Simple(r) GDP forecasting for the student?
Posted: Thu May 21, 2015 6:24 am
Hi all!
I am currently doing an intro macroeconometrics class (we use eviews 6) and we were asked to do a mini-project on a econometrics topic of our choosing.
I chose to try and make a forecast model for US quarterly GDP. Obviously being an undergrad my methods are incredibly basic, hence my posting here to seek your kind guidance and teachings!
What I have so far is a linear OLS model--I pulled a bunch of data from FRED (IPI, PCE, PMI, Retail, Sentiment, Claims) and regressed them on GDP. Then I used the static forecast function. It actually fits pretty well to past data, with a decent MAE/MAPE.
Though, the model has a hugee multicollinearity problem (PCE & Retail Sales, amongst others) and all of the variables are probably unit roots/autocorrelated (though I looked at model residuals and they were peachy). However, since I am only interested in a point estimate/forecast do all of these issues actually affect my purpose? I don't even want an efficient interval estimate, I just want a point estimate.
Secondly, I was wondering if there was any other methods of estimation that would be simple enough for me to learn by myself and perform in Eviews so I can improve my forecasts? I read up on alot of journal articles but so far much of them has been lost on me.
Cheers guys!
I am currently doing an intro macroeconometrics class (we use eviews 6) and we were asked to do a mini-project on a econometrics topic of our choosing.
I chose to try and make a forecast model for US quarterly GDP. Obviously being an undergrad my methods are incredibly basic, hence my posting here to seek your kind guidance and teachings!
What I have so far is a linear OLS model--I pulled a bunch of data from FRED (IPI, PCE, PMI, Retail, Sentiment, Claims) and regressed them on GDP. Then I used the static forecast function. It actually fits pretty well to past data, with a decent MAE/MAPE.
Though, the model has a hugee multicollinearity problem (PCE & Retail Sales, amongst others) and all of the variables are probably unit roots/autocorrelated (though I looked at model residuals and they were peachy). However, since I am only interested in a point estimate/forecast do all of these issues actually affect my purpose? I don't even want an efficient interval estimate, I just want a point estimate.
Secondly, I was wondering if there was any other methods of estimation that would be simple enough for me to learn by myself and perform in Eviews so I can improve my forecasts? I read up on alot of journal articles but so far much of them has been lost on me.
Cheers guys!