Diagnostics

[Pantera]

Hi,
Could anyone check out whether there is any bug when plotting N-step Forecast Test (Choose the follow steps: "view - Stability Diagnostic - Recursive Estimation (ols only) - N-Step Forecast Test".)

The plot from this test shows the recursive residuals but NOT the significant probabilities (based on the F-statistic) which should be visible in the lower portion of the diagram.

Regards,
Torbjørn

[EViews Gareth]

I cannot replicate this. Could you post the workfile you were using. Also the build date of your copy of EViews (available from Help->About EViews).

[Pantera]

Hi,
Please find attached the work file. Estimate AR(1)- model: "averskr c averskr(-1)". Then; "view - stability diasgnostic - recursiveestim. - N-Step Forecast". The recursive plot of the resi shows that 1956 res. crosses the conf.int. and 1968 is closeed to being significant - but there is no significant probabilities in the lower portion of the diagram - which according to the manual should be there.

Regards,
Torbj.

Ps: I'm using EView7 and it is updated 4th of Dec - automatic updating.

[EViews Gareth]

The red lines on the graph are simply +- 2*SE, and are not true confidence intervals. In your particular case, none of the probabilities are significant, and thus none are plotted on the graph.

[Pantera]

OK - thank a lot for your help Gareth.

Torbj.

[EViews Gareth]

I think you can calculate the probabilities with the following program:

Code: Select all

smpl @first+3 @last-1    'set sample to cut off end points to avoid errors

series dof = @trend-1        'rolling degrees of freedom
series robs = @trend+1    'rolling number of observations
series obs = 73                  'total number of observations

series ssr = @sumsq(r_res)    'total sum squared residuals

series csq  = @cumsum(r_res^2)  'rolling sum of squared residuals

series F = dof*(ssr-csq) / ((obs-robs)*csq)   'create F stat series

series FProb = @fdist(f,1,dof)   'pvalues for f-stat
show FProb


[Pantera]

Thank for your insightful suggestion – I look forward to test the program. Just a short comment: If you look at the N-step Forecast test (after estimating the AR(1)-model, please see the preceding emails) the plot shows a zero at the top of the left hand side y-axis. Why? However, when you calculate the One-step forecast test, EViews generate “nicely” the confidence interval and the plot of the one-step probabilities. Is that the idea also for the N-step Forecast test?

Regards,
Torbj.