Developing a tool to understand and utilize crypto asset portfolios
By, Kevin Le
This is a case study of a project I did back in 2018, where I developed a data analysis product for companies to better understand and analyze crypto assets. There are certainly things I would do differently with it today, but it remains a great experience that I wanted to share!
Note that this post does not represent trade advice, any decisions made based on it are accountable to you.
SUMMARY
Many companies nowadays have been creating expansive portfolios of crypto assets. However, because of the novelty and volatility of the market, many groups can have difficulty working with their crypto assets, resulting in compounding negative financial effects.
Here’s how I developed a software application to help groups better understand crypto assets and make data-informed decisions related to trades, portfolios, and risk.
CHALLENGE
Some groups with crypto assets rely on liquidating crypto assets on a regular basis to cover operational expenses. But, as generally new groups, they lack the time, capital, and technical expertise necessary to engage in conventional trading. Further exacerbating the issue is the high volatility of the crypto market, the technology’s current infancy, and the technology’s novelty. While no one’s really sure what to do, many are experimenting.
How could a tool be designed to help companies iterate crypto-related decisions without significantly increasing time and money expenses?
INSIGHTS
I collaborated with a group where the Strategy branch and Operations branches both wanted to explore and extract data-backed insights, but to different ends. Operations wanted an infrastructure and platform in place to easily explore crypto assets across basic to advanced statistical tests and simulation models. Strategy required immediate actionable insight that someone without a statistics background could understand. Both needed the ability to iterate and experiment. So, the tool had to satisfy two distinct users: the researcher and the business person.
SOLUTION
I created a software application called the “Crypto Explorer” using R, Shiny, and the CryptoCompare API. It supports every crypto asset and crypto comparison on CryptoCompare, and has two sections: the General section and the Playground section.
General Section
- Technical Analysis
- Quantitative Analysis
- Fundamental Analysis
- Risk Evaluation
Playground Section
- Standard Metrics
- Probability Distributions
- Portfolio Analysis
- Machine Learning Algorithms
- GARCH
- ARMA/ARIMA
The General section was designed for the business person. It shows visuals and simplified outputs that are immediately relevant to financial decisions.
The Playground section was designed for the researcher. It provides granular inputs and a wide variety of modular mathematical methods and backtesting.
Both sections are completely interactive, allowing any user to quickly and easily experiment as desired. For this case study, I will focus more on the General Section.
General Section
Technical Analysis
The biggest key consideration for the Technical Analysis was that users tend to have little to no experience in trading. Technical indicators are one of the most common tools in trading. In order to effectively communicate this without expecting people to research trading independently, I decided to focus on a few common technical indicators such as Relative Strength Index (RSI) and Moving Average Convergence/Divergence (MACD). Furthermore, I provided helper functions, such as toggling indicator explanations and making all graphs interactive.
In Technical Analysis, it is usually better to work with a few indicators and refine them from there. By limiting the subset of indicators and providing an easy input menu to set parameters, users at the company could avoid being overwhelmed by choices that they do not have time to parse through. Additionally, for the most part, the indicators selected are fairly easy to read. This further increases the accessibility of the application. However, the indicator types are diverse. For example, not every indicator is a momentum indicator. Thus, a reasonable analysis can still be conducted through corroboration. I also operated on the assumption that other traders use these indicators frequently because of how easy they are to read, resulting in self-fulfilling situations.
Now, Technical Analysis is not perfect nor always reliable. But, it can be iterated using the input menu. For example, should the parameters for a buy signal in crypto assets be the same as in a traditional stock? This can now be explored.
Quantitative Analysis
The Quantitative Analysis section does not match Quantitative Analysis in a traditional sense. Because of the decentralized nature of crypto assets, the market lacks public financial statements or sales revenue. Rather in this section, I have built in a quantitative decision model algorithm developed from the Playground section of the Crypto Explorer. Once again it is important to consider that users will not necessarily have a statistical or computer science background to understand the underlying process. As a result, the inputs are condensed to just what a user can understand — projecting days out. The output is also simplified down to a heat map describing the algorithm’s predicted probability distribution.
However, predictive algorithms are not always reliable, especially in a market so volatile. During the development of this application, one issue that began to arise was that users were treating the Quantitative Analysis independently of the Technical Analysis rather than together. Furthermore, users weighed the Quantitative Analysis higher because they felt it was more advanced and because of the convenience of just receiving a definitive answer.
To respond to this, I added what I call a “Lagging Predictions” indicator, which is a binary map of up and down projections for multiple different timestamps.
This way, projections could be viewed in a contextual and visual way — similar to a technical indicator — rather than in a definitive way. The map also makes it clear that the Quantitative Analysis is not entirely reliable because predictions have conflicts.
To further encourage the synergy between Quantitative Analysis and Technical Analysis, I made the Lagging Predictions available on the “Technical Analysis” page so that both Quantitative Analysis and Technical Analysis could be present in one place.
Fundamental Analysis
Out of all the primary analyses, Fundamental Analysis is, by definition, the most subjective. However, if done properly, it can fill in the gaps that Quantitative and Technical Analysis present (e.x. news events). Because of the subjective nature of Fundamental Analysis, it is not something that can be programmed. Additionally, Fundamental Analysis can take a very long time to perform — time which many groups do not have.
