Scribd Logo
Upload

Welcome to Scribd!

  • 155 views

    Sentiment Analysis On Transcript Calls

    Uploaded by

    Saurabh Sharma
    This document outlines a strategy to analyze earnings call transcripts using natural language processing and regressions to develop a scoring algorithm to predict stock performance. Key steps include obtaining transcripts, applying NLP to count keyword frequencies, running regressions to weight keywords, developing a buy/sell algorithm based on scores, and backtesting the strategy. The goal is to invest in small, cheap, highly leveraged stocks showing debt paydown and efficiency gains based on language in transcripts to outperform the market. Next steps include expanding the dataset and refining the language processing and scoring methodology.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Sentiment Analysis On Transcript Calls

    Uploaded by

    Saurabh Sharma
    155 views22 pages
    This document outlines a strategy to analyze earnings call transcripts using natural language processing and regressions to develop a scoring algorithm to predict stock performance. Key steps include obtaining transcripts, applying NLP to count keyword frequencies, running regressions to weight keywords, developing a buy/sell algorithm based on scores, and backtesting the strategy. The goal is to invest in small, cheap, highly leveraged stocks showing debt paydown and efficiency gains based on language in transcripts to outperform the market. Next steps include expanding the dataset and refining the language processing and scoring methodology.

    Original Description:

    Original Title

    Sentiment analysis on transcript calls

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    This document outlines a strategy to analyze earnings call transcripts using natural language processing and regressions to develop a scoring algorithm to predict stock performance. Key steps include obtaining transcripts, applying NLP to count keyword frequencies, running regressions to weight keywords, developing a buy/sell algorithm based on scores, and backtesting the strategy. The goal is to invest in small, cheap, highly leveraged stocks showing debt paydown and efficiency gains based on language in transcripts to outperform the market. Next steps include expanding the dataset and refining the language processing and scoring methodology.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    155 views22 pages

    Sentiment Analysis On Transcript Calls

    Uploaded by

    Saurabh Sharma
    This document outlines a strategy to analyze earnings call transcripts using natural language processing and regressions to develop a scoring algorithm to predict stock performance. Key steps include obtaining transcripts, applying NLP to count keyword frequencies, running regressions to weight keywords, developing a buy/sell algorithm based on scores, and backtesting the strategy. The goal is to invest in small, cheap, highly leveraged stocks showing debt paydown and efficiency gains based on language in transcripts to outperform the market. Next steps include expanding the dataset and refining the language processing and scoring methodology.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 22

    Using Earnings Call

    Transcripts to Predict Stock


    Performance
    Jonathan Khalfayan, Justin Kahl, Santiago Rodriguez, Matthias
    Schmitz, Sam Sklar, Juan Pablo Villarreal
    Roadmap
    1. Obtain Earnings Call Transcripts
    2. Apply Natural Language Processing algorithm
    3. Run regressions to weight keywords
    4. Develop scoring algorithm to determine when to buy/sell
    5. Run backtest evaluation of algorithm
    Earnings Calls
    Stock Criteria
    ● Small
    ○ Between the 25th and 75th percentile based on their market capitalization
    ● Cheap
    ○ Below 25th percentile EV to Ebitda ratio
    ● Highly Leveraged
    ○ Above mean LT Debt/EV
    Binary Variables
    LT debt year 1 > LT debt year 2

    Prior year returns below median:

    % Revenue Growth > % Asset Growth

    High Gross Profit/Assets relative to LT Debt/Assets


    Benchmark and Relative Performance
    Data Collection
    ● Earnings call transcripts from THOMSON REUTERS, manually scraped
    Data Collection
    Next Steps:

    1. Incorporate a larger number of equities in the dataset.


    a. Available extensive dataset still pending exploitation
    2. Enhance current dataset
    a. Fill in gaps of non-existent data
    Roadmap
    1. Obtain Earnings Call Transcripts
    2. Apply Natural Language Processing algorithm
    3. Run regressions to weight keywords
    4. Develop scoring algorithm to determine when to buy/sell
    5. Run backtest evaluation of algorithm
    Language Processing
    Methodology:

    1. Categorize words from .txt file into their corresponding fiscal quarter
    2. Account for discrepancies in length of transcript, case, and punctuality
    3. Count the frequency of specific words in each quarter
    4. Write csv files of organized data for regression analysis
    Language Processing
    Next Steps:

    1. Account for different tenses of words.


    a. Potentially use global vectors for word representation (GloVe research
    group at Stanford)
    b. Multi-Dimensional Matrixes with similar words mapped at neighboring
    locations within the Matrix, word frequency calculated for set of words.
    2. Investigate two word keyphrases such as “Pay Down”
    Word Selection Criteria
    ● We started analyzing the frequencies of 26 keywords that could possibly
    predict future growth.
    ● The presence of words like “cost-cutting,” “deleveraging,” and “growth” that
    we believe will lead to these small, highly-leveraged, companies to produce
    higher returns than the benchmark

    ● For every company’s earnings call transcript:


    ○ Frequency = # of appearances of keyword within specified fiscal quarter
    Data
    Screenshot table containing frequency values for one company for desired fiscal
    quarters and keywords:
    Roadmap
    1. Obtain Earnings Call Transcripts
    2. Apply Natural Language Processing algorithm
    3. Run analysis to weigh keywords
    4. Develop scoring algorithm to determine when to buy/sell
    5. Run backtest evaluation of algorithm
    Key Word Analysis
    Word Score Number of Occurrences

    Save 1.37 122

    Dividend .894 739

    cutting .805 92

    cut .772 473

    short-term .754 333

    pay .720 594

    demand .685 2053

    prudent .682 195

    reduction .657 1107


    Next Steps
    Roadmap
    1. Obtain Earnings Call Transcripts
    2. Apply Natural Language Processing algorithm
    3. Run regressions to weight keywords
    4. Develop scoring algorithm to determine when to buy/sell
    5. Run backtest evaluation of algorithm
    Algorithm
    Insight: By using the correlations found in the Earnings Call Transcript analysis, we
    can score companies by quarter as new transcripts are released.

    Methodology:

    ● When high frequencies of key terms occur (cost cutting, dividend, reduce,
    save) increment score.
    ● Add additional binary values for whether long term debt has decreased from
    last year (signalling debt paydown) as well as increasing Gross Profit/Assets
    ratio.
    ● Buy companies with highest scores that also meet binary metrics. Try on
    Quarterly vs Yearly holding period
    Roadmap
    1. Obtain Earnings Call Transcripts
    2. Apply Natural Language Processing algorithm
    3. Run regressions to weight keywords
    4. Develop scoring algorithm to determine when to buy/sell
    5. Run backtest evaluation of algorithm
    Stock Performance
    ● To prevent look-ahead bias, we will trade with one quarter lag.
    ● For instance, we will act upon data obtained from Q1 2010 at the beginning of
    Q3 2010.
    Hypothesis and Predictions
    ● Investing in small, cheap, highly-leveraged stocks with improving asset turnover and
    estimated debt paydown that we filter by keywords such as “cost-cutting” will lead to
    a strategy that will outperform the S&P 500 Index.
    Thank You!

    You might also like