BUSANA 7003 – Business Analytics

BUSANA 7003 – Business Analytics Project – Semester 2, 2024 

Final Project 
Background. 
You are starting a new job as a Business Analyst at AQR Asset Management, a global investment 
management firm focused on quantitative investment strategies. Your first task is to analyse the 
performance of US-listed securities to help the firm manage their portfolio. You have several datasets 
on US securities (stocks and ETFs), and you should help your employer with the following tasks: 
 
(1) Exploratory data analysis, visualisations, descriptive statistics 
Examine the dataset Stock_data_part1.xlsx. Characterise the performance of US securities between 
February 14, 2020, and March 20, 2020, and compare their performance to a similar period not affected 
by the COVID-19 pandemic. Justify how you choose the non-COVID period to make a fair comparison. 
Based on your analysis, comment on the effect of the COVID-19 pandemic on securities’ performance. 
In your analysis, you should consider the following: 
• Which variables to analyse? Consider the following: returns, volatility, Sharpe ratio, bid-ask 
spread, dollar volume, number of trades. You should do your own research and also use the 
course content to understand these variables, and potentially add others, which could be helpful 
for understanding the market dynamics during the COVID-19 pandemic and beyond it. 
• What is the data structure (unit of observation) that you need for this analysis? (Cross-sectional? 
Time series? Panel?) How to transform your data into the units of observation that you need? 
• How can you add value with your analysis? Can you offer relevant comparisons of returns or 
other variables across different groups (e.g., COVID vs non-COVID period, by industry, by 
company size, ETFs vs stocks etc.)? 
 
Use your own judgement to select the relevant descriptive statistics and visualisations. However, at a 
minimum, you should consider the following: 
• A table with descriptive statistics (mean, median, 25
th
 percentile, 75
th
 percentile, standard 
deviation, min, max) of the key variables you choose to analyse. (Add comparisons by group if 
relevant). 
• The time series of the key variables of interest (Add comparisons by group if relevant). Make 
sure you cover the entire time period of the sample. You can add analysis by sub-period (e.g., 
COVID vs non-COVID periods) as an additional point. 
 
 
(2) Supervised Machine Learning – OLS regressions 
Transform your data into a cross-section: for each security, compute returns between February 14, 2020, 
and March 20, 2020, and run regression analysis to help explain returns. Then, do the same for returns 
between December 14, 2019, and January 20, 2020. You may use any variables in the existing dataset, 
or variables outside the provided datasets (from external sources). For example, you may consider the 
following external sources: BUSANA 7003 – Business Analytics Project – Semester 2, 2024 
 
- Robinhood stock popularity history: https://www.kaggle.com/datasets/cprimozi/robinhoodstock-popularity-history?resource=download
(historical data about the number of users that 
hold each stock available on the Robinhood stock brokerage), 
- Stock exposure to the market, and to Fama-French factors. The time series data on Fama-French 
factors is available here: 
- https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html NOTE: you need to 
first run OLS regressions of stock returns on 3 Factors (Mkt, HML, SMB), and compute betas 
(factor exposures) for each stock. Then, you can use those betas as explanatory variables in 
your SML analysis. 
 
Please, make sure to split the dataset you use for analysis into training and testing samples, and comment 
on the model accuracy. Please, analyse the following: 
- Are you better able to predict returns between February 14, 2020, and March 20, 2020 or returns 
between December 14, 2019, and January 20, 2020? Why? 
- Are you better able to predict returns between February 14, 2020, and March 20, 2020 for stocks 
or for ETFs? Why? 
 
In your analysis, you should consider the following: 
• Which explanatory variables to use? 
• Over which period of time to compute the explanatory variables? Imagine it is February 14, 
2020, and you know nothing about the future returns, market caps, volumes, investor interest 
etc. You are tasked to develop a SML model to help predict returns between February 14, 2020, 
and March 20, 2020. How can you use the data from the past to predict the future? 
• Consider running separate models for stocks vs ETFs. Think about which explanatory variables 
are more relevant for each group. 
 
• What is the data structure (unit of observation) that you need for this analysis? (Cross-sectional? 
Time series? Panel?) How to transform your data into the units of observation that you need? 
• How well does your model perform? Can you offer relevant comparisons of your model 
accuracy across different groups (e.g., COVID vs non-COVID period, by industry, by company 
size, ETFs vs stocks etc.)? 
 
In your analysis, at a minimum, you should present the following output: 
• Regression coefficients from multiple models (at least five). 
• Model evaluation metrics: MSE, RMSE, MAE for each model. 
• Your assessment of which factors are most important for explaining stock returns between 
February 14, 2020, and March 20, 2020, and between December 14, 2019, and January 20, 
2020. 
• Your assessment of which factors are most important for explaining ETF returns between 
February 14, 2020, and March 20, 2020, and between December 14, 2019, and January 20, 
2020. 
 BUSANA 7003 – Business Analytics Project – Semester 2, 2024 
 
(3) Supervised Machine Learning – Logistic regressions 
Use the same dataset as in part (2), and introduce a dummy variable for whether a given security 
increased in price between February 14, 2020, and March 20, 2020. Model the probability of a price 
increase using any continuous or categorical variables you find relevant. You may use any variables in 
the existing dataset, or variables outside the provided datasets (from external sources). 
Please, make sure to split the dataset you use for analysis into training and testing samples, and comment 
on the model accuracy. 
 
