairs under matrix positivity and either B = λD′C or uniform positivity of D′D.For the singular case R ≡ 0, Theorem 4.3 gives positive solution pairs under two additional nonnegativity conditions.
- Equilibrium control: Theorem 4.4 states three cases guaranteeing a unique positive solution pair (M, N), and the resulting control u∗ given by (4.4) is an equilibrium.The verification also checks that the feedback coefficients are uniformly bounded, so u∗ belongs to the required square-integrable control class.
5. Mean-Variance Equilibrium Strategies in Complete Market.
The paper applies its open-loop equilibrium framework to mean–variance portfolio selection in a complete market, deriving explicit strategies for deterministic and stochastic risk premia. Random risk premia require a more involved FBSDE analysis and add hedging for parameter uncertainty.
- Model formulation: The portfolio model is a one-dimensional special case of the general stochastic LQ problem, with a wealth process governed by the market model.The market is complete under the stated volatility condition, and trading strategies are represented interchangeably by π and u.
- Model formulation: The objective balances conditional variance and conditional expectation, with the trade-off weight depending on current wealth.This state-dependent weighting is part of the model’s time-inconsistent structure.
- Stochastic risk premium: When the risk premium is stochastic, the analysis is not a direct application of the deterministic-coefficient case because coefficient randomness makes the analysis more involved.The paper therefore specializes the general FBSDE condition and solves the resulting BSDE system.
- BSDE solution: The BSDE analysis yields an indefinite stochastic Riccati equation because the driver contains M^-1, alongside boundedness and BMO-martingale properties for the relevant solutions.Existence is obtained through truncation, while uniqueness is established by continuation over successive time intervals.
- Equilibrium strategy: Theorem 5.4 constructs an equilibrium strategy from solutions of BSDEs (5.8) and (5.13), and the sufficient condition confirms that it is an equilibrium.The associated state and auxiliary processes satisfy the forward–backward system required by the general theorem.
6. Concluding Remarks.
The paper identifies unresolved questions around the solvability and broader extension of its game-theoretic framework for time-inconsistent stochastic control.
- The general solvability of the flow of FBSDEs (3.6) remains an open problem requiring systematic investigation.
- Adapting the generalized HJB approach to the open-loop framework warrants further study, even with deterministic coefficients.
- Extensions beyond LQ control and to probability-distortion-induced time inconsistency are identified as future research directions.