Personal Statement for GStat Accreditation from the American Statistical Association

In my senior year of high school, at age 17, I passed the AP Calculus BC Exam. In my first semester of college, at age 18, at University of California, Santa Cruz, I passed the multivariable Calculus III course. Through the Casualty Actuarial Society (CAS) and Society of Actuaries (SOA), I passed Exam 1/P (Probability) in 2014, Exam 2/FM (Financial Mathematics) in 2016, and Exam SRM (Statistics for Risk Modeling) in 2018. In May 2022, I graduated from Texas A&M University with an MS degree in Statistics, including an emphasis in Applied Analytics and Data Science, as well as electives in Biostatistics and Bioinformatics.

In 2015, I spearheaded a family reunion which filled a ballroom with 320 relatives. Throughout the year leading up to the event, I used Google Sheets to create a scatter plot of tickets sales. Next, I fit a trendline and used its polynomial equation as a predictive model to forecast the total ticket sales. This was important for planning purposes because the venue had a deadline for us to give them our final headcount. This point estimate would determine the amount of food that had to be prepared, along with other expenses that we had to budget for. So, this was a valuable real-world experience in using data to predict the future. This is what I refer to as looking into a scientific crystal ball, which is the passion underlying my choice to take a graduate course in Time Series Analysis as part of my MS degree in Statistics.

In August 2022, I began my data professional career at Credit One Bank in Las Vegas, Nevada. This fintech company is a digital credit card bank that specializes in subprime borrowers. I entered as one of the eight recruits in the inaugural cohort of their Management Associate Program (MAP). This was a two-year program in which we completed four consecutive six-month rotations with the job title of Management Associate II. The rotations that I completed were in Financial Planning & Analysis (Treasury & Finance), Customer Engagement (Marketing), Budgeting & Expense Management (Treasury & Finance), and Current Expected Credit Losses (Risk Modeling). After completing the program, I have remained on the CECL team as a Risk Modeler II.

In graduate school, most of my work was done using the R programming language, then using R Markup to output the results for coursework submission. Upon starting at Credit One Bank, I instead used Python. Here, I learned Python through self-study, using the Python for Time Series Data Analysis course in that language, taught by Jose Portilla, on the Udemy platform. I took the course on the job, but used the real data set for my work project rather than the sample data set provided by the course. The time series analysis that I learned in this process became the common thread of my work throughout the bank.

For my first MAP rotation, my time series project culminated in the forecasting of weekly data for the average payment per account on Mondays, for the bank’s 17 million active credit card members, at the time. The five forecasting models that I fit were additive triple exponential smoothing in Excel, then in Python, followed by ARIMA, recurrent neural networks (RNN), and the open-source Prophet tool developed by Meta. Using the “horserace method,” one of these would prevail with the lowest mean absolute percentage error (MAPE) for a given forecast. The metric that we chose to forecast had an MAPE of 9%.

For my second MAP rotation, I adapted the same Python code and forecasting models to the number of new member acquisitions (boards) at the end of a direct mail marketing campaign. This entailed the following data preparation. First, since the various campaigns ended with different numbers, we decided to transform the historical data to percentages by dividing by the final number of boards in each campaign. Second, since the campaigns overlapped in time, we changed the index from datetime to integer. Third, we linked the campaigns from end to end, and renumbered them altogether, starting at zero (in Python, rather than one in R). Surprisingly, the forecasting model with the lowest MAPE was the simple average across the top ten credit card products by dollar volume. This model now lives in a Tableau dashboard that the Chief Marketing Officer (Michael Coleman) regularly consults.

One day, during my third MAP rotation, there was an intermittent power outage. This resulted in a loss in gross credit card payments collected for that day. The Assistant Vice President (Tim Krueger) who I worked with in my first MAP rotation, asked if I could apply my forecasting models to estimate how much money in payments the bank would have normally collected that day. Prophet was selected as a parsimonious model with 12% MAPE. The resulting forecasted loss of $11.5 million was very close to the number that Tim obtained using his own methods, making it a mutually successful validation.

My fourth and final MAP rotation was in Loss Forecasting within Risk Modeling (in Risk Management). Starting with the SAS production code that generates the monthly values for CECL (current expected credit losses), I adapted that model to generate an historical data set of daily CECL values starting in 2022 (for 3+ years of data). Next, I fit a Prophet model in Python to forecast coverage ratio from mid-month to month-end, with an MAPE of 2%. At the time of this writing, the bank’s coverage ratio is approximately 17%, from a CECL value of $2.3 billion divided by aggregate outstanding cardmember balances of $13.5 billion.

Since developing the CECL daily forecasting model, I have automated the SAS and Python code to run daily. Currently, I am in the process of recreating the Jupyter notebook as a notebook written in PySpark within the Snowflake cloud environment. The next step will be to create a dashboard there using Streamlit. This foray into computer science has elevated my skillset beyond statistics, into the larger new world of data science. That said, my primary interest remains with my pedigree in the world of statistical data science.

List of Accredited Members