Paper: arXiv 2306.10582

Authors: Marc Chen, Mohammad Shirazi, Peter A. Forsyth, Yuying Li

Abstract

We propose a novel data-driven neural network (NN) optimization framework for solving an optimal stochastic control problem under stochastic constraints. Customized activation functions for the output layers of the NN are applied, which permits training via standard unconstrained optimization. The optimal solution yields a multi-period asset allocation and decumulation strategy for a holder of a defined contribution (DC) pension plan. The objective function of the optimal control problem is based on expected wealth withdrawn (EW) and expected shortfall (ES) that directly targets left-tail risk. The stochastic bound constraints enforce a guaranteed minimum withdrawal each year. We demonstrate that the data-driven approach is capable of learning a near-optimal solution by benchmarking it against the numerical results from a Hamilton-Jacobi-Bellman (HJB) Partial Differential Equation (PDE) computational framework.

Complexity vs Empirical Score

  • Math Complexity: 8.5/10
  • Empirical Rigor: 7.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced mathematics, including stochastic optimal control, HJB-PDEs, and neural network approximations, yielding a high math complexity score. Its empirical rigor is substantial due to a direct benchmark against a numerical PDE ground truth, robustness tests on out-of-sample data, and explicit discussion of implementation, though it lacks live trading or full dataset disclosure, placing it in the Holy Grail quadrant.

Research Flowchart

  flowchart TD
  A["Research Goal: Solve Optimal Decumulation<br>for DC Pension Plans"] --> B["Methodology: NN Framework<br>vs. HJB PDE Benchmark"]
  B --> C["Data & Inputs:<br>Market Scenarios & Stochastic Constraints"]
  C --> D{"Computational Process"}
  D --> E["NN: Data-Driven<br>Optimization & Training"]
  D --> F["HJB: Numerical<br>PDE Solver"]
  E & F --> G["Key Outcomes:<br>Near-optimal Strategy<br>Validated vs. Benchmark"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes a novel data‑driven neural network optimization framework to solve a stochastic control problem with constraints for DC pension decumulation.

Source sentence
We propose a novel data-driven neural network (NN) optimization framework for solving an optimal stochastic control problem under stochastic constraints.

Assumes 6

  • No taxes are incurred on rebalancing. execution
    Source sentence
    We assume no taxes are incurred on rebalancing, which is reasonable since retirement accounts are
  • Transaction costs are negligible. execution
    Source sentence
    In addition, since trading is infrequent, we assume transaction costs to be negligible (Dang and Forsyth, 2014).
  • Instantaneous rebalancing (no price changes between withdrawal and allocation). execution
    Source sentence
    We assume instantaneous rebalancing, which means there are no changes in asset prices in the interval (t−i ,t+i ).
  • Log of jump multiplier follows a double exponential distribution. model
    Source sentence
    We assume that log(ξs) follows a double exponential distribution (Kou, 2002; Kou and Wang, 2004).
Show 2 more
  • Controls depend only on wealth (no transaction costs). model
    Source sentence
    As formulated, the controls depend on wealth only (see Forsyth (2022) for a proof, assuming no transaction costs).
  • Synthetic data are generated from calibrated parametric SDE models. data
    Source sentence
    Any data set referred to in this paper as synthetic data is generated by parametric stochastic models (SDEs)

Data

  • CRSP monthly data (bond and stock indices) · CRSP market-prices commercial
    1926:1-2019:12 · CRSP 10‑year U.S. Treasury index (bond) and CRSP value‑weighted total return index (stock)
    Source sentence
    For the computational study in this paper, we use data from the Center for Research in Security Prices (CRSP) on a monthly basis over the 1926:1-2019:12 period.

Recognised sources: CRSP

Costs

Not modeled A trading or execution result is reported without costs.

Source sentence
In addition, since trading is infrequent, we assume transaction costs to be negligible (Dang and Forsyth, 2014).

Finds 4

  • The NN framework yields a solution that closely matches the ground‑truth HJB PDE results. theoretical
    Source sentence
    We demonstrate that the proposed NN framework produced solution accurately approximates the ground truth solution.
  • Out‑of‑sample tests show almost unchanged performance compared to training results. performance
    Source sentence
    Out‑of‑sample tests involve testing model performance on an unseen data set
  • The NN can accurately approximate a bang‑bang control despite its discontinuity. theoretical
    Source sentence
    the continuous function approximation from the NN framework is able to approximate a bang‑bang control with high accuracy.
  • The NN allocation control maintains a low‑risk profile, with the 95th percentile stock allocation never exceeding 40%. performance
    Source sentence
    the maximum stock allocation for the 95th percentile never exceeds 40%, indicating that this is a stable low‑risk strategy.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 18 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 1 paper in the archive

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.