The world’s leading publication for data science, AI, and ML professionals.

The Cryptographic States of America:

A Look at Bitcoin ATMs Across the U.S.

Purchased from artgraffiti on Anytask.com
Purchased from artgraffiti on Anytask.com

For this Data Storytelling project, I was interested in:

  • visualizing the distribution of Bitcoin ATMs across the U.S.
  • evaluating the possible negative effects of restrictive Bitcoin legislation and the influence on Bitcoin ATM availability

My research question– Did the BitLicense legislation that New York state passed in 2015 have a negative impact on Bitcoin ATM availability in the state?

Background

Bitcoin is a transparent and decentralized network and was created in 2009 by an individual under the pseudonym Satoshi Nakamoto. Bitcoin proponents frequently promote the idea that bitcoin allows everyone to "be their own bank".

Due to its’ decentralized nature, the Bitcoin network allows users to transact peer-to-peer with no need for the intermediaries we are used to: traditional banks. There is one deterrent to that idea, though; individuals need an easy way to purchase Bitcoin.

Centralized exchanges such as Coinbase allow individuals to purchase bitcoin but doing so requires a bank account. This results in a circular situation: to "be your own bank", you first need a bank account to kick-start the process.

In 2018, the Federal Reserve estimated that there are as many as 55 million unbanked adult Americans.¹ It’s more difficult for unbanked individuals to buy bitcoin; people with bank accounts have many more options.

Bitcoin appears to be transitioning into a store-of-value more so than an everyday currency; it has seen an unprecedented increase in USD value during a global pandemic. According to SEC Chairman Jay Clayton, "We determined that bitcoin was not a security, it was much more payment mechanism and store of value.²"

Paypal has recently allowed users to buy bitcoin directly, but currently, there is no way for customers to withdraw what they purchased. If unbanked individuals want to buy Bitcoin, they have few options to benefit from its store of value, and freestanding Bitcoin Atms can fill that need for unbanked and banked users alike. Despite the fees and rates associated with their use, Bitcoin ATMs are very convenient.

Bitcoin ATMs that allow users to walk up to the machine, buy bitcoin using cash, and have it immediately sent to a bitcoin address of their choice represent a valuable and convenient method to buy bitcoin. The CoinATMRadar dataset that I evaluated for this project contains Bitcoin ATMs that fit that criteria. (The alternative to freestanding Bitcoin ATMs require users log on to their website to redeem their purchased bitcoin.)

Dataset Description

For this Data Storytelling project, I evaluated a dataset detailing the locations and number of Bitcoin ATMs across the U.S. I received the main dataset directly from CoinATMRadar, and it includes freestanding Bitcoin ATMs in the United States as of November 16, 2020. You can view the corresponding code/IPython notebook for this project at the link at the bottom of this blog post.

CoinATMRadar is an online directory that lists Bitcoin ATM locations, and they have strict requirements for listings. Their directory also includes ATMs that buy and sell other cryptocurrencies including Ethererum, Monero, and Dash, among others. I did not evaluate alternative cryptocurrencies for this project; Bitcoin is the exclusive focus. There is a self-reporting mechanism to listing a Bitcoin ATM, and so there is probably a small gap of time before any potential errors are detected.

The specific limitations to this dataset:

  • Only current as of November 16, 2020
  • Does not contain CoinStar machines (they require redemption later with valid Driver’s license)
  • Does not contain LibertyX ATMs because they allow debit card purchases

I also downloaded a dataset with estimated 2020 population for all U.S. states from census.gov. and merged the two datasets so that I could evaluate the number of Bitcoin ATMs in relationship to each state’s population.

In 2015, New York state passed legislation that required any company who wanted to operate an exchange in the state or sell Cryptocurrency to apply for a BitLicense. Many people have speculated that this has had a negative effect on the industry, and I wanted to quantify and prove that effect.

Here I'm buying bitcoin from a local Bitcoin ATM listed in CoinATMRadar's dataset. I had to trek out in the rain, and this ATM was tucked in a corner of a small gas station.
Here I’m buying bitcoin from a local Bitcoin ATM listed in CoinATMRadar’s dataset. I had to trek out in the rain, and this ATM was tucked in a corner of a small gas station.

