Hi everyone
I have to regress some covariates on a set o states (0, 1, 2), using a multinomial logit. Assume now that each array of observations (one array for each covariate), is composed by several time series. So each considered country provides a time series for covariate, then the time series are ordinately put inside the same array (accurately checking for each associated date).
Is necessary to standardize data? If yes, we have to standardize the single time series, before building the final arrays, or should we standardize the final array, so the lined up time series?
Dear Matteo,
As I understand the question, you are facing panel data and you want to run a multinomial logit model.
In this case, you can control for the time by introducing time fixed effects into your multinomial logit model this will be good and will control for unobserved heterogeneity maybe you are discussing about it.
for instance:
y = 0, 1, 2 (tranquil times, financial crisis, post-crisis period)
x = CPI, REER, interest rates differential with US
for all the cited variables we have data available for Italy and France, from 1960 to 2000. So the idea is to line up variables, obtaining [1960-2000] y for France, and [1960-2000] y for Italy in the same column. Same process for the other variables, so indeed for each row we have observations for a given country at a given date.
Would be great to look at a table (a part of it) containing these values. At present the question is still not quite clear.
I too am unclear , does your response variable vary over time , or just your predictors?
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series