FMP
Sep 30, 2022 2:40 AM - Jack Dalton
Goal: To use Natural Language Processing to analyze earning call transcripts and consequent ticker quarter growth to predict performance as compared to respective index. This analysis can be added to pre-existing stock analysis.
Earnings call is a conference call between the management of a public company, analysts, investors, and the media to discuss the company's financial results during a given reporting period, such as a quarter or a fiscal year. An earnings call is usually preceded by an earnings report. This contains summary information on financial performance for the period.
Additionally, by SEC law, Officials are not allowed to lie or to spread false prospectus.
https://www.investopedia.com/terms/e/earnings-call.asp
Low to mid market cap technology companies on the NASDAQ. $300 M -> $10B
['ACIW', 'CEVA', 'CMTL', 'COMM', 'CPSI', 'CRUS', 'CSGS', 'CSOD', 'CVLT', 'DCT', 'DGII', 'DIOD', 'DMRC', 'DSGX', 'EBIX', 'EPAY', 'ERII', 'EVBG', 'EXTR', 'FEYE', 'FORM', 'LSCC', 'LTRPA', 'MANH', 'MARA', 'MDRX', 'MGRC', 'MIDD', 'MITK', 'MTSI', 'NATI', 'NH', 'NTCT', 'NTNX', 'NVEC', 'NXGN', 'OMCL', 'OSIS', 'PCTY', 'PDFS', 'PEGA', 'SLAB', 'SLP', 'SMCI', 'SMTC', 'SPSC', 'SPWR', 'SSYS', 'SWIR', 'SYKE', 'SYNA', 'TCX', 'TRIP', 'TRUE', 'TTEC', 'TTMI', 'TWOU', 'UCTT', 'UPLD', 'VECO', 'VIAV']
Index - VGT (Vanguard Information Technology Index Fund ETF); measures the entire technology sector as a whole.
Shameless plug - ETFs are amazing investments as they have low fees(as compared to mutual funds and other managed accounts) and they are extremely diversified(low risk).
TLDR: Harry Markowitz won a Nobel Prize by theorizing that diversification in the stock market is an efficient method of investing. His entire theory revolves around reducing risk(Std dev) by splitting assets while maximizing return.
Mutual funds and ETFs are built around this theory. What combination of weights minimizes risk while maximizing return.
When building models you cannot compare straight returns. Earning 10% return in a year is garbage if the market as a whole increased 15% that year. Most likely the investments were riding the bullish market.
1. Gather data
2. Feature engineering
3. Random Forest Classifier
4. Takeaways
Financial Modeling Prep API - For transcripts
https://financialmodelingprep.com/api/v3/earning_call_transcript/AAPL?quarter=3&year=2020&apikey=demo
Felsch-Kincaid : 100 = super ez, 0 = Difficult.
Gunning Fog : Grade level measurement for shorter passages.
SMOG: Grade level measurement for longer passages
Top 1000 features from transcripts
Included 2 ngrams
Loughran McDonald is a professor of accounting and finance at University of Notre Dame. Built a dictionary of words and scores based off of words in company earning statements
Sentimental Analysis: Positive, negative, superfluous, uncertainty, litigious
RF = RandomForestClassifier(bootstrap= True, max_depth= 4, min_samples_leaf= 35, min_samples_split= 8, n_estimators= 100)
Ran many grid searches on many combination of features. Best performing were the reading scores by themselves.
Used Precision Score to find best model. tp/(tp/fp)
Predicted positive = buy ticker for quarter. Sell after 90 days
Predicted negative = hold onto VGT
The model is measured using by assuming the investor either holds VGT. If model predicts that the stock will do better, the investor will sell VGT and buy the ticker in that quarter.
Closed Formula:
If predicted buy: ∑Δšššššš−Δššŗš
Else: ∑Δššŗš
Result: 0.0097
Or
Ticker outperformed VGT by 1% over the quarter.
The data was not perfect. The Transcripts have the operator, questioners, and chief officers responses recorded in each article. The model could be improved by filtering out non-business member speakers.
Did not account for technical or traditional forms of investment strategies. Therefore did not include key financial statistics such as market cap, EBITDA, EPS, etc.
Models explored were exclusively logistic regressions and Random Forest Classifiers. Other models to explore are Gradient Boosting Classifiers and Neural Networks.
This project was solely based on Small - Mid sized market cap tech companies from the NASDAQ, 2017-2020. Therefore the model is biased. Small to medium sized tech firms are characterized by their volatility and extreme growth. Additionally 2017 to 2020 saw multitude of extreme stock movements. To continue the project, additional exploration of other sectors and time periods are necessary.
This data can be used in conjunction with traditional company analysis to avoid troubled companies.
https://github.com/daniellkennett/Earnings_Call_NLP_Analysis
https://www.linkedin.com/in/daniel-kennett/
Sep 11, 2023 1:38 PM - Rajnish Katharotiya
Price to EarningsĀ is one of the key metrics use to value companies using multiples. The P/E ratio and other multiples are relative valuation metrics and they cannot be looked at in isolation. One of the problems with the P/E metric is the fact that if we are in the peak of a business cycle, earni...
Sep 11, 2023 1:49 PM - Rajnish Katharotiya
Price-to-Earnings ratio is a relative valuation tool. It is used by investors to find great companies at low prices. In this post, we will build a Python script to calculateĀ Price Earnings RatioĀ for comparable companies. Photo by Skitterphoto onĀ Pexels Price Earnings Ratio and Comparable Compa...
Oct 17, 2023 3:09 PM - Davit Kirakosyan
Shares of VMware (NYSE:VMW) witnessed a sharp drop of 12% intra-day today due to rising concerns about China's review of the company's significant sale deal to Broadcom. Consequently, Broadcom's shares also saw a dip of around 4%. Even though there arenāt any apparent problems with the proposed solu...