Today, investors, traders, and businesspeople are starting to realise the importance of qualitative data and the place of alternative data in the capital markets and business operations. Surely quantitative filings and financial analyses are still important, but how assets are perceived can be just as important for the current price. Because many financial and business decisions revolve around present and future prices, the influence of non-financial and non-quantitative factors should not be overlooked.
CityFALCON’s new Sentiment Analysis feature is one dimension of alternative data that improves due diligence and research quality. Our machine learning algorithms analyse the specific words used in content and calculate a score indicating whether that content carries a positive, negative, or neutral sentiment (tone of language).
We do this for every piece of content and every CityFALCON entity (company, person, sector, location, etc.) contained in the content, too. So, each story or Tweet or report has a score, and all 300,000 CityFALCON entities have scores calculated from the aggregation of content linked to them. Personal watchlists will receive their own unique scores, too, based on the constituent topics. But we do not stop there: with our Big Data lake of information, we can even offer sentiment analysis over time.
Psychological or Behavioural Finance
In finance classes and traditionally, stocks and markets were valued based on quantitative fundamental properties. This approach still thoroughly informs the price of assets on open markets, especially over the long term. It is impossible to escape the financial realities of a company, no matter the hype.
However, the way in which people perceive an asset, sector, or product informs the valuation to a significant degree. Because it is based more on the behavioural or psychology of the market participants, it is often called “behavioural finance” or “psychological finance”. This has only become more influential as more people participate in markets, and behavioural finance informs the valuation of assets from day-to-day, as people process news and announcements and the overall tone thereof. If everyone is lauding a company for its merits, the price tends to go up, regardless of the financial realities.
Thanks to information platforms like Twitter and social media, people are directly influenced by qualitative information flows. Companies can interact directly with customers and investors, and they can spin any announcement to suit their needs. This includes both pointing more attention to positive developments and blunting the impact of negative ones.
Influential investors and people, including heads of state and CEOs, can also speak their mind and promote their ideas or generate buzz around policies and products. This direct line of communication from companies and key people can sway stock prices and market outlook. Prominent cases in point from Twitter are Donald Trump and Elon Musk. Their Tweets may accelerate price movements that would otherwise be less dramatic based on the fundamentals alone.
Social trading platforms, too, play a hand in encouraging herd mentality. As more people follow an individual, their influence grows, and while the recommending individuals might do thorough fundamental quantitative analysis, their followers may be looking at less technical or quantitative approaches. Whether the herd is truly correct is not as important in the short term, because the demand pressure can alter the stock prices, as markets are not separable from their human participants.
Nor are contemporary markets separate from trading algorithms. Since it is known that sentiment and psychological finance are powerful short-term drivers of valuation, many of these algorithms take into consideration natural language understanding (NLU), including sentiment analysis, when making trading decisions. Now non-algo-traders can benefit from the same data via our API (consumer versions coming soon). With CityFALCON, it is not necessary to build your own NLU systems anymore to benefit from sentiment analysis.
How We Calculate Sentiment Analysis
Sentiment Analysis breaks text down into its constituent sentences and maps the language of those sentences to the corresponding level of optimism or pessimism. Words and phrases like “fail”, “filed bankruptcy” and “collapse” are negative, while “exceed”, “analyst upgrades to buy”, and “record profit” are all positive. Other phrases and words will have neutral sentiment, like “reports earnings tomorrow”.
Our algorithms calculate these scores for individual sentences then aggregate all of that for the entire document, whether it is simply the meta description and title of an article or the entire article or report. The final score for that piece of content is then delivered to the user.
Very popular topics that receive a lot of media attention may have hundreds or even thousands of associated stories. The more content available, the more accurate the sentiment score is for that entity. And because humans cannot read thousands of stories, this approach provides a level of analysis unavailable before Big Data.
Without getting too technical – and without giving away our intellectual property – we also break down sentences into their clauses. For example, a sentence like
Google beats earnings estimates while Microsoft stumbles
Will split into two clauses, one for Google and one for Microsoft. This single headline then becomes just one part of the large pool of headlines and descriptions that are fed into the overall sentiment score algorithm for both Google and Microsoft.
This is achieved in part by our natural language understanding (NLU) entity extraction. This technology identifies entities, like Amazon (company) or Elon Musk (person) or Energy (sector), so we can calculate the sentiment score for each entity in the targeted text (article, Tweet, etc.). Then we aggregate all scores related to each entity to give each entity an appropriate sentiment score. In the same way we can gather all the stories associated with Amazon for your watchlist, we calculate the aggregate sentiment score for all Amazon-related content in our system and attach it to Amazon for you.
