I have an ARDL bound test model that consist of the following variables: CPI, private credit as a percentage of GDP, broad money M2 as a percentage of GDP, real GDP, and trade openness index. My reference put each of those variables in their ln form to avoid heteroscedasticity (attached). My professor (being not so sure about what he said) told me that we cannot put ratios/percentage in their ln form, while also asking me to look it up. Any suggestions?
Please visit: https://www.researchgate.net/post/Log_transformation_of_variables_in_Rates_or_percentage
You may consider this link useful: https://stats.stackexchange.com/questions/341365/log-transformation-interpretation-ardl-model
It is always better to consider the variables in ratios as it is without transforming in logarithmic values as logarithmic values would give just difference the variables in ratios and changing the results in different version where one has to be very careful in interpreting the results.
However, in worst cases one can consider the variables in the natural values of ratio variables. I think it is high time that models and packages should be built to take care second order difference stationary in cointegrating procedures.
@Hrushikesh Mallick I am bit confuse about whether we should take log of a ratio variable or not as on the one side researchers argue that ratios are generally asymmetric in nature and taking log will help in conducting linear regression. However , as per your suggestion taking log ( a/a+b) can change the interpretation. It means log a- log (a+b).
In my case my dependent variable is in the ratio form (a/a+b), standard deviation =0.39 and there is no skewness. Can you suggest, whether we should take log or not for the dependent Variable.
- Badges
- Science topic
- Similar topics
- Philosophy
- Esthetics
- Interpretation