I instead decided to focus on providing a support tool for users at groups with crypto assets to iterate their ability to do Fundamental Analysis more efficiently. Key to this is the fact that these groups are immersed in the blockchain industry landscape and news, providing them a higher ability to properly value blockchain products compared to the average person.
I decided to set up an Events Log — pictured above — as a lightweight framework for backtesting fundamental analysis. The idea is that as events come up, users can input what they think will happen and later check their assumptions. Over time, users can better discern what events actually make a significant difference for crypto assets, and perhaps even see patterns. Additionally, this log is shared by everyone — thus creating a cumulative, centralized source of backtesting fundamental analyses.
Adding to the Events Log is fast and users can be as comprehensive in their explanations as they want. At minimum, the log simply asks for the date of the event, the source, the actual event, the expected market change, a short rationale, and name. I included name as a required field to encourage collaboration and conversation between employees. For example, if I disagree with someone’s analysis and want to know more, then I know who to reach out to.
Finally, I also included a news feed that users can search through and filter to make the Fundamental Analysis tab a one-stop shop.
Risk Evaluation
The last part of the General Section is the Risk Evaluation tab. I decided to focus on two risk measurement and response practices: Value at Risk (pictured above) and Portfolio Analysis. I used Value at Risk as it is a common and effective practice in finance that most business people can understand. Likewise, it is easier to negotiate with potential clients using a metric that is generally understood and accepted, even if it might be flawed for crypto assets. Many clients request doing full or partial payments in tokens. In such a volatile market, it is critical to price such tokens properly. In the Crypto Explorer, daily Value at Risk is measured using GARCH (generalized autoregressive conditional heteroskedasticity), which was tested in the Playground section. Longer-term Value at Risk is measured in a more traditional way, using Historical Value at Risk.
To this end, I also allowed for parameters to be set for liquidity adjustment — position size, bid price, ask price — so that whoever handles new business can simulate the outcomes of different negotiations quickly.
The Portfolio Analysis optimized for minimal variance (i.e. risk) for any combination of crypto assets based on historical data. Users could also set additionally parameters, such as weight constraints and box constraints, to further iterate how the company (or any user) balances their portfolio.
Playground Section
The Playground Section is named as such because it is a place to “play around” with crypto assets in a much more comprehensive manner than the General Section. This is where the models for Quantitative Analysis and Risk Evaluation (from the General Section) are put to the test, along with many other insights that are useful for financial purposes. Below are the features of the Playground Section!
Standard Metrics
- Confidence interval for mean
- Confidence interval for variance and standard deviation
- Max Drawdown
- Sharpe Ratio
Probability Distributions
- Histogram of [Log] Returns or Price
- Normality plot of [Log] Returns or Price
Portfolio Analysis
- Full investment optimization based on historical data (selected time range)
- Optimized to minimize standard deviation
- Interactive plot of randomly generated portfolios (Std. Dev. vs. Mean Returns)
Machine Learning Algorithms
- Summary of indicators/potential features
- Mean, Standard Deviation, Kurtosis, Skewness → Gives an idea of what the data looks like
- Correlation plot of indicators
- Feature Selection
- Train, Compare, and Combine Models
GARCH
- GARCH Bootstrap Summary
- GARCH Bootstrap Plot
- GARCH Rolling Estimation Summary
- GARCH Rolling Estimation Plot
- GARCH Bootstrap Backtest
- Conditional Coverage and Unconditional Coverage Hypothesis Tests
ARMA & ARIMA
- Log Returns Plot
- Augmented Dickey-Fuller Test
- ACF Plot
- PACF Plot
- Actual Returns & Forecasted Returns from ARMA / ARIMA
- ARMA & ARIMA accuracy evaluation for up/down movement
RESULTS
During the development of the project, I was presented with various real financial decisions. Using the Crypto Explorer, I helped advise decisions which resulted in a a significant amount of money saved.
However, this could have all come down to luck. At the end of the project, I set up a trade log with BTC and ETH (the two most common assets moved) in a public spreadsheet using a simulated account of $1000. Every move would be left immutable, and my decisions would be based on my use of the Crypto Explorer. I maintained this log for 18 days.
Market Conditions during Trade Log
- Bearish (down) or sideways.
- Worst drop occurred in first week, where ETH dropped by about 30%.
End Result
- Final Return: 13.15%
- (Control Situation) Return if no trade moves were made: 6.01%
I made mistakes during this time, and not all of my moves were optimal. However, I significantly outperformed the control situation while acting as a casual and amateur trader, which best demonstrates many companies’ situation and supports the use of the Crypto Explorer as an effective exploration tool.
Ultimately, users could now efficiently analyze countless crypto assets and crypto asset comparisons from simple to advanced ways, with a defined section for meaningful outputs. Smart, data-backed decisions could now be iterated to drive business, reduce risk, and navigate the most volatile and exciting market of the past century.
Thank you for reading! This is my blog where I share thoughts and experiences. You can learn more about me at my website here.