If you are interested in entity extraction for your own organisation, we provide the service for enterprise use on internal content. See this blog post or contact us for more information.
CityFALCON has yet another advantage: our sectors hierarchies and aggregated entities. Thanks to the hard work of our financial analyst team, who tirelessly curate the data and train our machine learning models, we know that Microsoft and Google are both part of the Technology sector, an aggregate of all companies within itself. The example headline above impacts the score for the Technology sector. However, Google falls under the Communications industry while Microsoft falls under the Software industry, so this same sentence positively impacts our score for Communications while negatively affecting the score for Software and neutrally impacting the score for the Tech sector. Each one of these aggregated entities (Sectors, Industries, etc.) get their own scores.
A highly simplified scenario is laid out in the image below. Take note that aggregate categories, like Sectors, will be weighted averages – that is, they will skew towards entities that have more articles, Tweets, etc. to contribute to the pool. In the below example, CityFALCON is part of the Tech sector, but we would be much less influential on the sector’s sentiment score than companies like Apple and Amazon, each of which garners far more media and press coverage than we do (for now!).
What Receives Sentiment Scores?
Every named topic in our database receives a sentiment score. This could be a person, a company, an organisation, a country, a city, a sector, or any other entity type. We cover at least 300,000 topics, assets, and locations. Browse all of them in our Directory.
Every single story, report, and Tweet also has a sentiment score, and each day sees up to one million new pieces of content to be analysed and added to our systems. Sometimes the amount of incoming content is even higher if global chatter spikes, as happened with Brexit.
Watchlists also receive sentiment scores, built from the component entities chosen by the user. Thus, every watchlist has an aggregate score, every entity has a score (including aggregated entities like Sectors and Locations), and every story, tweet, or report has a score. We attach scores to pretty much anything that one could logically attach a score to, and this wealth of information affords our users a strong competitive edge.
We even provide sentiment over time. This lets users track how sentiment moves as businesses develop and could lead to investment or business insights.
Use Cases and Applications of Sentiment Scores
Sentiment Analysis is not simply an interesting number to know. As mentioned earlier, alternative data carries importance in the proper valuation of assets, particularly on shorter timescales when psychological finance is more influential.
One use case for sentiment is proper due diligence. When making long (buy) investments, it is important to understand the negative perceptions and circumstances surrounding the company, sector, or other investment target. Before Big Data sentiment analysis, it could be very difficult to find dissenting voices in the media. With sentiment analysis across 5000+ sources, though, CityFALCON lets you discover negative sentiment, even if most analysts and other investors ignore it. In some cases, this negative news can prevent a poor investment, while in other cases, the investor can rest assured that all due diligence bases have been thoroughly covered.
For the contrarian investors, traders, and businesspeople, sentiment helps identify overbought or overhyped territory, too. When all news has been positive over the last three months, the likelihood of euphoria distorting markets is higher, and when markets are distorted, contrarians gain the edge. Bubbles have warning signs, and contrarian whispers, marked by against-the-odds sentiment, are among those signs of distortion. Sentiment analysis scores, especially over time, make identifying the distortion easier. Moreover, during end-of-the-world or hyper-confidence scenarios, sentiment analysis can help discover rational voices that would otherwise be drowned out or to find companies with the potential to resist the trend.
Yet another way to leverage sentiment is in momentum trading. Sudden shifts in the sentiment of content coming from social media and traditional outlets may signal a momentum play. By closely monitoring sentiment over a period of time, traders may be able to identify changing winds before the broader market notices, opening opportunity.
Get Started with Sentiment Analysis
As one of our most sophisticated and sought-after features, we have provided sentiment analysis on the API already. Because of the high value of this feature and the ability to locally store sentiment data forever, we are only offering it at the enterprise level. Send us a message to discuss your situation. Remember, academia, healthcare, and non-profit organisations receive a 50% discount on many CityFALCON products, so open a discussion with us if you’re part of those groups!
Enterprises will need to build their own dashboards and interfaces, but we will provide the fundamental data. A well-designed internal dashboard for employees or clients may look like the cover image of this blog, but clients are naturally permitted to create dashboards however they like.
For our retail users – that is, those who use the CityFALCON website or mobile apps – sentiment analysis will be integrated into the website and mobile app designs shortly. Sentiment will be displayed by default, so whenever you visit the website, you will notice the change right away. The mobile app may require an update if you do not automatically receive updates.
Sign up for our periodic newsletter to be notified of new features and blogs in the Preferences > Notifications section of your CityFALCON account (desktop users can use the form on the right of this page to sign up).
Leave a Reply