Consumer Complaints: Financial Products and Regional Analysis

-------------------------------------------------
Introduction
As a result of the financial crisis, the financial services industry has been under the spotlight in recent years. It has been alleged that financial institutions have poor business practices, leading to increased risk being transferred to consumers. These poor business practices would be expected to be reflected in consumer complaint data for these services. Public unrest with the financial services industry has also mainly been focused on certain products, such as mortgages, and large public financial corporations, such as Bank of America, or J.P. Morgan Chase.

The goal of our analysis is to understand the actual underlying structure and trends of consumer complaints regarding consumer financial instruments in order to understand if these assumptions hold true. This includes which financial products receive the most complaints, which geographic areas these complaints originate from, and whether a link between company size and consumer complaints can be identified.

-------------------------------------------------
Related Work
While multiple studies have been conducted to analyze consumer complaints in relation to behavior, strategy, satisfaction, and response; our approach to review consumer complaint data through financial product geographic segmentation can be deemed unique.

-------------------------------------------------
Processes
To fulfill the goals of this analysis, our team explored the following three research questions:

1. For each year in the consumer complaint dataset, which three products recorded the most complaints?

2. What region has the highest proportion of complaints in relation to the population of complaints for each year in the consumer complaint dataset? Within this region, which state has the highest proportion of complaints, and within this state, which zip code has the highest proportion of complaints?

3. Can a correlation between the market capitalization of the company and the rate of complaints received be identified?

To address these questions, we wrote multiple python programs utilizing the pandas library to analyze and visualize the data set.

Prior to starting our analysis of any of our purposed questions, our first step was to confirm data integrity and perform any data cleansing that would be necessary. Right away we discovered that our dataset did not contain complete data for all the years listed. The team concluded that to ensure our analysis was not skewed; the original dataset had to be reduced to only the available years of complete data, 2012 and 2013. From this point, we were able to dig into the data and start answering questions.

1. For each year in the consumer complaint dataset, which three products recorded the most complaints?

The team utilized Python to import the entire dataset into a pandas DataFrame. From the original import we created two additional DataFrames which contained the separated 2012 and 2013 complaint data. We then performed a groupby sum function on both DataFrames, with the grouped criteria being the financial product (ex. Mortgage, Credit Card, etc.). This provided a very clear picture of the top three financial products in each year that received the most complaints. We then chose to graphically display these results by plotting a bar graph. To ensue the clarity of our results, we differentiated the top three complaint products by color.

2. What region has the highest proportion of complaints in relation to the population of complaints for each year in the consumer complaint dataset? Within this region, which state has the highest proportion of complaints, and within this state, which zip code has the highest proportion of complaints?

In order to effectively perform analysis for our second inquiry, we needed to expand our data to encompass population demographics. The original complaint data set provided us with the state and zip code where the complaint was received. This allowed us to merge our existing DataFrame with the following additional datasets:

Regions_Map.csv – This dataset provided a breakdown of the US states and territories into ten different geographic regions. http://www.eia.gov/consumption/commercial/census-maps.cfm

Pop_By_Zipcode.csv – This dataset provided the population count for 33,092 zip codes. http://blog.splitwise.com/2013/09/18/the-2010-us-census-population-by-zip-code-totally-free/

Before we were able to merge these datasets, it was discovered that there were duplicate zip codes, each with their own population value listed in the file Pop_By_Zipcode.csv. Using a pandas DataFrame, we imported this dataset and used a pivot table to sum the populations in duplicated zip codes. Upon completing this task we merged these two external datasets with our original consumer complaints datasets of year 2012 and 2013.

Here we encountered our next obstacle. When reviewing the populations in relation to zip codes it was discovered that some populations in the areas defined by zip code were too large, while populations in other zip code areas were too small. Our concern was that the values on both extremes would skew our calculations and result in misleading findings. To solve for this we determined that segmenting the dataset into four segments based on population percentile would be our best course of action. This would allow us to still encompass the entire dataset while also avoiding skewed statistics being provided. Leveraging the numpy percentile function, the four segments were grouped into: 0-25th percentile, 25-50th percentile, 50-75th percentile, and 75-100th percentile.

The final step was to identify the highest complaint rate for each percentile grouping by geographic region, state within the region, and zip code with the state. Using our merged DataFrame, new columns were added to calculate the complaint percentages, defined by: Regional, State, Zip Code = Complaint count / Population count

This methodology was repeated for the remaining three population percentile groupings. Visuals were leveraged in all of our findings through Python. Again, to ensure clarity of our results, the highest value was additionally differentiated by color.

