Carmen Wang MS-DSPP'22

About Me

Hello! Welcome to my portfolio for ANLY 560 (Time Series)!

My name is Carmen Wang. I'm currently a second-year student from the McCourt School of Public Policy. I will be graduating from Georgetown in May 2022 but will be re-joining Georgetown graduate school as a full-time employee. My research area is mainly around higher education accessibility and student achievement. I earned my B.BA in Finance from Seattle University.

I am a student, a higher education practitioner, and a big fan of sports. I love working in student services, but for this project I'd like to do something different; thus, this project is all about the sporting goods industry.

Introduction

Athletic and sporting goods stocks

According to the industry analysis on Statista, athletic and sporting goods, as a broad industry, includes several different types of goods, such as athletic apparels, equipment, and licensed merchandises. We have witnessed both opportunities and challenges to this industry. For instance, when covid-19 hit, people had an increased awareness of health risk and wanted to change their lifestyles by exercising more, which increased the demand of certain sporting goods, such as at-home exercising equipment and apparels. Meanwhile, the industry has been experiencing issues with the supply chain as the global pandemic keeps affecting the economy and various business sectors.

This is an interesting industry to study, not only because of its opportunities and challenges, but also because of some stocks with strong performance over years. With a good understanding of the historical performance, we could better analyze the factors that affected the stock performance and shaped the industry trend and better predict the industry's future.

Important Questions:

1. What has been the general trend of sporting goods stock prices over time? Can we observe similarities across firms?
2. How do major shifts in stock prices reflect major economy events?
3. How do we describe the stock performance over time for individual companies?
4. Does the industry perform differently across regions (e.g the U.S, Canada, EU)?
5. Can we observe seasonal changes in the stock prices?
6. Do sporting goods stock prices reflect the business cycle?
7. Are there specific subsectors (niche sports) that perform better than others over time?
8. Can we tell the competitive landscape from different firms' stock performances?
9. Do we have outliers in the industry (i.e. stocks perform in the opposite direction from others)?
10. What are the lessons learned from historical data that can be used in the forecast? What recommendations could we suggest to firms?

Data Sources

All the stock price data are retrieved from Yahoo Finance. I will be using the adjusted close price for my analysis, and I will be accessing the data via the Yahoo Finance API. Since Peloton is a relatively young stock which only went public in late 2019, to make a consistent analysis, I will be using the prices from late 2019 to Febuary 2022.

DICK's Sporting Goods Inc. (DKS)

The dataset is retrieved from Yahoo Finance and contains the 5-year performance of the stock (DKS).

Lululemon Athletica Inc. (LULU)

The dataset is retrieved from Yahoo Finance and contains the 5-year performance of the stock (LULU).

Peloton Interactive, Inc (PTON)

The dataset is retrieved from Yahoo Finance and contains the performance of the stock from September 2019 to Feburary 2022 (PTON).

NIKE, Inc (NKE)

The dataset is retrieved from Yahoo Finance and contains the 5-year performance of the stock (NKE).

Under Armour, Inc (UAA)

The dataset is retrieved from Yahoo Finance and contains the 5-year performance of the stock (UAA).

Big 5 Sporting Goods Corporation (BGFV)

The dataset is retrieved from Yahoo Finance and contains the 5-year performance of the stock (BGFV).

Data Visualization

I used 5-year performance data for 6 companies: Lululemon Athletica, Dick's Sporting Goods, Big 5 Sporting Goods, Nike, Under Armour, and Peloton. Datasets cover the adjusted close prices from February 2017 to January 2022, with the exception of Peloton which went only public in 2019; therefore, for Peloton, the data cover the adjusted close prices from the first day of IPO till January 31, 2022.

The dashboard below shows the stock performance. In each graph, there is a difference in colors, which represents the trading volumes. The darker the color is, the higher the volume is. Overall, the line charts show an upward trend for the prices, except for Peloton, which is understandable since it's still a relatively young stock. One of the common patterns that can be observed on this graph is all stock prices declined in the beginning of 2022. In addition, the dent in March 2020 is consistent with the hit of the global pandemic. Relatively speaking, the sporting goods stocks have similarities in terms of the trend and investor's expectations.

Besides the trend, it is hard to observe seasonality from the graph. However, it is possible that seasonality does exist, given that scholars and practitioners have studied the "January anomaly" and "days-of-the-week anomaly". It will be studied in more depth in the EDA section of the portfolio.

Exploratory Data Analysis

ARMA/ARIMA/SARIMA Models

Spectral Analysis and Filtering

TBD

Financial Time Series Models (ARCH/GARCH)

Based on the plots, I will be fitting a Garch(1,0) model for Lululemon stock returns. The GARCH (1,0) model returns both alpha 1 and beta 1 significant at the 99.9% confidence level, and the model computes an AIC of 6.72 and a BIC of 6.75. The residuals are not normally distributed though, based on the Ljung-Box test.
The next is Big Five. I will be fitting Garch(2,2), Garch(1,1), Garch(1,2) and Garch(2,1) and then select the best performing model.

GARCH(2,2)

The GARCH(2,2) model computes an AIC of 1.47 and a BIC of 1.51.

GARCH(1,1)

The GARCH(1,1) model returns an AIC of 1.51 and a BIC of 1.54.

GARCH(1,2)

The GARCH(1,2) model returns an AIC of 1.52 and a BIC of 1.56.

GARCH(2,1)

The GARCH(2,1) model returns an AIC of 1.48 and a BIC of 1.52.

Although GARCH(1,1) model does not return the lowest AIC and BIC , all the coefficients are significant at the same confidence level; therefore, GARCH(1,1) model is more appropriate in this study for the Big Five Sporting Goods stock prices.


To summarize, Lululemon stock returns can be modeled with a GARCH(1,0) model, Big Five stock returns can be best modeled with a GARCH(1,1) model, and Peloton returns can be modeled with an Arima(1,2,0) model as explained in the previous section. However, it is important to note that based on Ljung-Box test, all p values are very close to 1, meaning that the residuals are not normally distributed.

Deep Learning for TS

Conclusions

The sporting goods have experienced great growth over the past few years, even amid of the global pandemics, as more and more people developed an increased awareness of health and fitness. However, we do see a significant decrease this year. Why is this the case? Are sporting goods stocks overvalued? That requires a thorough valuation which is out of the scope of this project. Nonetheless, there are several takeaways from this project that could provide some answers to the bigger question.

1. Stock prices are hard to predict. Given my background in Finance, I know that it is not a good idea to use historically data to know the future performance. Also, stocks usually move irregularly, and that is why the random walk model is good for studying stocks. The sporting goods industry is no exception. I tried using complicated ARIMA models and GARCH models, but most of the time, the ARIMA (0,1,0), aka the random walk model, performs better. Thus, simple models are not inferior compared to complicated models.

2. The predictions based on ARIMA models tell us that the sporting goods industry is likely to keep growing. Despite the drops recently, the industry still has a rosy outlook as the heath awareness is likely to help people to keep good habits which are beneficial to the sporting goods industry.

3. Before starting this project, I would like to see if the trend and patterns would persist across different regions. However, it is very challenging to obtain data from other markets. Besides, the companies chosen in this project are more representative in North America. Thus, it is safe to conclude that the patterns would vary across markets. The results from this project cannot be generalizable.

4. Lastly, deep learning could be a powerful tool to make predictions. However, given the small sample size and great fluctuations in the stock market from 2019 up till early 2022, deep learning models did not outperform ARIMA models in this project. Again, sometimes simple models would outperform complicated models, so it is always important to consider use cases for different projects.