ES95r:Blockchain Inaugural, Fall 2017

We began our exploration with the intention of building a product or service not purely investment- or trading-related, to see if we could use the technology to create value that went beyond providing investors better returns. A such, we first took a wide survey of potential spaces, focusing on the system-level -- real estate, the legal system, healthcare, industrial equipment, and social.

Though we found inefficiencies in these systems, most could be solved by non-blockchain innovation, especially given the technical maturity of available blockchain tools. This led us to focus on giving others the power to use smart contracts more easily, via a sort of IFTTT interface, and seeing what they came up with, though we found difficulty in making onboarding users an easy process. Throughout, we also taught ourselves Solidity in order to build our technical prototypes.

Ultimately, we focused most on building our understanding of the technical space to allow for lower-level innovation in the future, as our exploration taught us that the biggest potential in the blockchain space still seems to be at the protocol level: creating oracles, optimized blockchains for specific use-cases, or integrations with other blockchains.

We began with research and technical exploration into the space. Those of use who took a more technical path (or who chose to do both technical and research) taught themselves solidity, while others researched current opportunity spaces. After landing on the IFTTT for smart contracts concept, we completed user research with small business owners and venmo super-users, developed a working backend prototype for triggering smart contracts based on events, and created a UI.

To continue this project, we would want to do further user research, beyond discussions and UI walkthroughs with local businesses and students. We envision a larger-scale simulation game we hold for ~20 people that puts our assumptions about social trust, and the number of trusted versus untrusted parties one needs in a trustable interaction, through extreme stress.

Smart contracts, despite their elegance, are quite limited by the fact that the blockchain has to be deterministic, such that the result of a computation must be completely verifiable on-chain. This means that smart contracts usually can’t have bearing on the outside world without an “oracle.” I thought that a great way to expand smart contract usage would be to allow an real physical items to be traded using smart contracts.

Utilizing existing APIs for logistics networks (FedEx, Uber) and oracles to link them up to contracts in the Ethereum blockchain. Through the process, I was able to set up a rudimentary oracle on oraclize, although the actual functionality of it was limited due to costs.

I implemented a connection between live, third-party APIs and the ethereum blockchain. Since it was difficult to set up complex API queries on the blockchain using Solidity, I instead opted to set up an outside server that took a simple request and a code and sent a corresponding request to FedEx, then had oraclize send a request to this server I set up.

In the process, I learned a lot of Solidity, and the specific security concerns that developers must consider. One of the biggest challenges I faced was trying to figure out bugs in my contract code, especially since there is not much literature online about setting up oracles in a consistent way.

The next step would be to create a convenient web-app to be able to create smart contract code without learning solidity. This will allow the everyday user to create logistics contracts, which will certainly increase the number of people using smart contracts.

We began by pinpointing the characteristic advantages of the blockchain and the corresponding weaknesses of the current investment processes in emerging and frontier markets. We determined that investors fail to (or rarely) invest in these areas given the lack of data they have about the regions as well as the lack of security in capital transactions. The blockchain, we decided, provided a good mechanism for secure transactions and was a viable source of legitimate data.

Given that we had no prior expertise on smart-contract implementation as well as design best practices, the majority of the first half of the semester consisted of finding the right resources to teach ourselves how the language operated, and how the different elements of decentralized applications operated with each other. In parallel, we had to align our product strategy with current limitations and potential roadblocks to perceptions and technical difficulties involving the blockchain technology. Cryptocurrency adoption was another issue we had to design our application around so that the volatile and unpredictable nature of the currency would not be a deterrent for institutions and businesses to adopt our technology.

Throughout the rest of the semester, we began to implement different aspects of our platform, including a working currency minting and destroying smart contract (Solidity), a user interface framework (JavaScript, Web3), part of a digital signature encryption mechanism (JavaScript), as well as a piece of a financial-institution registration platform (JavaScript).

Throughout this entire process, a lot of our time was used looking for the right resources that would teach us best design and implementation practices.

We have learned a lot and hope to continue our progress in this endeavor, but would have liked some support in our technical intelligence. We, however, understand that this is a nascent field, and are excited for what Harvard will offer in the future.

Our group completed a two-fold project: one was application-based and the other was research based.

Our current neural network predictor is an LSTM that uses historical price data as the input features. Since we have our general DQN architecture built, we next plan to expand on this predictor to take in more features that might be helpful in predicting expected reward. Our current predictor performs best on hourly data, so we are also interested in modifying this predictor to better handle the volatility of 10 minutely price data.

