Readme

This is a dummy trading bot. I have no idea about trading (yet), this effort is basically centered in clean implementation and composing the infrastructure.

In the future I will try to implement different strategies when possible using this as a sort of skeleton.

Implementation

This python implementation uses multi-threading extensively. While the GIL is still there I still have a dream that it will really go again in a while.

The multi-threading implementation is based in the ThreadPoolExecutor. As it simplifies thread managemennt and reduces thread creation overhead with limited over-subscription control.

Inter thread communication uses Queue which basically is a lock protected fifo with condition variables to avoid consuming resources in dangling threads.

The code uses 2 helper threads and a configurable number of workers (-w option in the command line).

1. A pooling service that keeps a cache updated information of the interesting asset prices. (pooling_prices)

2. An external listener that keeps checking input signals from redis. When a signal arrives this threads copies them into the fifo Queue where the worker threads are waiting in a blocking call. (redis_reader)

3. The workers receive the signal information and handle them with the proper strategy function. (worker)

   After the new order is submitted this threads also keep checking until the order is filled or canceled. This is a design choice at the moment because the orders are ioc. (pooling_check_order) But otherwise it will be delegated to another pooling service.

At the program beginning an important and potentially heavy initialization takes place. Such initialization includes checking current prices and account initial situation, but also compute the assets weight in the operations.

Alpaca API

As we use the alpaca API and I (not sure why) decided not to use the Alpaca library for python. This project implements two levels of abstractions over the Alpaca API.

1. AlpacaAPIClient: This is a direct interface for the alpaca API, which submits all the REST requests. In case this project grows, this class could be replaced with the official Alpaca API.

2. AlpacaAPIWrapper: This is a wrapper class that acts as a connector between the user code and AlpacaAPIClient. This class performs optimizations, protections and checks hidden to the user. It caches the prices and position values in order to direct access them without network delays. At the same time the class provide the update functions needed to update those values safety from the user code. Also include some optimizations to reduce latency and lock retention.

Signal Format

Signals appear in the Redis stream 'nvda' with the following fields:

• Ticker: NVDA
• Direction: "b" (buy) or "s" (sell)

Signals are generated randomly during US market hours (9:30 AM - 4:00 PM ET) with a 0.5% probability each second.