Actuarial Software Now - A Supplement to Contingencies

Working collaboratively, the actuarial science and computer science departments at Central Washington University have created a high-performance platform named Cluster-Distance Sampling for Tail Estimation of Probability (CSTEP), which can be used to develop optimal probability model fittings for analyzing outcome distribution.

CSTEP is available as open-source software for actuaries and researchers.

What’s CSTEP?

CSTEP is a desktop-based application to assist efficient stochastic modeling through scenario-reduction techniques and studies. It provides actuaries with an HPC (high-performance computation) research platform for selecting a sample set of pivot risk scenarios to represent the entire scenario population ( 1 million scenarios tested with 4,500 time periods each) with the best accuracy at the distribution tails.

The open formula editing dialogue box takes into account the importance of backing assets runoff or liability attributes so that the business block can be modeled by employing an editable-input distance-formula box. That, in turn, allows stochastic models to be more accurate, time-efficient, and reliable in tail distributions. It lets practicing actuaries and researchers examine and analyze the probability distributions of time-prohibitive large blocks of integrated business. CSTEP offers a drastically reduced run time through a scenario-reduction technique that is researchable. The output of this platform program is clustered (optionally nested) and distance measured. It selects pivot samples of risk scenarios from the user-imported scenario universe, which allows stochastic modelers to test a reduced number of scenario runs.

Why CSTEP?

CSTEP is engineered according to techniques from a paper I published in the North American Actuarial Journal in 2002 and an associated open-source software SALMS (Stochastic Asset Liability Modeling Sampling). It’s designed to increase the functionality and capacity in SALMS and let users research their input-customized distance formulas. It’s been difficult for actuaries to program these published sampling algorithms without computer engineering training or experience. CSTEP replaces spread-sheet-based programs that require days of computing to sample from a moderate universe size (such as 10,000 when the advanced pivoting process is performed). With CSTEP’s user-friendly interface and automatic file management, research time can be spent on studying the model technique rather than building tools and manually processing data using a transparent spreadsheet format.

In addition, CSTEP can serve as an HPC platform for analysis in a number of areas, including:

n;The size of the original universe that can expose (probabilistically), present, and estimate the tail distributions and the means;

n;The size of the sample, justified with research given the size of the original universe and the business blocks and risk characteristics;

n;A comparison among various scenario-reduction techniques, refining their technique efficiency and effective implementation;

n;The creation of composite/nested risk scenarios on path-dependent scenario simulations for more accurate financial reporting and enterprise risk management;

that’s difficult to model effectively at the tail distribution;

n;More efficient stochastic modeling incorporating optimal parametric probability model fitting;

n;Improved model-efficiency techniques for bias-corrected maximum likelihood estimators and mixed probability models.

The stochastic model outcome distribution from a reduced-sample run can be analyzed to obtain tail estimates of probabilities and conditional expectations. Such scenario-reduced empirical model distribution can be used to test the effectiveness of pivot-scenario approaches through open-entry distance formulas—a new functionality for academic and company research and implementation. When the distance metric is more firmly connected to the relation between the scenario and its stochastic model projection, the extreme scenarios can be detected more easily. Once achieved, the pivot-scenario approach, enhanced by editable and researchable distance metrics, can help actuaries and executives make and analyze decisions on cash flow testing, risk-based capital, economic capital, principle-based reserving, capital budgeting, solvency, and pricing.

The CSTEP distance formula-building platform allows for continual tests and improvement in these effective scenario-reduction techniques so as to keep up with advancing computation technology. In the future, when companies have the capability to run tens, hundreds, or even thousands of nested-risk scenarios, CSTEP will still be able to search out extreme pivot scenarios from a nested universe of at least 8,388,608.