BUSANA 7003 – Business Analytics Project

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 investmentmanagement firm focused on quantitative investment strategies. Your first task is to analyse theerformance of US-listed securities to help the firm manage their portfolio. You have several datasets

 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 betweenFebruary 14, 2020, and March 20, 2020, and compare their performance to a similar period not affectedby 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-askspread, dollar volume, number of trades. You should do your own research and also use thecourse content to understand these variables, and potentially add others, which could be helpfulfor 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 orother variables across different groups (e.g., COVID vs non-COVID period, by industry, bycompany size, ETFs vs stocks etc.)?Use your own judgement to select the relevant descriptive statistics and visualisations. However, at aminimum, you should consider the following:

• A table with descriptive statistics (mean, median, 25th percentile, 75th percentile, standarddeviation, min, max) of the key variables you choose to analyse. (Add comparisons by group ifelevant).

• The time series of the key variables of interest (Add comparisons by group if relevant). Makesure you cover the entire time period of the sample. You can add analysis by sub-period (e.g.,OVID 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/robinhood

stock-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 interestetc. 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 variablesare 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,

• 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 securityincreased in price between February 14, 2020, and March 20, 2020. Model the probability of a priceincrease using any continuous or categorical variables you find relevant. You may use any variables inthe 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 commenton the model accuracy.In your analysis, you should consider the following:

• How does the industry variable affect whether a given security increased in price betweenFebruary 14, 2020, and March 20, 2020?

• How does being a stock or an ETF affect whether a given security increased in price betweenFebruary 14, 2020, and March 20, 2020?

• Imagine we are facing a replay of the COVID-19 代 写BUSANA 7003 – Business Analytics Projectpandemic now, and you are asked to predictwhether your testing sample securities will increase or decrease in price, based on what youlearnt from your analysis over February 14, 2020, and March 20, 2020. Comment on whichfactors 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 securityincreased 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 combineStock_data_part2.xlsx dataset with any other data (e.g., returns) fromthe previous exercise. The keypart of this analysis is finding clusters if stocks with similar characteristics (financial ratios), andshowing whether their returns duringbetween February 14, 2020, and March 20, 2020 weresignificantly different. How about theirreturns 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 variablesor clustering, and why?

• Cluster Characteristics: What are the average values of the selected financial ratios for eachcluster? Present a table showing the average values of the selected financial ratios for eachidentified 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 twoselected financial ratios on the axes.BUSANA 7003 – Business Analytics Project – Semester 2, 2024In 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 - March20, 2020, vs December 14, 2019 - January 20, 2020.

• Your assessment of investment implications. Which stocks should AQR invest in, if marketconditions 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 PrincipalComponents. The Principal components will be used to explain averagemonthly returns on S&P500ndex. 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 principalomponents 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 amountof variance explained by each component. This should include a scree plot or a tablesummarizing the eigenvalues and the percentage of variance explained by eachprincipalcomponent, helping to identify the most significant factors affecting monthly returns.

• Factor Loadings: For each principal component identified as significant, provide the factorloadings of the original variables (e.g., valuation ratios, market returns, Fama-French factors,

nterest 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 couldrepresent 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 monthlyreturns on S&P500 index. This could involve analyzing how movements in the principalcomponents are associated with changes in market returns, providing insights into theunderlying 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 securitiesin test groups were affected in terms of relative bid-ask spread. Use monthly data from

tock_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 alternativehypothesis.

• 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 andontrol 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 theyfound 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 youin providing recommendations based on those models?

- What does your analysis reveal about the market dynamics around the COVID-19 period, andhow it differs across different groups of securities?

- Which variables from the dataset are most important in their contribution to the variation inS&P500 returns?

- What did you learn about the way the effect on liquidity, as a result of the Tick Size Pilot?