The Review — 15/05/2018 at 10:00

The Next Era of Factor Investing

by

Goldman_Sachs

For decades, investors have used certain stock attributes known as “factors” — such as size, value or momentum — to predict investment returns over long periods. Today, factor-based investing is getting a boost from big data and technology. Nicholas Chan, of the Quantitative Investment Strategies team in Goldman Sachs Asset Management, explains the evolution of factor-based investing.

Nicholas, can you explain factor-based investing and how this type of investing has evolved over the years?

Nicholas Chan: Factor-based investing has actually been around for decades. This type of investment strategy has its roots in academic research which found that certain traits of equities — such as size, volatility, value and momentum — were shown to help predict returns over long periods. For example, small company stocks typically outperform large company stocks over the long term. Today, technologies such as machine learning (ML) are opening up vast new areas of data to analyze. We’re seeing a new era of factor investing.

How is this different from the type of analysis that investors already conduct?

NC: Certain machine learning techniques have made it possible for investors to collect, analyze and track data at a much deeper — and infinitely faster — level than humans could otherwise achieve on their own. In fact, if you think about data more broadly, about 80 percent of new data that is created today is considered unstructured, or alternative, data. This is essentially data that isn’t in numerical form, but rather consists of text, audio, imagery and even video. For example, using a form of ML known as natural language processing (NLP) — which is essentially teaching computers to read text — we’re able to review tens of thousands of news articles per day in multiple languages; tens of thousands of earnings call transcripts per year; and hundreds of thousands of analyst research reports per year.

Can you provide some examples of emerging areas for analysis?

NC: In the same way that technology firms and digital content providers are forming hyper-personalized views on people by deducing connections based on consumers’ digital footprints, we’re also using ML and technology to form hyper-personalized views — but on companies. The analysis we undertake is to find ways that connect companies. Some company connections are intuitive such as the relationship between customers and suppliers. But others are more subtle and can be gleaned by analysis of large datasets like media chatter, which we do using advanced techniques. Say, for example, that certain stocks are moving higher. The economic intuition says that other stocks that are highly sensitive to that stock will get a halo effect and move higher in tandem.

Factor-based investing is sometimes described as quantitative or systematic investing since the strategies rely on computer models. How much transparency do investors have into the models?

NC: Factor-based investing doesn’t have to be a black box. On our team, we have more than 170 professionals, approximately half of whom are technologists and engineers, who are researching and testing fundamentally based, economically motivated signals that drive our investment strategies. We think of the investment process like a barbell, heavily weighted by human analysis and oversight at the front and back of the strategy. Humans research and determine the strategies’ factors at the start, which are then implemented through a systematic process in the middle. People then return at the end to review every trade and position, or to intervene during certain periods of extreme stress in an effort to reduce or contain risk. In fact, we recently launched a new series, QuantinomicsTM , that aims to explain the “factors” in factor-based investing and to give a behind-the-scenes look into how we leverage technology and apply human judgment to make data-driven investment decisions.


Read this excellent Interview on BRIEFINGS from Goldman Sachs


Leave a Reply

— required *

— required *