Hi, I have a question about calibrating the correct innov values for stochastic equations in a model. Is it the estimated standard error of the equation, or of the variable? For example, if I have an equation in a model that is specified as:
dlog(gnq_pcr) = 0.033432218067729 + 0.0719655003020915* (log(gnq_pcr(1))  log(gnq_rpdi(1)) ) + 0.363199702910733 * dlog(gnq_rpdi)
should I calculate the innov value as the standard deviation of
dlog(gnq_pcr)dlog(gnq_pcr)_hat
or as
gnq_pcr  gnq_pcr_hat ?
Does it depend on the type of add factor that has been declared, or is it always one or the other?
Thanks very much!
Dawn
Innovations in stochastic scenarios
Moderators: EViews Gareth, EViews Moderator

 Fe ddaethom, fe welon, fe amcangyfrifon
 Posts: 12158
 Joined: Tue Sep 16, 2008 5:38 pm
Re: Innovations in stochastic scenarios
It should be on the equation, not the variable.
Follow us on Twitter @IHSEViews
Stochastic simulation: How EVIews computes confidence bounds in case of text equations
Dear Madam, Dear Sir
I have a rather complicated nonlinear macroeconomic model for which some blocks are estimated as system objects with FIML (supply block, monetary block, housing block) and several remaining equations are estimated as single equations. All equations are then combined as text into a model object. Stochastic model simulations and forecasts are technically computed without any problems  with the expected message that the equations are treated as nonstochastic (see below). Obviously, as EViews receives no information about the coefficient estimates in case of text equations, the simulations cannot account for coefficient uncertainty. Nevertheless, confidence bounds for all endogenous variables are computed. These must be based somehow on the residual distributions of the individual equations (the estimated standard errors of the equations). Now my question: Where does EViews find this information in case of text equations? Are the residuals and the corresponding standard errors of the equations computed within the model object? Probably yes. More generally speaking, a more transparent technical outline of what happens exactly in stochastic simulations would be welcome.
Many thanks for your answer,
Peter (pestal@bluewin.ch)
Model: SIMMOD
Date: 05/01/19 Time: 11:55
Sample: 2019Q1 2025Q4
Solve Options:
DynamicStochastic Simulation
Solver: Newton
Max iterations = 5000, Convergence = 1e06
Requested repetitions = 10000, Allow up to 25 percent failures
Solution does not account for coefficient uncertainty in linked equations
Track endogenous: mean, standard deviation, 75% confidence interval
Calculating Innovation Covariance Matrix
Sample: 1990Q1 2018Q4 (this is the estimation period)
Insufficient IME innovations  Equation treated as nonstochastic
Insufficient YC innovations  Equation treated as nonstochastic
Insufficient LCAP innovations  Equation treated as nonstochastic
Insufficient LSUP innovations  Equation treated as nonstochastic
Insufficient LFPOT innovations  Equation treated as nonstochastic
Insufficient POP innovations  Equation treated as nonstochastic
Insufficient UROFF innovations  Equation treated as nonstochastic
Insufficient WAGE innovations  Equation treated as nonstochastic
Insufficient PGDP innovations  Equation treated as nonstochastic
Insufficient WINCI innovations  Equation treated as nonstochastic
Insufficient PCONSP innovations  Equation treated as nonstochastic
Insufficient PCI innovations  Equation treated as nonstochastic
Insufficient PCONSG innovations  Equation treated as nonstochastic
Insufficient PIME innovations  Equation treated as nonstochastic
Insufficient PIMTOT innovations  Equation treated as nonstochastic
Insufficient POIL innovations  Equation treated as nonstochastic
Insufficient PHR innovations  Equation treated as nonstochastic
Insufficient PICNSTR innovations  Equation treated as nonstochastic
Insufficient PEXTOT innovations  Equation treated as nonstochastic
Insufficient YPRIMB innovations  Equation treated as nonstochastic
Insufficient CONSP innovations  Equation treated as nonstochastic
Insufficient CONSG innovations  Equation treated as nonstochastic
Insufficient ICBUS innovations  Equation treated as nonstochastic
Insufficient PHOUSE innovations  Equation treated as nonstochastic
Insufficient IHOUSE innovations  Equation treated as nonstochastic
Insufficient EXTOT innovations  Equation treated as nonstochastic
Insufficient IMTOT innovations  Equation treated as nonstochastic
Insufficient LRATE innovations  Equation treated as nonstochastic
Insufficient EVEUROFR innovations  Equation treated as nonstochastic
Insufficient MRATE innovations  Equation treated as nonstochastic
Insufficient PIMCC innovations  Equation treated as nonstochastic
Insufficient PEXCC innovations  Equation treated as nonstochastic
Insufficient EXS innovations  Equation treated as nonstochastic
Insufficient EXCC innovations  Equation treated as nonstochastic
Insufficient IMS innovations  Equation treated as nonstochastic
Insufficient IMCC innovations  Equation treated as nonstochastic
Matrix scaled to equation specified variances
Scenario: Fcst_BASE
Solve begin 11:55:36
Repetitions 13200: successful 11:55:45
Repetitions 32016400: successful 11:55:53
Repetitions 64019600: successful 11:56:02
Repetitions 960110000: successful 11:56:03
Solve complete 11:56:03
10000 successful repetitions, 0 failure(s)
I have a rather complicated nonlinear macroeconomic model for which some blocks are estimated as system objects with FIML (supply block, monetary block, housing block) and several remaining equations are estimated as single equations. All equations are then combined as text into a model object. Stochastic model simulations and forecasts are technically computed without any problems  with the expected message that the equations are treated as nonstochastic (see below). Obviously, as EViews receives no information about the coefficient estimates in case of text equations, the simulations cannot account for coefficient uncertainty. Nevertheless, confidence bounds for all endogenous variables are computed. These must be based somehow on the residual distributions of the individual equations (the estimated standard errors of the equations). Now my question: Where does EViews find this information in case of text equations? Are the residuals and the corresponding standard errors of the equations computed within the model object? Probably yes. More generally speaking, a more transparent technical outline of what happens exactly in stochastic simulations would be welcome.
