Python and Machine-Learning for Asset Management with Alternative Data Sets

The consumption module introduces students to the basics of consumption-based alternative data. By aggregating online and offline consumer purchase activity and behavioral datasets including geolocation data (e.g., cell locations, satellite imagery etc.), transaction data (e.g., credit card transaction logs and point of sale data), as well as consumer interaction with brands and products on social media, researchers can learn about company performance ahead of official company earning announcements. Such information may be extremely useful and can provide investment and risk management advantages. This module reviews the theoretical aspects of various consumption datasets, and provides practical demonstrations of relevant data analytics.

Module 2 is an introduction to text mining as well as a demonstration of how to get from data retrieval (web scraping) to financial market insights. Some of the classic text mining methodologies are covered such as vectorization of text (the bag of words approach), stop words for filtering, and term frequency-inverse document frequency (TF-IDF). Students will learn how text can be mathematically represented, and regularized/filtered to reduce noise. Measures of text-similarity will be covered in theoretical and practice sessions. Lab sessions go through examples of web scraping data, regularizing with the described techniques and finally, insights will be derived from the textual data.

Module 3 is a practical extension of the text mining lessons to 10-K and 13-F, two of the most commonly researched corporate filings. This type of data can be extremely daunting when used by individual analysts due to the sheer size of the documents, but module 3 describes the methodologies for quantitatively analyzing these documents with Python code. Both the 10-K and 13-F documents are worked through, and within the lab sessions it is demonstrated how one can automatically pull this kind of data as well as define metrics around them. We investigate implementations of research in this field around similarity of given companies 10-K statements over time as well as similarity between fund holdings from the 13-F in the lab.