A Generative Deep Learning Framework for Emotion-sensitive Robo-advisors in Personal Wealth Management

Description

On 28th March, 2018 the Hong Kong Securities and Futures Commission (SFC) has published the public's comments regarding the SFC's guidelines on the provision of automated roboadvice given the trend that there are increasingly more robo-advisors being deployed to various financial institutions in Hong Kong.  For example, the HKSE-listed Yunfeng Financial Group launched its first robo-advisor named “Youyu” who could help individual investors manage their global assets more effectively at a lower cost in 2017.  Moreover, Magnum Research launched its robo-advisor named AQUMON that managed 200 exchange-traded funds (ETFs) in the Hong Kong market while 8-Securities deployed its robo-advisor called Chole to support their electronic trading platform. In the mainland, the robo-managed assets are expected to reach 6 trillion yuan by 2020, and active players such as the Pintec group, Ant Financial, China Merchants Bank, ICBC Bank, etc. have all developed their robo-advisors.Though robo-advisor is one of the hot topics in the field of Financial Technology (Fintech), the design of robo-advisors is still at the infant stage only. The functionalities of existing robo-advisors are relatively limited. For instance, these robo-advisors cannot detect and respond to users’ emotional feedback, whereas human advisors do. It is essential for a truly intelligent robo-advisor to be sensitive to the emotions of its users in order to provide “suitable” advice. For example, if an inexperienced client demonstrates an intense level of “fear”, and asks her robo-advisor to sell out an extraordinary large amount of equities, a truly smart robo-advisor should recognize such an emotion and remind its client to calm down first before making an abrupt investment decision. In fact, according to the HK SFC's proposed guidelines for the provision of online distribution and advisory platforms released in May 2017, robo-advisors should generate “suitable” advice by considering a client's specific personal circumstances.There have been very few published articles about the design of intelligent robo-advisors, not to say emotion-sensitive robo-advisors. Though deep learning, one of the core artificial intelligence (AI) methods, is promising for developing the multimodal emotion detection modules of robo-advisors, it requires a large number of manually labeled examples to train deep neural networks for multimodal emotion detection. Moreover, it is extremely time-consuming to train deep neural networks. For the design of practical emotion-sensitive robo-advisors, we must address these fundamental problems. Guided by the Design Science research methodology, the proposed GRF project is just able to fill the aforementioned research gaps by developing a novel generative deep learning framework for a more efficient and more effective semi-supervised deep network training to facilitate the practical development of the multimodal emotion detection module of emotion-sensitive robo-advisors.In the foreword of the HK Chief Executive’s 2018 policy address, our HK chief executive explicitly states that “from enhancing our position as a financial centre, which includes revision of listing regulations and application of financial technologies (Fintech), to forging ahead the development of innovation and technology (I&T), such as the establishment of artificial intelligence and robotics technologies, …..”.  The proposed research project is just strategically aligned with the HK government's appeal of leveraging AI, Fintech, and robotics technologies for enhancing Hong Kong as an international financial hub. To ensure real business and societal impacts, we plan for technology transfer of our research via our industrial collaborator, the Pintec group which is located in the greater China area.