Many thanks for your answer,
Peter (pestal@bluewin.ch)
Model: SIMMOD
Date: 05/01/19 Time: 11:55
Sample: 2019Q1 2025Q4
Solve Options:
DynamicStochastic Simulation
Solver: Newton
Max iterations = 5000, Convergence = 1e06
Requested repetitions = 10000, Allow up to 25 percent failures
Solution does not account for coefficient uncertainty in linked equations
Track endogenous: mean, standard deviation, 75% confidence interval
Calculating Innovation Covariance Matrix
Sample: 1990Q1 2018Q4 (this is the estimation period)
Insufficient IME innovations  Equation treated as nonstochastic
Insufficient YC innovations  Equation treated as nonstochastic
Insufficient LCAP innovations  Equation treated as nonstochastic
Insufficient LSUP innovations  Equation treated as nonstochastic
Insufficient LFPOT innovations  Equation treated as nonstochastic
Insufficient POP innovations  Equation treated as nonstochastic
Insufficient UROFF innovations  Equation treated as nonstochastic
Insufficient WAGE innovations  Equation treated as nonstochastic
Insufficient PGDP innovations  Equation treated as nonstochastic
Insufficient WINCI innovations  Equation treated as nonstochastic
Insufficient PCONSP innovations  Equation treated as nonstochastic
Insufficient PCI innovations  Equation treated as nonstochastic
Insufficient PCONSG innovations  Equation treated as nonstochastic
Insufficient PIME innovations  Equation treated as nonstochastic
Insufficient PIMTOT innovations  Equation treated as nonstochastic
Insufficient POIL innovations  Equation treated as nonstochastic
Insufficient PHR innovations  Equation treated as nonstochastic
Insufficient PICNSTR innovations  Equation treated as nonstochastic
Insufficient PEXTOT innovations  Equation treated as nonstochastic
Insufficient YPRIMB innovations  Equation treated as nonstochastic
Insufficient CONSP innovations  Equation treated as nonstochastic
Insufficient CONSG innovations  Equation treated as nonstochastic
Insufficient ICBUS innovations  Equation treated as nonstochastic
Insufficient PHOUSE innovations  Equation treated as nonstochastic
Insufficient IHOUSE innovations  Equation treated as nonstochastic
Insufficient EXTOT innovations  Equation treated as nonstochastic
Insufficient IMTOT innovations  Equation treated as nonstochastic
Insufficient LRATE innovations  Equation treated as nonstochastic
Insufficient EVEUROFR innovations  Equation treated as nonstochastic
Insufficient MRATE innovations  Equation treated as nonstochastic
Insufficient PIMCC innovations  Equation treated as nonstochastic
Insufficient PEXCC innovations  Equation treated as nonstochastic
Insufficient EXS innovations  Equation treated as nonstochastic
Insufficient EXCC innovations  Equation treated as nonstochastic
Insufficient IMS innovations  Equation treated as nonstochastic
Insufficient IMCC innovations  Equation treated as nonstochastic
Matrix scaled to equation specified variances
Scenario: Fcst_BASE
Solve begin 11:55:36
Repetitions 13200: successful 11:55:45
Repetitions 32016400: successful 11:55:53
Repetitions 64019600: successful 11:56:02
Repetitions 960110000: successful 11:56:03
Solve complete 11:56:03
10000 successful repetitions, 0 failure(s)
Who is online
Users browsing this forum: No registered users and 1 guest