Predicting and Trading on the Cryptocurrency Markets using Alteryx

• Analytics & Modeling - Predictive Analytics
• Application Infrastructure & Middleware - Blockchain

• Equipment & Machinery

• Maintenance
• Procurement

• Movement Prediction
• Smart Campus

• Data Science Services
• System Integration

Predict Crypto, a project by a student at the University of Colorado Boulder, aimed to predict, trade, and research the cryptocurrency markets. The project required a fully automated solution capable of dynamically pulling the latest data from a database, producing a new set of predictive models daily, and executing real trades on the live cryptocurrency markets every hour. The challenge was to navigate the volatile world of cryptocurrencies and observe patterns within the markets to make medium-term predictions. The goal was to cash in on these trends better than a simple 'HODL' (Hold On for Dear Life) strategy, which involves putting the investment away and forgetting about it for a couple of years.

The customer in this case study is a student at the University of Colorado Boulder who created the Predict Crypto project. The project aimed to predict, trade, and research the cryptocurrency markets. The student sought to create a fully automated solution that could dynamically pull the latest data from a database, produce a new set of predictive models daily, and execute real trades on the live cryptocurrency markets every hour. The goal was to navigate the volatile world of cryptocurrencies and make medium-term predictions to cash in on market trends better than a simple 'HODL' strategy.

The solution involved a four-step process. First, data was extracted from a MySQL database hosted on Google Cloud Platform. The 'train' dataset consisted of all the data collected relating to the cryptocurrency markets in the past 100 days. Second, data was prepared by extracting it from several different sources within the database and performing data manipulation to join all the data together and prepare the dataset for the predictive modeling step. Third, individual models were created for each exchange using the XGBoost framework programmed in R using the R developer tool. Additional models using the R Predictive tools found within Alteryx were also created. Fourth, trades were executed using the models created in step 3. The predictions were finalized applying an 80% weight to the predictions made by the XGBoost models, and 20% weight to the predictions made by the Alteryx models. The data was then passed to a Python tool, which leveraged the Shrimpy developer API to execute trades on each of the exchanges individually.

• The use of Alteryx provided a robust framework for the Predict Crypto project. It made the process less challenging to maintain than using R or Python alone. The time savings from a development perspective allowed for the creation of a wide variety of tests and the ability to organize them and enable/disable different ones easily. The project also benefited from the ability to connect the data stream between tools in Alteryx, providing a more powerful solution than using R Studio and the Reticulate package. The project built out of Alteryx removed limitations to data capabilities as the project evolved and created a huge amount of possibilities in terms of what could be achieved in a day compared to what could be done programming in R or Python.