PcGive 12: enhancements and solved problems

Fixed and improved in PcGive 12.1

• General cointegration restrictions with restriction on alpha, but not in all rows: could give indexing run-time error.
• Logit estimation did not take selected sample (start,end) into account. It only did sample selection using a Select by variable. Store predictions in database gave run-time error.
• Generated Ox code when specifying `less forecasts' as part of the estimation sample: this is not taken into account when writing the SetSelSample command in the Ox code.
• PcGive now installs Arfima and DPD packages
• Autometrics name for combined dummies are wrong (the values of the created variables are right, but the S and D names are misleading). Now just writing the name as the sum or difference of the two dummies,
• Forecasting beyond the end of the database with 52 weeks/year as frequency and Seasonals goes wrong, because when the database is extended, the wrong seasonal is constructed.
• Generated GARCH code would try to load garch.dll, which doesn't exist.
• PcNaive: ability to use PcGive and Autometrics.
• PcNaive loading experiment from file: would lose some parameters if there are no lagged Z's in the DGP.

New in PcGive 12.0

• Now a module in the new OxMetrics 5 system.
• Autometrics^tm for automatic model selection.
• Generate Ox code. The required oxo files are now installed, and the generated code can be run with Ox Professional.
• Enhanced status handling in Formulate dialog.
• Improved output when variable names are very long.
• Autometrics GUM now stored in history for recall (removed checkbox about Use final model in next formulation dialog).

Problems in PcGive 11.1 that have been fixed in PcGive 12

• PcNaive: SetPcGiveDimensions didn't check rows in Pi. Start formulating matrix DGP, cancel, switch to PcFiml DGP: would forget to allow for constant and trend.
• Garch: could not have lagged Y in variance.
• Batch nonlinear/ml estimation would fail if there is a variable in the database that has all missing values (or the sample size could be restricted if some variable(s) have shorter sample).
• Further output, LaTeX of nonlinear model should use \&
• Restricted cointegration: in some cases wrongly decided that restrictions on alpha were simple, in which case rescaling was applied when it shouldn't be. As a consequence, the alpha matrix would not reflect the restrictions.