Paper: arXiv 2308.10556

Authors: Kristoffer Andersson, Cornelis W. Oosterlee

Abstract

In this paper, we propose a machine learning algorithm for time-inconsistent portfolio optimization. The proposed algorithm builds upon neural network based trading schemes, in which the asset allocation at each time point is determined by a a neural network. The loss function is given by an empirical version of the objective function of the portfolio optimization problem. Moreover, various trading constraints are naturally fulfilled by choosing appropriate activation functions in the output layers of the neural networks. Besides this, our main contribution is to add options to the portfolio of risky assets and a risk-free bond and using additional neural networks to determine the amount allocated into the options as well as their strike prices. We consider objective functions more in line with the rational preference of an investor than the classical mean-variance, apply realistic trading constraints and model the assets with a correlated jump-diffusion SDE. With an incomplete market and a more involved objective function, we show that it is beneficial to add options to the portfolio. Moreover, it is shown that adding options leads to a more constant stock allocation with less demand for drastic re-allocations.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 5.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper involves advanced mathematics like jump-diffusion SDEs, time-inconsistent stochastic control, and neural network approximations, yet it focuses on backtest-ready methodology with realistic constraints and numerical experiments, though it lacks live trading metrics.

Research Flowchart

  flowchart TD
  A["Research Goal: Incorporate Options in Time-Inconsistent Portfolio Optimization"] --> B["Methodology: Deep Learning w/ Neural Networks"]
  B --> C["Inputs: Correlated Jump-Diffusion SDE Data"]
  C --> D["Computation: NN Allocation & Options Pricing"]
  D --> E["Constraints: Rational Preferences & Trading Limits"]
  E --> F["Outcomes: Improved Returns & Constant Stock Allocation"]

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 machine‑learning algorithm (D‑TIPO) that uses neural networks to determine time‑varying allocations to stocks, a risk‑free bond, and options (including strike prices) for time‑inconsistent portfolio optimization.

Source sentence
Besides this, our main contribution is to add options to the portfolio of risky assets and a risk‑free bond and using additional neural networks to determine the amount allocated into the options as well as their strike prices.

Assumes 6

  • The market is incomplete, modeled with a correlated jump‑diffusion SDE for the assets. market
    Source sentence
    model the assets with a correlated jump‑diffusion SDE.
  • Trading occurs only on a finite set of discrete dates. execution
    Source sentence
    From now on, unless otherwise is stated, we assume that T is a set of finite trading dates
  • Proportional transaction costs are incurred on each trade. market
    Source sentence
    Transaction costs: Assuming a proportional transaction cost we can write the sum of the transaction costs for stock k as
  • A non‑bankruptcy constraint forces liquidation when portfolio value becomes non‑positive. execution
    Source sentence
    The non‑bankruptcy constraint is formulated such that the first time the sum of the values of the stocks and the bond is non‑positive, the portfolio is liquidated.
Show 2 more
  • Synthetic data are used to train and test the algorithm. data
    Source sentence
    Since our data is synthetic we could have generated new data instead of reusing data in each epoch.
  • The objective function is an empirical version of a more realistic utility (beyond mean‑variance). model
    Source sentence
    The loss function is given by an empirical version of the objective function of the portfolio optimization problem.

Data

  • Synthetic asset paths synthetic unknown
    Source sentence
    Since our data is synthetic we could have generated new data instead of reusing data in each epoch.

Costs

Modeled Costs are part of the reported headline results.

Proportional cost C expressed as a percentage of the transaction size (100×C).

Source sentence
Here 100×C ∈R+ represents the transaction costs as a percentage of the size of the transaction.

Finds 3

  • Adding options to the portfolio yields higher empirical mean returns and lower expected shortfalls compared with strategies without options, while also reducing transaction costs by more than 60% relative to the D‑TIP strategy. performance
    Source sentence
    By using the D‑TIPO strategy, we see that the transaction cost is decreased by more than 60% compared to the D‑TIP strategy and by more than 90% compared to the MV‑strategy.
  • The D‑TIPO strategy achieves a more constant stock allocation and requires less drastic re‑allocations than strategies without options. performance
    Source sentence
    that it is beneficial to add options to the portfolio. Moreover, it is shown that adding options leads to a more constant stock allocation with less demand for drastic re‑allocations.
  • The neural‑network based allocation reproduces the reference solution with high accuracy, as shown by convergence of the loss and close matching of empirical PDFs and CDFs. theoretical
    Source sentence
    Upper left: Convergence of the loss to the analytic counterpart with respect to the number of training epochs

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 28 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 2 papers in the archive (1 by independent authors)

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.