EViews.com

Hello everybody,

I apologize if my questions would be hilarious but I'm a newbie regarding GMM.
I'm trying to replicate a previous study regarding the short time interest rate. I have the equations for the moment conditions - first and second order but I don't know what should be the procedure to estimate those parameters. The study says just GMM with Newey West methodology and provides 2 moment conditions and a list of instrumental variables.

1. My versions would be:
a. A GMM system with 2 equations (both the moment conditions) - time series HAC (the one in the workfile)
b. Estimating each equation separately with GMM - time series HAC and replacing the coefs from the first one to estimate the second.

2. According to user manual GMM Time series HAC provides a a weighting matrix that is robust to heteroskedasticity, contemporaneous correlation of unknown form and autocorrelation of unknown form. My question would be: do I get Newey West procedure to estimate the variance-covariance matrix of coefs (robust to heteroskedasticity and autocorrelation) only by thicking the checkbox "Identify weighting matrix in estimation"? If yes how would a prewhitening would affect the estimation? (the theoretical effect of Newey West + prewhitening - empirically I would observe).

I will attach the workfile and the model. The study is built for more than one country so there are more series.
Thank you

P.S. I am using Eviews 7.

As long as your estimation method is "GMM-Time series (HAC)", then you will get Newey-West standard errors.

The check box you are talking about determines whether the point estimates are estimated using a GMM approach (i.e. using the Newey-West matrix as part of the point estimation, not just the standard errors), or using a TSLS approach (i.e. without using the Newey-West matrix as part of the point estimation).

My apologize for being insistent (and have a poor handling over the concepts), but just to confirm: in both cases (ticking or no the box) you get a robust matrix to heteroskedasticity, contemporaneous correlation of unknown form and autocorrelation of unknown form? And these properties will be kept until the convergence?

- ticking the box the estimation starts with the identity matrix and performs TSLS (updating that matrix and coefs)
- without ticking the box the Newey-West matrix is used in the beginning and updated according to the number of iterations performed to convergence.

As someone mentioned on an early post the 2 methods may not converge to the same thing.

Regarding my equation to be estimated. What do you think it will be the best approach? Those two moment conditions treated as a system or equation by equation?

Thank you.

hi allan and Gareth,
nice to meet you guys.

This is the first time i encounter GMM equation.
I try to redo the research of CKLS (1992).

The equation is simpler than allan's equation, where (pls look at my attachment for better view)
rt-rt-1 = a + brt +e
E(e)=0 and E(e^2)=variance^square*rt*gamma

In this case rt is interest rate.

In eviews, i wrote as follows:
@inst c uk_level uk_squared
uk_level(+1)-uk_level=c(1)+c(2)*uk_level
uk_level(+1)-uk_level=c(1)+c(2)*uk_level +c(3)*(uk_squared^c(4))

My problem is:
1. How do we actually choose the instrument list?
Some source says that we simply put the entire variable in equations and more. Do we have to consult literature or can we technically get the entire instrument we need from the equation we build?
2. How many instruments do we need in GMM? Should it be as much as the parameter we want to find out?
3. As it produce near singular matrix, what option can I take to get the results? Have i done mistakes in writing the equations?

I attached my data and model in the attachment.
Looking forward to your help sir.