Study on probabilistic and computational approaches to risk modeling, analysis and forecasting

Abstract

Scope and Method of Study: A statistically-based yet probabilistically-concluded and computationally-implemented approach to modeling and evaluation of likelihood for events of interest to occur with a focus on risky events.

Findings and Conclusions: This study introduces a method that can facilitate the extension of the Multiple Regression with Dependent Dummy Variable (MRDDV) Model to provide a way of estimating the likelihood of any event of concern by probability. MRDDV employs a dependent dummy variable in its regression model as primary inputs for estimation. However, MRDDV is not proper to provide a probability-based estimator because it violates the definition of probability. To overcome this, a method, namely Logit Transformation, is employed to facilitate the probabilistic manipulation of MRDDV. By using Logit Transformation, the estimation of risk in MRDDV is, stably, represented in probabilistic domain (e.g., in the range beyond 0 or 1). Simulation results showed that Logit Transformation-based MRDDV Model improved the basic scheme significantly. And, a user's risk defining system is, also, introduced. The enhanced Logit Transformation-based MRDDV Model is probabilistic and robust in risk tracking.