• Markets are incredibly noisy, so extracting the signal from the noise can be a challenge.
• Machine learning has helped the Vanguard Quantitative Equity Group (QEG) to identify patterns in complex financial data.
• Vanguard’s QEG understands the rationale behind the algorithms’ recommendations for each investment.

Geopolitics, economic uncertainty or the direction of monetary policy can be a distraction for many investors. Amid this noise, machine learning in our actively managed equity funds is helping the Vanguard Quantitative Equity Group (QEG) to identify patterns in complex financial data. 

Vanguard’s QEG has always used quantitative models¹ that replicate the decision-making process of a skilled fundamental stock picker, but in a systematic and scalable manner. The team is one of the managers we use for our range of active equity funds globally. It co-manages a portion of the Vanguard Global Equity Income Fund, for example. 

While the portfolio managers remain responsible for all investment choices, the tools they use to process the data have naturally evolved. In recent years, they have incorporated machine learning into their model-driven process. The fund managers believe that this level of robust quantitative analysis of reams of data gives them a competitive edge when it comes to making buy, sell and rebalancing decisions. 

The successful incorporation of machine learning is a result of the QEG project team and members of Vanguard’s “AI Garage”—a community of artificial intelligence (AI)/machine learning practitioners from across the company—which found that machine learning consistently added value in simulated portfolios before they applied it to real funds. 

They found that machine learning helps to clarify complex relationships between economic and financial market conditions and the share price performance of individual companies. In some instances, the new machine learning-driven models produce the same output as QEG’s ‘legacy models’, but they can vary.   

How machine learning generates alpha by extracting signals from the noise

Machine learning can enhance alpha, or returns above a portfolio's benchmark, in a number of ways. QEG has always considered classic market factors, such as value, momentum and volatility, as alpha sources, but machine learning helps to examine such factors more thoroughly. The team expects the signals, i.e. the recommendations produced by the team's models, to continue to evolve over time as a result of machine learning.

Despite the integration of machine learning, portfolio turnover and aggregate measures of portfolio risk, such as tracking error relative to a benchmark, have not significantly changed. 

Certainly, the introduction of machine learning to a portfolio management process is no guarantee of stronger results and machine learning-based recommendations are not perfect. 

However, QEG believes that if, through the use of its various models (including machine learning), it can produce an outcome where it is right 55%–60% of the time over extended periods, that this can add significant long-term value for investors.

Integrating machine learning into an established investment process

For some products, Vanguard’s QEG employs an ensemble model - a collection of dozens of models tracking tens of millions of underlying economic and financial market interactions. Overall, half the inputs can come from the team’s traditional approach, refined over decades - and half from machine learning. 

A key new element is machine learning’s continual interpretation of the macroeconomic environment as context for company- and stock-specific changes.

Making sense of machine learning-based investment recommendations

Before incorporating machine learning into the management of Vanguard fund assets, it was important for Vanguard’s QEG to be able to understand the rationale behind the algorithms’ recommendations for each investment. 

Vanguard’s QEG won’t trade on signals that it doesn’t understand. Everything has to tie back to a fundamental rationale. That’s why, as part of the machine learning project, the team built an “interpretability model” that gives portfolio managers - as well as Vanguard’s independent portfolio review and risk-management teams - explanations of:

• The drivers of individual stock predictions.
• The aggregate-level interpretations of model behaviour (such as the sensitivity of the signals to macroeconomic changes).
• The model health analytics (such as the volatility of scores and outliers in daily model changes), which help the team to assess model inputs and outputs.

Vanguard’s QEG believes that the foundations of its approach to machine learning are solid as it looks to build on its strong basis of knowledge to help deliver positive investment outcomes for clients over the long term.

¹ Algorithms that crunch tremendous amounts of data in search of exploitable patterns.