Speakers

Dr. Richard McCreadie is a Senior Lecturer within the School of Computing Science at the University of Glasgow. He specialises in real-time IR technologies to enable users to efficiently and effectively find information in big data streams; the research and development of automatic (assistive) agents to detect events, extract knowledge and summarize information within big data streams; as well as methodologies to evaluate systems that process such streams. He leads the Financial Analytics Research theme within the School, which represents a cross-cutting group of researchers working on technologies to collate, analyse and apply financial data in real-time and at scale for customers.

Keynote Talk: The Customer is Not Always Right: Lessons when Learning from Past Investments

In more traditional recommendation domains, such as movie or product recommendation, the strongest signal that AI models can learn from is the past transactions of that customer (i.e. what they watched or bought). Translating to the financial domain, we might expect that what customers previously invested in would provide a strong signal for training effective AI investment recommendation agents, like robo-advisors. In this talk I will examine this idea, the underlying assumptions being made, and demonstrate why despite this being intuitive it does not work in practice on a real customer dataset from a large bank. During the talk I will touch on important challenges and their implications when training financial AI models, specifically issues with how we measure the ‘success’ of an investment, the quality of the investor pool, and difficulties in interpreting training data without knowledge of the investor intentions and strategy.

Dr Will Cong is the Rudd Family Professor of Management, a tenured Professor of Finance, and the founding director of FinTech at Cornell Initiative and the Digital Economy and Financial Technology (DEFT) Lab. He is also a Finance editor at the Management Science, Research Associate at the NBER, cofounder of two international research forums (ABFR and CBER), and was formerly a Kauffman Junior Fellow, Poets & Quants World Best Business School Professor, George P. Shultz Scholar, and Lieberman Fellow. He previously taught at the University of Chicago after earning his Finance Ph.D. and MS in Statistics from Stanford, and A.M. in Physics jointly with A.B. in Math and Physics from Harvard. He pioneered interdisciplinary research on tokenomics, AI for finance, Web3 economics, blockchain forensics and design, fintech, and how digitization and big data interact with and influence competition, growth, and entrepreneurship. His work has been recognized with numerous best paper prizes and grants and has been widely circulated and adopted in the industry. He is highly sought-after not only as keynote speaker at various international conferences and forums, but also as advisors for leading FinTech firms and quant funds, as well as government and regulatory agencies around the globe. 

LinkedIn, Website

Keynote Talk: Building AI Models for Investment via Goal-Oriented Search

I discuss how the core theme in recent advances in AI can be adapted to finance research. In Cong, Tang, Wang, and Zhang (2020), we build the first large investment model to directly optimize the objectives of portfolio management via Transformer-based deep reinforcement learning---an alternative to conventional supervised-learning paradigms that routinely entail first-step estimations of return distributions or risk premia. We develop multi-sequence, attention-based neural-network models tailored for the distinguishing features of financial big data, while allowing interactions with the market states and training without labels. Such AlphaPortfolio models yield stellar out-of-sample performances (e.g., Sharpe ratio above two and over 13% risk-adjusted alpha with monthly re-balancing) that are robust under various market conditions and economic restrictions (e.g., exclusion of small and illiquid stocks). We further demonstrate AlphaPortfolio's flexibility to incorporate transaction costs, state interactions, and alternative objectives, before applying polynomial-feature-sensitivity and textual factor analyses to uncover key drivers of investment performance, including their rotation and nonlinearity.

Greedy search presents a more interpretable alternative to reinforcement-learning-based heuristic search, In Cong, Feng, He, and He (2022), we develop a new class of tree-based models (P-Tree) for analyzing (unbalanced) panel data utilizing global (instead of local) split criteria that incorporate economic guidance to guard against overfitting while preserving interpretability. We grow a P-Tree top-down to split the cross section of asset returns to construct stochastic discount factors and test assets under the MVE framework, generalizing security sorting and visualizing (asymmetric) nonlinear interactions among firm characteristics and with macroeconomic states. When applied to U.S. equity, P-Trees significantly advances the efficient frontier relative to those constructed with common factors and basis portfolios. Transparent and diversified P-Tree test assets demonstrate significant unexplained alphas against various factor models. P-Trees also outperform most known observable and latent factor models in pricing individual stocks and test portfolios, while delivering transparent trading strategies and risk-adjusted investment outcomes. Beyond asset pricing, P-Trees offer effective alternatives to large, deep-learning-based AI models for analyzing panel data via economically guided, goal-oriented searches in a large clustering space.

