SAS Dynamic Actuarial Modeling Features List

Premium modeling process

Premium modeling process

  • Supports the end-to-end pricing process – from data preparation to modeling and deployment, both online and in batch mode.
  • Enables both pure premium and gross premium modeling.
  • Enables users to create and include underwriting rules, and to verify the impact of the new premium model, including new underwriting rules.
  • Different users with different responsibilities/roles and rights can work on the same premium modeling process in a traceable and auditable environment.
  • Makes machine learning more accessible to actuaries by offering them their choice of models and guidance along the pricing process.
  • Dramatically reduces time to market for new pricing models.

Data management

Data management

  • Lets you load data from various sources into the system.
  • Enables the definition and application of data quality checks, and reports of any abnormalities that lie outside predefined ranges. An easy-to-use interface guides users in providing preconfigured rule sets and predefined ranges.
  • Allows users to make changes in the data of interest, creating new columns based on formulas.
  • Enables data visualization, correlation analysis, box plots and the ability to run prototype models for validating variables for use in the final models.

Modeling

Modeling

  • Model estimation
    • Provides templates for performing frequency/severity/aggregate modeling. Users can also build their own models and/or model templates.
    • Enables the use of new modeling techniques – machine learning model, generalized additive model (GAM) or another SAS Visual Data Mining and Machine Learning model – through the user interface without the need to write SAS code into the pipeline.
    • Enables the use of R or Python models in the modeling process.
    • Supports model recalibration using different input data/assumptions.
    • Allows benchmarking of the results of all selected models against the classical generalized linear model (GLM).
  • Interactive grouping node (IGN)
    • Enables easy and automatic variable grouping, providing users with complete visibility and control of the results. Users may also interactively change results coming from the automated model.
    • If the grouping structure is already known, users may import it from a local file, saving time and ensuring consistency with previous runs.
  • Optimization node
    • Allows users to simulate renewal pricing scenarios and see the impact in the current portfolio profitability.
    • Speeds decision making with a user-friendly interface for adding constraints and configuring the objective function, along with a visual reporting interface for exploring scenarios and making the data transparent.
  • Ratemaking node
    • Provides a dedicated component for modeling the frequency and severity of claims using classical GLMs and more advanced models, with a final comparison of their performances.
    • Enables the definition of the use of different variables in different models, and trains them simultaneously.
  • Explainable AI
    • Explainable AI elements are available as global and local interpretability indicators. These elements make the interpretation of the machine learning models easier and more transparent.
    • The explainable AI indicators are:
      • Local interpretable model-agnostic explanations (LIME).
      • Individual conditional expectation (ICE).
      • Partial dependence (PD).
      • Shapley additive explanations (SHAP).
      • At the variable level, a bias analysis for avoiding unfairness in pricing can be applied to any technique (including GLM).
  • Fairness & bias
    • Enables analysis of the fairness and biases of the modeling, giving the right insights directly in the results of the models.

Business rules

Business rules

  • Enables the definition of underwriting and pricing rules via an edition of generic predefined rules or the creation of new ones.
  • Allows users to deploy the score code created in the previous step.
  • Enables simulation/testing of the full new premium structure on user-provided data. Results may be reviewed and differences with previous premium structures disclosed in a reporting tool, which includes maps and other segmentation variables.

Integrated reporting

Integrated reporting

  • Enables post-modeling modification of premium modeling parameters.
    • Cap and floor risk factor coefficients, with modifications tracked automatically in the report.
    • Fix risk factor coefficients to a specific level, with modifications automatically displayed in the relativity plots and tracked in the report.
  • Presents the ratebook with parameter estimates for each variable in a tabular format. Users may select any rate range, and manually modify an individual rate or range of rates.
  • Allows users to discard any changes made and restart from the original output table, or save the modified table as a scoring code.

Automatic deployment

Automatic deployment

  • Enables one-click porting to production – both online and in batch mode – using the business rules defined within the premium modeling process.