Overview

The AXIS™ actuarial system supports tools and techniques to reduce the run time of large and computationally intensive models. They may be categorized into compression techniques and proxy functions.

Compression is based on filtering out redundant or irrelevant inputs. Compressed models can produce rapid, accurate estimations of full model results, and can be run repeatedly to allow extensive analysis. The two main applications of compression techniques in the AXIS system are seriatim data compression and scenario compression. Seriatim data compression reduces the number of policies or coverages to run, with minimal loss of information. Scenario compression reduces the number of scenarios to be run while preserving the statistical properties of the scenario set. AXIS built-in tools include data dimensionality reduction, statistical analysis, principal component analysis, clustering algorithms and customizable scenario selection tools.

While scenario compression techniques can reduce the number of scenarios that have to be run, proxy functions can in some cases eliminate the need for stochastic simulations altogether. AXIS includes tools for configuring proxy functions, generating training data, and calibrating model parameters. Generalized Linear Models (GLM) and Neural Networks (NN) models are currently supported. Model configurations and parameters can also be imported from external sources. AXIS allows proxy functions to be called at key points embedded in the projections to eliminate bottlenecks by speeding up calculations of intermediate results such as projected prices of complex derivatives or structured securities, which do not have simple closed form solutions.

Optimization tools in the AXIS system may be combined in flexible fashion to build customized solutions which balance speed and accuracy. Ease of use through tight integration with existing AXIS features is a key part of their design.