Measurement error in Maximum Likelihood Estimation

[emma]

Hi,

1.In Dynare, if we estimate the model using maximum likelihood estimation, is it ture that dynare adds measurement errors implicitly?

2. In some examples, measurement errors are added to model explicitly as exogenous variables. In this case, will Dynare add measurement errors again? Does Dynare know that measurement errors are already explicitly modeled ?

Thanks

Emma

[StephaneAdjemian]

Question 1: No. You have to specify measurement errors in the estimated_params. For instance, if you think that there may be a gap between the observed variable Y and the theoretical Y, you can add a measurement error on this variable by writting:

stderr Y, UNIFORM_PDF,,,0,10;

in the estimated_params block

Question 2: I do not understand your question.

Best,
Stéphane.

[emma]

Hi Stéphane,

I add the measurement error explicitly following Ireland(2004) " a method for taking models to the data". (no beyesian methods involved)

In the model sector, measurement error ey is added to endogenous variable y, so the observed variable is oy. ey is declared as exogenous
variable.

oy=y+ey;

And in the estimated_params sector,
stderr ey, 0.1;

Am I right?

Emma

[StephaneAdjemian]

You're right Emma. You can alternatively introduce measurement errors this way. If I remember correctly, Ireland assumes that the measurement errors are modelled with a VAR(1). This can be done, if I follow your example, by declaring ey as an endogenous variable and by adding

ey = phi * ey(-1) + eey;

in the model block (with eey an exogenous variable). And then you just have to estimate the size of eey and the autoregressive parameter.

Best, Stéphane.