SAS/ETS 13.2 introduces many new estimation features, including Bayesian estimation for count data models, new compound distribution model features, a new data interface engine, and many other enhancements.

SAS/ETS 13.2 includes a new utility for simulating compound distribution models (CDMs): the HPCDM procedure enables you to simulate aggregate loss distribution models, which are useful in assessing operational risk and defining appropriate capital requirements in situations in which rare but catastrophic events occur. The HPCDM procedure combines results of two other SAS/ETS procedures, the COUNTREG and SEVERITY procedures, giving the modeler extensive flexibility with which to model the number of events and size of losses. This flexibility can take the form both of distributional assumptions on the loss and of count data models, each of which contains different regressors. The HPCDM procedure includes the following features:

- simulation of aggregate loss models
- what-if analysis
- parallelized simulation that takes advantage of multicore machines
- ODS output for visual inspection of result

Quandl data can now be accessed from new SASEQUANDL interface engine. This interface engine conveniently enables you to dynamically query the most up-to-date information from the Quandl databases and then include these data to enrich and enhance your models.

PROC VARMAX now provides *p*-values for the Johansen cointegration rank test and for multistep forecasts for
the multivariate GARCH models.

The COUNTREG procedure now provides Bayesian estimation methods. You can now use Bayesian simulation techniques to estimate Poisson (zero-Inflated) models and negative binomial (zero-inflated) models.

The main features of the Bayesian methods include the following:

- choice of prior distributions
- tools to initialize and tune the MCMC algorithm
- multithreaded Metropolis sampling
- convergence diagnostic tools such as Raftery-Lewis, Heidelberger-Welch, and Geweke
- prior and posterior predictive analysis

A variety of new model specification tests has been added to the PANEL procedure. These tests check the model's statistical
assumptions about stationarity, cointegration, and structural change, and they include *p*-values
that are generated by high-performance simulation methods. Many software packages report only selected critical values
for these tests. For panel data models, the PANEL procedure supports the following new test statistics and methods:

- first-differenced methods for one-way and two-way models
- panel poolability tests
- Lagrange multiplier (LM) test for cross-sectional and time effects
- locally mean most powerful (LMPP) and standardized Lagrange multiplier (SLM) tests
- Gourieroux, Holly, and Monfort Lagrange multiplier test

The SEVERITY and HPSEVERITY procedures now support a new OUTSCORELIB statement to create scoring functions. Scoring eliminates the need to write a complex DATA step that reads the estimates from the OUTEST= data set. Both procedures also now support a new OFFSET= option in the SCALEMODEL statement to model the scale parameter per unit, which is a measure of exposure.

The SEVERITY procedure also supports CLASS statements. You can specify a wide variety of regression effects, such as singleton continuous effects, polynomial continuous effects, main CLASS variable effects, and more.

The COUNTREG procedure now supports the STORE statement and an enhanced CLASS statement. The STORE statement enables previously estimated models to be used for out-of-sample prediction and other post-estimation routines. The procedure is also multithreaded for faster computation.

The QLIM procedure has added an automated algorithm that searches for a good representation of the posterior distribution for Bayesian estimation.

The system generalized method of moments (GMM) estimator proposed by Blundell and Bond for dynamic panel models is now available in the PANEL procedure.

The X12 procedure now enables you to control the following:

- size of forecast confidence limits
- difference in critical values for almost outliers
- alpha value for outlier detection
- method of calculating the critical value for outlier detection, which is based on the alpha value and the number of observations in the span that is used for analysis
- number of level-shift outliers to consider for forming a temporary level shift
- method of adding outliers at each iteration of model estimation
- rate of decay for temporary change outliers
- moving average filter for each period

For more information, ask your organization’s SAS representative to contact the SAS Customer Interaction Center at 1.800.727.0025.