The way to generate alpha through AI is by combining infrastructure and methodology that capitalises on computing power and automation.
When did Seeyond decide to become an active quant investor?
With the use of proprietary algorithms and quantitative tools dating back to 2007, but also given the technological progress the industry has known over the past years, it was only logical to us to evolve from Active Quant 1.0 to Active Quant 2.0. As an innovation driven firm, we are continuously looking for advanced techniques and data to bring added value.
In a nutshell, what is the key difference between traditional modelling and machine learning?
Put simply, traditional modelling is a ‘handmade’, human-centred process, based on statistical inference, so it’s hypothesis driven. Machine learning, on the other hand, is an agnostic, computer-centred process – so, it’s based on autonomous learning – and is based on KPIs, so it’s objective driven.
How does AI change the role of traditional portfolio managers?
Nevertheless, we think that AI power needs to be used carefully and with great awareness. At Seeyond, we consider it our fiduciary responsibility as an expert asset manager to provide our clients with an optimal risk return assessment and a deep comprehensive understanding of the models we use and their outcomes. These bringreadability and reliability to our clients. This is what we call being an ‘Accountable Portfolio Manager’.
How would you describe your AI journey as an asset manager?
One of our projects is to explore several AI use cases based on a robust machine-learning pipeline with a fintech specialized in machine learning infrastructure. With this framework, we aim to be able to quickly generate signals and explore new extra financials data sources.
This cross-expertise approach is another growth driver for Seeyond and enables us to work in a more agile ecosystem. We learn fast on computer science expertise while they benefit from our business experience.
How have you seen AI and machine learning techniques develop over time?
This is a machine learning-based indicator that analyses US macroeconomic time-series through eight stages. We then apply signal treatment’s techniques to model each cluster’s evolution in time by a level and a trend time-series.
Finally we use a machine-learning algorithm to screen history and detect periods where the US economy was similar as it is today.