Terminus Research Hackathon

How to run the code

Prerequisites

• Install Poetry.
• Install Python 3.9

Installation

Run the following in your shell after downloading the code and unzipping

cd hackathon
poetry install

For executing the jupyter notebook playground.ipynb run

poetry run jupyter notebook

For executing for the evaluation script run

poetry shell
python evaluation_script.py

The task

Implement a trading strategy for a (secretly constructed) TRCoin given its historical data. This typically has two components:

• Forecasting a suitable future price.
• Efficiently turning the price forecasts into positions.

For details of the trading behaviour, read carefully through simulator.py.

See also the example strategy in strategy.py.

Data

• Training data covers the period from 2018-01-01 to 2022-01-01
• The simulator can run strategies on any date range in between 2018-01-01 and 2022-03-31
• The final evaluation will be performed by executing submitted code on the data in the range from 2022-04-01 to 2022-07-01

A few tips

• The simulator can be executed also for earlier periods for more careful validation of your strategies.
• Positions held can also be negative.
• Strategies that hold large positions for long periods of time are typically not good trading strategies (e.g. naive HODLing) and tend to generalize poorly. High frequency of position changes is a good (but not sufficient) indicator. Inspect the corresponding chart regularly (see playgroung.ipynb)
• Think twice about running expensive parameter tuning algorithms as it might be very easy to overfit to the short validation period.

Evaluation

There is a limit for submission per day (12). Profit calculation takes some time after uploading. There is limitation of submission archive size = 50Mb.

The key metric is profit at the end of the trading window (see for example evaluation_script.py). This will determine the winner of the hackathon assuming no irregularities (access to external data, tampering with the simulator etc.) in the submitted code are found.

For the purposes of hiring to Terminus Research, we will pay much greater attention to the inventiveness, technical quality of implementation, and overall robustness of the strategy. Feel free to submit additional models you developed as well. If your strategy looks great to us but got "unlucky" in the testing period, we will contact you!

The reviews will be performed by our employees who are world-class experts in algorithmic trading as well as in modern machine learning.

Intermediate leaderboard

During the competition, we will maintain a leaderboard corresponding to performance in between 2022-01-01 to 2022-04-01 (as specified in evaluation_script.py). This should give a general idea of the performance of the strategy. However, beware that your strategy might be overfit to the particular time period. In that sense, the leaderboard will not be predictive of the final test performance.

For this purpose, submit the sim_result.csv file (the output of evaluation_script.py). We will trust the fairness of your submission. However, our inability to reproduce the result once the code is submitted will lead to immediate disqualification. )

Final evaluation

For final evaluation, we will require the submission of source code. In particular,

• An updated pyproject.toml file that allows installation of specific python libraries you will require for your strategy.
• Your implementation of strategy.py
• Optionally a directory called tools which containts additional python scripts needed in strategy.py.

On the other hand, do NOT submit:

• any of the remaining python files (e.g. simulator.py or evaluation_script.py), we will provide them in their original form.
• any jupyter notebooks
• any files with data (again, they will be provided)
• any files containing trained ML models. Training should be included in the execution of evaluation_script.py.

The evaluation will proceed as follows:

• The data files will be extended to contain newer data points (both for training and for simulation).
• The simulation period will be adjusted in evaluation_script.py.
• The submitted files will be copied in the directory with the original hackathon code (overwriting the original files)
• The following will be executed:

There will be a limit of 90 minutes on training and 2 hours on the evaluation of the simulator, on a computer with 32GB RAM.

cd hackathon
poetry install
poetry shell
python evaluation_script.py

Please guarantee success of the procedure by preparing your submission correctly. Disfunctional submissions will not be evaluated.

Good luck!