Parimutuel wagering is a significant source of revenue for many state governments. MonitorPlus is a surveillance system for parimutuel operators and regulators. Using industry expertise and best practices, MonitorPlus examines each and every wager and account transaction for evidence of fraud, crime, and money laundering. Alerts are generated in real-time. In forensic discovery mode, MonitorPlus is designed to collaborate with skilled analysts to discover more complex suspicious wagering patterns. MonitorPlus utilizes machine learning, so its risk profiles are current: its knowledge base improves with time. Each alert is accompanied by an automatically generated, rule-based explanation. This is critically important if an event rises to the level where legal action is required. Our development and deployment strategy is based on a new paradigm of a secure surveillance utility, where real-time alerts and dataintensive forensics support multiple regulatory jurisdictions. We believe this surveillance paradigm can be applied to other application domains such as lotteries, casinos, online gaming, and financial services.

The problem that concerns us is the cost-effective computation of the expected value of a derivative security. One should not separate the method of computing the expected present value of a structured security from its ultimate computing topology. In particular, the network infrastructure is as least as important a factor in cost-effective computing as the algorithm design and its processor implementation. We investigate the network issues involved with deploying sophisticated derivative analytics on a modern computer network. We show that same technology that can be used to exploit parallelism can also be used to deploy sophisticated analytics to authorized users in a cost-effective way that is secure, easily updatable and relatively machine-independent. We put these ideas to practice by extending our derivative computation system, which was used to compare the derivative valuations on various computing network architectures. The benchmark problem computes an American "put" option under various interest rate scenarios using a combination of binomial lattice and Monte Carlo methods. We rebuilt the system as an executable derivative calculator applet. It is currently viewable on any Java-enabled Web Browser on the World Wide Web, independent of the computer processor or operating system. It also exploits parallelism: it uses any processor available on its local host to automatically speed itself up.

The concept of domain testability of software is defined by applying the concepts of observability and controllability to software. It is shown that a domain-testable program does not exhibit any input-output inconsistencies and supports small test sets in which test outputs are easily understood. Metrics that can be used to assess the level of effort required in order to modify a program so that it is domain-testable are discussed. Assessing testability from program specifications and an experiment which shows that it takes less time to build and test a program developed from a domain-testable specification than a similar program developed from a nondomain-testable specification are also discussed.>

A description is presented of a methodology, called knowledge-based tuning, that allows a human analyst and a knowledge-based system to collaborate in adjusting an analytic model. Such a methodology makes the model more acceptable to a decision-maker, and offers the potential for making better decisions than either an analyst or a model can make alone. In knowledge-base tuning, subjective judgments about missing factors are specified by the analyst in terms of linguistic variables. These linguistic variables and knowledge of the model error history are used by the tuning system to infer a specific model adjustment. A logic programming system was developed that illustrates the tuning methodology for a macroeconometric forecasting model. It empirically demonstrates how the predictability of the model can be improved.>

A Gram-Charlier series involves the expansion of one probability density function in terms of the derivatives of another density function. This paper generalizes the original series introduced by Gram, an expansion that minimizes a weighted square error. Two new Gram-Charlier expansions for the density of a finite variate are derived and recursion formulas for series coefficients are given.