I am using the GARCH-DCC estimation-based bootstrap to predict one-month (22 days) cumulative returns. The details are as follows:
1. The GARCH-DCC model is fitted to the original return series to obtain estimated parameters and residuals.
2. Residuals are standardized by conditional standard deviations.
3. The standardized residuals are resampled with replacement to form pseudo samples of GARCH-DCC innovations.
4. Use the pseudo samples as inputs of the DCC and GARCH filters to obtain pseudo samples of GARCH-DCC log returns from period T+1 to period T+22.
5. Compute the cumulative arithmetic return for each pseudo sample.
My question is how to deal with the bad draws in cumulative returns? Can I manually delete some extreme values contained in standardized residuals before bootstrapping? I have checked the data I used in the estimation, some big financial firms do experience significant changes in their returns within a short period. I guess this might be the reason for having some extreme values in standardized residuals.
Many thanks in advance.
A filtering method proposed by Laurent et al. (2016) might work for your case. According to this approach you first need to detect jumps in financial returns or disturbances using a semi-parametric test for additive jumps, and then estimate your DCC model using the filtered series. Literature shows that the jump-filtered returns or disturbances also display excess skewness, kurtosis and conditional heteroscedasticity.
See:
Laurent, S., Lecourt, C., Palm, F.C., 2016. Testing for jumps in GARCH models, a robust approach. Comput. Stat. Data Anal. 100, 383-400.
Dear Akram Shavkatovich Hasanov, many thanks for the reference paper, I will look into it.
Hi Weidong Lin , see this article. It addresses exactly the problem you are facing.
Trucíos, C., Hotta, L. K., & Ruiz, E. (2018). Robust bootstrap densities for dynamic conditional correlations: implications for portfolio selection and value-at-risk. Journal of Statistical Computation and Simulation, 88(10), 1976-2000.
Very interesting paper :
Trucíos, C., Hotta, L. K., & Ruiz, E. (2018). Robust bootstrap densities for dynamic conditional correlations: implications for portfolio selection and value-at-risk. Journal of Statistical Computation and Simulation, 88(10), 1976-2000. … Read more
Recommend Share
- Badges
- Science topic
- Similar topics
- Mathematics
- Applied Mathematics