Timeline for PCA and risk bucketing
Current License: CC BY-SA 4.0
5 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Nov 11, 2019 at 15:44 | comment | added | tgeorge | Now we get to my confusion: how can I reduce to the tenors I'm interesting?? So far, I have worked out the most granular level I could have i.e 3M->20Y. How can I collapse to the three buckets I want? Or four, or five.... | |
| Nov 11, 2019 at 9:36 | comment | added | demully | I suspect the confusion (at least mine personally) with the outputs above is that you happen to be eg paying 3s and receiving 2s? In which case, the Pv01s you're modelling are already a function of the portfolio's exposures. PCA on these would only really try to describe a pattern to the risks you happened to be running at present. Which is different from modelling the risk of your positions (and thus to the portfolio) of broad-sweep rate and curve shifts. | |
| Nov 11, 2019 at 9:32 | comment | added | demully | Then from the sounds of it, you've already done the hard work here! You have reduced all the dimensions of "interest rate risk" down to the three that really matter to you. You know what +/-1bp does on 10s or 5s10s etc. does to these 3 factors, which are independent so additive. And from these, you can predict what should happen to 1y2y, 3y6y, 8y etc. The portfolio impact is the sum of these times position size and current pv01 (without the need for additional PCA). | |
| Nov 10, 2019 at 17:32 | comment | added | tgeorge | Hi demully I already have an analysis to the three major components (level, slope, curvature). The idea behind is that instead of having remember my pv's to all tenors, what if i could have a three tenor approach (5,7,10 as attack suggests) and could have a quick calc of how a +/- 1 bp shock would affect my total pv. | |
| Nov 7, 2019 at 18:02 | history | answered | demully | CC BY-SA 4.0 |