Asset manager of the year: Goldman Sachs Asset Management

Risk Awards 2019: Firm’s algos pick through earnings call transcripts to figure out what analysts really think

Nick Chan
Nick Chan, Goldman Sachs Asset Management

Analysts aren’t known for their combative style in earnings calls with company managers. But even if the questioning can sound sycophantic, the nuances of what’s said can reveal something about what analysts really think. Goldman Sachs Asset Management is checking for those clues.

Or rather, Goldman’s robots are checking.

The asset manager’s computers are using machine learning to ‘read’ transcripts and assess the tone of questions during unscripted Q&A sessions as part of analyst calls, comparing the phrasing of questions with each analyst’s usual style. (The content in the scripted parts of calls is often priced into the market very quickly.)

“When you think about the psychology of the relationship between analysts and companies, usually analysts want to preserve a good relationship with companies so they’ll say things like ‘great job’, ‘great quarter’, ‘congratulations’,” says Nick Chan, a managing director on GSAM’s Quantitative Investment Strategies (QIS) team.

“What we’re trying to calibrate for each analyst is their propensity to praise management when they ask questions over many, many years. Let’s say an analyst today is complimenting and praising much more than their average. To us, that is an unconscious tell that they’re perhaps more bullish. They may not even realise they are telegraphing that information.”

GSAM is doing this calculation in real time, not just for one or two analysts, but across many hundreds of analysts across tens of thousands of companies, globally, throughout the year – to detect the tiniest change in sentiment and get ahead of the market.

It’s one example of how the asset manager is exploring beyond the best-known types of alternative data, such as credit card transaction data, satellite imagery and footfall traffic. In addition to its work on scouring analyst transcripts, the firm has spent much of the past year training algorithms to read all sorts of text using a form of machine learning called natural language processing.

While many fundamental and quantitative asset managers are pumping alternative data for alpha, the sheer scale of the operation at GSAM sets it apart. Its algorithms churn out an alpha score for more than 13,000 companies every day.

“We have a view every day of every publicly traded stock in the world, so the scale there is very different from a traditional manager who might have sector analysts who know a couple of dozen stocks, each very, very well. But it would be humanly impossible to know 13,000 stocks exceptionally well,” Chan says.

“The scale and grandness of the data we’re using allows us to play the depth game and the breadth game,” he adds – in other words, Goldman can get to know a lot about a lot of companies.

Another area where GSAM has exploited alternative data is to identify cross-company linkages.

Most people are familiar with the idea that some companies’ fortunes can be linked so that their stock prices are sensitive to each other, Chan says. The most intuitive example is customer-supplier relationships. But GSAM believes there are thousands of other relationships with a similar kind of knock-on effect on pricing that mostly go unseen.

“They’re not as obvious as traditional customer-supplier relationships, and in fact, we believe most people are not even noticing them,” Chan explains.

We have a view every day of every publicly traded stock in the world, so the scale there is very different from a traditional manager
Nick Chan, Goldman Sachs Asset Management

The firm has built algorithms to detect these linkages by sifting through news articles, television news transcripts, press releases, analyst research and even patent filings – looking at how often companies appear together. Patents, for example, often cite other companies’ patents or similar research.

What’s driving GSAM’s interest in alternative data? Partly, the firm is responding to the commoditisation of factor investing; due to investor crowding, the proliferation of data and technology, evolution in investor behaviour, and sometimes for unknown reasons, many of the alpha signals of yesterday have lost their potency.

Take traditional price momentum, which has become a widely traded, potentially crowded, plain vanilla factor, Chan says. GSAM now has close to a zero allocation to momentum in US equities.

“The whole idea of this alpha to beta continuum… We’ve seen it happen in front of our eyes. So that’s partly the motivation always to innovate and explore new research,” he adds.

Fifteen years ago, the story was very different; GSAM had around a dozen different factors in its equity models. Today, that number is more like 200, and 50–60% of them use some kind of alternative data or machine learning.

To ensure GSAM is not caught relying on yesterday’s alpha, the asset manager monitors the volatility and crowdedness of each factor or signal daily, to determine the weighting of each in the firm’s investment strategies.

“It’s not like there’s a cliff where suddenly [a signal or factor is] not effective anymore. We monitor this over time, and for the factors we perceive to be less and less effective, we may remove or down-weight them over time,” Chan says.  

Sometimes, though, threats arise more quickly. And despite its systematic approach, Chan says his team’s investment process starts and ends with human input. The firm will, and has, overridden its machines during periods of market stress if it feels models are unlikely to perform so well. Take the unrest in Turkey over the summer.

“There was a period in Turkey this year when the lira was falling, the equity market was doing poorly… and there was a lot of geopolitical risk … where we did exactly this kind of intervention. We implemented an overlay from a risk perspective that we didn’t want to add to any of our long positions there,” Chan says.

Even if sometimes it still seems better to switch the machines off, clients welcome Goldman’s tech-fuelled approach. Its Insight funds have won four- and five-star ratings from Morningstar in recent years.

And by allocating less risk to traditional factors such as value and momentum, and applying machine learning to more unusual, unstructured sources of data, the firm has created an array of new signals that set it apart in a sector that, to many, looks crowded.

  • LinkedIn  
  • Save this article
  • Print this page  

You need to sign in to use this feature. If you don’t have a Risk.net account, please register for a trial.

Sign in
You are currently on corporate access.

To use this feature you will need an individual account. If you have one already please sign in.

Sign in.

Alternatively you can request an indvidual account here: