Principal Components of the RSI | Machine Learning Trading Strategy in Python

Summary

This video explores a method to extract valuable insights from the Relative Strength Index (RSI), a popular technical indicator. It delves into the RSI's various uses and periods, demonstrating its correlation across different timeframes. The script employs principal components analysis to reduce the high dimensionality and overlapping information, preserving the most variance. A predictive model is then built using the principal components to forecast market movements, with an emphasis on the predictive power of extreme values. The model's performance is tested with a walk-forward approach, examining its robustness across various parameters, and suggesting its potential for trading strategies.

Takeaways

• 📈 The video demonstrates a powerful technique to extract important information from the Relative Strength Index (RSI).
• 📊 The RSI is a popular technical indicator used by traders in various ways, including short periods for small moves and longer periods for spotting divergences and overbought/oversold conditions.
• 🔄 The script visualizes the behavior of the RSI across many period values using hourly Bitcoin data.
• 🔍 Most RSI period values exhibit a bell-shaped distribution, with the 2-period RSI being an exception.
• 🔗 There is a strong correlation between RSIs of different periods, particularly between the 2-period and 24-period RSIs.
• 🧮 Principal Component Analysis (PCA) is used to reduce the dimensionality of the RSI data while preserving maximum information.
• 📐 The video explains how to compute principal components using the RSI data and apply them to build a predictive model.
• 📉 The model uses a least squares linear regression to predict market movements, focusing on extreme values for predictive power.
• 🚀 The out-of-sample model performance is tested using a walk-forward method to simulate real-time trading scenarios.
• 🔧 The robustness of the model is evaluated across various parameters, showing that fewer principal components and shorter look-aheads generally perform better.
• 💡 The technique of computing principal components can be applied to other correlated indicators, enhancing trading strategies.

Q & A

What is the main focus of the video?

The video focuses on demonstrating a technique to extract important information from an indicator using the Relative Strength Index (RSI) as an example and building a trading strategy with a predictive model.

What is the RSI and why is it significant in the video?

The RSI is a well-known technical indicator that measures the speed and change of price movements. It is significant in the video as it serves as the basis for the technique discussed to extract information and build a trading strategy.

What is the 'period' in the context of RSI and how is it used in the video?

The 'period' is a parameter for the RSI that determines the time frame it covers. In the video, different period values are used to analyze the behavior of the RSI and to identify patterns.

Why might different traders use different RSI periods?

Different traders might use different RSI periods to suit their trading strategies. Some may use shorter periods for scalping small moves, while others use longer periods to spot divergences or identify overbought and oversold conditions.

What is the purpose of using principal components analysis (PCA) in the video?

PCA is used to reduce the dimensionality of the 23 different RSI periods while preserving the maximum amount of information, making it easier to analyze and build a predictive model.

How does the video script visualize the behavior of the RSI across different period values?

The script visualizes the behavior by plotting histograms for each RSI period and showing the correlation between different RSI values, demonstrating how they move together and the degree of their correlation.

What is the target variable used in the predictive model in the video?

The target variable is the change in the log closing price over the next six candles, which the model predicts based on the principal components derived from the RSI periods.

How does the video script handle the high correlation between different RSI periods?

The script acknowledges the high correlation and uses PCA to extract principal components that capture the most variance, effectively summarizing the overlapping information.

What is the walk-forward model mentioned in the video, and what is its purpose?

The walk-forward model is a method used to test the predictive model out of sample, ensuring that it works effectively with new, unseen data and is not just overfitting the in-sample data.

How does the video script determine the number of principal components to use in the model?

The script initially uses an arbitrary number of components, such as four, but later adjusts this number based on the performance of the model, finding that two or three components often yield the best results.

What does the video suggest about the robustness of the model across different parameter combinations?

The video suggests that the model appears to be fairly robust across a wide range of parameters, with the best performance observed when using a look-ahead of one candle and two or three principal components.

How can the technique demonstrated in the video be applied to other indicators?

The technique of computing principal components of highly correlated indicator series can be applied to other indicators with a lot of overlap, such as the Average Directional Index (ADX) or volatility indicators, to summarize information and reduce dimensionality.