CSTEP Scenario Reduction for Model Efficiency

CSTEP Cluster-Distance Sampling for Tail Estimation of Probability Open-Source software for Risk Scenario/Attribute Sampling/searching and Model Efficiency Strategies CWU Computer Science Department http://www.cwu.edu/~ahmadyt/stochastic.html Stochastic Team: Alan Chandler, Nathan Wood, Eric Brown, Temourshah Ahmady; Advisor: Dr. James Schwing CWU Math Department UCONN, Dr. Charles Vinsonhaler, Dr. Jeyaraj Vadiveloo Edward F. Cowman, FSA, MAAA “As to credits - I don't need any face time for this. Access to your new software will be reward enough for my efforts over the past several weeks. However, just so you will know that I'm a real person I'm attaching a short bio that our firm uses in our brochure.” Acknowledgement Representative-> Pivot Sampling -> Optimal Parametric Distribution Fitting ->BC-MLE Practice: TAS TRITON MG-ALFA GC GCC RSS 1996 1998 1999 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Research: NY 7 [Longley-Cook] Representative scenarios [Christiansen] Scenario sampling SALMS [Chueh] Parametric model outcome fitting AMOOF [Chueh] Scenario reduction research tool CSTEP [Chueh] BMLE [Johnson & Chueh] Model Efficiency Techniques CSTEP (Distance) Cluster Sampling for Tail Estimation of Probability Distribution of Financial Model Outcome Distance formulas, sufficient sample size, controlled sampling bias and error—stratified, pivot probability, replications General scenario sampling, Algorithms & Procedure Composite/nested scenario selection, Nested stochastic analysis (1M, 4500 periods) Economic scenarios as well as risk attributes ;random v.s. deterministic Model Efficiency Using CSTEP CSTEP : Cluster (-Distance)Sampling for Efficient Estimation of Tail Probability Distribution of Financial Model Outcome Upgrade from SALMS—Editable distance formula and scenario format 8,388,608 scenarios, 4500+periods universe capacity, flexible sample size, reversible & reusable sampling Risk scenario study and reduction(e.g. interest rate, equity return, quantifiable credit risk and ph behaviors…) Introduction to CSTEP CSTEP : Over one month 30+ projects submitted and 5 selected Replace spreadsheet-based platform and SALMS that has been used by modelers since 2003 Open-source computational tool (software) for scenario sampling studies and implementations Introduction to CSTEP Model outcome, as a complicated function of stochastic risk scenarios, not possible to have a closed-form; however, has been proved to be continuous if uniform bounds on number of projection periods and projected cash flow amounts The farther/closer the two scenarios, the farther/closer their model outcome values Extreme scenarios are where the functions extrema/tail distributions are projected Extreme scenarios replicate tail distributions, making modeling efficient Algorithms Significance Method With economic value parameters Ck=1 Simple operation Time: Θ(n) Memory: Θ(n) Algorithms Euclidean Distance Method More complex Time: Θ(n*m) Pivoting process Memory: Θ(n) Algorithms Present Value Distance Method With economic value parameters More complex Time: Θ(n*m) Pivoting process Memory: Θ(n) Algorithms Present Value Distance Method With economic value parameters More complex Time: Θ(n*m) Pivoting process Memory: Θ(n) Algorithms Significance Method With economic value parameters Ck=1 Simple operation Time: Θ(n) Memory: Θ(n) Algorithms GUI GUI PROCESS Yvonne Chueh Programing Languages C# Graphical User Interface C++ Sampling algorithms Lua Formula Scripts Requires Windows (XP, Vista, 7) Questions? Yvonne Chueh chueh@cwu.edu http://www.cwu.edu/~chueh Director of Actuarial Science Program Professor of Central Washington University