It is a significant challenge to collect data on historical mass litigation for the purpose of back-testing. Most lawsuit outcomes are covered by confidentiality agreements and therefore dollar amounts are largely unavailable. In addition, while data on filings from United States federal courts are available, similar data from most state courts are not.
Nonetheless, Praedicat has assembled a great deal of published data to benchmark its results, including data from asbestos, tobacco, pharmaceutical, medical device and environmental litigation. The information is used to parameterize the models and calibrate the results.
In addition, Praedicat has back-tested pieces of its model in the following ways:
- The general causation projections of our forecasts of future scientific literatures, which drive our probabilistic model, can be back-tested in a traditional manner. Praedicat has performed back-testing for all Litagion® agent-harm combinations at 3 and 6 year projections. The results inform and improve our GC projection algorithm.
- Based on their historical claims experience, our strategic partners have provided input that assisted with validation of Praedicat’s approach to the valuation of bodily injuries.
- Regulatory actions, such as California’s Prop 65, IARC risk assessments and REACH, have provided data about the relationship between science and risk and our scoring has proven to provide a credible early warning.
- While sparse, data is available via verdicts, published opinions and academic studies on previous litigation and we have incorporated all of this into our models to inform variables such as damages, claiming behavior, available defenses, financial incentives and defense costs.
- Since the model is based on copious scientific information and not just historical claims information, we have calibrated our model using literatures that have not generated litigation. For instance, there are many Litagion agents with mature hypotheses of bodily injury for which claims have not yet emerged. We have profiled these Litagion agents and our models are built to fit both the sparse data we have on actual litigation and also the large amounts of data that we have on non-events. We call this calibrating against zeros. For example, to date there has not been significant litigation over bodily injury from mold. As a result, when we project the science for a risk, it must look different from the mold literature for our model to predict that litigation is possible.