logo_2 (2)

Using Unstructured Data In Financial Services

KEY CONCEPT

Financial analysts and investment firms spend significant time analyzing a firm’s earnings reports and monitoring real time performance metrics to assess investment potential of the company. Manual analysis of these earnings reports and performance metrics is time-intensive and the resulting analysis could be incomplete given the multitude of information channels.

Advances in natural language processing (NLP) techniques are enabling firms to more effectively identify investment opportunities by unearthing hidden information from unstructured data. Unstructured data is data that is not stored in any fixed record length format unlike structured data. Examples of unstructured data include free form text in documents, social media feeds, and digital pictures and videos. Unstructured data holds much more useful information than structured data leading to more qualitative insights that cannot be provided by structured data i.e. a pure quantitative approach.

About 80% of data of an organization processes daily is unstructured data” – Gartner 2019

To really understand and evaluate future equities performance, a holistic approach that combines qualitative analysis with quantitative analysis is essential. Qualitative analysis can be derived from unstructured data sources like a firm’s earnings statements, analysts coverage on the firm, social media feeds etc. while quantitative analysis can be derived from structured data sources like a firm’s historic performance metrics.

UNSTRUCTURED DATA SOURCES FOR EQUITY ANALYSIS

Individuals, equity analysts and trading firms have been using quantitative techniques for several decades now for equities analysis. What has been lacking up until recently was a holistic mechanism for qualitative analysis leveraging unstructured data.

With the explosion of Big Data and increased usage of AI techniques like NLP, several government and commercial entities are providing unstructured financial data to public for analysis. Further, with growing social media channels investors have access to unprecedented textual data that can be analyzed for investment opportunities.

To really understand and evaluate future equities performance, a holistic approach that combines qualitative analysis with quantitative analysis is essential. Qualitative analysis can be derived from unstructured data sources like a firm’s earnings statements, analysts coverage on the firm, social media feeds etc. while quantitative analysis can be derived from structured data sources like a firm’s historic performance metrics.

UNSTRUCTURED DATA SOURCES FOR EQUITY ANALYSIS

Individuals, equity analysts and trading firms have been using quantitative techniques for several decades now for equities analysis. What has been lacking up until recently was a holistic mechanism for qualitative analysis leveraging unstructured data.

With the explosion of Big Data and increased usage of AI techniques like NLP, several government and commercial entities are providing unstructured financial data to public for analysis. Further, with growing social media channels investors have access to unprecedented textual data that can be analyzed for investment opportunities.

UNSTRUCTURED DATA BUSINESS USE CASES IN FINANCIAL SECTOR

Financial sectors offers a plethora of publicly available unstructured data that can be harnessed by businesses to create tangible value for customers. Ready availability of massive data makes NLP based machine learning algorithms more precise overtime as they learn from training data. Unlike several other verticals, finance industry has relatively fewer barriers (ex:data availability, regulatory hurdles, etc.) and has been leading the overall industry in AI adoption.

Some of the prominent use cases that can be leverage from this unstructured data in the finance domain include:

1. SENTIMENT ANALYSIS AND PREDICTING VOLATILITY RANGE OF STOCKS

By overlaying sentiment analysis of a stock based on real-time social feeds, corporate earnings and macro economic reports with historical HLOCV (High/Low/Open/Close/Volume) data, investment managers and analysts will get a holistic view of a equities future earnings power.

For example using sentiment analysis on news feeds sites like Dow Jones several insights could be gleaned:

  • Relevance: Is there correlation between macro sentiment and particular equity performance
  • Novelty: Is this ground breaking news or similar news reported already. For example Reuters News Tracer filters tweets through machine learning algorithms to pick up on breaking news before it’s reported elsewhere.
  • Type: Merger event type is more important than a routine dividend issue event
  • Sentiment: Positive/Negative or neutral coverage of the equity

Several hedge funds and firms offering investment services use these to:

  • Execute trades and make bets based on real-time sentiments
  • Identify market patterns
  • Evaluate trading strategies
  • Offer Robo advisory services with minimal human intervention

2. NATURAL LANGUAGE GENERATION (NLG) OF FINANCIAL REPORTS

Financial market data is shared in real time and analysts are overwhelmed with numerical reports from different agencies (quarterly earnings reports, employment numbers, etc). Natural Language Generation allows financial services companies to distill this content from all the various sources and create consolidated summary reports in real-time.

Some typical use cases with NLG include:

  • Asset management portfolio summary: Automatically generate portfolio commentary on a quarterly, monthly, yearly or on demand basis.
  • Personalized wealth management services: Get automated, personalized investment performance reports, on-demand and integrated into client and advisor-facing portals.
  • Generate Dynamic Narratives: Dynamic Narratives transform data and visualizations into insightful stories—making insights easier to consume and act upon.

3. PATTERN RECOGNITION TO OPTIMIZE INVESTMENT OPPORTUNITIES

By monitoring industry sector sentiment patterns in real-time through social media and online channels predictive models can alert fund managers and analysts of impending trends in equity movement. Several large financial institutions use pattern recognition algorithms for optimal profit generation in currency trading. Advances in this area include applying technical analysis to subjective analysis based on unstructured data from multiple channels.

4. INSIDER TRADE DETECTION AND COMPLIANCE VERIFICATION

Through text analysis (NLP) of employee stock purchase reports available through SEC filings it becomes possible to monitor insider trading which otherwise could have been undetected. This helps firms to monitor their internal activities and can also be leveraged by analysts and investment managers for clues about future impacts on stock performance.

By plugging into various publicly available databases financial institutions can instantly perform verification and background checks for activities like driving and criminal records on applicants. In addition Natural Language Processing on regulatory and legal text can identify compliance issues.

5. IMAGE RECOGNITION USING DEEP LEARNING

Firms like JPMorgan Chase with their Contract Intelligence (COiN) platform are using image recognition software with deep learning techniques to analyze legal documents and extract important data points and clauses in seconds, compared to the 360,000 hours it takes to manually review 12,000+ annual commercial credit agreements.

Image recognition using deep learning is also used heavily for spotting patterns, object/human/face recognition, scene understanding, and activity detection with capabilities that can be achieved with scale and accuracy not possible with human involvement.

KEY TAKEAWAYS

Natural Language and Image processing techniques when applied to Unstructured data has countless applications in the financial services ecosystem that are poised to transform the industry in the next several years, including detecting and analyzing brand sentiment; providing investment insights; making banking more efficient and less risky, and identifying fraud.

With advances in AI technologies like deep learning, specialized GPU based hardware and availability of open source datasets processing large scale unstructured data is getting easier. This will open plethora of new business use cases in finance and other industry verticals currently not foreseen.