Regressing changes in yield/yield curve

If I'm regressing changes in individual points along a yield curve and measures of changes in level/slope/curvature of that yield curve against the returns of some random variable then do I want to use % changes in yields or absolute changes in yields (eg yield change from 1% to 0.95% would be -0.05)?

I am new to fixed income and my experience with volatility tells me that I should use absolute changes since it is already measured in %. For example, the R^2 is much higher when regressing 1-month vix futures changes in points against spx returns than % change against spx returns. However, since I'm new to fixed income I'm running myself in circles (because I don't know what I don't know).

Thanks in advance for any help!

• What I usually see (and do) is working with absolute changes, i.e. $r(t)=R(t)-R(t-1)$. This depends on the rates regime as well, of course. I have seen insurance companies (historically) model their rates with lognormals (r(t)=\ln(R(t)/R(t-1))\$ until 2011/2021, then with a shifted lognormal model until 2018+, and with normals (your suggestion) now.... Feb 11 '21 at 8:12
• There's an old RBS piece by Rebonato from 2010, "Dependence of Magnitude of Rate Moves on Rates Levels" which is relevant to your question. At the present time, I would use differences as described by @Kermittfrog. Feb 11 '21 at 14:02
• You may find my answers here and here relevant. Mar 15 '21 at 1:09