# Time Series - Using log or not?

I am testing some time series on Eviews, which are 10-year treasury constant maturity rate and 3-month treasury bill. I need to test for the order of integration of the former, and the cointegration between them. But I'm not sure whether to transform them into logarithm form or not. I read somewhere that if it's interest rate then we should leave it as it is. Is this correct?

• You're correct in what you've read. It's usual not to takes logs of interest rates. On the other hand, it's usual to take logs of economic time-series such as output, consumption, etc. One reason for these choices is interpretation. Usually, one will be interested in percentage changes of economic time-series, but, in terms of interest rates, one is interested in percentage point changes. If you estimate a few different models and try to interpret the coefficients, or do shock analysis, then you will see this in action. – Graeme Walsh Jan 7 '18 at 14:15
• From a purely time-series perspective (as opposed to an economist perspective), taking logs can help to normalize residual autocorrelation, but that seems to be a separate issue here. In this case, where you just want to build an adequate model, say, for forecasting, and less for interpretation purposes, I don't see anything wrong with log transformations - or any Box-Cox transformation. Hopefully someone can provide a full answer. Cheers. – Graeme Walsh Jan 7 '18 at 14:19