Artificial Hedging Intelligence: a Reinforcement Learning Approach to Hedge Fund Management
Artificial Hedging Intelligence (AHI) describes a machine-learning approach to hedge-fund management. Using Reinforcement Learning, AHI models probability using the discipline of econo-physics. Econo-physics finds in the phenomena of the physical world, from Newtonian physics to Einsteinian and quantum systems, analogs for economic phenomena. AHI describes phenomena of each of the three classes algorithmically. Then, observing sales data, AHI predicts future sales events by applying econo-physical principles. Errors cause AHI to adjust the geometric model governing the weighting, calculation and combinatorics of its model, and recursively rewrites itself, from its most foundational principles to its most case-specific protocols, to learn adaptively from its mistakes and its successes. Part I describes the overall heuristics driving AHI thematically in the form of numbered desiderata. This explains what the model is, how it is structured, and offers detailed explanations for why it is shaped and structured in the way it is. It also narrates the process of Reinforcement Learning AHI would use to teach itself. It also details the physical parametrizations it will use. Part II offers a series of algorithms translating economic phenomena, including fear, doubt, and optimism, into physical phenomena. These algorithms are offered in the form of mathematical formulae; it is left to the user to express these algorithms programmatically. AHI is offered in the spirit of a proof of concept of modelling probability geometrically. Readers assume any and all responsibilities for the ramifications of any use or application of any part of parts of AHI however modified, adapted, or combined. The University of Houston, the UH Libraries, the DRC, the author, and their agents and representatives explicitly disclaim responsibility for any consequences arising from such use, as AHI is offered only in the spirit of an academic exercise in the physical modeling of probability through particular phenomena.