3. Can a correlation between the market capitalization of the company and the rate of complaints received be identified?

In order to effectively answer this inquiry the team decided to create a metric to normalize the complaint rate. This would allow us to adequately determine whether there is there a correlation between the market capitalization of the company and the rate of complaints a company receives. The normalized complaint rate metric is calculated as: the number of complaints the company receives divided by the market capitalization of the company, normalized on a scale of 0 to 1, where 1 is the greatest complaint rate. This would help to determine whether the rate of complaints increase with the size of the company, or if a stable number of complaints is expected per unit of market capitalization.

To generate this metric, the team would again have to obtain outside data sources to be merged with our original set. To attempt to match companies that were listed in the complaint data to those that were publically traded on the NASDAQ, NYSE, and AMEX we obtained the following datasets:

NASDAQ.csv – Contains all companies within this exchange as well as stock and financial data. http://www.nasdaq.com/screening/company-list.aspx NYSE.csv – Contains all companies within this exchange as well as stock and financial data. http://www.nasdaq.com/screening/company-list.aspx AMEX.csv – Contains all companies within this exchange as well as stock and financial data. http://www.nasdaq.com/screening/company-list.aspx We fist utilized pandas merge function to combine the three exchange databases together and then compared the list of public companies to our compliant database. This resulted in a limited amount of companies due to the inability correctly to match companies based on their names provided in the files. In order to produce a more robust analysis, the research team took the list of limited matches, and added any publicly listed company with over 1000 complaints in the Customer Complaint dataset.

Calculating the normalized complaint rate was next to be completed. We added a column to our DataFrame and divided the total number of complaints for a specific company by the company’s market cap (in billions). We then sorted in ascending order by complaint rate. Finally, we wanted to graphically show the companies market cap in billions against their calculated normalized complaint rate.

-------------------------------------------------
Results
1. For each year in the consumer complaint dataset, which three products recorded the most complaints?

In 2013, our analysis concluded that the leading three financial products in complaints were: 1. Mortgage 2. Credit Reporting 3. Bank Account or Service

Our analysis for 2012 yielded similar but not completely identical results: 1. Mortgage, 2. Credit Card 3. Bank Account or Service

2. What region has the highest proportion of complaints in relation to the population of complaints for each year in the consumer complaint dataset? Within this region, which state has the highest proportion of complaints, and within this state, which zip code has the highest proportion of complaints?

a. 0-25th Percentile
In 2013, the highest complaint proportion breakdown is as follows:

Region: Pacific
State: California
Zip Code: 90071

In 2012, the highest complaint proportion breakdown is as follows:

Region: South Atlantic
State: Florida
Zip Code: 34101

b. 25-50th Percentile
In 2013, the highest complaint proportion breakdown is as follows:

Region: Mountain
State: Utah
Zip Code: 84528

In 2012, the highest complaint proportion breakdown is as follows:

Region: Mountain
State: Utah
Zip Code: 84521

c. 50-75th Percentile
In 2013, the highest complaint proportion breakdown is as follows:

Region: Pacific
State: California
Zip Code: 93675

In 2012, the highest complaint proportion breakdown is as follows:

Region: New England
State: New Hampshire
Zip Code: 3809

d. 75-100th Percentile
In 2013, the highest complaint proportion breakdown is as follows:

Region: South Atlantic
State: Washington, D.C.
Zip Code: 20005

In 2012, the highest complaint proportion breakdown is as follows:

Region: South Atlantic
State: Washington, D.C.
Zip Code: 20015

3. Can a correlation between the market capitalization of the company and the rate of complaints received be identified?

Market capitalization of the company and the complaint rate of the company are not highly correlated. Correlation between the market capitalization of the company in billions of dollars, and the normalized complaint rate is -.333

-------------------------------------------------
Conclusion
While the analysis provided insights into the structure of consumer financial complaints in terms of geographic concentration, financial product composition, and company-specific metrics, our dataset prevented a deeper analysis of the data. One of the questions which is unaddressed is the composition of the complaints for specific products received for each company relative to the proportion of revenue these products represent for the company. This would provide insight into whether inefficient business practices can be identified at the product level.
Insufficient information regarding revenue also prevented the team from analyzing the dataset at a greater level of granularity. While market capitalization served as a proxy measure for company size in this analysis, revenue would provide a more precise underlying measure for the normalized complaint rate.