Dear all, I want to assess the effects of financial instability on growth. My model is based on Beck and Levine (2004). But when I run it I cannot reject the null hypothesis of AR1 that there is no autocorrelation (My data are in log form).
Are my results relevant?
When I use data without transformation into log form, my results are all good.
What Is it better to do?
Thank you!
Data is transformed into log form to remove the base effect that may exist while studying returns on share prices, etc. which may lead to deceptive results due to the continuously shifting base. If such is not the case & you do not have a constantly shifting base, I don't see the logic of using log data. Autocorrelation can influence results as it shows that later data is a lagged version of the earlier time-series data. Correctness of the results are an issue if there is doubt about independence of data. Hope it helps you.
My dependant variable is growth rate of real GDP per capita not in log.
My explanatory variables are all in log form, and are: initial level of GDP per capita, inflation rate, openness rate, government consumption expenditures (over GDP), average years of schooling, private credit to GDP ratio and an index of financial stability I have constructed. All
I think you should revisit the reasons why you transformed the data into logarithms. Data in logs is usually meant to observe volatile relations of the time series especially when plotting. If not not, then why then did you transform your data?
I hope this helps Dr Mbassi
I think the way you transformed your data is the problem. Or the problem may emerge if your time series is long.
Hi Mbassi,
Could you write the GMM commands you used to estimate your model? That will help us assist you to identify any problem.
Hello, Christophe. You could try to introduce an AR(1) element in your model, in order to deal with autocorrelation.
- Badges
- Science topic