Data Transformation for regression

[zobee]

Hi, im doing this regression which involves inflation rates, interest rate and stock index. im having a problem regarding how to transform my data for the regression. like for inflation rate im using consumer price index on monthly basis, so how do i transform my data i.e. either to use monthly percentage change i.e. (x2-x1)/x1 (which i believe is inflation rate)or just simply x2-x1 (change in index value) or log(x2-x1) (x2 is current month index value and x1 is previous month index value). similarly for interest rate i have 3-month tbill interest rates, do i input values into variables as it is i.e. percentage values, or take log values of differenced value or simple monthly change (differenced values)? and if i want to use interest rate differential (subtracting 3-month tbill and 10yr bond interest rates) how do i transform these values so the series are stationary? im really confused!
quick help will be highly appreciated. thanks

[trubador]

Percentage change and log-difference will produce similar results as long as the change is small (i.e. r→0 => log(1+r)≈r). Although I do not know what exactly your model and study are about, I think it would be sufficient for you to use percentage change of the index values of stock prices and consumer prices while leaving others as is (unless they are significantly non-stationary).

[zobee]

thanks for your reply, it was really helpful in understanding. i calculated the percentage change for indices......for stock index, the percentage change is good, as it is stationary series, but for cpi percentage change, it has unit roots for all except with 'none' at 5% but becomes stationary when i do ADF test with first difference.
also, for interest rate, series are non-stationary, but when i take difference they are stationary. do it make any sense to use difference of monthly interest rate? also with diferential series, they are non-stationary, values are actually not rejected i.e. have a unit root but pretty close to be rejected.
basically i'm doing granger causality test for stock index and house price index where inflation, short-term interest rate and interest differential are exogenous variables. im really thankful for your help...

[trubador]

If the non-stationarity is due to a structural break, then you can remove it via including a dummy variable into your equation. Otherwise, you should try taking the first difference of the non-stationary variables. Working with difference series will change the interpretation of your results, and it will not be a problem as long as you are aware of it.

[zobee]

yup.....there arent any structural break so i think i'll go with the differenced values.....thank you so much for help. now, i have got much better understanding of it!

[joebrandom]

Hello all, ive been reading this conversation and find it very helpful however I was hoping for some clarification.

What are/is the implications of differenced and 2nd differenced data to induce stationary data series?

How do I interpret differenced and 2nd differenced data?

Is it ok to use the 1st difference of variables such as fertility rate if percentage growth of a population?

I would really appreciate any help on the matter

kind regards

joe

[spycy]

Percentage change and log-difference will produce similar results as long as the change is small (i.e. r→0 => log(1+r)≈r). Although I do not know what exactly your model and study are about, I think it would be sufficient for you to use percentage change of the index values of stock prices and consumer prices while leaving others as is (unless they are significantly non-stationary).

Hi Trubador.

Is it same interpretation when use Percentage change and log-difference or not ?

for example :

log difference koefisien 0.04 its mean 4 persen
Percentage change 0.04 its mean 0.04 persen

which one ?

thank you

[waqaskashif]

if data is monthly of crude oil price and yearly percentage change(another name as growth rate) is use for making data stationary so this technique could be utilize or not for further analysis?