ANNOUNCEMENT: Expert Investor is now PA Europe. Read more.

How much insight can sentiment data really offer?

Especially when it comes from non-traditional sources like social media and news articles

|

David Burrows

While many equity investors commit to certain fixed beliefs in their investment approach – such as a focus on buying cheap companies, or those with a healthy balance sheet – the focus at NN IP is to follow an adaptive, strategy-agnostic philosophy, says Tjeerd van Cappelle, head of automated intelligence equity investing.

In other words, they aim to adapt to changing market circumstances in the way they invest, which information they use, and which industries they invest in at different points in time.

For this, van Cappelle insists, sentiment data is an invaluable tool.

Data points

It can originate from ‘traditional’ sources, such as fund positioning and the opinions of sell-side analysts, but modern-day sentiment data is more likely to come from news sources, social media, or online blogs.

“There are all sorts of people sharing their opinion on companies,” says van Cappelle. “And we use that in our investment process to see how they feel about companies and how that will influence future returns.”

Richard Peterson, whose company MarketPsych provides sentiment data based on media analytics, explains how technological advancement enables asset managers to gather conversations on news and social media in real time.

Also, computing capabilities and an advanced understanding of behavioural economics make it possible to digitise these conversations and distil them into the elements that drive human decision-making.

This is bolstered by the accumulation of huge amounts of data that goes back decades – which gives a deeper, different insight into what an unexpected development, such as a spike in uncertainty, might mean.

“We can test that statistically and say, ‘does this correlate with the future price action of the company? And can we as investors take advantage of that in a predictable, systematic way?’,” Peterson says.  

Harvesting this data is not a single-step process; it requires careful programming and a deep understanding of human biases, for example to weed out spam or irrelevant comments.

Peterson adds: “On social media you get a lot of people posting irrelevant content, but if it’s about that company and they are an investor, it may ultimately be relevant because they are likely participating in the share price movements.”

Applying the data

Investors need a high level of analytical skill to properly apply the data in their investment process, Peterson says. “The data itself can certainly be used to find simple correlations, but actually deploying it in an investment strategy is quite challenging.”

It often takes an entire team of developers, theorists and applied investors to put this data to work. For the successful application of sentiment data, it’s also necessary to have a framework for understanding markets historically, as well as human behaviour in markets. 

When van Cappelle and his team receive sentiment data, the next step is to carefully analyse how it translates into future returns that are relevant to portfolio construction. This doesn’t simply mean investing in the companies or industries that receive the most attention.

“Typically, a lot of attention goes to a few companies or a few sectors, but in the end we need to create a balanced portfolio. We don’t want to be fully invested in one company.”

Based on the information, van Cappelle and his team make two types of decisions: one on industry positioning, and the second on which specific companies to invest in, based on what the information tells them about which companies will outperform their peers.

“What we see is that there’s a lot of value in the extremes,” van Cappelle says. “Companies that get an extreme level of attention perform well, and that also lasts quite a while. It’s not immediately priced in. It can take months.”

No novelty

Tancredi Cordero, chief executive of Kuros Associates, argues that the use of sentiment data in the investment industry is far from a novelty and for years traders have closely watched other market participants, taking into account behavioural patterns as well as additional non-financial data.

But like van Cappelle he believes that what is novel is the way modern technology allows investors to process a virtually infinite amount of data (sentiment or numeric) more accurately in order to predict market or stocks movements.

“From the likes of Bloomberg to small start-ups, financial software providers already allow their clients to monitor how many tweets or Google searches a company or a topic is receiving. We recently saw hedge fund managers making millions by monitoring the Reddit chats discussing GameStop and AME Entertainment,” Cordero explains.  

He also points out that pure quantitative money management institutions like Renaissance Technology and AQR are increasingly implementing their algorithms with more sentiment data rather than traditional numeric data.

“We are entering a new phase in financial markets where the information driving returns will be available publicly to the masses, as opposed to the elite few which means managers need to have an arsenal of algorithms able to process them,” Cordero says.

He stresses that sentiment data is a very useful tool to incorporate into the investment process whenever possible and appropriate.  

“As a rule of thumb, we think that sentiment data are more reliable for momentum stocks rather than value ones. For instance, stocks with high valuations that don’t have considerable hard assets on the balance sheet and that may also have negative EPSs typically have prices that are more market flow driven than fundamentally driven.”

In essence, the price moves more accordingly to the promises of a rosier financial future. Hence since market flows are mostly determined by the short-term behaviour of market participants, sentiment data becomes extremely valuable when it comes to forecasting price movements of momentum stocks.

Conversely, stocks of companies that are more established, asset rich, profitable, larger market cap, will be more fundamentally driven and stock price movements will be less influenced by sentiment.

There are exceptions to this, as Cordero points out, for instance when you have catalysts like M&A or litigation, but by and large he sticks to the general rule when it comes to using sentiment data. 

He concludes: “I think sentiment data will be used more and more because AI software will continue to prosper as an everyday tool within the financial industry. Using AI and data as much as possible is today a requirement for firms to survive, not an option”.