Airflow-Crypto-Buying-Bot

A cryptocurrency purchasing bot using a crude Dollar Value Averaging strategy.

Introduction

The motivation for this program is to (1) automate the purchasing of cryptocurrency using Airflow, and (2) use a Dollar Value Averaging strategy to potentially improve the return compared to a Dollar Cost Averaging strategy. The return is compared to the theoretical DCA return with the same investment amount and timing.

The following two investment strategies are useful in understanding this motivation.

DCA

Dollar Cost Averaging (DCA) is basic investment strategy used to reduce the affect of market volatility on investments. By breaking an investment into many equal size investments over time, it minimizes risk of losing value if the market declines after investing. Multiple small investments averages out the volatility, producing a more consistent investment.

DVA

Dollar Value Averaging (DVA) is another strategy. Similarly, a large investment is split into smaller amounts, but DVA differs in that the amount invested is modulated based on how good the current price seems. In theory, this can greater reduce the chances of losses and increase the potential return. There exist more complex interpretations of this strategy, but this sufficiently describes how it is implemented.

Trading Strategy

The bot is composed of several tasks which make up a DAG, pictured below.

The first task, determine_value_task, looks at the current and past prices to ascribe a "value" to the current price. If the price is high, it will have a low value; if the price is low, it will have a high value. The value is calculated by comparing the current price to the average price over the past several days. The next task, place_order_task, places an order for an amount of cryptocurrency based on the determined value. The limits of the amounts to purchase and the length of days to average are defined by constants in crypto_buying_bot_dag.py. Next, find_filled_orders_task will collect all the orders placed by the bot and count_open_orders_task will count the orders that have yet to fill. The resulting information will be used by analyze_trades_task to calculate metrics on the orders. These metrics include the number of orders placed, the return, and the theoretical return if the same amount had been invested using the simpler DCA strategy.

Performance

After running this bot for one month on a 3-day interval, it executed 11 orders. The return is currently 5% higher than the theoretical return using a DCA strategy.

How to Run

To run this automated purchasing bot, the following is required:

• A Gemini exchange account setup for use with the API
  • Details on setting up the API is on their website: https://docs.gemini.com/rest-api/#introduction
• A server running Apache Airflow
  • Airflow must be configured with the Gemini API key variables
  • The API keys can be imported to Airflow using variables.json
• Optional: in order to use email alerts for run failures, an SMTP host must be initialized and configured in the airflow.cfg file
  • The chosen email address can also be imported to Airflow using variables.json

The constant variables in crypto_buying_bot_dag.py can be adjusted to the desired specifications.