SAS Econometrics Features List

Hidden Markov models

  • Fits and applies hidden Markov models to time series data.
  • Does fitting (or learning), smoothing, filtering, decoding and scoring.
  • Supports univariate or multivariate models, regime-switching regression models and regime-switching autoregression models.
  • Supports discrete state Gaussian models.
  • Provides methods for efficiently handling very long (big data) series.
  • Automates selection of the number of states and the number of lags.
  • Provides regime-switching autoregressive model in mean-adjusted form.
  • Provides stochastic gradient descent (SGD) optimization algorithm for all models.

Spatial econometrics modeling

  • Supports the following:
    • Linear models.
    • Linear models with spatial lag of X (SLX) effects.
    • Spatial autoregressive (SAR) models.
    • Spatial Durbin models (SDM).
    • Spatial error models (SEM).
    • Spatial Durbin error models (SDEM).
    • Spatial moving average (SMA) models.
    • Spatial Durbin moving average (SDMA) models.
    • Spatial autoregressive moving average (SARMA) models.
    • Spatial Durbin autoregressive moving average (SDARMA) models.
    • Spatial autoregressive confused (SAC) models.
    • Spatial Durbin autoregressive confused (SDAC) models.
  • Provides output tables that enable you to fully understand and interpret the impact of individual variables in the model.

Other econometric models

Count regression models for integer-valued dependent variables

  • CNTSELECT procedure models the frequency with which events may occur during a time period.
  • Supports:
    • Poisson, negative binomial and Conway-Maxwell-Poisson (CMP) regression.
    • Zero-inflation models conditional on covariates.
    • Overdispersion models conditional on covariates (with CMP model).
    • Random-effect panel data models for counts.
    • Spatial count data models.
    • Bayesian estimation.
  • Provides automated variable selection methods.
  • Includes many diagnostic tests and plots, including plots for focused visualization of specific parts of the fitted probability distribution.
  • Display tables to assess covariance and correlation among estimated model parameters.

Severity regression models

  • Fits the distribution to the size or severity of losses or other events.
  • Supports:
    • Regression models for severity distribution scale parameter.
    • Left censoring and right truncation (e.g., deductibles and coverage limits).
    • Many distributions, including:
      • Burr.
      • Exponential.
      • Gamma.
      • Generalized Pareto.
      • Wald.
      • Log-normal.
      • Tweedie.
      • Weibull.
  • Provides the ability to program additional distributions.
  • Fits multiple distributions and automatically selects the best.
  • Provides many diagnostic tests and plots, including plots for focused visualization of specific parts of the fitted probability distribution.
  • Includes display tables for assessing covariance and correlation among estimated model parameters.

Qualitative and limited-dependent variable regression models

  • CQLIM procedure estimates regression models for univariate qualitative and limited-dependent variables.
  • Supports:
    • Censored and truncated models.
    • Logit, probit and tobit models, and bivariate probit and tobit models.
    • Models with heteroscedasticity.
    • Univariate limited-dependent variables models.
    • Bivariate and multivariate limited dependent variable models.
    • Bivariate and multivariate discrete response variable models.
    • Multivariate linear equation models.
  • Estimates stochastic frontier production and cost models.
  • Heckman sample selection model.

Copula models

  • Simulates copula models of the multivariate dependency structure among sets of potentially many variables.
  • Supports simulations from the following copulas:
    • Normal.
    • t.
    • Clayton.
    • Gumbel.
    • Frank.

Regression models for panel data

  • Analyzes relationships between the past and the future using a large number of observations and more than one observation per time period.
  • Supports:
    • One-way and two-way models.
    • Fixed-effects, random-effects and hybrid models.
    • Autoregressive and moving average models.
    • Dynamic panel models.
  • Provides Hausman-Taylor and Amemiya-MaCurdy estimators.
  • Provides different kinds of heteroscedasticity and autocorrelation consistent (HAC) covariance matrix estimators.
  • Fits and compares multiple models.
  • Includes many diagnostics and tests.

Economic capital modeling

  • Combines results from frequency, severity and copula modeling.

Compound distribution modeling

  • Provides a graphical summary of aggregate loss distribution from large, distributed simulated samples.
  • Includes simulation modes.
  • Provides a flexible way to specify count distributions.
  • Enables more realistic simulations of loss modeling using stochastic variables.
  • Enables perturbation analysis for estimating mean and variability of aggregate loss distribution statistics.

Forecasting models for time series analysis

  • Enables you to programmatically create forecasting models on time series data.
  • Lets you create time series models:
    • User-defined ARIMA.
    • Exponential smoothing models (ESM).
  • Lets you create time series analysis, decomposition models and diagnostic testing.
  • Provides output tables with parameter estimates of fitted models, multistep variable forecasts and modeled variable information.

Distributed, open & cloud-ready

  • Runs on the SAS Viya platform, a scalable, distributed in-memory engine.
  • Distributes analysis and data tasks across multiple computing nodes.
  • Provides fast, concurrent, multiuser access to data in memory.
  • Includes fault tolerance for high availability.
  • Lets you add the power of SAS Analytics to other applications using SAS Viya REST APIs.

SASEMOOD Interface Engine

  • Retrieve time series data from the Moody's Analytics Data Buffet. Access over 600 sources of global historical statistical data and 40 forecast databases – over 220 million time series.