Trading rules using machine learning

by jojothepizza

Version 1.6, 1.7

• This is my financial trading system using ML.

• I'm still working on this project. If you're interested, feel free to contact me.

Experiments

• This is the results of my trading strategies (there's full code in Notebooks).
• The trading rule is based on Triple-Barrier Method introduced in Lopez De Prado (2018).

Data

• BTC-USD (30 minutes data)

First model 1 (get trend signals and position entries)

SMA

• long when 10-days-SMA > 30-days-SMA

Force Index

• long when FI>0

RSI

• long when RSI>50

CUSUM Filter

• detect structural change using cumsum of volatility
• filter out entry points

Combined filter

First model 2 (set up target rates of each position and exit of it)

• Triple-Barrier method

Results of the first model

Second model

• It is a model that predicts whether the outcome of each bet is profit or loss.
• It corrects bets of the first trading strategy.

Features

• Technical indicators and price history
• Use MinMaxizer scaler and PCA

Results of second model (Random forest classification)

Bet sizing

• Decide how much to buy
• Bet sizing based on Sharpe ratio & Gaussian distribution (Lopez de Prado, 2018)
• Use probabilities of Random forest classification predicted probabilities
• Scale with minmaxizer

Results

• Here are Annualized Sharpe Ratio and Cumulative Returns of three strategies (only long strategy which excludes short selling):

1. First trading Strategy: This is a primary model using only technical analysis. (predict when to buy)
2. Second Meta-Label: This is a secondary model using ML algorithm on the trading strategy. (predict whether its bet is profit or loss/ correct bets of the trading strategy) /I used Random Forest and LSTM here.
3. Bet Sizing: This is a sizing model using predict probabilities of ML algorithm used in Meta-Labeling ()
4. Buy and hold: Buy-and-hold for an entire period.

And those strategies were introduced in Lopez de Prado, Advances in Financial Machine Learning.

Other assets results

Samsung Electronics (Daily data)

SK Hynix (Daily data)

LG Chemicals (Daily data)

Hyundai Cars (Daily data)

Naver (Daily data)

With LSTM

BTC-USD (30 minutes data)

Samsung Electronics (Daily data)

References:

• Advances in Financial Machine Learning, Lopez de Prado (2018)
• mlfinlab, https://github.com/hudson-and-thames/mlfinlab
• ta, https://github.com/bukosabino/ta