Well-established practice exists for trading traditional currencies. Via foreign exchange, you can, for example, swap the U.S. dollar for the euro, Japanese yen or other global currency — potentially turning a profit. Can foreign exchange practices also be applied to the cryptocurrency space, wherein there exist thousands of currencies of vacillating value? Nicholas Miller, Timothy Chen and Sophie Deng, students in the University of Michigan School of Information’s (UMSI) Master of Applied Data Science (MADS) capstone class, used data science and machine learning techniques to find out. 

For students in UMSI’s 100% online MADS program, SIADS 697: Capstone is a culmination of training on diverse data science concepts and methods. The MADS capstone is a project-based course in which students propose and build end-to-end data science projects in their domains of interest. 

“By the time MADS students get to the capstone, they’ve covered a lot of topics in data science and tend to have ambitious ideas for how they can 'make their mark' on the field,” says UMSI lecturer and research investigator Elle O’Brien. In addition to demonstrating mastery as data scientists, capstone students produce a creative, original and technically rigorous portfolio piece. 

MADS graduate Nicholas Miller, a Shibuya, Tokyo-based software engineer and data scientist for Trust Lab, sat down to discuss the capstone project he completed with fellow graduates Timothy Chen and Sophie Deng, “Creating and Evaluating Successful Cryptocurrency Exchange Strategies.” He dives into what they learned about predicting buy and sell opportunities and the exchanging of cryptocurrencies, and reflects on his capstone experience and how the project has served him post-graduation. 

What is cryptocurrency?

Nicholas: Cryptocurrency is basically electronic money that’s decentralized. What that means is it doesn’t have any brick and mortar stores — there are no institutions that run and dictate how the currency is issued. 

This decentralized currency started off pretty small, but it’s grown into multiple different currencies. The big one that started it was Bitcoin. Ethereum is probably the next biggest, and then there’s several other smaller ones that also exist. It’s very easy for someone who is familiar with cryptocurrency to create a cryptocurrency. So there are thousands and thousands of cryptocurrencies out there, but they’re not all of equal value. 

Can you provide a basic overview of your capstone project? 

Nicholas: Think of traditional currencies across the globe, for example, the U.S. dollar, the Canadian dollar or the Japanese yen. There’s a well-established practice for trading traditional foreign currencies, and there are different strategies to do that — you can find tons of tutorials on foreign currency exchange, or “forex.” 

The question is, can you apply these same ideas to cryptocurrency and all the different currencies in that space? Can you do the same thing and grow your cryptocurrency? 

It’s something that has been looked into by researchers and is being used by many people today. A lot of financial institutions are probably looking into it because it’s kind of a sidestep to their businesses that currently do foreign currency exchange. Now they want to do it in the cryptocurrency space because there is money to be made in exchanging money.

Our project examines that space wherein different cryptocurrencies are traded to see if we can use data science and machine learning techniques to predict when is a good buy opportunity and when is a good sell opportunity and therefore make a profit. 

Which a-ha moments from your work have stuck with you?

Nicholas: One of the struggles I had was I knew I could see the Bitcoin pricing charts and how they moved, but I needed to come up with a way to test my algorithm. I had to figure out how far back it’s going to look at the cryptocurrency exchange to predict the next price. And then, was it actually going to predict the price? Because that’s typically what people do, they say, “OK, based on this history, the next price is going to be here.” 

I was wondering if I should do it that way, but then what I discovered through my research was that you can actually set a threshold with two sell limits. It’s basically a boundary box so if the value of your cryptocurrency either gets too high or too low, you sell. We visualize those boundaries, calculated from historical 15-minute chunks of data, in candlestick charts. You plug that data into the algorithm, which then tells you based on this moment whether it’s a good time to buy, or it will automatically sell. That was the machine learning part.

Figuring that part out was a unique challenge, but it brought a huge benefit. I was able to narrow a guessing game into a binary problem. The program automatically sells when your crypto value hits the upper threshold target price or lower threshold stop-loss price, so you never have to make a decision within the program to sell — the problem is cut in half; you don’t need to worry about when to sell, just about when to buy. It really helped me focus my efforts. 

What were your team’s conclusions? 

Nicholas: What we found is that the market is very random. We tested out a lot of models, and some models were profitable, but most models are not profitable. And over time, even some of the profitable models dipped below nonprofit, and I’m sure some of the nonprofit models would eventually become profitable — but that’s kind of the randomness, I think. 

One area I wish we could have spent a bit more time on is the statistical arbitrage, which was Sophie’s work. The interesting idea behind that is if you see two currencies moving up, then you can probably assume that this other currency will start moving, if it hasn’t already. We did play with statistical arbitrage, but we weren’t able to get a very strong model with it. I think if we had more time, it would have been an interesting place to explore. 

How have you been able to cultivate community in the MADS program?

Nicholas: That’s one thing that’s quite unique about the MADS program. The remote aspect of this master’s program allows me, as well as my peers, to actually participate in it. I’m based out of Tokyo, and my team members are in Singapore and Hong Kong. 

As we worked on the capstone project together, we never physically met each other. We did everything via Zoom or Slack, chat or email, and that’s how we communicate. We have established friendships over multiple projects to the point where we know each other’s personalities and how we work. 

How has your capstone project propelled you forward?  

Nicholas: Capstone really set the stage for how to go about addressing new problems and how to explore the space around something you’ve never done before. We tried to take up something new, challenging and interesting to us, even though we had zero experience with cryptocurrency, and we tried to see what we could learn and presented our results that way. 

I think that experience has really prepared me for my current work with social media measurement, which is something I’ve never done before. I’m drinking from a firehose when it comes to learning new stuff, but that’s what I enjoy: exposure to new content, new methods. It’s been a great learning experience, and I’ve brought that approach into my new role. 

What are your greatest takeaways from your MADS experience?  

Nicholas: I enrolled in the MADS program because I was really interested in artificial intelligence and machine learning. What really surprised me about the program and myself is that while I didn’t expect to be interested in other aspects of data science — the analysis and manipulation of data, data engineering, exploration and visualization of data, et cetera — I fell in love with data, and now I love being a data scientist. Now I look at data engineering pipelines and that’s something I want to work on. I’ve learned to appreciate all aspects of the science. 

I also really enjoyed the diversity of the MADS program, and there’s many different aspects of diversity: diversity of location, for example, and diversity of work background. My peers worked in health care, education, finance, and I’m in tech. It made the program much more interesting. And the faculty are energetic, friendly, open and always there to help. 

 

View Nicholas, Timothy and Sophie’s final project here.

Learn more: Q&A on MADS capstone course with lecturer Elle O’Brien