I am doing a time-series data analysis. The idea is to produce a forecast from the regression output. I am regressing Air traffic passengers of country A with GDP/capita of country A.

I am getting very high R² (0,97) by regressing absolute values of Air traffic passengers of country A with GDP/capita of country A (p.value<0,05)

I am getting low R²(0,35) by regressing the same variables but that time, by taking the %growth of those variables. (p.value<0,05)

What is the most accurate in this case? regress absolute values or regress the %growth?

  • $\begingroup$ You are right, absolute values is not precise enough and can have other meanings. When I said "absolute" I meant the real value of the gdp/capita and the real number of passengers (in units) $\endgroup$ – IlvarNourtan Jul 30 at 10:29

Use the observed values as there may be lag structure needed for GDP (X) that may be different than the lag structure need for air traffic passenger data (Y) . You can always subsequently express the change in passenger forecast as a percentage of the previous value as it relates to the change in GDP as it relates to the previous value of GDP .

Finally to get correct/honest prediction limits for passenger data (Y) you may/will need to account for the uncertainty in GDP (X) predictions and possible re-occurring anomalies.

You might want to read https://autobox.com/pdfs/regvsbox-old.pdf which discusses some of the opportunities that are available when you have time series data and some of the pitfalls when you use simple OLS.

I surmise you ran a simple OLS of Y versus X and obtained .97 as the Rsq AND then you created two new series Y1 and X1 where Y1=[Y(t)-Y(t-1)]/Y(t-1) and X1=[X(t)-X(t-1)]/X(t-1) and obtained .35 when you regressed Y1 and X1 . This specification is probably suspect for a ton of reasons whereas a SARMAX https://autobox.com/pdfs/SARMAX.pdf is much more general approach. Your specification of the model (two ratios) is a possible model but not probable because it is based upon presumption.

  • $\begingroup$ Many thanks for this IrishStat. though, I don't undersand why use absolute values and not %growth would allow a lag between passengers and GDP/capita? $\endgroup$ – IlvarNourtan Jul 30 at 10:08
  • 1
    $\begingroup$ It could BUT it is always IMHO much cleaner to use observed data rather than derived data (secondary data) due to possible obfuscation (KISS). $\endgroup$ – IrishStat Jul 30 at 10:14
  • $\begingroup$ Noted, never heard about the obfuscation. Thank you again for that. $\endgroup$ – IlvarNourtan Jul 30 at 10:21
  • $\begingroup$ If you are happy with my response please accept it to close the question $\endgroup$ – IrishStat Aug 15 at 8:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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