Dr Jongha Jeon is a Senior Data Scientist in the Big Data & AI LAB at the Hana Institute of Technology. His interests lie in developing credit scoring and bankruptcy prediction models for financial or alternative data (person, corporate and industry) and analyzing the phenomenon using machine learning and conventional mathematical analysis. He received his PhD in Mathematical Sciences from the Ulsan National Institute of Science and Technology, where he studied reservoir computing networks incorporating mathematical analysis.

Presentation: Explainable AI driven Credit Scoring Technology (ExACT) in Hana Financial Group

With the remarkable development of AI, financial institutions around the world have also actively developed credit scoring models using various AI algorithms and methods. These models are being applied to loan products to determine whether or not to approve loans to customers or to determine loan limits. Although the performance of the model was improved due to the use of a complex model, the explainability of the model was inevitably reduced due to the trade-off between explainability and model performance. As we know, there have been various attempts (Transparency model, Post-hoc expansion and etc.) to overcome these points. Recently, it is known that post-hoc explanation (a kind of permutation method) is used as the main algorithm for computing reasons of the models. These methods have a conflict in which the speed of computing reasons slows down as the quality of the reasons is improved. In addition, various alternative data, as well as cps (credit profile service) data, are used for developing credit scoring models for thin-filers. The performance of the model is better using the alternative data, but the variables are reluctant to explain to customers for a number of reasons. In this presentation, we would like to talk about what methods and data (cps or alternative data) financial institutions in Korea use to develop credit scoring models and secure explainability of the model. Furthermore, we would like to briefly introduce ExACT(Explainable AI driven Credit Scoring Technology) in Hana Financial Group.

Mr Sungmin Jang is a researcher with the Investor Understanding AI team at NCSoft. He is deeply engrossed in exploring the application of ML-based investment solutions tailored for individual investors. A graduate of KAIST, he majored in Industrial and Systems Engineering and delved into portfolio theory. His academic pursuits were primarily centered on causal inference within alternative asset classes. Additionally, while working in Tokyo as a data scientist, he played a pivotal role in the development and operation of an AI-driven credit rating service for Sumitomo Mitsui Banking Corporation.

Presentation: AI/ML powered portfolio rebalancing support system for individual investors

As the complexity of a portfolio increases, the more asset classes, currencies, and stocks it contains, the more difficult and lengthy rebalancing becomes. In particular, retail investors need to consider labor and capital income, while also taking into account their daily consumption when managing their portfolios. In addition, behavioral finance suggests that retail investors may be subject to various biases and emotions when executing rebalancing transactions. We introduce an AI/ML powered portfolio rebalancing support process that takes these unique circumstances of retail investors into account.

Mr Paul Nixon heads up behavioural finance for Momentum Investments in South Africa. He established an applied behavioural finance capability after experiencing both client and adviser investment behaviour for 20 years with various South African insurers and Barclays Bank in London. Paul began his career in neuromarketing and “atmospherics” establishing links between the olfactory sense (smell) and behaviour and returned to behavioural sciences in a finance context with Absa / Barclays and Momentum. Paul holds an MBA (with distinction) from Edinburgh Business School and recently completed a Masters degree (with distinction) where he researched ‘risk behaviour’ at Stellenbosch University using machine learning. This paper was published in the international Journal of Behavioural and Experimental Finance. Paul is a published author on financial psychology and is a registered member of the Swiss-based Global Association of Applied Behavioural Scientists (GAABS) where he co-leads the Middle East and Africa regions. He is a PhD candidate at the University of Stellenbosch in South Africa. 

Accepted paper: Using Machine Learning to Predict Investors’ Switching Behaviour with co-author Evan Gilbert (Momentum Investments)

While there is some evidence that an individual’s preference for risk is relatively stable over time, they are exposed to a continuous stream of stimuli that influences their perception of risk (the way risk feels) at any given moment along their investment journey. This ultimately influences their risk propensity or inclination to take investment risk (or not). This often leads to a behaviour tax or lower investment return from switching to different investments (mutual funds) because of shifting risk perception. During the COVID pandemic, South African investors destroyed ≈ $36 million (behaviour tax) from switching between mutual funds. This behaviour has perpetuated post the Pandemic and is evidence that heightened perceptions of risk have persisted. To reduce this behaviour tax by delivering timely and appropriate nudges, the random forest algorithm (supervised machine learning) is applied in this paper to over 12 million observations and 87,592 Momentum clients between 2018 and 2022. The algorithm also identifies features of investors that switch and takes these onward to predict switch behaviour. A commercially acceptable predictive value as measured by the area under curve [AUC] and Gini Index statistics are produced as evidence that the identified features are useful in predicting switch behaviour. Ultimately this charts the way forward for financial services providers (D2C as well as advised clients) and administrative investment platforms to engage with those investors and advisers proactively.