Data Wrangling and Feature Engineering

My data wrangling process included cleaning the column that listed city locations. There were some misspellings that needed to be corrected. Also, the city column included punctuation and state abbreviations in some cases. All of those needed to be removed. Using Python’s split() method allowed me to easily remove the commas in many rows and any two-letter abbreviation that appeared after it.

I also used Python’s RegEx module. In total, it took me 16 lines of code to clean a dataset that contained 9,550 rows. In future units, I look forward to expanding my data wrangling skills and being able to accomplish the process in fewer lines of code and employing the least-compute intensive process possible.

I also added a new feature to the dataset including a column that calculated the per capita ATMs per 100k residents. Here is a visualization showing those calculations.

To create the above visualization, I needed to define a data dictionary and map it to each state, resulting in a new column in the dataset that contained the two-letter abbreviations for each state.

For those who prefer to see the states by name, here is another visualization that indicates the top-ten states for Bitcoin ATMs per 100,000 residents.

My statistical methods included creating a scatter plot and overlaying a regression line, which showed an apparent positive linear correlation between population and number of Bitcoin ATMs, with an outlier at 1e7.

I then calculated a correlation coefficient to measure the strength and direction of this relationship. The result was 0.93 which represents a strong correlation. New York state is our nation’s 4th most populated state, yet only has 113 Bitcoin ATMs. This much lower than the mean of 198, which we wouldn’t expect from a state that size.

Next, I wanted to investigate my initial idea that the restrictive BitLicense legislation that New York enacted in 2015 has had a negative impact on the cryptocurrency industry as evidenced by limiting the number of Bitcoin ATMs in New York state beyond what we would typically expect from a state that populated. So, I created a linear regression model and used it to calculate three separate predictions.

The linear regression model showed that 86% of the variability in number of Bitcoin ATMs can be accounted for by population, so I was unsure if this project would be able to quantify my initial idea. However, the resulting predictions came as a pleasant surprise!

Results and Conclusion

I first calculated a prediction for the number of Bitcoin ATMs for a state with a population equal to the size of New York and also a prediction for both New Jersey and Connecticut.

The predicted number of Bitcoin ATMs for a state whose population is equal to New York state’s population is 634.69, which is 521.73 more than the mere 113 Bitcoin ATMs that New York has!

New Jersey has 286 more Bitcoin ATMs compared to neighboring New York state, even though New Jersey is less than half as populated.

The linear regression model predicts that a state the size of New Jersey would have 270 Bitcoin ATMs and Connecticut would have 67.3. According to our dataset, New Jersey has 399 Bitcoin ATMs and Connecticut has 154.

It’s certainly possible that businesses that would have opened Bitcoin ATMs in New York went to neighboring New Jersey and Connecticut and opened Bitcoin ATMs there to avoid the high cost of applying for a BitLicense.

This could account for New Jersey having roughly 129 more Bitcoin ATMs than our linear regression model predicted and 286 more Bitcoin ATMs than New York state, even though NJ is less than half as populated. It could also account for Connecticut’s actual number of Bitcoin ATMs being more than double the predicted number. Compared to New York, Connecticut has 41 more ATMS even though Connecticut is less than one-fifth as populated!

Further areas of research include periodic similar evaluations on a more recent dataset. Also, cross-referencing Bitcoin ATM locations from multiple sources before comparison could further refine the results. Legislation involving Bitcoin and the cryptocurrency industry as a whole is evolving, and future research and analysis can take into account these changes as they develop.

Also, I did a lot of feature engineering to explore the possibility of quantifying the term "bitcoin-friendly state". This exploration didn’t make it into the final IPython notebook, but it could be a sizeable and separate area of research by itself!

Here is a link to the IPython notebook corresponding to this blog post: https://github.com/cryptobellum/DS-Unit-1-Build/blob/main/The_Cryptographic_States_of_America.ipynb

Works Cited

(1) https://www.bloomberg.com/news/articles/2019-06-04/why-cleveland-wants-to-bring-back-postal-banking

(2) https://news.bitcoin.com/us-cryptocurrency-regulation-sec-chairman-jay-clayton-bitcoin/


Related Articles