Error when estimating State Space

[birtab01]

I am trying to estimate a state space model with 3 unobserved variables.
The first equation states the equation: Break even rate = inflation expectations + inflation risk premium - liquidity premium.
I have proxy variables for the unobserved variables: cpi_vol for inflation risk premium, survey5 for inflation expectations and volatility for liquidity premium.

Here is my output in state space in the object Sspace

param C(8) 0.758 C(9) -1.622 C(10) 2

@Signal brk5 = (C(10))*inflation_expectations + inflation_riskp - liquitity_proxy+ [ENAME = E7, VAR = (C(1))^2]
@Signal cpi_vol = inflation_riskp + [ENAME = E2, VAR = (C(3))^2]
@Signal volatility = liquitity_proxy + [ENAME = E5, VAR = (C(2))^2]

@State inflation_expectations = inflation_expectations(-1) + [ENAME = E3, VAR = (C(5))/(C1))]
@State inflation_riskp = (C(8))*inflation_riskp(-1) +[ENAME = E4, VAR =(C(3))/(C(1))]
@State liquitity_proxy = (C(9))*liquitity_proxy(-1) + [ENAME = E6, VAR =(C(2))/(C(1))]
Restriction is imposed on the variance as suggested by the reaserch above


Here are the equations for signal variables
〖brk5〗_t=π_(t,t+n)^e+ ρ_t^π- ρ_t^SÁ
〖survey5〗_t=π_t^e+ ε_t
〖std5〗_t=ρ_t^π+ ξ_t
〖volatility〗_t=ρ_t^SÁ + ϑ_t

Here are the equations for the state variables
π_t^e=π_(t-1)^e+ ω_t
ρ_t^π=ρ_(t-1)^π+ u_t
ρ_t^SÁ =ρ_(t-1)^SÁ + ψ_t


I have starting values obtained from OLS for inflation risk premium and liquidity premium.
I am trying to replicate the research https://www.sciencedirect.com/science/a ... 6911000123 on extracting inflation expectations from bond data but I have a proxy on inflation expectations.

I have also tried posing a restriction on the standard deviation like the research above does but then I get the message "near singular matrix"
When I try and run this model i get the message "unmatched parenthesis"
I have tried different starting values and changing the model but I end up with "Failure to improve likelihood"
I attached my data.

[EViews Glenn]

Code: Select all

@State inflation_expectations = inflation_expectations(-1) + [ENAME = E3, VAR = (C(5))/(C1))]


has an incorrectly specified C(1) coefficient.

I also get failure to improve with non-zero gradients using the default optimizer. What I was able to do was to use the EViews legacy estimator, which seems to do a better job at not getting stuck for this particular problem. Then I used those values as starting values for the other optimizers as a stability check. But even so, I think that this particular problem is difficult to estimate and I would suggest you use caution.