[PDF] Robo-Advising: Enhancing Investment with Inverse Optimization and Deep Reinforcement Learning | Semantic Scholar (2024)

This paper proposes a new approach for developing long-term investment strategies using an ensemble of evolutionary algorithms and a deep learning model by taking a series of short-term purchase decisions, and provides empirical evidence of superior performance using the ensemble approach.

This survey paper aims to review the recent developments and use of RL approaches in finance, including optimal execution, portfolio optimization, option pricing and hedging, market making, smart order routing, and robo‐advising.

A lower bound for regret minimization in finite-horizon MDPs with LDP guarantees is established which shows that guaranteeing privacy has a multiplicative effect on the regret.

This work considers the shuffle model of privacy and shows that it is possible to achieve a privacy/utility trade-off between JDP and LDP and presents an algorithm with regret bound, while guaranteeing both central and local privacy.

By leveraging shuffling techniques, this paper presents an algorithm that, depending on the provided parameter, is able to attain any privacy/utility value in between the pure JDP and LDP guarantee.

A systematic literature review and evaluated 97 publications on the determinants of conventional and digital investment advisory decisions found five main determinants were identified that are important for investment advisory decisions.

It is demonstrated how direct reinforcement can be used to optimize risk-adjusted investment returns (including the differential Sharpe ratio), while accounting for the effects of transaction costs.

This work presents a framework for hedging a portfolio of derivatives in the presence of market frictions such as transaction costs, liquidity constraints or risk limits using modern deep reinforcement machine learning methods and shows that the set of constrained trading strategies used by the algorithm is large enough to ε-approximate any optimal solution.

This paper presents a novel approach of measuring risk preference from existing portfolios using inverse optimization on the mean-variance portfolio allocation framework and allows the learner to continuously estimate real-time risk preferences using concurrent observed portfolios and market price data.

This work devise a scalable and data-efficient RL algorithm and conduct large scale empirical tests using data from the S&P 500 stocks, finding that the method consistently achieves over 10% annualized returns and it outperforms econometric methods and the deep RL method by large margins.

A novel framework in which a robo-advisor interacts with a client to solve an adaptive mean-variance portfolio optimization problem is introduced and it is argued that the optimal portfolio’s Sharpe ratio and return distribution improve if the robos counters the clients’ tendency to reduce market exposure during economic contractions when the market risk-return tradeoff is more favorable.

This work proves that the optimal feedback policy for the continuous‐time mean‐variance portfolio selection with reinforcement learning must be Gaussian, with time‐decaying variance, and then proves a policy improvement theorem, based on which an implementable RL algorithm is devised.

This paper develops an online learning algorithm that uses an implicit update rule which can handle noisy data and proves that the algorithm converges at a rate of $\mathcal{O}(1/\sqrt{T})$ and is statistically consistent.

Across a large range of implementation choices, this result indicates that RL-based portfolios are very close to the equally-weighted (1/N) allocation, which implies that the agent learns to be agnostic with regard to factors.

It was shown that Modern Portfolio Theory remains the main framework used in RAs worldwide and it was revealed that the AuM volumes tend to be higher for the systems applying newer and more sophisticated methods.

This work presents an actor-critic, model-free algorithm based on the deterministic policy gradient that can operate over continuous action spaces, and demonstrates that for many of the tasks the algorithm can learn policies end-to-end: directly from raw pixel inputs.

...

...