Backtesting-Trading-Strategy

---

This project is an environment for backtesting a trading strategy based on crypto-currency market. The aim is to have a customizable environment where testing can be done quickly and easily for identifying a profitable strategy.

It does not provide any ways for collecting data, does not use Twitter data or any textual data and it is not a trading bot either.

Strategy

---

Description

The strategy use 2 assets :

• a stable coin called SC (for example : USDT, BUSD, USDC, ...)
• an unstable crypto-currency called U (for example : BTC, ETH...).

And at each period, predict an action depending on the last position :

• if long U at last_position:
  • if beneficial:
    • do nothing
  • else:
    • short U against SC
• else:
  • if beneficial:
    • short SC against U
  • else:
    • do nothing

NB : all the amount of SC or U is trade if the position change.

What beneficial means ?

Due to transaction fees, the notion of beneficial cannot mean only next expected return of U > 0 or < 0, so it depends to rate of transaction fees.

First, I compute for each period :

• optimal actions if we long on U (label U).
• optimal actions if we long SC (label SC).

Then, I fit a binary classifier algorithm (like logistic regression for example) on label U and an another on label SC. These fitted classifier will predict probability for each period.

Finally, probabilities are used with different aggregation rules for defining actions.

Installation

---

The easiest way is to use Anaconda3 or Miniconda3 on Linux distribution.

You can download Anaconda3 at this link : https://www.anaconda.com/products/individual#Downloads or Miniconda3 at this link : https://docs.conda.io/en/latest/miniconda.html. (recommend python3.8).

Next,

• Download or git clone project.

• On script located in bin/install_with_conda.sh : modify the PYTHONPATH with your root of project according to your file system and save.

• launch with terminal :

   bin/install_with_conda.sh
  
   conda activate backtesting

Usage

---

1 - Collect data:

Use this project : https://github.com/binance/binance-public-data for catching klines Binance historical data and save them into data folder in the root of this project.

NB : if you collect data by any other means you must insert your data in the database by your own. Keep in mind that the SQL table have this structure:

market	frequency	open_time	open	high	low	close	volume	close_time	quote_asset_volume	number_of_trades	taker_buy_base_asset_volume	taker_buy_quote_asset_volume	ignore

2 - Insert collected data into database:

In conf/database.conf, modify parameters 'frequency', 'market' and 'path_historical_klines' according to your collected data.

launch :

python src/database/add_binance_vision_zip_to_db.py

This script will open zip files and insert the data in 'path_historical_klines' to SQL database.

3 - Preprocessing

In conf/preprocessing.conf, modify parameters like you want.

launch:

python src/preprocessing/preprocessing.py

This script will do :

• request SQL database according to parameters specified in preprocessing.conf.

• load data.

• compute features.

• compute labels (see src/preprocessing/cl_labelling.py for more details).

• save all these data in 'save_path' into csv format.

NB : if frequency is different than 1h or 1min, create your own feature_engineering_frequency function in src/preprocessing.py. but column "rdt_crypto_frequency" is required.

4 - Machine learning

In conf/predict.conf, modify parameters like you want.

launch:

python src/ml/generate_predictions.py

This script will fit or refit many times 2 binary classifier on last data and save predicted probabilities into csv format.

NB : model and theirs parameters must be change in src/ml/generate_predictions.py directly.

5 - Find strategy

In conf/strategy.conf, modify parameters like you want.

launch:

python src/strategy/find_and_evaluate_strategy.py

This script will do :

• open .csv of predicted probabilities for both models.
• decide actions according to simulation parameters.
• find the best simulation.
• evaluate the performance of the best simulation.
• generate a customizable evaluation report (.txt) and selected figures.

NB :

• the parameters of simulation must be change in src/strategy/find_and_evaluate_strategy.py directly.
• details of each probabilities transformation rule are provided in src/strategy/cl_strategy_simulator.py.
• it can take long time for doing simulations so I speed up with numba.
• depending on your hardwares and DIC_PARAMS you can met MemoryError.

Troubleshooting

---

please open an issue

License

---

MIT

Donation

---

If you find this project helpful, feel free to make a donation at this address : 0x21eAa2A429735410733320065D800ad14Aa6F5DB