Cove Capital manages a mid-cap equity fund focusing on international companies. Over the past five years, the company has maintained a portfolio of 50 stocks (selected from a pool of 5,000 stocks). These holdings are chosen through a combination of screening based on various market and financial data and the insights of Cove Capital’s team of three analysts and one portfolio manager.
Despite solid investment performance, the Chief Investment Officer, Emily Thompson, considers revamping the investment process to achieve even better returns. Inspired by data science workshops and vendor interactions, Emily believes that machine learning (ML) can enhance Cove Capital’s investment selection. Notably, many of their past successes resulted from stocks that eventually became acquisition targets. After some research, Emily sends the following email to Cove Capital’s CEO.
Subject: Investment Process Revamp
I propose that we continue managing a portfolio of 50 international mid-cap stocks but restructure our process to incorporate machine learning (ML) while still allowing human insight to contribute domain knowledge. Additionally, we should focus on identifying potential acquisition targets. I suggest a four-step process to be repeated every quarter.
Phase 1 We use ML techniques on a model with fundamental and technical variables (features) to predict each of the 50 stocks’ returns for the upcoming quarter. The 10 stocks with the lowest estimated return are identified for replacement.
Phase 2 We apply ML techniques to categorize our investable universe of about 5,000 stocks into 10 distinct groups based on relevant financial and non-financial characteristics. This approach aims to avoid unintended portfolio concentration by selecting stocks from each of these groups.
Phase 3 For each of the 10 different groups, we use labeled data to train a model that predicts the three stocks (in any given group) most likely to become acquisition targets within the next year.
Phase 4 Our three experienced analysts are each assigned three or four of the groups, and they select their best stock pick from each assigned group. These 10 “high-conviction” stocks will replace the 10 underperforming stocks sold in Step 1.
Additional comments related to the above:
The ML algorithms will need extensive data.
We should first explore using free or low-cost historical datasets and evaluate their usefulness for the ML-based stock selection processes before deciding on using subscription-based data.
As we progress, we expect to discover more ways to apply ML techniques to refine Cove Capital’s investment processes.
What are your thoughts?
0 of 10 Questions completed
You have already completed the quiz before. Hence you can not start it again.
Quiz is loading…
You must sign in or sign up to start the quiz.
You must first complete the following:
0 of 10 Questions answered correctly
Time has elapsed
You have reached 0 of 0 point(s), (0)
Earned Point(s): 0 of 0, (0)
0 Essay(s) Pending (Possible Point(s): 0)
“It’s not going to be easy, but it’s going to be worth it.”
In Phase 2, which unsupervised machine learning technique is best suited for dividing the investable universe of stocks into distinct clusters?
In Phase 3, what is the target variable for the labeled training data most likely to be?
Assuming a Decision Tree model is used in Phase 3, which technique is most likely to help improve the accuracy of predictions?
For Phase 4, which of the following statements best describes the role of the senior analysts in the new investment process?
In Phase 1, which of the following performance metrics is most appropriate for evaluating the regression model predicting monthly stock returns?
In Phase 2, how does the choice of the number of clusters (k) in K-Means Clustering affect the quality of the clusters?
In Phase 3, which of the following methods would be helpful in addressing class imbalance in the labeled training data?
What is the main advantage of using an ensemble learning method in Phase 3, compared to relying on a single Decision Tree model?
In Phase 4, what is the primary benefit of involving senior analysts in selecting their best stock picks from the clusters identified by the machine learning model?
Why is the language of the Learning Outcome Statements (LOS) different from the curriculum?
The LOS are protected under the CFA Institute's copyright, and we don't have permission to duplicate them verbatim. Therefore, we've rephrased the LOS and included alphabetical labels (a, b, c, …) to simplify cross-referencing with the original LOS in the curriculum when needed.
Now available for all Level I topics! Try it now!
Enter a search term (e.g. ‘LIFO reserve’), or bookmark the glossary page!