1. Ordinal, units-free algorithms. These are systems that train experts to rank abstract political variables across countries. In an ideal world, experts agree on what constitutes a “4″ on an arbitrary 7-point scale. Ultimately it doesn’t matter whether it’s a 7-point or a 20-point scale, until different variables are integrated into a single algorithm. The algorithm assigns weights to different variables. And again, ideally, the algorithm is fitted statistically against the particular type of project the customer has, to be sure that your index is a good predictor of future risks to that firm. Prototype: ICRG model.
2. Actuarial models. These systems rely on expected rates of future events that affect revenue streams to investors. Models in this category seek to measure objective event probabilities and expected losses from those events. These models seek to provide objective forecasts of the specific events that determine cash flows, and recently, the cash flows themselves (rather than the precursor events). These models, unsurprisingly, lend themselves best to insurance and credit products, providing the analysis is sufficiently accurate to permit underwriting over a long time horizon.
3. Threat models. Subjective probability of future events is a different type of analysis from actuarial calculation. Instead of using the past as a reasonably comprehensive guide to future events, threat models look for some combination of threats, vulnerabilities, and consequences. These terms are used with very different, although often precise, meaning by different firms and different industries. Threat models may refer to specific groups of individuals and their capabilities, or to generic motivations for unspecified hypothetical adversaries. Vulnerabilities may refer to the nature of the firm’s project, or in IT security, the comprehensive IT systems and policies designed to protect assets, or to the likelihood that an attack will be successful if attempted. These probabilities are subjective in that they describe the behaviors of actors with free will in a competitive, strategic, adversarial game with the firm. In the language of Knightian uncertainty, these probabilities attempt to quantify uncertainty and not risk.