Aggregating Security-Level OAS into a Portfolio Spread
Summary
The document asks how to estimate the option-adjusted spread (OAS) of a portfolio when only the OAS of each security can be calculated. It identifies several possible weighting bases, including balance, duration, and dollar duration, but does not establish a preferred method or compare their results.
The central issue is that averaging individual-security OAS values may lose information about the portfolio’s joint cash flows and embedded options. The author suggests that valuing the entire portfolio in a Monte Carlo framework would be the more direct way to obtain a portfolio OAS, and flags model risk as a concern. No numerical example, empirical evidence, industry convention, or research findings are provided, so the document leaves the question open. It is useful as a framing of the aggregation problem, but not as implementation guidance; a suitable method would depend on the portfolio’s structure and valuation model.
Key ideas
- A portfolio OAS may differ from a weighted average of its securities’ OAS values.
- Possible weighting bases include balance, duration, and dollar duration.
- Aggregating security-level spreads can discard information and introduce model risk.
- Joint portfolio valuation using Monte Carlo is proposed as a more direct approach, but no method is specified.
Tags
Full text
# "Correct" way to average OAS of multiple securities? # "Correct" way to average OAS of multiple securities? Suppose one wants to compute an OAS on a portfolio of securities, but one can only compute the OAS of the individual securities. Is there a "best" way (under some metric) for one to go about doing this? One could simply take a weighted average by balance, duration, dollar duration, or something else. I get the feeling that no matter what averaging technique is used, some information would be lost and there would be some inevitable model risk involved. Ideally one could run a Monte Carlo simulation on the entire portfolio at once to compute the portfolio OAS - this would be the "correct" OAS. In sum, I suppose I have the following question(s): - Is there any industry standard or "best" way to combine OASs? - Is there any research out there in attempting to quantify the model risk involved in doing such a thing?
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.