Scribd Logo
Upload

Welcome to Scribd!

  • 56 views

    Alpha in Quant Trading: Hoàng Tùng

    Uploaded by

    Le Doan
    This document discusses two ideas for alpha generation in quant trading: 1) Using web scraping, natural language processing, and machine learning to analyze corporate filings from the SEC EDGAR database to identify sentiment shifts and changes in language over time. 2) Analyzing earnings estimates data from Estimize to identify strategies like buying stocks before earnings if estimates are rising or selling those with high analyst dispersion. Accuracy increases with more analysts, those closer to the announcement, more experienced analysts, and upward estimate revisions.

    Copyright:

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

    Alpha in Quant Trading: Hoàng Tùng

    Uploaded by

    Le Doan
    56 views19 pages
    This document discusses two ideas for alpha generation in quant trading: 1) Using web scraping, natural language processing, and machine learning to analyze corporate filings from the SEC EDGAR database to identify sentiment shifts and changes in language over time. 2) Analyzing earnings estimates data from Estimize to identify strategies like buying stocks before earnings if estimates are rising or selling those with high analyst dispersion. Accuracy increases with more analysts, those closer to the announcement, more experienced analysts, and upward estimate revisions.

    Original Title

    Webinar_Tung

    Copyright

    © © All Rights Reserved

    Available Formats

    PPTX, PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    This document discusses two ideas for alpha generation in quant trading: 1) Using web scraping, natural language processing, and machine learning to analyze corporate filings from the SEC EDGAR database to identify sentiment shifts and changes in language over time. 2) Analyzing earnings estimates data from Estimize to identify strategies like buying stocks before earnings if estimates are rising or selling those with high analyst dispersion. Accuracy increases with more analysts, those closer to the announcement, more experienced analysts, and upward estimate revisions.

    Copyright:

    © All Rights Reserved
    Download as PPTX, PDF, TXT or read online from Scribd
    Download as pptx, pdf, or txt
    56 views19 pages

    Alpha in Quant Trading: Hoàng Tùng

    Uploaded by

    Le Doan
    This document discusses two ideas for alpha generation in quant trading: 1) Using web scraping, natural language processing, and machine learning to analyze corporate filings from the SEC EDGAR database to identify sentiment shifts and changes in language over time. 2) Analyzing earnings estimates data from Estimize to identify strategies like buying stocks before earnings if estimates are rising or selling those with high analyst dispersion. Accuracy increases with more analysts, those closer to the announcement, more experienced analysts, and upward estimate revisions.

    Copyright:

    © All Rights Reserved
    Download as PPTX, PDF, TXT or read online from Scribd
    Download as pptx, pdf, or txt
    You are on page 1of 19

    ALPHA IN QUANT TRADING

    HOÀNG TÙNG
    ALPHA IN QUANT TRADING
    • Alpha is a measure of the active return on an
    investment
    • Idea1: Edgar web scraping + NLP+ Machine learning
    Edgar web scraping
    • Majority of information related to the world of investing is
    unstructured: text, image, audio, or video.
    • How to transform unstructured information into numeric data
    in real time:
    • Integrated systems, web scraping, data collection, distributed
    parallel computing, advanced Natural Language Processing
    (NLP), machine learning techniques.
    • In this research, I demonstrate how NLP and machine learning
    can be used in processing corporate filing data from the
    EDGAR database
    Edgar web scraping
    • EDGAR (the Electronic Data Gathering, Analysis, and Retrieval)
    system, as a source of corporate filing information.
    • The EDGAR is used by the US SEC (Securities and Exchange
    Commission)
    Edgar web scraping
    • Web scraping EDGAR
    • EDGAR provides daily/quarterly master index files to
    effectively download company filings from its website.
    • A sample master index file for Apple
    Edgar web scraping
    • MAP-REDUCE FRAMEWORK FOR TEXT MINING
    • The mapper functions classify and extract relevant keywords, pairs and
    numbers from the downloaded text files.
    • Build a customized reduction function that aggregates words based on
    company, type of document, date of filing and reporting period and
    section.
    Edgar web scraping
    • Natural language processing
    • Create a Corpus which is a collection of documents containing
    natural language text.
    • Remove punctuation and numbers, and further convert all the
    text to upper cases.
    • Remove all the English stop words from the text document,
    remove words which are not relevant for our needs, such as
    generic words, geographies.
    • Stemming is the process of reducing words to their base or roots,
    so that to group together words with similar meaning.
    • At the end of this exercise, we should have a list of words and
    their frequency count for each text section.
    Edgar web scraping
    • Word cloud of textual documents
    Edgar web scraping
    • SENTIMENT AND TONE ANALYSIS
    • Positive sentiment = total positive words/total words
    • Negative sentiment = total negative words/total words
    • Net tone = positive sentiment – negative sentiment
    • Naturally more negative words than positives
    Edgar web scraping
    Edgar web scraping
    Edgar web scraping
    • DISTANCE MEASURES (YOY CHANGE IN CORPORATE FILING LANGUAGE)
    Edgar web scraping
    • American Express is an example
    Analyst
    Idea2: Analyst
    • Dataset: Estimize : crowd source earnings and revenue estimates from
    financial professional and non financial.
    • Coverage: 1600 companies in Russell 3000 (2014).

    Some conclusion:
    • More analysts => More accurate
    • Closer to earning announcement day => better estimate
    • Analyst experience. More experience => more accurate
    • Analyst skill: estimation error = |Actual EPS – Estimated EPS| / Stock Price
    Analyst
    Pre-earning announcement strategy
    • Before earning announcement day, if the estimated EPS go up => it’s likely
    the actual EPS will beat the estimated EPS => Price will go up
    Analyst
    Earning Revision

    • Upward (downward) EPS revision => positive (negative) return in


    announcement day return
    Post-earning announcement drift
    • 3 weeks after earning announcement, stock price will follow the return of
    announcement day.
    • The signal now decay quickly in US
    Analyst
    Low risk strategy

    • Higher dispersion => Higher risk and volatility and lower return on
    announcement day.
    • Remove those stocks have high dispersion => Improve performance

    You might also like