The traditional Quant research process starts with an idea backed by an economic narrative, and is followed by months, sometimes years of data and statistics driven investigation, at the end of which a very small percentage of ideas are found to be viable. Those viable ideas (or alphas) become known as investment styles that, as they leak to competitors, turn into `smart-beta' risk factors and become less effective. As markets mutate those alphas into smart-betas, new patterns emerge and, in so doing, give rise to new alphas.
To preserve their edge, absolute-return quantatitive investment managers need to detect fading alphas, and quickly find new ones. Driven by trading ideas found by Quants, this exploration cycle can be pretty slow. However, discovering alphas early on in their life cycles comes with a big first-mover advantage on returns, and discovering alphas in larger numbers can greatly improve risk management.
The AI Quant Paradigm
The AI Quant paradigm aims at boosting the alpha exploration cycle through model-based automation, so as to seek and maintain a first-mover advantage on returns, and a robust risk management.
This new paradigm is however fraught with danger; the risk of overfitting. Financial markets, unlike physical systems traditionally dealt with by Machine Learning researchers, are extremely noisy. While some consider the AI Quant paradigm to be inherently too challenging for the aforementioned reason, for us this is an opportunity to differentiate ourselves. By combining a good command of mathematical modelling with thinking from first principles, questioning long-held econometrics dogmas if needed, and first-class experience in Quant Trading, we are tackling the challenges of the AI Quant paradigm head-on.
After over 5 years of award-winning research, we have designed an AI that continuously detects new 'pure-alpha' investment styles, some of which we've been trading for over a year. The findings of our AI are always complemented with human oversight to get some insights into found opportunities prior to acting on them.
Example Insight: Non-Linear Risk Premium
An example investment style found by our AI is the non-linear risk premium.
Traditional measures of risk used by market participants in the portfolio optimization process (e.g. Pearson's correlation matrix), by in large ignore critical aspects of risk such as non-linearities (i.e. tail-risk) and temporal dependencies (i.e. cascading effects). However, as you'll read on our insights page, modern Machine Learning techniques can be used to properly account for both tail-risk and cascading effects.
Our AI has found alphas whose performances are strongly correlated to the extent to which returns of traded assets exhibit non-linearities, or said differently, they capture a non-linear risk premium.