Timeline for t-statistic varies widely when used for bootstrap sampling process
Current License: CC BY-SA 4.0
8 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jun 3, 2021 at 14:56 | vote | accept | kyc12 | ||
| Jun 2, 2021 at 22:54 | comment | added | BruceET | Yes. Ratio of geometric means. See my answer just posted. | |
| Jun 2, 2021 at 22:53 | answer | added | BruceET | timeline score: 1 | |
| Jun 2, 2021 at 22:30 | comment | added | kyc12 | Do you have any suggestions on which other metric could be used? | |
| Jun 2, 2021 at 21:49 | comment | added | kyc12 | Regarding using B = 100, I can use 1000 but I see a similar behavior. Also, the t-stat jumps in the range of 7-11. | |
| Jun 2, 2021 at 21:46 | comment | added | kyc12 | I'm using geometric means because the samples represent collection of annual returns of a portfolio. For example - returns for each year between 1981-2000. At the end of the day I want to compare and see if annualized return for this period (geometric mean of 20 yearly returns) for two strategies is significantly different or not. No this is not a course assignment. I'm free to chose any method. | |
| Jun 2, 2021 at 21:15 | comment | added | BruceET | (1) $B = 100$ bootstrap re-samples may not be enough for stable bootstrap results. (2) I wonder why you're using t statistic for a bootstrap of geometric means. (3) You don't say anything about your data, but geometric means are only for non-negative data. (4) Partly on theoretical grounds and partly based on experience, I prefer to use bootstraps for CIs, and permutation tests for testing hypotheses. It seems to me you're mainly interested in a test. // Is this a course assignment in which the method is specified, or a practical application in which you're free to choose methods? | |
| Jun 2, 2021 at 20:39 | history | asked | kyc12 | CC BY-SA 4.0 |