Part III: Alpha Vertex Launches Alta to Create Investment Signals From Our Speech, Chaos and Language
By Mutisya Ndunda, CEO of Alpha Vertex
Part III of our series on Alta and using alternative data to gain an advantage in the markets. As a background, check out Part I here of our series and Part II here.
In my earlier posts, I introduced Alta. The product is built from advanced natural language processing (NLP) and machine learning tools to extract unique, high-value information — such as investment signals — from unstructured, text-based data sets with broad coverage.
In this post, I’ll share a systematic investing use case using Alta.
BACKGROUND
Conference calls and investor presentations are the primary mechanisms for direct communication between company executives, their shareholders and analysts. By carefully studying these events, we can provide valuable insights to model driven investors to help them put context around the numerical data presented by companies in their financial statements.
In this case study, we demonstrate how information communicated in conference calls can be used as high value alpha signals in systematic strategies.
WHAT MAKES A GOOD SYSTEMATIC ALPHA GENERATOR?
The most defining characteristic of an alpha generator is its profitability. Profitability is measured as the performance spread between portfolios with high to low exposure to the factor. Generally, portfolios with a high exposure to a factor should outperform portfolios of stocks with a low exposure to the factor (decile analysis).
Breadth
Breadth refers to the number of stocks for which the alpha generator is predictive. Alpha generators with good breadth have a high investment capacity and do erode quickly. Most alternative datasets tend to have a short history, poor coverage and limited capacity.
Reliability
The reliability of an alpha generator refers to its frequency of success. The more reliable the factor, the more often its returns are positive. Alpha generators with the same profitability can have different reliability characteristics, with some repeatedly delivering consistent returns and others delivering skewed returns.
Symmetry
High value alpha factors perform similarly across multiple sectors and market capitalization groups.
FACTOR PERFORMANCE
To test whether conference calls can be used to predict future stock returns or provide other useful risk indicators, we generate over 100 unique features that offer insights on everything from executive character traits, sentiment and the inner workings of a firm.
The most traditional and widely used method for implementing factor based portfolios is the hedged portfolio approach, pioneered and formulated by Fama and French. In this approach, the universe of stocks is ranked by its factor value from lowest to highest. This universe of stocks is then chunked into equal size groups of five (quintiles) or ten (deciles). Stocks in each group are equally weighted to form ten portfolios. Each portfolio’s average return is examined (see decile plots below).
A long/short hedged portfolio is constructed from the lowest and highest bins and the performance of this portfolio is tracked over time. This is the most intuitive approach as the backtesting performance resembles a real life trading strategy.
Drawbacks to the hedged portfolio approach include the following:
- Information contained in the middle bins is wasted as only the top and bottom groups are used to form a hedged portfolio.
- There is an implicit assumption that the relationship between the factor and returns is linear or at least monotonic.
- The hedged portfolio requires managers to short stocks.
- Transactions costs, liquidity and other institutional constraints are not considered.
Another metric that can be used to evaluate the performance of a signal is the Spearman Rank Correlation (IC). It measures the strength and direction of monotonic associations between two ranked variables. Good factors typically have an absolute rank IC score of 2%
A monotonic relationship is a relationship that does one of the following: (1) as the value of one variable increases, so does the value of the other variable; or (2) as the value of one variable increases, the other variable value decreases. Examples of monotonic and non-monotonic relationships are presented in the diagram below:
1. Language Features
As shown in the figure below, a weekly rebalanced long/short portfolio based on the language factor has delivered an average return of 7.7% and a Sharpe ratio of 1.4x, pre-cost. Additionally, the model has a low turnover rate of 6.7%.
2. Sentiment Features
These portfolios are formed based on the sentiment of management in earnings calls when discussing various financial concepts such as guidance, dividends, regulatory matters, etc. Portfolio 1 is formed with companies exhibiting the highest amount of sentiment while Portfolio 10 is formed with companies having the lowest level of sentiment in their earnings call. The figure below also shows the annualized return of these equally weighted portfolios.
We find that portfolio performance deteriorates as the level of sentiment decreases. This is consistent with our expectation. In fact, companies with the highest sentiment outperform companies with the lowest sentiment by approximately 10% annually.
3. Performance of the Alta Composite Factor
Creating a composite factor from the individual factors described above results in a unique signal with very low turnover and consistent performance across multiple market cycles. This is unlike most traditional factors, which show substantial decay.
The composite factor delivers superior alpha even in a highly concentrated portfolio; therefore, fundamental and discretionary managers can use Alta to supplement their investment process. The Alta composite factor is useful either as a tool to source potential companies to invest in or as a failure model to select stocks to short.
As shown in the figure below, a weekly rebalanced long/short portfolio based on the language factor has delivered an average return of 10.7% and a Sharpe ratio of 1.6x, pre-cost. Additionally, the model has a low turnover of 5% and a high rank IC of 4.7%.
Our aim with Alta at Alpha Vertex has been to bring clarity and truly meaningful signals to investors trying to make sense of messy human data and hard to quantify spoken and written data. And that’s the wrap on our three part series!
Disclaimer: The content of this report is to be used solely for informational purposes and should not be regarded as an offer, or a solicitation of an offer, to buy or sell a security, financial instrument or service discussed herein. Opinions in this communication constitute the current judgment of the author as of the date and time of this report and are subject to change without notice. Information herein is believed to be reliable but Alpha Vertex, Inc. (“Alpha Vertex”) makes no representation that it is complete or accurate.
The information provided in this communication is not designed to replace a recipient’s own decision-making processes for assessing a proposed transaction or investment involving a financial instrument discussed herein. Recipients are encouraged to seek financial advice from their financial advisor regarding the appropriateness of investing in a security or financial instrument referred to in this report and should understand that statements regarding the future performance of the financial instruments or the securities referenced herein may not be realized. Past performance is not indicative of future results. This report is not intended for distribution to, or use by, any person or entity in any location where such distribution or use would be contrary to applicable law, or which would subject Alpha Vertex to any registration requirement within such location.
