Should I use 2 assets, then use a Kalman filter based on the 2 assets, then use the OU process to get a spread, or use the 2 assets to get the OU spread, then use the Kalman filter on the OU spread?
Hello Emma,
It depends on a variety of factors. But from what I can recall,
The advantage of using Kalman Filter on Two Assets First can help extract a more accurate estimate of the underlying state variables (e.g., mean and volatility) from noisy measurements of the two assets.
It can handle correlations and dependencies between the assets more effectively.
The OU process meanwhile might model the mean-reverting behavior between the two assets directly, which can simplify the modeling process.
It may require fewer parameters and assumptions compared to using a Kalman filter on two separate assets.
You might have to consider using both situation and applying it on unseen data to see which performs better.
Using the OU spread is wild because it changes each time I run a backtest. I think using the parameters sigma, theta, kappa the entry and exit values is sufficient. Is this normal? Or should you use the OU spread and not the Kalman-filtered spread.
Hi,
t's normal for the Ornstein-Uhlenbeck (OU) spread to change with each backtest run, especially if you're using historical data where market conditions can vary widely. The OU process is stochastic, meaning it incorporates randomness into its calculations, which can lead to different results each time you run the analysis.
Using the parameters sigma, theta, and kappa to determine entry and exit values can be a practical approach, especially if you're looking for a more stable and predictable trading strategy. These parameters govern the mean-reverting behavior of the OU process and can help you define when a spread is overbought or oversold, making it suitable for pair trading strategies.
Ultimately it depends on which method is giving you better results and are replicated in a live or paper trading environment.
I deeply appreciate your help.