Is it acceptable in PhD thesis to use data that have been converted from low frequency (annually data) to higher frequency (quarterly data). I'm asking because I have only few observations (GDP growth for 12 years) and would like to convert then into quarterly data so that I can run my model and conduct all validation tests.
changing frequency of data
- Thread starter siddig
- Start date
- Tags identification var
If you're doing this simply to produce a larger apparent sample size, then no...
But sometimes this kind of thing needs to be done for other reasons. E.g., in my phd thesis I was looking at the relationship between daily temperature and daily counts of suicide, self-harm hospitalisations, and assault. I needed to use population size as a control variable - but only had annual observations for that. So I used linear interpolation to convert the annual population figures to daily figures, to line up the temporal resolution with my IVs and DVs. So it depends a bit on what you're doing - more info might be helpful.
But sometimes this kind of thing needs to be done for other reasons. E.g., in my phd thesis I was looking at the relationship between daily temperature and daily counts of suicide, self-harm hospitalisations, and assault. I needed to use population size as a control variable - but only had annual observations for that. So I used linear interpolation to convert the annual population figures to daily figures, to line up the temporal resolution with my IVs and DVs. So it depends a bit on what you're doing - more info might be helpful.
Thank you colleagues Mmercker and CowboyBear. I intend to investigate the impact of oil price shocks on GDP growth over a specific period, which is 2000-2011. I'm running unrestricted VAR model at level with one lag. I have no problem to estimate the model using OLS and performing Granger causality test, Impulse Response Functions and Variance decomposition analysis. The problem I'm encountering is related to performing model validation/adequacy tests. I succeeded to perform the stability and normality tests, but failed to perform auto-correlation and heteroscedasticity tests due to insufficient number of observations.
No, you can not just divide the yearly data by 4 and pretend that is is quarterly data.
But many countries have quarterly data that is possible to get.
But step down from the ivory tower...
,,, and think about how many observations you have. In 2000 -2011 there has been an increase in in oil price from ca: $80 to $140 and then a decline from $140 to $80 and then a decline to $40. So that makes 2 - 3 observations.
You will not get essentially more information just by getting quarterly data. You will still have the same historical pattern. If you get data from say 1970, then you would have more info. (And that is easy to get.) Then you would get the information from the two oil shocks in the 1970:ies. (That was the experience that formed your hypothesis that oil prices can have an influence on GDP.)
Even if you could get reliable information on monthly or even daily basis, you would not have more information.
By the way, I don't think that you can "investigate the impact of oil price shocks on GDP" by estimating an unrestricted VAR model. Var models can be used for forecasting, but I doubt that you can investigate the the true causal impact of oil price on GDP. Read about the identification problem.
But many countries have quarterly data that is possible to get.
But step down from the ivory tower...
,,, and think about how many observations you have. In 2000 -2011 there has been an increase in in oil price from ca: $80 to $140 and then a decline from $140 to $80 and then a decline to $40. So that makes 2 - 3 observations.
You will not get essentially more information just by getting quarterly data. You will still have the same historical pattern. If you get data from say 1970, then you would have more info. (And that is easy to get.) Then you would get the information from the two oil shocks in the 1970:ies. (That was the experience that formed your hypothesis that oil prices can have an influence on GDP.)
Even if you could get reliable information on monthly or even daily basis, you would not have more information.
By the way, I don't think that you can "investigate the impact of oil price shocks on GDP" by estimating an unrestricted VAR model. Var models can be used for forecasting, but I doubt that you can investigate the the true causal impact of oil price on GDP. Read about the identification problem.
Thank you GretaGarbo> In fact transforming data frequency from lower o higher frequency or vice versa is conducted using mathematical formulas available in most of the statistical and econometric software e.g. EViews. Currently, the VAR model is widely used in understanding the relationship between oil price shocks and macro-economy! there are hundreds of research papers using VAR model to investigate such relationship e.g. Prof. Sims (1980) {by the way, this is the scholar who developed the VAR model}, Prof. Hamilton (1983, 1996, 2001, 2009), Prof. Mork (1989, 1994), Zhang (2011), Berument (2010) Akin and Babjidi (2011) among others. Just google "impact of oil price shocks"!With regard to the definition of "oil price shock", there are many definitions in the literature and how to calculate them, check these research papers: Mork (1989), Hamilton 1983, 1996), Federer (1996), Jimenez-Rodriguez (2008).