Question about estimation of 3-variable VAR model

For econometric discussions not necessarily related to EViews.

Moderators: EViews Gareth, EViews Moderator

Thewad
Posts: 3
Joined: Tue Jul 14, 2015 5:58 am

Question about estimation of 3-variable VAR model

Postby Thewad » Tue Jul 21, 2015 2:53 am

Hello everyone.
Currently, I am doing a thesis about the effects of public investment on macroeconomic variables, to be more precise, on GDP. Basically, I want to estimate multiplier of public investment consumption for Czech Republic (using EViews 7.1).

Just as suggested in recent IMF working paper "Fiscal multipliers in Ukraine" (https://www.imf.org/external/pubs/ft/wp/2015/wp1571.pdf), I included three variables in my VAR model - Total state revenues, Total expenditures of general government (net of taxes and interests) and GDP. I use quarterly real and seasonally adjusted data (Q12000 - Q42014). I found that all of my series are I(1). I also included a few exogenous variables, different than authors suggested in mentioned paper. I included two spike dummy variables, quadratic trend and seasonally adjusted GDP of EU15 (since Czech economy is notably affected by economic performance, i.e. cycles of main EU countries).

The Lag Lenght Criteria was a bit ambiguous, AIC suggested 8 lags, SIC suggested 1 lag and HQIC suggested 3 lags. Since I have small sample, I have chosen to include 3 lags, which is in line with HQIC recommendation.

Multivariate Normality test showed that residuals are normally distributed, with P-values > 0.10, and I think the results of Autocorrelation LM test are quite good. But what confuses me, are correlograms. As can be seen from the photo, I think I have a problem with autocorrelation in residuals (marked with red).
Normality.jpg
Normality.jpg (75.52 KiB) Viewed 2130 times

autocorrelationLM.jpg
autocorrelationLM.jpg (44.24 KiB) Viewed 2130 times

Correlogram.jpg
Correlogram.jpg (99.43 KiB) Viewed 2130 times


My question here is: how should I proceed my work? Should I change my specification maybe, or eliminate the autocorrelation in some other way. Just for the record, my goal is, after I finish this part, to estimate SVAR just like in the mentioned paper, which is based on Blanchard and Perotti (2002) identification process, in order to get multiplers of goverment revenues and expenditures (aggregated and also disaggregated).

Return to “Econometric Discussions”

Who is online

Users browsing this forum: No registered users and 30 guests