1 of 24

Team AGILE

2 of 24

Problem Statement

How can we help portfolio managers analyse earnings calls in a less arduous manner to identify risks and opportunities plus what insights can be derived from them and how they can be presented in a usable manner

3 of 24

Introduction

Earnings calls - Their importance

Analyzing earnings calls can quickly extract novel and useful insights to:
Assess potential issues and business activities of companies.
Show how companies reflect on their performance and conditions.
Be an incremental source of information on companies’ results and numbers.
Act as key factors in the decision making process.

@Krystal, please read. Thx.

Their importance

Offers the potential to identify potential issues faced by companies & how they plan to deal with those issues as well as their business activities like their revenue stream
Reveal the notions and perceptions of company executives on their companies’ performance and conditions
Q&A Sessions in calls can be a potential source of incremental/supplementary information regarding the results and quarterly figures.
Insights obtained overall can be part of key factors in the decision making process of investments.

Why it matters

Opening the potential for transcripts to be analyzed using Natural Language Processing (NLP), making them useful for the following reasons

Identify possible opportunities and risks companies face that could affect stock performance.

Deeper understanding of companies’ directions and perceptions in relation to their performance overtime.

Companies can disclose supplementary information in such calls that aren’t easily conveyable via traditional financial reporting channels.

4 of 24

Proposition

5 of 24

Our solution - InvestSight

Quantitative + Qualitative Insights for Corporations in the US Stock Market

6 of 24

InvestSight

Analyses earnings call transcripts with Machine Learning and Natural Language Processing

7 of 24

Qualitative Analysis Workflow

Getting Data

Cleaning

Processing

Visualisation

Analysis

8 of 24

Getting Data + Cleaning

Data will be mainly gathered from Finnhub.io’s Earning Calls API.
Basic cleaning to be done, removes courtesies and captioned sounds

AAPL Earnings Call Sample:

9 of 24

Processing

Cross Reference using TF-IDF (term frequency–inverse document frequency)

Word scores to be used in subsequent calculations

Usage of Scikitlearn’s TfidfVectorizer >>

10 of 24

Analysis - Part 1

Every sentence will go through MonkeyLearn’s Sentiment API

Given a “good”, “neutral” or “negative” sentiment

MonkeyLearn API Demo >>>

11 of 24

Analysis - Part 2

Calculation of the final score of the sentence:

Where:

Sentiment is a set value, “Good” = 1, “Neutral” = 0.5, Bad = “0”

Confidence = Machine Learning Model Confidence

12 of 24

Analysis - Final Part

Final Score Formula:

Where:

Maximum possible score = Total Score, assuming all sentiments are “good”

13 of 24

Visualisation - Part 1

The final score over time will be shown to the user:

14 of 24

Visualisation - Part 2

The top 5 most negative and positive sentences will be shown to the user.

15 of 24

Visualisation - Part 3

Word cloud: The words with the highest TF-IDF score and occurrence will be shown as the ‘largest’

16 of 24

Visualisation - Part 4

Full transcript:

Contains a breakdown of sentences highlighted in red, green and black.

Indicating negative, positive and neutral respectively.

It will also pick out which words contributed the most to the score of the sentence.

17 of 24

Benefits

Shorter time taken to analyse earnings call transcripts
Historical transcripts are analysed, no need to worry about missed calls
Direct insights into companies
Identify potential opportunities and/or risks
More informed investment decisions

This will be done through a machine learning model using textual analysis algorithms and Natural Language processing will analyse the transcripts retrieved from an API to identify important words, phrases and sentences and determine their sentiments (Positive, negative or neutral). The results and findings from the model will then be displayed to users in a dashboard.

Lesser time taken to analyse earnings call transcripts

InvestSight seeks to lend a helping hand to users in by extracting critical information from an unstructured dataset like a call transcript and present it in a succinct manner which would make it easier for users to gain insights from earnings calls whereby instead of having to sit in to listen to an earnings call and detect the insights and signals yourself, InvestSight does the analysis for you and points out the key information about companies allowing you to analyse them in a shorter period of time.

Also, if you missed a call you wanted to hear? Don’t worry, InvestSight will carry out analysis on past, present and newly released transcripts which you can access at your own time.

Additionally, we are now providing a solution that can allow users to cut through the noise and derive the key insights from companies themselves

These insights in turn will act as metadata that can support in identifying potential issues, risks and/or opportunities in companies that eventually can

become part of key factors or factors themselves to consider to make more informed investment decisions.that can subsequently affect the investment outcome

Identify potential opportunities and/or risks highlighted in earnings calls
Insights into companies’ activities and directions

This seeks to lend a helping hand to portfolio managers in obtaining key insights from various areas of discussion that highlight the activities and performance of companies they are trying to assess..

So to summarise, our solution aims to, that within a shorter period of time should their agenda be unable to cater time for them to sit in fully to earnings calls.

Features a Sentiment Analysis Dashboard, implements machine learning model with NLP

Highlight key words, phrases and sentences

Derive sentiment of sentences

18 of 24

Cost Effectiveness

Comparison to existing application

Alphasense

Market intelligence software that utilises machine learning and natural language processing to analyse earnings calls.

19 of 24

Problem with Alphasense

Catered to large enterprises for their own data
Sentiment Analysis is only a small feature of Alphasense
Starts from $10,000 for teams of 3 or more.

20 of 24

Cost Effectiveness

The following table shows the APIs and monthly running costs required:

Price ($)	Price ($)
Financial metrics Financial Modelling Prep	75
Earning Call Transcripts Finnhub.io	100
Sentiment Analysis MonkeyLearn	299
Hosting and Deployment	100
Total Cost	$ 574

21 of 24

Implementation Timeframe

Front-end development
Training of Sentiment Analysis Model

July - August

2020

August - September

2020

September - October

2020

October - November

2020

Back-end development
Training of Sentiment Analysis Model

Back-end development
Training of Sentiment Analysis Model

Final touch-ups
training of Sentiment Analysis Model

5 Nov

Delivery of MVP

22 of 24

Conclusion

Arduous analysis and time consuming earning calls.

InvestSight:

Reduce Time Taken
Gives more information about investments

Leads to better and more accurate investments = positive returns

Slower investment decisions = missed potential opportunities

23 of 24

Q&A

24 of 24

References