2054. Interactive Robo-Advising through Scoring Mechanism
Invited abstract in session WD-54: Stochastic Models and Optimal Control , stream Stochastic modelling.
Wednesday, 14:30-16:00Room: Liberty 1.08
Authors (first author is the speaker)
| 1. | Yuwei Wang
|
| Statistics, Chinese University of Hong Kong | |
| 2. | Hoi Ying Wong
|
| Statistics, Chinese University of Hong Kong |
Abstract
We investigate discrete-time predictable exploratory forward performance processes (PEFPPs) within a reinforcement learning framework, motivated by the practical consideration that a client's chosen investment strategy may deviate slightly from one that precisely corresponds to her risk aversion. By employing Tsallis entropy to quantify the level of model exploration and the client's uncertainty about her risk aversion, we construct and analyze PEFPPs for both the CARA and CRRA classes, along with the corresponding density function of the optimal investment strategies. This provides a theoretical foundation for our scoring mechanism, which is based on an acceptance-rejection method and the optimal control distribution induced by the PEFPPs.
We demonstrate that the optimal exploratory control attains its maximal density at the optimal strategy derived in the absence of exploration. Furthermore, we establish a one-to-one correspondence between the client's score and the risk aversion parameter, so that for any fixed guess, the client's feedback score uniquely determines actual preferences. Furthermore, we propose an interactive robo-advising framework that, instead of requiring the client to specify risk aversion as a fixed input, infers her preferences by eliciting scores for recommended strategies. Using these scores, we iteratively refine our recommendation, aiming for the proposed strategy to exceed the client's acceptance threshold, while accounting for feedback errors.
Keywords
- Control Theory
- Financial Modelling
- Optimization in Financial Mathematics
Status: accepted
Back to the list of papers