Machine Learning - We use the raw commodity price data (open/high/low/close) plus technical indicators such as Moving Average, Bollinger Bands, RSI, and more, to create an AI model that generates buy and sell predictions. The other strategies on this page then try to improve on this basic model.
X In A Row - Only enter the trade when the commodity goes up or down X number of bars in a row.
Big Move - Only enter the trade when the commodity has made a big recent up or down move.
Gaps - Trade based on the size and direction of the overnight gap (how much the price went up or down).
Price Bar Length - What time period works best for the commodity quotes? Should we trade every minute? Every hour? Every day?
Pair Trading - Out of a basket of commodities, the AI predicts the best and worst one. We then buy the best commodity and sell short the worst commodity. This helps hedge against the randomness of the overall market move.
Time of Day - Only enter the trade during times when the backtests show the AI model is highly accurate. For example, should we only trade during the opening hour of the day when the market is most volatile?
Day of Week - Only enter the trade on days when the AI model is most profitable.
Feature Matrix - We try a strategy where we only trade when the technical indicators are at unusually high or low level levels.
Data Cleaning - We experiment with different ways of handling the commodity price data, such as how to remove outliers, how to handle missing prices, and how to deal with imbalanced data.
Social Media Signals - We create an indicator derived from postings about a commodity on Twitter, Facebook, Instagram, and TikTok, using a machine learning technique called sentiment analysis to decide if each posting is good, bad, or neutral.
Multiclass - Instead of only predicting Up and Down, we add a 3rd category named "No Move", where if no significant price move is predicted, we don't trade.
Regression - Instead of generally predicting Up or Down move, it predicts an exact price. This is similar for example to how Zillow predicts house prices.
Time Series - Instead of training on each price bar individually, it looks at the stock price data sequentially over time. Typical methods include RNN, LSTM, and GRU.
Reinforcement Learning (RL) - Turns commodity trading into a game, where the AI wins if it makes trades that are profitable. It learns to buy and sell whenever it thinks is best, just like a human would.
Evolutionary Algorithms - Inspired by Darwin's theories about nature and humans, it mimics biological evolution. A random group of possible solutions is initially created and with each new generation, the best solutions fight for survival until eventually a winner is chosen.
Neural Architecture Search - A technique that improves the design of a neural network by automatically discovering the architecture with the highest accuracy for your specific dataset.
Dynamic Time Warping - A method based on the theory that history repeats itself, where it finds time periods in the historical data similar to the current one, to predict if the same thing will happen again.
Chart Pattern Detection - A more modern version of Dynamic Time Warping where image recognition (computer vision) is used to detect historical chart patterns that are similar to the current one.
Symbolic Regression - A type of genetic programming that tries to find the best mathematical expression to represent the data. The resulting model is a plain mathematical formula, no machine learning program needs to be installed to run it.
AutoML - Dozens of different machine learning methods are tested automatically to see which one obtains the best accuracy.