How to know which forward-looking variables are

Dear Prof Johannes,

test_RBC.mod (3.0 KB)

I’ve got a problem as Dynare always says “There are 5 eigenvalue(s) larger than 1 in modulus for 6 forward-looking variable(s)”. I know this could be the timing problem but as in your replication examples, only capital and bond are predetermined.

Dynare can find accurately steady states since I provide an exact “steady state block”. I checked it many times but I do not know why timing matters.

How can I know which vars are classified as forward-vars when Dynare attempts to solve the model? Could you do me a favor by looking at the mod file attached!?

The model is just small open RBC with habit in consumption. I compare my file with you AG(2007), and GPU(2010) but I cannot fix the error.

Thank you very much,
Kind regards,

I think your Euler equation is wrong. Marginal utility lambda should not have anything forward-looking in there. Start with an easier specification without habits where you can easily check the equation.

Dear Prof Johannes,

I am not sure but today if I rewrite the model without predetermined variable k, b and detrend accordingly, it’s fine! That is I am writing K = (1-delta)K(-1) + … and B/(1+r) + Y = … + B(-1), and divide by X(-1).

One question is in AG(2007) and GPU(2010) they detrended by dividing X(t-1), if we use the end of period notation then B(-1) will be detrended as:

(B(-1)/X(-2)) * (X(-2)/X(-1)) = b(-1) / g(-1)

The same applies to K(-1). Is this right?

Thank you and best regards,

Sorry to ask again my above simple question. Is it correct detrending end-of-period K(t-1) and B(t-1) with X(t-1) as above post?

Since I read Johannes’s guide showing that k(t) = K(t) / X(t) as well as k(t-1) = K(t-1) / X(t-1) but in AG they detrended with X(t-1) with an explanation about information set at t-1 of variable(t) will be preserved after dividing by X(t-1).

What is the difference? Does timing convention matter here?
Thanks in advance