We looked at different approaches to algo-trading cryptocurrencies. The first stage was to get data: we coded and ran a data-logging server on the EC2 cloud which logged the minutely order book data across 17 different crypto exchanges. Next, we started an initial survey of differences between exchanges. Key findings included significantly different transaction fees (ranging from -0.1% to 1%), trading in different real currencies (USD, ZAR, CNY), differences in volatility and trading volume, and significant differences in the valuation of BTC based on exchange rates (for example, some exchanges traded BTC 10% higher than other exchanges – arbitrage trading this away is challenging, because capital needs to be moved between different real currencies, and deposited into a different bank account, for each trade).

We began a preliminary analysis of naive trading techniques: using nearest neighbors and linear regression to create estimators for future price changes, and trading accordingly. We found that transaction fees can make a significant difference to the level of profitability of high frequency trading algorithms.

Next we created a live paper trader to test these algorithms in real time – we found that the real time trading performance (vs performance on historical data) was much lower: this was because of small assumptions made when running the algo trader on historical data (for example, not taking into account that the bid price does not equal the ask price on the exchange).

After this, we started to look at more sophisticated time series models, including VAR models and the use of recurrent neural networks. While these models would give positive return, these tests were done during the time the BTC was increasing rapidly, so it was difficult to evaluate the success of these trading algorithms relative to a realistic baseline.

We want to continue testing and refining our algorithm, especially during a period when bitcoin is not increasing rapidly. Otherwise, it’s difficult to separate the effectiveness of our algorithm against the overall increase of the market, and it often also makes the most sense to just hold rather than actively trade currency during those price surges.

We started by identifying our group’s goal – we were all interested in investing and wanted to learn more about how to best invest our funds and where we can find more information about investing. We began by mapping out the most popular cryptoasset investments and analyzing their growth, value proposition, application, and tentative futures to see if there were any patterns amongst the cryptoassets that had grown to be the most popular.

As we moved deeper into the cryptoasset analysis, we decided to form an investment group to map out how well our newly-gotten information and instincts would fare in the cryptoasset investment market. However, upon midterm check-ins, we realized that this approach didn’t leave much lasting impact - we’d be the only ones to benefit from this type of approach. We decided to pivot to addressing a key point that had consistently come up during our research - it’s very difficult to cement the fundamentals of blockchain and set up an investing flow, so we decided to build an education platform that helped with exactly that.

We built a website, written for the complete novice, about how to invest in cryptocurrency, with the goal of making it a more accessible investing space.

We want to continue testing and refining our algorithm, especially during a period when bitcoin is not increasing rapidly. Otherwise, it’s difficult to separate the effectiveness of our algorithm against the overall increase of the market, and it often also makes the most sense to just hold rather than actively trade currency during those price surges.

Our group began this semester with some exposure to blockchain and cryptocurrency but little formal technical training in the space. Initially, we sought to focus on smart contracts and applications; in this vein, we set out to learn smart contract programming with Solidity on the Ethereum blockchain. We quickly realized that we would be more interested in studying the space at a higher level and developing a better understanding of the crypto landscape in terms of coin valuations and typologies. This motivation was in part spurred by the guest speaker and current HBS student Homan Mohammadi, who has spent the last few years studying the space more in depth.

After a few meetings, we began working closer with Homan on a few key areas: 1) establishing a more fundamental approach to cryptoasset valuations, and 2) developing a typology for the different kinds of coins being made available. There is immense opportunity in applying traditional economic and financial thinking to the cryptoasset space, and the dearth of research in this nascent field makes the opportunity relatively accessible.

We focused on building a better metric, by evaluating a metric called NVT (Network Value to Transactions Ratio) that is analogous to the PE ratio for a stock. The PE ratio represents a ratio of price to “health” of a given financial instrument, where health for stocks is calculated as the company’s earnings. For a cryptoasset like bitcoin, “health” can be thought of as network transaction volume. If people are using Bitcoin for more of their transactions, this justifies a higher price.

The fundamental value of a cryptoasset like Bitcoin can be thought of as either a medium of exchange or a store of value. Some of the ways in which we considered refining the NVT metric were: using circulated supply vs total supply of coins; and changing how we represented transactions (to avoid overcounting in cases where the transaction was represented as going to two addresses).

We see considerable opportunity for Harvard academics to weigh in on some of the valuation techniques and economic implications of the technology. As discussed, the emerging nature of this field of research provides an opportunity for newcomers to quickly get up to speed and to bring along interdisciplinary insights that can positively affect the space.