Explain a few fintech applications of Big data in the
industry?
Impacts and Fintech Applications of Big Data in
Financial Services Industry
• Text analytics
• Natural language processing (NLP)
• Robo-advisory services , AI and Simulation
for Diversified Portfolio Construction
• Risk analysis
Impacts and Fintech Applications of Big Data in
Financial Services Industry-Cont.
❖ Text analytics-the use of computer programs to analyse and
derive meaning typically from large, unstructured text- or voiced-
based datasets, such as company fillings, written reports, quarterly
earnings calls, social media, email, internet postings, and surveys
to help identify indicators of future performance such as
consumer sentiments.
❖ Natural language processing (NLP)-focuses on developing
computer programs to analyse and interpret human language (eg.,
translation, speech recognition, text mining, sentiment analysis,
and topic analysis), evaluate what people are saying –via their
preferences, opinions, likes or dislikes, may be used to monitor
analyst commentary, assign sentiment ratings ranging from very
negative to very positive to aid investment decision making.
Robo-advisory services , AI and Simulation for Diversified
Portfolio Construction
❖ Robo-advisory services-automated asset allocation, trade
execution, portfolio optimization, tax-loss harvesting, and
rebalancing for investor portfolios (eg. Stashaway using
Economic Regime Based Asset Allocation (ERAA)
❖ There are several problems with the way modern portfolio theory
(MPT) and portfolio diversification are often implemented in
practice.
Risk analysis
• Risk Analysis-Big data may provide insights
into real-time and changing market
circumstances to help identify weakening
market conditions and adverse trends in
advance, allowing managers to employ risk
management techniques and hedging practices
sooner to help preserve asset value (eg. helping
analysts to detect buying or selling pressure in
the stock).
Risk analysis-Cont.
• A real-time ESG analytics process drives ESG insights and
scenario analysis with news and media screening. Adverse
event monitoring further enables investors to stay in
control of both reporting and assessment. Advanced NLP
technologies monitor unstructured data in news and social
media articles. The insights can be used in conjunction with
the ESG scores from the rating agencies to provide a holistic
view and thus improve decision making based on
techniques.
• Investors and fund managers use these assistive
technologies to detect and mitigate ESG fraud, such as
Greenwashing. Fraud and anomaly detection platforms that
monitor firm disclosures and are powered by a
comprehensive greenwashing detection framework will
drive investor confidence and flow of funds to truly green
entities—one of the biggest challenges for ESG investors
today.
Risk analysis-Cont.
• NLP-based detection of fraudulent disclosures, news, and
communications is the first line of defense against companies
misrepresenting brown assets to solve their problem of stranded
assets and loss of valuation. Quantitative and measurable data are
needed to enable comparability. There can be real-time
assessments of issuer operations.
• In all these NLP-driven use cases, it becomes clear that large
language models can be greatly beneficial from an engineering
perspective because these models can be adapted to various tasks
with a few shots with examples and prompt engineering (Gopani
2022).
• Over the past few years, some leading software and solution
companies have been established, making use of accelerated
computing to be able to quickly process vast amounts of data.
Some of them have developed numerous AI-based services and
data around NLP for ESG information
Natural language processing (NLP)
-focuses on developing computer programs to
analyse and interpret human language (eg.,
translation, speech recognition, text mining,
sentiment analysis, and topic analysis), evaluate
what people are saying –via their preferences,
opinions, likes or dislikes, may be used to
monitor analyst commentary, assign sentiment
ratings ranging from very negative to very
positive to aid investment decision making.
Common NLP Task
• Summarization
• Topic extraction
• Search/information retrieval
• Question answering
• Sentiment analysis
• Named entity recognition
Common NLP Task
• Summarization
• News aggregators use this technique to summarize long articles to send digestible
content to inboxes. We may also wish to summarize corporate filings or
presentations for quick consumption. The key challenge with this task is
understanding and capturing the key relevant points of the large document. This
challenge is made more difficult because different investors may weigh the
content of the text differently, and as a result, the optimal summary depends on
the user or usage.
Common NLP Task-Cont.
• Topic extraction
• The key themes and topics in an article can be extracted using supervised, semi-
supervised, or unsupervised methods. In the supervised approach, the model is
trained to look for specific keywords related to a predefined theme, whereas the
unsupervised approach attempts to infer the themes being discussed. The semi-
supervised approach is a hybrid approach in which seed words can be used as
starting points to identify themes.
• One example is in the context of central bank statements. Common themes
debated during central bank meetings in the United States include inflation and
growth.
• We may seed the topic modeling process with these words and then apply various
language techniques to find new words and phrases resembling these themes to
derive the final topics from the documents.
Common NLP Task-Cont.
• Search/information retrieval
• We may be interested in looking for specific terms or references from the text.
Examples may include competitor or product references in a document. An analyst for
Apple may be interested in finding all references to the iPhone in corporate
documents or news articles.
• Question answering
• Similar to the search task, we are interested in finding information in a document.
However, the QA task is focused on answering specific questions rather than
returning references to those questions.
• In the previous example regarding the Apple analyst, he may be interested in the
actual iPhone units sold rather than simply references to the phone. Another example
may be extracting the interest rate or the covenants in a loan document. A specific
application of this question answering task is the chat box, where specific and
relevant responses are needed to reply to questions.
Common NLP Task-Cont.
• Sentiment analysis
• Should the text be viewed positively or negatively? This will again depend on the
user and application. For example, certain words, such as “debit,” “liability,” and
“resolve,” may have different meanings depending on the context. The first two
terms are normally viewed as negative words in common parlance but tend to be
more neutral in a financial setting since they are common terms in financial
statements. “Resolve” may be viewed as positive in most settings, but “did not
resolve” should be viewed as negative.
Common NLP Task-Cont.
• Named entity recognition
• Extracting entity names from text is a common but important task. Entities include
countries, organizations, companies, individuals, places, and products. By
identifying the entities in a document, investors can link the article to other
information on the entities and thereby create a comprehensive view before
making decisions. Client interactions are no different: Determining client-relevant
text is critical to a complete understanding of a client. While this type of data is
critical, the problem is difficult; for example, separating articles regarding the
technology company Apple and the fruit apple is not trivial.