In your analysis, you should consider the following: 
• How does the industry variable affect whether a given security increased in price between 
February 14, 2020, and March 20, 2020? 
• How does being a stock or an ETF affect whether a given security increased in price between 
February 14, 2020, and March 20, 2020? 
• Imagine we are facing a replay of the COVID-19 pandemic now, and you are asked to predict 
whether your testing sample securities will increase or decrease in price, based on what you 
learnt from your analysis over February 14, 2020, and March 20, 2020. Comment on which 
factors are affecting this. 
 
In your analysis, at a minimum, you should present the following output: 
 
• Regression coefficients from multiple models (at least five). 
• Model evaluation metrics: accuracy, precision, recall for each model. 
• Your assessment of which factors are most important for explaining whether a given security 
increased in price between February 14, 2020, and March 20, 2020? 
 
 
(4) Unsupervised Machine Learning – K-means clustering 
Examine the dataset Stock_data_part2.xlsx to perform the k-means clustering. You may combine 
Stock_data_part2.xlsx dataset with any other data (e.g., returns) from the previous exercise. The key 
part of this analysis is finding clusters if stocks with similar characteristics (financial ratios), and 
showing whether their returns during between February 14, 2020, and March 20, 2020 were 
significantly different. How about their returns during December 14, 2019 - January 20, 2020? 
In your analysis, you should consider the following: 
• Variable Selection: Which two financial ratios did you select as the most important variables 
for clustering, and why? 
• Cluster Characteristics: What are the average values of the selected financial ratios for each 
cluster? Present a table showing the average values of the selected financial ratios for each 
identified cluster. 
- Cluster Visualization: How can we visualize the distribution of the clusters in a twodimensional
 space using the selected financial ratios? 
- Provide a scatter plot where each point represents a stock, coloured by cluster, using the two 
selected financial ratios on the axes. BUSANA 7003 – Business Analytics Project – Semester 2, 2024 
 
In your analysis, at a minimum, you should present the following output: 
 
• Elbow plot and cluster visualisation in two-dimensional space. 
• How returns differ across cluster, with comparison of two periods: February 14, 2020 - March 
20, 2020, vs December 14, 2019 - January 20, 2020. 
• Your assessment of investment implications. Which stocks should AQR invest in, if market 
conditions are similar to those on February 14, 2020, vs those on December 14, 2019? 
 
(5) Unsupervised Machine Learning – Principal Component Analysis 
The lead asset manager asks you to distil a variety of market-wide factors to the core Principal 
Components. The Principal components will be used to explain average monthly returns on S&P500 
index. You may combine Stock_data_part3.xlsx dataset with any other data (e.g., valuation ratios, 
market returns, Fama-French factors (external), interest rates (external) etc.), and analyse the principal 
components to understand the factors driving market returns. 
In your analysis, you should present the following output: 
• A detailed summary of the principal components extracted from the PCA, including the amount 
of variance explained by each component. This should include a scree plot or a table 
summarizing the eigenvalues and the percentage of variance explained by each principal 
component, helping to identify the most significant factors affecting monthly returns. 
• Factor Loadings: For each principal component identified as significant, provide the factor 
loadings of the original variables (e.g., valuation ratios, market returns, Fama-French factors, 
interest rates). This will show how each original variable contributes to the principal 
components, indicating which factors are most influential in driving returns. 
• Factor Interpretation: An interpretation of what each significant principal component represents 
in the context of market-wide and company-specific factors. For example, the first principal 
component might be interpreted as overall market risk, while the next components could 
represent size and value factors, sector exposures, or interest rate sensitivity. This section 
should bridge the mathematical output of PCA with intuitive financial concepts. 
• Implications for Monthly Returns: Discuss how each principal component influences monthly 
returns on S&P500 index. This could involve analyzing how movements in the principal 
components are associated with changes in market returns, providing insights into the 
underlying risk factors or market conditions that impact asset prices. 
 
 (6) Analysing experimental evidence using Difference-in-Difference regressions 
AQR wants to analyse the effects of the SEC Tick Size Pilot program, and find out how the securities 
in test groups were affected in terms of relative bid-ask spread. Use monthly data from 
Stock_data_part3.xlsx and the list of treatment and control securities (https://www.finra.org/rulesguidance/key-topics/tick-size-pilot-program)
to investigate this question. 
In your analysis, you should present the following output: 
• A regression specification with detailed explanation of the null hypothesis and alternative 
hypothesis. 
• Regression output and interpretation of coefficients. BUSANA 7003 – Business Analytics Project – Semester 2, 2024 
 
• Implications for AQR trading strategy. How should AQR adjust their trading in treated and 
control stocks to minimise their transaction costs? 
 
(7) Summary 
Based on your earlier analysis, draw conclusions for AQR that tackle the following overarching 
questions: 
- What are your recommendations to AQR about their stock and ETF selection strategy, if they 
found themselves in market conditions similar to pre-COVID (e.g., February 14, 2020)? 
- How accurate are your ML models in predicting the price changes, and how confident are you 
in providing recommendations based on those models? 
- What does your analysis reveal about the market dynamics around the COVID-19 period, and 
how it differs across different groups of securities? 
- Which variables from the dataset are most important in their contribution to the variation in 
S&P500 returns? 
- What did you learn about the way the effect on liquidity, as a result of the Tick Size Pilot?