Time Series Reversibility | Algorithmic Trading Indicators in Python

Summary

The video examines methods for measuring time irreversibility in financial time series using Bitcoin price data. It covers the relative asynchronous index, which compares visibility graphs of a time series played forwards and backwards, and an ordinal pattern method that compares permutation pattern distributions. These measures attempt to identify chaotic or nonlinear behavior and could help design trading systems. The presenter finds the measures useful for filtering mean reversion trades and thinks they may benefit machine learning strategies since they have little correlation with common market indicators.

Takeaways

• 😊 The video discusses using time reversibility as an indicator for financial time series analysis
• 🔄 A time series is reversible if its statistical properties stay the same when reversed
• 📈 Two methods are introduced to measure time irreversibility: the relative asynchronous index and permutation pattern comparison
• 📊 The relative asynchronous index uses visibility graphs and measures differences in link counts between a time series and its reverse
• 👀 The permutation pattern method compares probability distributions of patterns in the original and reversed series
• 📉 Bitcoin price data is analyzed using these irreversibility measures, with more irreversibility observed during volatile periods
• 🌡️ When calculated on hourly data, these indicators may help filter mean reversion trades
• 💡 The measures introduced could be useful features for machine learning models in trading systems
• 📜 A survey paper of different irreversibility measures for time series analysis is referenced for further reading
• 🎥 Overall, the video demonstrates quantitative techniques to detect chaotic dynamics in financial data

Q & A

What is time reversibility in the context of financial time series?

Time reversibility refers to whether the statistical properties of a financial time series are the same when the series is reversed in time. A reversible series will have the same properties and dynamics whether analyzed forward or backward.

What is the relative asynchronous index and how is it calculated?

The relative asynchronous index is a measure of time irreversibility in a time series. It is calculated by comparing the number of outgoing links in the horizontal visibility graph of the original series versus the time-reversed series.

How can irreversibility measures be useful for trading strategies?

Irreversibility measures can help identify periods where a financial time series displays more chaotic or nonlinear behavior. Some trading strategies may perform better or worse during high or low irreversibility regimes.

What is the ordinal pattern reversibility measure and how does it work?

This measure computes and compares the probability distributions of ordinal patterns in the original and time-reversed series. The callback liebler divergence between the two distributions quantifies the irreversibility.

Why use an embedding dimension of 3 for ordinal patterns?

An embedding dimension of 3 is used because higher dimensions would require an extremely long lookback window. 3 allows calculating irreversibility in a reasonable window while capturing some nonlinear dynamics.

How can these indicators be used in trading systems?

They may be useful features for machine learning models to improve prediction. They could also help filter mean reversion or trend following strategies during certain irreversibility regimes.

What causes the relative asynchronous index to have high noise?

The complex dynamics of financial data combined with estimating irreversibility in a finite rolling window introduces noise. Smoothing can help reduce this.

Why does the ordinal pattern measure need long windows?

It needs enough data to accurately estimate the probability of all patterns occurring at least once. Too little data leads to zeros in the distributions which break the calculation.

Do the two irreversibility measures correlate?

Yes, they tend to be positively correlated as they aim to quantify the same time asymmetry property. But they have some differences coming from their distinct approaches.

Are there other ways to measure time irreversibility?

Yes, there are many other proposed irreversibility measures as surveyed in the paper referenced at the end. The two covered here are not at all the only options.