Dirichlet Policies for Reinforced Factor Portfolios

Abstract

This article aims to combine factor investing and reinforcement learning (RL). The agent learns through sequential random allocations which rely on firms' characteristics. Using Dirichlet distributions as the driving policy, we derive closed forms for the policy gradients and analytical properties of the performance measure. This enables the implementation of REINFORCE methods, which we perform on a large dataset of US equities. Across a large range of implementation choices, our result indicates that RL-based portfolios are very close to the equally-weighted (1/N) allocation. This implies that the agent learns to be agnostic with regard to factors. This is partly consistent with cross-sectional regressions showing a strong time variation in the relationship between returns and firm characteristics.

The Effect of Firm Location on Corporate Dividend Policy: Study in Indonesia Capital Market

“…The second one, which is known as the bird in the hand theory, is based on investors' preference for dividends for the quality of certainty (Lintner, 1962). The last of the three views holds that investors are not in favor of dividend payment because the tax factor that they assume for the dividends paid is greater than the capital gain (Litzenberger & Ramaswamy, 1982;Febrianti & Zulvia, 2020;Kovalev & Drachevsky, 2020;André & Coqueret, 2020;Hakim & Kusmanto, 2020;Nguyen et al, 2021).…”

Robo-Advising: Enhancing Investment with Inverse Optimization and Deep Reinforcement Learning

“…Our work handles the no-shorting, no-borrowing constraint using deep neural network approximations with a special structure for the input data. A similar investment optimization problem under Dirichlet policies has been recently studied in [51]. It is interesting to note that their algorithm also generates equal-weight strategy in most cases, although the algorithm is trained using firm-level data rather than just market data.…”