News and Sentiment Analysis in Finance – A Primer

by Hugh Taggart
  • The field of news and sentiment analysis is only just beginning to take hold in finance.
News and Sentiment Analysis in Finance – A Primer
Photo: Bloomberg
Join our Telegram channel

The field of news and sentiment analysis is only just beginning to take hold in finance, even though some of the larger firms have been at it for over five years. But what, exactly, is it?

Most people, if not all, in Forex are familiar with technical analysis – the study of patterns in price and volume data with the aim of forecasting future price behavior. Sentiment or news analysis is a bit like that – except the underlying data is derived, by computers, from text published about the markets. So, it’s about finding patterns in news content that may be able to forecast future price behavior.

At its most basic level, news or sentiment analysis could just be about counting the number of times an entity, e.g. a forex pair, is mentioned in the news – or the number of positive versus the number of negative words (from a specific financial dictionary). That might give you an indication of volatility and perhaps Liquidity , but it’s a bit crude.

The technology currently available to us is now much smarter than that, thanks to a large degree to the likes of Google and the NSA (and, of course, their lower profile equivalents). In laymen’s terms there are a couple of approaches to this problem that the current vendors use. Of course, vendors will lay claims to all manner of uniqueness, but I’ll ignore the nuances for now given this is a primer.

One approach is to look at the ‘mood’ of the market surrounding particular keywords or phrases. Mood is usually related to some of the basic human emotions like fear, greed, positivity, negativity, hype and so on. So, you might choose a phrase like “Non-Farm Payrolls” and ask your software to return mood metrics from all the articles containing your key word or phrase. Thus you’ll get a picture of how the market’s mood toward your phrase changes over time. And you could interpret things like rising fear as a volatility warning, rising positivity or negativity as directional signals and hype as a reversal warning.

Another way to skin the cat is to analyze the text to detect financial events – usually events from a predetermined list – and then use the event type to determine the sentiment toward an entity (e.g. a forex pair). The frequency of that event type being mentioned for a particular entity can also be used as a proxy for the event’s importance. This event-based approach to sentiment analysis is perhaps a little more difficult to work with because it doesn’t plot nicely (events can be few and far between), but it tends to be more accurate, particularly over the short-term, because the sentiment presented is specific to a particular entity at a point in time.

I have introduced some pretty basic concepts about news and sentiment analysis in this post, but I will in future posts talk about these approaches in more detail and discuss things like which content sources might give the best signals. Until then…

The field of news and sentiment analysis is only just beginning to take hold in finance, even though some of the larger firms have been at it for over five years. But what, exactly, is it?

Most people, if not all, in Forex are familiar with technical analysis – the study of patterns in price and volume data with the aim of forecasting future price behavior. Sentiment or news analysis is a bit like that – except the underlying data is derived, by computers, from text published about the markets. So, it’s about finding patterns in news content that may be able to forecast future price behavior.

At its most basic level, news or sentiment analysis could just be about counting the number of times an entity, e.g. a forex pair, is mentioned in the news – or the number of positive versus the number of negative words (from a specific financial dictionary). That might give you an indication of volatility and perhaps Liquidity , but it’s a bit crude.

The technology currently available to us is now much smarter than that, thanks to a large degree to the likes of Google and the NSA (and, of course, their lower profile equivalents). In laymen’s terms there are a couple of approaches to this problem that the current vendors use. Of course, vendors will lay claims to all manner of uniqueness, but I’ll ignore the nuances for now given this is a primer.

One approach is to look at the ‘mood’ of the market surrounding particular keywords or phrases. Mood is usually related to some of the basic human emotions like fear, greed, positivity, negativity, hype and so on. So, you might choose a phrase like “Non-Farm Payrolls” and ask your software to return mood metrics from all the articles containing your key word or phrase. Thus you’ll get a picture of how the market’s mood toward your phrase changes over time. And you could interpret things like rising fear as a volatility warning, rising positivity or negativity as directional signals and hype as a reversal warning.

Another way to skin the cat is to analyze the text to detect financial events – usually events from a predetermined list – and then use the event type to determine the sentiment toward an entity (e.g. a forex pair). The frequency of that event type being mentioned for a particular entity can also be used as a proxy for the event’s importance. This event-based approach to sentiment analysis is perhaps a little more difficult to work with because it doesn’t plot nicely (events can be few and far between), but it tends to be more accurate, particularly over the short-term, because the sentiment presented is specific to a particular entity at a point in time.

I have introduced some pretty basic concepts about news and sentiment analysis in this post, but I will in future posts talk about these approaches in more detail and discuss things like which content sources might give the best signals. Until then…

!"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|} !"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}