6. Design and Development

6.1. Help Structure

_images/Help_Structure.jpeg

Documentation roadmap.

The documentation is structured around application, access, implementation, and theory.

6.2. Design Philosophy

  • Work at the correct speed order…

  • …but don’t worry about speed until it becomes a problem

  • Save everything until space becomes an issue

  • Offer sensible defaults for (almost) everything

  • Separate internal naming from user naming and offer standard dataframe renamer dictionaries

  • Use sensible number formats

6.3. History

I have built several iterations of software to work with aggregate distributions.

  • Late 1990s: a Matlab tool for CNA Re with a graphical interface. Computed aggregates by business unit and the portfolio total. Used to discover the shocking fact there was only a 53 percent chance of achieving plan…which is obvious in hindsight but was a surprise at the time.

  • Late 1990s: a C++ version of the Matlab code called SADCo in 1997-99. This code sits behind MALT. It won the CAS award for an online contribution.

  • Early 2000s: another C++ version with an implementation of the Iman-Conover method to combine aggregates with correlation using the (gasp) normal copula, SCARE. Used by SCOR.

  • 2003-2005: I worked on Aon Re Services’ simulation based tools called the ALG (Aggregate Loss Generator) which simulated losses, and Prime/Re which manipulated the simulations and applied various reinsurance structures. ALG used a shared mixing variables approach to correlation.

  • Late 2000s: At Aon Re, I also built related tools

    • The Levy measure maker

    • A simple approach to multi-year modeling based on re-scaling a base year, convolving using FFTs and tracking (and stopping) in default scenarios

  • 2010s: At Aon Benfield, I was involved with ReMetrica, a very sophisticated, general purpose DFA/ERM simulation tool.