Hello everyone. I am currently working on implementing quantile connectedness analysis for my data, utilizing monthly frequency. However, I am facing challenges in selecting an appropriate rolling window value (10 , 30 , 40 or 50). I would greatly appreciate any insights or recommendations regarding the choice of rolling window values, particularly when working with different data frequencies (weekly, monthly, and annual). ?
Thank you in advance for your assistance?
Steps:
1. Data Frequency: The choice of rolling window value should align with the frequency of your data. Since you mentioned that you are working with monthly frequency data, it's generally recommended to select a rolling window value that represents a reasonable time span but is not too short. For monthly data, values like 10, 20, 30, or 40 can be considered.
2. Sample Size: The number of observations in your dataset is another important factor. With a limited number of observations, it's advisable to choose a smaller rolling window value to avoid excessive data loss. If you have 322 observations, a rolling window value of 10 or 20 might be more appropriate to ensure an adequate sample size within each window.
3. Trade-Off: The choice of rolling window value involves a trade-off between capturing short-term dynamics and providing enough data points for robust estimation. Smaller rolling window values capture short-term changes more effectively, but they may result in higher noise and instability in the estimation. On the other hand, larger rolling window values provide a smoother estimate but may overlook shorter-term dynamics.
4. Sensitivity Analysis: It's a good practice to conduct a sensitivity analysis by trying different rolling window values and examining the stability and consistency of results across different window lengths. This can help you identify the optimal window size that provides meaningful insights without sacrificing stability.
5. Cross-Validation: If you have a sufficient number of observations, you can consider using cross-validation techniques to evaluate the performance of different rolling window values. Split your dataset into multiple folds, estimate the quantile connectedness using different window lengths on each fold, and assess the stability and predictive power of the results across the folds.
Hope it helps
Mohammad Imam Thank you, Professor Mohammad, for your detailed response. It's very helpful! I'm also exploring weekly data with 250 observations and was wondering about the optimal rolling window value in this context. Any insights you could share would be greatly appreciated!
see the link
https://www.sciencedirect.com/science/article/abs/pii/S0378437121004003
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts