Foreign exchange (FX) traders are starting to turn to sentiment data as an additional input to their trading strategies, as firms seek new insights to deliver an edge for alpha generation and hedging purposes, with research conducted by sentiment data specialists and market participants showing that a sentiment overlay on traditional currency factors can deliver significant outperformance. Matt Weller, global head of research at trading solutions provider Gain Capital, says that as this data improves—though evolution has been slow in FX—more clients are asking to use sentiment as part of their investment process. 

“We’re seeing growing interest in sentiment as a third data type, alongside technical and fundamental data,” says. “It’s something we’re looking at very closely. We are looking at the opportunity to provide clients with a suite of fundamental and technical tools, and we think sentiment data is an input that clients may want to incorporate into their trading strategies.”

Yet while interest is growing, progress has been slow. Whereas traders have applied sentiment data to equities trading for some time, its use in other areas, such as foreign exchange, fixed income, and commodities, is still in the early stages, says Yue Malan, senior analyst at Aite Group.

“The equities markets are saturated in some ways. There is so much sentiment and other alternative data already used in the equities markets, so firms are looking for something else to create an edge,” she says. “Applying sentiment to FX trading is definitely happening, but is still very niche and in the early stages.”

One such early adopter is Alexandre Giulietti de Barros, a data science consultant based in Spain, who recently adopted sentiment data for a Brazilian financial firm that wanted to incorporate machine learning into its US dollar–Brazilian real (USD–BRL) trading operations.

“I did some research on what kind of alternative data could give them an edge, such a scraping data from websites and creating my own sentiment data. I was pretty sure there was a close connection between news and forex, especially because USD–BRL is so volatile,” Giulietti says.

Incorporating sentiment delivered measurable returns almost immediately—with a combination of sentiment data on the Brazilian real and other emerging-markets news producing the best results—in contrast, other inputs only provided “marginal improvements,” he says.

The firm decided to use sentiment to support its long-term FX strategies—meaning forecasts for two months ahead—though Giulietti says he’s sure there is also alpha to be gained from utilizing sentiment on an intraday basis because of its real-time nature and potential to impact short-term strategies.

An even earlier adopter of sentiment for FX is Saeed Amen, an independent currency trader and founder of Cuemacro, which performs quantitative analysis of macroeconomic markets, and provides FX-focused strategies, analytics, and indexes. Amen has been researching the impact of news sentiment on FX markets since 2014, and has assessed offerings in this space from vendors, including Bloomberg and RavenPack.

Both Giulietti and Amen started out creating their own datasets—collecting raw data, cleaning the data to remove HTML tags and invalid observations, structuring the dataset to add descriptive tags, applying sentiment analysis to text, filtering the dataset, creating an indicator, and applying trading rules to the indicator—which Saeed says is a “hassle” to do by yourself. “Originally, I had to do a lot of work to extract a trading signal,” including aggregating topics and creating his own sentiment indicators, he says.

Thus, one obstacle to broader adoption of sentiment in FX markets, says Aite’s Malan, is a lack of commercial datasets specifically designed for use by FX traders.

“The idea of news sentiment data has been around for some time, but now there are more providers. And most of the focus has been on equities—there has not been so much focus on asset classes like FX or commodities,” Amen says, suggesting that this may be because the impact of sentiment is more obvious on equities, and while an equities trader may have a portfolio or watchlist comprising a large number of stocks, a currency trader may only trade a couple of currency pairs.

That’s where Spanish news sentiment data provider RavenPack comes in: The vendor has specialized in applying sentiment analysis to news since it was founded 17 years ago. Last year, RevenPack’s chief data scientist Peter Hafez turned his attention to applying its data to currency trading, and—after six months of analysis—recently published a whitepaper demonstrating how sentiment can be applied to FX trading, and its potential benefits. According to the paper, applying sentiment on its own to FX strategies produced “semi-decent” results, but delivered more powerful results—outperforming an equally-weighted benchmark strategy by 156 basis points per year—when used along with other factors, and was able to “enhance both long and short baskets,” Hafez says.

When looking at sentiment as an overlay to maintain, value, and carry factors, “we were able to enhance all three factors quite consistently over time,” Hafez says. In addition, sentiment acted like a rising tide that lifts all boats, spreading gains across all currencies, rather than benefitting some more than others.

RavenPack tracks more than 20,000 sources of news, from which it creates analytics based on just headlines, rather than a story’s full text—reflecting the “headline-driven” nature of the market, Hafez says—and also looking at the volume of headlines compared to normal levels in the economies associated with currencies. The vendor normalizes these headline counts to account for the fact that not every economy will generate the same amount of news. RavenPack also excluded news specifically relating to currencies. “What drives currency prices is not news that mentions currencies. What really drives this is economic and [geopolitical] news that doesn’t mention a currency, but where you can infer the impact on a currency,” Hafez says.

Instead, the vendor decided to focus on macroeconomic issues such as economic guidance, elections, and employment figures. Hafez says that though different firms may focus on different factors, the sentiment is likely to have a high correlation with whatever factors they use.

He adds that in the few weeks since publishing the whitepaper, the vendor has already seen a lot of interest from firms seeking to incorporate sentiment into their trading models. For interested parties, RavenPack can provide access to all its resources—including access to its data science team—for a trial period, with the aim of turning those trials into paid subscriptions to its data.

Although Giulietti began his currency trading analysis by creating his own dataset, he approached RavenPack last year, and began working with the vendor’s data last December.

“I told RavenPack that my interest was to apply their data to a new market. In fact, they were already working on currency markets, though I didn’t know that when I contacted them,” Giulietti says. “When I first got the data from RavenPack, it was easy to set up my first model, and after a couple of months, I had everything set up in such a way that I didn’t need to change anything about the sentiment data.”

However, Gain’s Weller sounds a few notes of caution about being too bullish about sentiment data. For example, the FX markets are driven by different factors than equities markets, and there are other datasets that may deliver more practical insights into investors’ trading decisions.

“There is a big opportunity, but that opportunity will always be greater for speculators. The FX markets are much larger and more commercially driven than equities markets. Large multinational companies might have to hedge large amounts of currencies, regardless of where they think the price will move … so the potential for excitement and emotion to drive sentiment is less.

He continues: “I find positioning data is more valuable than pure sentiment. Sentiment is not as reliable as positioning data, which shows who has long or short positions. There’s more signal in what people are doing, rather than what they’re saying. In addition, generally, we found that sentiment indicators are most valuable at the extremes when markets are most bullish or bearish. That middle 80% of the time is more difficult to interpret, so the biggest risk is that traders might put too much emphasis on it and ignore more established indicators.”

Amen also warns against traders piling into using sentiment alone as a basis for trading decisions—not only because of risk, but because the data could quickly become worthless. “If a lot of people are using the same specific dataset, the returns may diminish over time. The way to prevent that is to use it in conjunction with other datasets,” he says.