Will artificial intelligence transform investment research?

November 15, 2018

A blog post by Rohit Kataria and Doug Dannemiller, investment management researchers at the Deloitte Center for Financial Services

Investment research and analysis are evolving rapidly, with proliferating data sources and expansion of AI applications. Portfolio managers and analysts rely on financial statements, earnings call transcripts, press releases, investor presentations, blogs, news articles, and sell-side reports for investment research. Synthesizing information originating from multiple sources and building proprietary quantitative models takes enormous human effort and time. AI tools not only enable large-scale data processing at a rapid rate, but also integrate traditional data sources with new ones such as web traffic, web search trends, and social media data. Application of AI to these data helps portfolio managers and analysts save time and uncover hidden signals, contributing to improvements in forecasting, investment decision-making, and idea generation.

Natural language processing (NLP) and natural language generation (NLG) are the branches of machine learning that enable computers to understand and generate natural human language. NLP processes natural language by transforming text into structured data, while NLG interprets and analyzes structured data and converts it into readable format. Application of these technologies results in a machine-generated report that conveys insight from computation of the data. It is able to make sense of spoken and written language. This approach overcomes some inherent limitations associated with rule-based algorithms, which struggle with processing unstructured data and lack the intelligence built from thousands of corrective iterations that machine learning conducts. Traditional rules-based algorithms don’t self-correct.

Augmenting research with NLP

Investment managers are integrating NLP capabilities into their analytics platforms. NLP tools can augment investment research in the following ways, among others:

• By interpreting management sentiment during earnings calls to predict a company’s future performance. 
• By parsing sell-side reports for wording to gauge changes in analysts’ projections.
• By sifting through volumes of unstructured data sources, such as blogs, news reports, social media, and sentiment data to identify trends and potential investment ideas.

Some investment management (IM) firms are trialing NLP technology for investment decision-making. They are using NLP technology to score each piece of information a portfolio manager consumes into positive and negative groups. A positive score indicates the likelihood of a rise in company performance or corporate value, and a negative score means it is unlikely to rise. Trials also translate textual data from websites and blogs into quantitative scores. The goal is to augment the investment decision-making ability of portfolio managers by increasing throughput and reducing bias and other errors prone to humans.

Considerations before implementing NLP/NLG

NLP can also be used to generate investment ideas. Using NLP enables firms to reduce the time spent conducting initial research on one company the current average of four to five hours to 30 to 45 minutes.

Margin compression and regulatory mandates are driving investment managers to pay for research directly, meaning the buy-side investment research landscape is likely to undergo a profound change. Investment managers may expand in-house research and analysis capabilities by making long-term investments in advanced technologies like NLP and NLG to reduce their dependency on external research. The pace at which the natural language application of AI is accelerating. Automation of the investment research and analysis function at scale could soon be a possibility. Business leaders at IM firms may need to take the following factors into consideration before starting an NLP/NLG implementation:

1. Piloting: Undertaking pilot projects/proofs of concept before full implementation to test whether desired results can be achieved.
2. Deployment: Integrating NLP and NLG tools into the data analytics platforms accessed by analysts and portfolio managers to enable widespread deployment across the firm.
3. Data format: Data sourced from a vendor in a structured form can be fed directly into an NLG process. While NLP is required as a preliminary step for unstructured data.
4. Talent: Assigning a team of domain experts, or hiring external specialists to champion implementation.

In the coming years, computers will likely be able to process text and speech, enabling them to generate narratives about potential investments based on thousands of times more information than analysts alone can read. This development could completely transform the investment research and analysis function at IM firms.