Cai’s field of expertise is predicting extreme events that happen very rarely, such as an earth quake or flood. The large magnitude of these events makes them important to study. However, since they are so rare, there is usually not enough data to study them carefully. Using the available data and econometric methods for extrapolation, Cai aims to make the most accurate estimates of extreme events.
Econometricians and statisticians can be of great value when it comes to predicting extreme events. Cai explains: “The world is complicated and there is a lot of uncertainty. We often can’t determine exactly how likely it is that some event will happen. Econometrics provides us with the tools needed to interpret data and make informed decisions even in the face of uncertainty.”
Collaboration with KNMI
Cai has collaborated with the Royal Netherlands Meteorological Institute (KNMI) to improve their models for forecasting extreme weather. “Standard weather models are not tailored to deal with extreme weather conditions,” says Cai. “They are very good at predicting normal rainfall, but not at accurately predicting extreme rainfall for which there is not a lot of data. There are extra steps needed to apply the weather models to extreme rainfall, and we helped KNMI with these extra steps, using econometric tools.”
From predicting extreme weather to cure chances
In the coming years, Cai will develop new models for estimating survival and cure chances, with many practical applications. The Dutch Research Council (NWO) has awarded a research grant from the Open Competition Domain Science - M-2 to Cai and her research partner, Eni Musta, assistant professor at the University of Amsterdam.
“Our research could be valuable for health care providers and patients, for example. Many medical studies only follow patients for a limited amount of time. If patients are not observed until the end of their life, it is hard to estimate whether a treatment is curative or only life prolonging. For example, some forms of cancer are very aggressive and only very few patients are cured. These cured patients are not always observed later in their life – for instance, 20 years later– so that we never know if they are actually cured or if the treatment only prolonged their life. With our methods, we help estimate if the treatment worked in the long term.”
Practical applications
Two practical applications that will be part of the study are evaluation of treatment strategies for childhood osteosarcoma patients, focusing on cure and not only survival prolongation, and default prediction in credit scoring. Other examples of possible applications include fertility — assessing the likelihood that a treatment is effective for specific individuals — and criminology, where the focus is on predicting the probability of recidivism among convicted individuals.