I have some work to do on the drivers of government bond spreads - ie. across terms (not across governments) of the yield curve, say 5yr and 20yr bond spreads from the same government issuer - and am having a bit of conceptual difficulty with this.

I have done some reading and get a sense that potential drivers could be: credit risk (measured by Credit default swap (CDS) rates); liquidity risk (measured by bid-ask apreads); risk aversion; crisis-period variable (a indicator variable assigned a value 1 during the latest financial crisis);

Should I just take the difference of these variables (and assume the differencing renders the variables stationary) and then run OLS regression of the bond spreads on the drivers?

Are there better ways to go about this? Would PCA have any application here?

Thanks in advance for any advice


2 Answers 2


A good piece of literature on this is Colin-Dufresne et al. (2001), "The Determinants of Credit Spread Changes", Journal of Finance, 56, 2177-2207.

I think differencing all variables is a good idea in this case. Not only because of stationarity concerns but also because of unobserved time-invariant issuer-level characteristics.

I do not see the case for a PCA here. Usually you use PCA to reduce dimensionality of sometimes highly correlated variables. Here you have variables at hand that seem to capture different risk premia.


The list of risk drivers that you provide are good. I would add significant macroeconomic factors:

  • risk aversion indicator
  • growth and inflation
  • => rate expectations
  • money supply
  • QE

Also add more curve factors:

  • duration or term premia i.e. 10y term premium ( ... )
  • cross-market curves
  • swap spreads

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.