Timeline for Michaud's Resampled Efficient Frontier - Out of Sample Simulation Testing
Current License: CC BY-SA 3.0
15 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Oct 4, 2012 at 8:33 | vote | accept | CommunityBot | ||
| Oct 4, 2012 at 8:33 | history | bounty ended | CommunityBot | ||
| Oct 2, 2012 at 10:57 | comment | added | user2921 | That's what I thought. Thank you for clearing everything up. | |
| Oct 2, 2012 at 10:43 | comment | added | vanguard2k | OK now I know what you mean (sorry about the misunderstanding). Intuitively, you would score the weights to get the objectively scored portfolios on the frontier. Then you would average all these frontiers. Look at the descriptions of the charts (also in the paper). They underline this: "The bottom solid curves in Exhibit 6.3 display the average of the true, out-of-sample, risks and returns of the optimized portfolios." and not the scored average weights. | |
| Oct 2, 2012 at 10:29 | comment | added | user2921 | I'm talking about the OOS simulation study. After $N$ simulations in the OOS study (i.e. I have $N$ MVO frontiers and $N$ REF frontiers), am I averaging the $N$ MVO weights and averaging the $N$ REF weights, or am I averaging $N$ $\sigma$'s and $E[r]$'s for MVO and REF? | |
| Oct 2, 2012 at 10:25 | comment | added | vanguard2k | @Harokitty Of course there is a difference between averaging the weights and the scoring results. (as you mentioned volatility depends on it in a nonlinear fashion) Look at page 10 and Step 4 in the Michaud^2 Paper you cited. It says to get the RE optimal portfolio you should average the weights. To generate the comparison chart above you should average the efficient frontiers (which, in the case of the REF - which consists of already scored averaged portfolio weights). | |
| Oct 2, 2012 at 9:27 | comment | added | user2921 | Can you answer this question? If not I'll just transfer the 207 points now in approximately 19 hours when I'm able to. | |
| Oct 1, 2012 at 15:00 | comment | added | user2921 |
I'll report back later. However, in the explanation Michaud says In each simulation ... score the portfolio. Why would we score each portfolio in each simulation if we're just going to average the weights then score the mean weights vector after all the simulations??
|
|
| Oct 1, 2012 at 14:45 | comment | added | vanguard2k | You should average the weights for each point of the efficient frontiers and then score the portfolios with the population parameters $\mu$ and $\Omega$. Just think of how the "average portfolio estimated with method $x$" performs compared to the "average portfolio estimated with method $y$" which one is better under the assumption of parameters $\mu$ and $\Omega$? | |
| Oct 1, 2012 at 12:53 | comment | added | user2921 | Ok I've concluded that it does matter when you average the weights before scoring or just average the $\sigma$ and $E[r]$. It doesn't matter for the case of finding the return but for finding variance it does. | |
| Oct 1, 2012 at 11:51 | history | edited | vanguard2k | CC BY-SA 3.0 |
edited body
|
| Oct 1, 2012 at 11:39 | comment | added | user2921 |
Nice one! I have a query about your response to 6). When you (and the Michauds) say "average", do you mean that I average the weights $\textrm{AND THEN}$ score, or do you mean I score the $N$ portfolios and then average each $\sigma$ and each $\mathbb{E}r$?
|
|
| Oct 1, 2012 at 11:39 | history | edited | vanguard2k | CC BY-SA 3.0 |
added 1 characters in body
|
| Oct 1, 2012 at 11:28 | history | edited | chrisaycock | CC BY-SA 3.0 |
Removed signature, as per rules in the FAQ
|
| Oct 1, 2012 at 11:25 | history | answered | vanguard2k | CC BY-SA 3.0 |