Hi, I am fairly new to EViews and have a few questions about implementing a SVAR model. using EViews 10.
I am doing a study similar to this one https://researchportal.port.ac.uk/porta ... ing_GF.pdf except with updated data just for the UK. In the study a SVAR framework is used with 7 variables, economic activity (gea), GDP (y), inflation (pi) government spending (gov), money supply (ms), interest rates (i) and stock market return (sm) with the vector yt:
[ gea_t
y_t
pi_t
gov_t
ms_t
i_t
sm_t]
My first question is:
When creating the initial VAR model in EViews, if i enter the endogenous variables in the order gea, y, pi, gov, ms, i, sm, then will this create a VAR model in the same order as the vector yt described above? (sorry if this is a silly question) I just want to make sure that the short run restrictions I apply later on are applied to the correct variables.
My second question is about inputting short run responses:
I need to impose short run restrictions on A0 (shown on page 12 in the link above). I know that I need to go onto the VAR model, proc > estimate structural factorisation. However I'm not sure whether to add these restrictions to A, B or S, or which restriction preset to choose.
When chose the restriction preset Recursive factorisation and added the restrictions to A, it came up with the response "the structural VAR objective function cannot be evaluated at the initial parameter values" and did not produce the SVAR model.
When I chose restriction preset custom and added the restrictions to S, it produced a SVAR model, however I'm confused because the Estimated S matrix gave values in places where I had specified to be 0 in the restrictions.
Hopefully this makes sense! Thank you for your help.
Implementing short run restrictions on SVAR model
Moderators: EViews Gareth, EViews Moderator

 Posts: 4
 Joined: Mon Jul 30, 2018 10:45 am

 EViews Developer
 Posts: 290
 Joined: Thu Apr 25, 2013 7:48 pm
Re: Implementing short run restrictions on SVAR model
Hello,
To your first question, yes, the order in which you list the endogenous variables determines their order within the VAR.
To your second question, it appears that the authors of that paper are using what EViews calls an AB model. Matrix A_0 in the paper corresponds to EViews' A matrix, while matrix D in the paper corresponds to the outer product of EViews' B matrix, i.e., D = B * B'. However, it's not clear that the authors estimated D (and attempting to do so would result in an underidentified model), so for your purposes I believe you can assume that B is an identity matrix. I'd recommend starting with the "Recursive factorization" preset, modifying the restrictions on A to match the paper, and then modifying the restrictions on B to make it an identity.
To your first question, yes, the order in which you list the endogenous variables determines their order within the VAR.
To your second question, it appears that the authors of that paper are using what EViews calls an AB model. Matrix A_0 in the paper corresponds to EViews' A matrix, while matrix D in the paper corresponds to the outer product of EViews' B matrix, i.e., D = B * B'. However, it's not clear that the authors estimated D (and attempting to do so would result in an underidentified model), so for your purposes I believe you can assume that B is an identity matrix. I'd recommend starting with the "Recursive factorization" preset, modifying the restrictions on A to match the paper, and then modifying the restrictions on B to make it an identity.

 Posts: 4
 Joined: Mon Jul 30, 2018 10:45 am
Re: Implementing short run restrictions on SVAR model
Hi Matt,
Thank you for your response! I did what you said and added the restrictions to A and made B the identity matrix, however it came up with the response "The structural VAR objective function cannot be evaluated at the initial parameter values". Does this mean that there is an issue with the data?
Thank you again.
Thank you for your response! I did what you said and added the restrictions to A and made B the identity matrix, however it came up with the response "The structural VAR objective function cannot be evaluated at the initial parameter values". Does this mean that there is an issue with the data?
Thank you again.

 EViews Developer
 Posts: 290
 Joined: Thu Apr 25, 2013 7:48 pm
Re: Implementing short run restrictions on SVAR model
It's more likely that the initial values of the variables being estimated are numerically problematic. On the "Optimization Control" tab of the SVAR specification dialog, try one of the random draw options.

 Posts: 4
 Joined: Mon Jul 30, 2018 10:45 am
Re: Implementing short run restrictions on SVAR model
Thank you again for your help, the model works now!
I have another question about the study I linked (https://researchportal.port.ac.uk/porta ... ing_GF.pdf). On page 14 it discusses the graph shown on page 27 (Figure 3), however I don't really understand the findings. The study states that positive government expenditure and interest rate shocks (shocks 4 and 6) cause a decline in the stock market (the bottom row of graphs, referred to as R_IND, i.e. the graph showing shock in government expenditure to stock market prices is titled 'Accumulated response of R_IND to shock 4').
1) Am I right in understanding that the authors have come to this conclusion because the zero line is not within the confidence intervals, therefore the result is significant?
The reason I am unsure is because the study then goes on to say that the only other determinant of the stock market, as evidenced from the impulse response functions, is GDP (shock 2). Positive GDP shock = slight increase in stock market. However I can't see why this would be from the graph (Accumulated response of R_IND to Shock2), as zero is within the confidence intervals?
Also, if possible, could you briefly run me through how to interpret these graphs? I.e. later on they discuss how two of the variables (R_GOV and R_INT) react in a countercyclical manner in response to shock 3 etc. I have created my own study with different data and produced similar graphs and will need to be able to interpret them.
Thanks again!
I have another question about the study I linked (https://researchportal.port.ac.uk/porta ... ing_GF.pdf). On page 14 it discusses the graph shown on page 27 (Figure 3), however I don't really understand the findings. The study states that positive government expenditure and interest rate shocks (shocks 4 and 6) cause a decline in the stock market (the bottom row of graphs, referred to as R_IND, i.e. the graph showing shock in government expenditure to stock market prices is titled 'Accumulated response of R_IND to shock 4').
1) Am I right in understanding that the authors have come to this conclusion because the zero line is not within the confidence intervals, therefore the result is significant?
The reason I am unsure is because the study then goes on to say that the only other determinant of the stock market, as evidenced from the impulse response functions, is GDP (shock 2). Positive GDP shock = slight increase in stock market. However I can't see why this would be from the graph (Accumulated response of R_IND to Shock2), as zero is within the confidence intervals?
Also, if possible, could you briefly run me through how to interpret these graphs? I.e. later on they discuss how two of the variables (R_GOV and R_INT) react in a countercyclical manner in response to shock 3 etc. I have created my own study with different data and produced similar graphs and will need to be able to interpret them.
Thanks again!

 EViews Developer
 Posts: 290
 Joined: Thu Apr 25, 2013 7:48 pm
Re: Implementing short run restrictions on SVAR model
Hello,
Yes, that the 2 S.E. error bands (almost) exclude the zero line indicates statistical significance (at ~5% level). Regarding the authors' claim about R_IND and Shock2, that mystifies me as well. There is a statistically significant impact of Shock2 on R_INT, perhaps the authors' mistook the two graphs?
Yes, that the 2 S.E. error bands (almost) exclude the zero line indicates statistical significance (at ~5% level). Regarding the authors' claim about R_IND and Shock2, that mystifies me as well. There is a statistically significant impact of Shock2 on R_INT, perhaps the authors' mistook the two graphs?
Who is online
Users browsing this forum: No registered users and 8 guests