<Previous version: PcGive 13>

PcGive 14: enhancements and solved problems

Fixed and improved in PcGive 14.2

• Multivariate robust Hedgehog plots had variables scrambled.
• Hedgehog plots use a different color in the forecqast period
• Hedgehog Levels forecasts beyond estimation sample: not integrated
• Levels forecasts with gap: created cGap extra forecasts Also used wrong levels if cGap > 0.
• Recursive hedgehog would omit early part when there are dummies
• Fixed Ox issue with find, affecting forecasting

Fixed and improved in PcGive 14.1

• GetVarRf for FIML,CFIML when there are no estimated parameters returned JJ' instead of J&lt;>J' (which cannot be computed). Now it returns a matrix of zeros of the same dimension as JJ', so that forecasts with and without parameter uncertainty are the same. (NRE)
• Autometrics 1.5g: using standard instead of default for 0.05 in output Using Renew instead of Append in DummySaturation
• Small change in handling of lagged Y's that are classified as U_VAR: now only transforming back to Y_VAR after Autometrics run. Otherwise we get different results when using the PcGive class for estimation only with Autometrics.
• Autometrics block search now works with components (always adding all components into the model)

New and improved in PcGive 14

• Autometrics related:
  • Autometrics default now set to 0.01 instead of 0.05
  • Added SIS, IIS+SIS, DIIS, DIIS+IIS dummy saturation
  • Now have PcGive batch command createinterventions and Ox function PcGive::CreateInterventions to construct intervention dummies from saturation in the database.
  • reclassification of lagged dependent variables from X,U to Y,I is now done at a later stage. This allows forcing of lagged dependent variables when running Autometrics on a system (OLS) or single equation (OLS, IVE)
  • additional instruments for IVE (denoted with an A) now is different from unrestricted (U). U is used for unrestricted: not to be removed by automatic model selection.
  • removed variables from an IVE model are only made into additional instruments if reduction of the reduced form was used first (which is the default).
  • Writing initial GUM when entering k>T.
• Much extended dynamic forecasting:
  • Automatic detection of certain types of transformations of the dependent variable Y:
    fn(Y), Dfn(Y), DDfn(Y), DD_sfn(Y), D_sfn(y)
    where fn(Y): Y, log(Y), logit(Y)
    also detects on rhs corresponding fn(Y), Dfn(Y), D_sfn(y)
  • Based on this a levels forecast will be available in two forms:
    1. median forcasts: inverse transformation (exp(y), logistic(y))
    2. mean forecasts: based on lognormal and logitnormal distribution.
    The 5% and 95% quantiles, based on two standard errors simply arise from the transformation, e.g. for log(Y): exp(y-2SE), exp(y+2SE)
  • Robust forecasts. Robust forecast are derived from using the model on the differenced data, which is then reintegrated.
  • Optionally, forecasting can start a specified number of observations after the end of the estimation sample.
  • Hedgehog graphs: forecasts starting from each observation in the estimation sample:
    1. fixed parameters: the same parameters are used for each forecast
    2. recursive parameters: if the model is estimated recursively, the recursive parameter estimates are used. So the model is re-estimated each time for the new forecasts.
  • Estimating an empty model now does not give an error message anymore, facilitating DDD
  • Add documentation to header files, DoForecasting doesn't store SEs, nor are there functions to extract forecasts.
• Added X12-ARIMA
• Markov-switching models:
  • MS_Component: Markov-switching with component structure for mean and variance
  • MS_GARCH: Markov-switching GARCH
  • MS_MV: Multifractal volatility models
  • MS_VAR: Multivariate Markov switching
  • Refactored code to implement improvements in structure.
  • Removed singularity in analytical derivatives when probability is (very close to) zero.
  • Changed starting values for transition probabilities.
  • Probability mask: avoid diagonal when setting 1-sum element
  • Ability to set lower pound for probability of staying in same regime (if this is zero, we have identified an outlier - this can be on the boundary of the parameter space with a lower log-likelihood than the interior solution. Default minimum for diagonal of transition matrix (prob of staying) is 0.01.
  • reporting sdev(y) not var(y)
  • updated NBER info
  • Forecasts: not reporting naive forecasts anymore
  • Forecast se: only simulating for dynamic models
  • Impulse responses with standard errors

Fixed in PcGive 14

• exit batch command crashes OxMetrics
• fixed infinite reload loop in when loading help in Chrome
• CFIML: not forcing rows/columns of restricted parameters to zero anymore in VarCovar matrix. (fixes problem with forecast se when parameter uncertainty is included).
• IVE withholding forecast, store fitted in database gave concatenation warning (but worked fine)
• recursive IVE: default no inits was zero, which didn't work for IVE
• CFIML then FIML: was still estimating CFIML.
• Dummy saturation (IIS) would go haywire if a dummy is already in the model