Although the money management industry is one in which the nascent machine learning field has not yet seen great success, early evidence suggests that machine learning tools could serve portfolio managers well within asset management, according to a report from AQR Capital Management.
The report, "Can Machines 'Learn' Finance?" suggests that financial machine learning could be "the next leap forward in quantitative investing."
The report raises two points essential for understanding machine learning's current state within the asset management industry. The first is that research in the field is in very early days. The second is that early research suggests potential economically and statistically significant improvements in the performance of portfolios that leverage machine learning tools.
"However, the gains are evolutionary, not revolutionary," the report says.
As the report notes, although machines can now recognize, say, images of cats, that doesn't necessarily translate to being able recognize which stocks to pick. While image recognition has a high signal-to-noise environment, or a high degree of predictability, financial markets do not.
"The signal-to-noise ratio is not just weak, but it will always be pulled toward zero ... because financial markets are extremely noisy," says the report. "The best stock or investment portfolio in the world will, on any given day, quarter, or year, experience wild swings in performance due to unanticipated news. Second, the signal in financial markets is expected to be low and will be kept low."
However, while accurately predicting returns poses a challenge for machine learning, there are in fact other areas within the realm of finance that could benefit more from machine learning. For example, portfolio implementation, including risk management, transaction cost management and factor construction.
The ideas behind machine learning — leveraging new data sets to identify strong additive portfolio performance and using quant methods to systematically extract information — are the M.O. of quant investment processes. For decades, asset managers have used human-intensive, decentralized statistical learning.
Machine learning offers a systematic approach to investing that mechanizes that process, allows managers to metabolize information from more new sources faster, including unstructured data previously untapped, and provides tools to search through increasingly flexible economic models that seek to better capture complex realities of financial markets.
The report argues that applying machine learning techniques to the money management industry is the next natural step for investment research, and one that will continue to be explored.