Anticipate Market Moves With News Sentiment Analysis

Use Cases & Projects, Dataiku Product John McCambridge

Everyday, thousands of newspapers, news websites, specialist blogs, paid news networks, and many other sources publish financial news. They may cover stocks and other financial instruments movements as well as company announcements and journalistic investigations. This information is crucial to trading desks and related functions managing funds across banks and other financial institutions. 

Which stocks are most likely to move based on current news sentiment? What are the underlying news events driving volatility for a specific ticker? What historical insights can be gained through systematic analysis of past news events? These are the kind of questions that traders, equity analysts, or portfolio managers ask themselves. They might want the answer to these questions to simply perform some ad hoc analysis or as part of a complex automated system that may require a constant feed of information. 

Either way, efficiently managing the flow of information fueling decision making is a process that welcomes automated solutions, especially when the task is done repeatedly. In this article, we will look at one powerful approach: using an automated alert system based on a model that links news headlines with stock price movements. We will then see how a dedicated Dataiku industry solution can help you quickly start getting value from this system.

absolutvision-WYd_PkCa1BY-unsplash

A New Approach to Leveraging News in Trading

We could imagine a room full of people equipped with computers with one objective: continually review all available news platforms and find useful information on a certain stock. We can also easily imagine how costly, repetitive, and inefficient such work would be. What modern solution out there can help provide us with real-time insights into market-moving news articles, allowing traders to quickly identify and investigate stocks and articles of relevance? 

This solution would ideally be able to help identify stocks that are most likely to move or become volatile based on current news. This would create immediately actionable real-time insights. We would also need to keep comprehensive and transparent historical insights if we want to allow the creation of more informed responses to future news events. 

The solution Dataiku has built intersects a specified stock universe against news articles in real time. Using the resulting dataset and a historical anomaly detection model, stocks likely to experience atypical movement and the potentially driving articles are ranked by volatility score. Let’s look at what you might find in this solution exactly, but first a reminder of what an Industry Solution actually is.

How Can Dataiku's Industry Solutions Help You Reach Full Potential?

Industry Solutions are Dataiku add-ons accelerating the journey to realize advanced or foundational industry-specific use cases within your organization. They are an operational shortcut to achieve real-world business value. Taking advantage of Dataiku’s core features, they are built to be fully customizable and entirely editable.

They come with:

  • A user-friendly interface that enables fine tuning to match specific business requirements
  • Ready-to-use dashboards that can be customized
  • Documentation and training materials

Dataiku industry specialists develop solutions for every vertical, among which:

As a result, business professionals experience a boost in AI productivity and can rationalize their resources.

How Does It Work in Practice?

The News Sentiment Stock Alert System solution provides a reusable project wireframe to accelerate the development of analytics tailored to your data and business structure. It includes a Dataiku application that eases parametrization of the stock universe and news sources.

With this solution, traders, equity analysts, and portfolio managers can access:

  • Ready-to-use volatility scores at the ticker level based on news sentiment analysis
  • Immediately actionable real-time insight into ticker-level market movements
  • Comprehensive and transparent historical analysis which enables the creation of more informed responses to news events
  • Anomaly detection and stock price analytics including Principal Component Analysis (PCA)

From a user perspective, the solution is made of three easy-to-use components: 

1. Real-Time News Scoring

Individual news and news aggregated at the ticker level are scored and ranked. A first view shows the stocks with their volatility score and, the closer it is to 100, the more likely it is to exhibit an anomalous move.

real-time alerts

Then, the user can investigate all the news that was published on this source for a selected ticker and see how each individual piece of news was scored. The link directs to the full article for further analysis.

real time news score

2. Case Study

Each past anomaly can be displayed to have a deeper understanding of how the algorithm labels the data. The stock prices are plotted around the move that was considered an anomaly (represented by the vertical black line). When news articles are published about this stock during the neighboring days, they are also displayed on this tab.

past anomaly analysis

stock price

3. Historical Price Anomaly Detection

To analyze stock price evolution on a longer term horizon, another tab allows the user to plot the stock prices on the time window of their choice. Vertical lines indicate labeled anomalies in the dataset.

price evolution

4. Historical News Scoring

To explore past news and their scores, each individual news item is scored. Across time, it lets the user understand how the news scoring model performed on the whole dataset.

most impactful news

5. Stock Price Analytics

The covariance matrix of stock price log returns can be reduced using PCA. Using this decomposition, stocks can be visualized on a two-dimensional plane with the two first components. Each dot represents a ticker and is colored according to its sector.

stock price analysis

Thanks to this spatial representation, distance between stocks can be defined and clustering applied. This further layer of analysis gives additional insight on how stocks and sectors should be grouped together using only stock prices as input information.

labels and sector in Dataiku

Start implementing your stock alert system in minutes, with these simple requirements:

Data needed:

  • Stock universe: a simple list of tickers/stocks to evaluate (Note: project uses Yahoo Finance for pricing data, with no credentials required)
  • News Feed: pulled from Yahoo Finance using the EOD API that requires an API key to run
  • Dataiku version: 9.0 or later

You May Also Like

Alteryx to Dataiku: Working With Datasets

Read More

Demystifying Multimodal LLMs

Read More

I Have AWS, Why Do I Need Dataiku?

Read More

Why Data Quality Matters in the Age of Generative AI

Read More