Object Reference : Object View and Procedure Reference : Var
Computes (n-period ahead) dynamic forecasts of the VAR or VEC equation.
forecast computes the forecast for all variables and all observations in a specified sample. In some settings, you may instruct forecast to compare the forecasted data to actual data, and to compute summary statistics.
var_name.forecast(options) f_pattern [se_pattern]
You should enter a naming suffix for the forecast series and, optionally, a naming suffix for the series containing the standard errors. Forecast standard errors are currently only available for non-Bayesian VARs, and are computed via simulation.
Not currently available for switching VARs
General Options
Graph the forecasts in individual graphs - one per dependent variable.
Graph the forecasts in a combined graph.
Produce the forecast evaluation table.
f = arg (default= “actual”)
Out-of-forecast-sample fill behavior: “actual” (fill observations outside the forecast sample with actual values for the fitted variable), “na” (fill observations outside the forecast sample with missing values).
Force the dialog to appear from within a program.
Print view.
Non-Bayesian Options
Number of simulation repetitions. Only applicable if a se_pattern is provided.
Fraction of failed repetitions before stopping. Only applicable if a se_pattern is provided.
BVAR Options
Perform classical forecasting – forecast based upon the posterior means of the coefficients as if they were calculated from a classical VAR. If omitted Bayesian sampling is used.
If “classical” is not specified, the following Bayesian forecasting options are available:
Store the mean of the draws from the sampler. If omitted the median is stored.
draws=integer (default= 100000)
Number of draws.
burn=arg (default=0.1)
Proportion of initial draws to discard.
Random number seed.
Drop any draws that produce unstable coefficients.
Produce distribution graphs.
Produce fan graphs.
Store the individual draws in a new page.
Use posterior mean as the point estimate. The posterior median is used if usemean is not included in the options list.
Show credibility intervals (bands).
cilevels = arg
(default = "0.95")
Set credibility levels. For multiple levels, enter a space-delimited list of values surrounded by quotation marks, e.g., "0.3 0.5 0.8".
Use lines instead of shading for credibility intervals.
seed = int
Set the random seed. EViews will generate a seed if one is not specified.
rng = arg
(default = “kn” or method set via rndseed)
Set random number generator type. Available types are: improved Knuth generator (“kn”), improved Mersenne Twister (“mt”), Knuth’s (1997) lagged Fibonacci generator used in EViews 4 (“kn4”), L’Ecuyer’s (1999) combined multiple recursive generator (“le”), Matsumoto and Nishimura’s (1998) Mersenne Twister used in EViews 4 (“mt4”).
The following lines:
smpl 1970q1 1990q4
var var1.ls 1 3 con inc
smpl 1991q1 1995q4
var1.forecast(m) _f _se
estimate a VAR over the period 1970Q1–1990Q4, and then computes dynamic forecasts for the period 1991Q1–1995Q4, and plots the forecasts as line graphs.
smpl 1970q1 1990q4
var var2.bvar(prior=inw) 1 3 con inc
smpl 1991q1 1995q4
var1.forecast(m, draws=50000, burn=.05, dgraph, page=draws) _f
estimates a Bayesian VAR with an independent normal-Wishart prior over the same period, and then forecasts that VAR taking 50,000 draws of a Gibbs sampler, discarding the first 2,500 draws, producing a distribution graph of the forecasts and storing the draws into a new panel page called DRAWS.
See “Forecasting” for a discussion of forecasting from VARs and VECS.
See also Var::fit.