Welcome to PcGets

Contents

PcGets

The PcGets system

Autometrics

PcGive 12 was released in October 2007 as part of OxMetrics 5. PcGive contains Autometrics^tm for automatic econometric model selection, including facilities for automatic detection of outliers and breaks, as well as more variables than observations. PcGive 12 with Autometrics replaces PcGets 1, which is only available for GiveWin 2.

PcGets

PcGets^tm is an automatic econometric model selection program. PcGets is designed for modelling economic data when the precise formulation of the equation under analysis is not known a priori. The current version is for models that are linear in variables.

PcGets is a revolutionary new approach to model building, based on recent advances in the understanding of model selection procedures. Experiments show that PcGets 'outperforms' even the most experienced econometrician.

Efficient model formulation

• Dynamic econometrics involves creating and naming lagged variables, controlling the available sample and forecast period etc., and assigning the appropriate status to all variables, so such operations are either automatic or very easy. The basic PcGets operator is a lag polynomial, so long-run solutions, the significance of lagged variables (or groups of lags), and roots of lag polynomials are all calculated.

• The ordering of the menus and dialogs is determined by the econometric theory: first establish a data coherent, constant-parameter model, then simplify it, ensuring by diagnostic tests that all reductions remain valid.

• Estimation methods currently supported include ordinary and recursive least squares, instrumental variables and recursive instrumental variables. Models are easily revised, transformed and simplified; up to 15 models are remembered for easy recall and progress evaluation.

• Automatic outlier detection and removal is available, with user-selected definitions.

Automatic model selection

• The recommended general-to-specific approach to model construction is automatically adopted, the sequence of reductions is monitored, and F-tests, information criteria etc. are reported.

• The pre-selection screening tests quickly eliminate irrelevant variables, using loose significance levels.

• Multi-path searches check for `hidden' relations, and highlight the relevant explanatory variables, while ensuring that all reductions are acceptable, with diagnostic tests remaining insignificant

• All tentatively-selected contending models from pre-selection and path searches are retained and evaluated against each other and the joint model by encompassing.

• The final selection utilizes all the search, encompassing and information criteria collected during the analysis.

• Extensive Monte Carlo simulations and theoretical analyses have demonstrated the remarkable properties of PcGets in model selection. Not only does PcGets remove the hard work of model selection, it usually selects relevant variables with probabilities close to those operating when the model is correct and eliminates irrelevant variables with probabilities close to nominal significance levels.

• Nominal significance levels can be automatically adjusted for sample size.

• Two `pre-specified' selection strategies, denoted `liberal' and `conservative', make for simple yet powerful automated modelling, either to minimize the chances of omitting relevant variables, or to minimize the chances of including irrelevant variables. The `expert' strategy allows all the program parameters (namely, significance levels of all the selection criteria) to be designed at choice.

• Once a levels simplification has been found, cointegration can be investigated, and the model reduced to a stationary, near-orthogonal representation, checked for parsimonious encompassing.

Thorough evaluation

• Evaluation tests are automatically calculated for the GUM and during all model reductions. A comprehensive and powerful range of mis-specification tests is offered to sustain the methodological recommendations about model evaluation. Equation mis-specification tests include residual autocorrelation, ARCH, heteroscedasticity, functional form mis-specification, and non-normality. Constancy tests are computed automatically, and recursive procedures are available.

• Graphical diagnostic information includes plots of residual correlograms, residual density functions and histograms, and QQ plots.

• Recursive estimators provide voluminous output (coefficients, standard errors, t-values, residual sums of squares, 1-step residuals and their standard errors, constancy tests etc. at every sample size), but can be graphed for easy presentation (up to 36 graphs simultaneously). The size of models is only restricted by the available memory, as long as fewer than 100 variables are involved. Both forward and backward recursions are available.

• All estimators provide graphs of residuals, fitted and actual values and their cross-plots, as well as 1-step forecasts or forecast errors with 95%confidence intervals shown by error bars, bands or fans. Full graphical editing facilities can be applied to any graph (e.g., adding regression lines etc.).

With PcGets all the drudgery has gone - you choose the variables, then PcGets selects sensible and statistically-valid model(s), allowing you to concentrate on the variable choice and interpretation of the model(s).

The accompanying book transcends the old ideas of `textbooks' and `computer manuals' by linking the econometric theory with empirical modelling: the tutorials walk the user through the steps from inputting data to the final selected econometric model of the variables under analysis. The econometrics chapters explain the theory and methods with reference to the program. The statistical output chapters carefully define all the estimators and tests used by PcGets.

PcGets uses GiveWin for data input and graphical and text output. GiveWin has its own help system.

Even though PcGets is largely written in Ox, it does not require Ox to function.

The PcGets system

PcGets^tm uses GiveWin for data input and graphical and text output, and is part of the OxMetrics family.

OxMetrics^tm is the name of a family of software packages providing an integrated solution for those requiring econometric analysis of time series, financial econometric modelling, or statistical analysis of cross-section and panel data. The core packages of the family are GiveWin, which provides the user-interface, data handling, and graphics, and Ox Professional, which provides the implementation language.

GiveWin^tm (version 2) is the front-end for all the members of the OxMetrics family. GiveWin displays reports and graphics, which can be manipulated on screen, offers a calculator and algebraic language for transforming data, and enables the user to open multiple databases. A batch language allows for the automation of many of these tasks. GiveWin provides nearly 50 types of graphs, ranging from time-series plots to cross-plots, ACF, density, 3-D plots, automatic graphing of logs and growth rates, seasonal subplots, error fans and many others. For example:

© HM Krolzig This file last changed .