- UID
- 223458
- 帖子
- 310
- 主题
- 9
- 注册时间
- 2011-7-11
- 最后登录
- 2014-8-7
|
5#
发表于 2011-10-3 03:36
| 只看该作者
I haven't looked at the link, but my guess is that you use a regression that predicts yields given whatever risk characteristics seem relevant or measurable (duration, industry, financial ratios, collateralization features, embedded options, etc.) and then look at the residual (actual yield - regression predicted yield). Those with a positive residual are cheap, those with a negative residual are rich (low yield -> high price).
Z score might come in when trying to choose the cheap and rich bonds. standardize the residuals (i.e. subtract the mean (0 in this case) and divide by the standard deviation), and then buy the ones that have a standardized residual greater than, say 1.5 and sell the ones that have a standardized residual less than -1.5.
A lot of people might try to buy all bonds that have a standardized residual > 2 and sell the ones that have a s.resid < -2, which would buy the 2.5% cheapest bonds and sell the 2.5% richest bonds, but my guess is that that is a bad strategy, because the very extreme bonds are likely to have reasons that they are so far out of line. For a long/short strategy, I think you might want to buy things that have s.resid > 1 and < 1.7 (chosen subjectively) and sell things that have s.resid < -1 and > -1.7. These are more likely to be mispriced due to mean reverting factors, and those that aren't can be diversified away (or eliminated later if the residual gets too extreme). |
|