**Dr. Wouter Verbeke** - KU Leuven (website)

"An introduction to causal machine learning for business decision-making"

Machine learning is a powerful tool to support business decision-making. For instance, predictive models can be learned from data to anticipate the future and to make informed decisions, with the eventual objective of optimizing the efficiency and effectiveness of business operations.

Even better than having predictive models, which tell you what will happen, is to have prescriptive models, which tell you what to do so as to optimize the outcome of interest. To this end, prescriptive analytics and operations research develop simulation models, which are typically crafted by an human expert modeler in the form of a series of mathematical equations. As an alternative approach, causal machine learning can be adopted to learn to predict the future as a function of the decisions that are made. In other words, causal machine learning models estimate the net effect on the outcome(s) of interest that would be caused by various potential business decisions. As such, these models directly indicate the optimal decision.

In this talk, I will demonstrate the use and need for causal machine learning by discussing on a business case. I will discuss on the challenges in estimating causal effects and learning a simulation model from data, and introduce some basic causal machine learning methods.

]]>

Risk theory: data-driven models

Abstract:

Ruin theory studies the riskiness of an insurance portfolio by investigating the evolution of an insurer's surplus. The existing models often assume stationary increments of the surplus process, which is not always appropriate to describe an actual experience. In this thesis, we consider some modifications that are inspired by a real-life insurance data set to the existing risk models and investigate how these modifications affect ruin theory results.

We first explore potential surplus modelling improvements by investigating how well the available models describe an insurance risk process. To this end, we obtain and analyze a real-life data set that is provided by an anonymous insurer. Based on our analysis, we discover that both the purchasing process and the corresponding claim process have seasonal fluctuations. Some special events, such as public holidays, also have impact on these processes. In the existing literature, the seasonality is often stressed in the claim process, while the cash inflow usually assumes simple forms. We further suggest a possible way of modelling the dependence between these two processes. A preliminary analysis of the impact of these patterns on the surplus process is also conducted. As a result, we propose a surplus process model which utilizes a non-homogeneous Poisson process for premium counts and a Cox process for claim counts that reflect the specific features of the data.

Next, we study a risk model with stochastic premium income. It is assumed that both the premium arrival process and the claim arrival process are modelled by homogeneous Poisson processes, and that the premium amounts are modelled by independent and identically distributed random variables. After reviewing various known results of this model, a simulation approach for obtaining the probability of ultimate ruin based on importance sampling is derived. We demonstrate this approach by examples where the distribution of the sampling random variable can be identified. We then give other examples where we use fast Fourier transform to obtain an approximation of the sampling random variable. The simulated results are compared with known results in the existing literature.

In the last part of the thesis, we consider a risk model where both the premium income and the claim process have seasonal fluctuations. We obtain the probability of ruin based on the simulation approach presented in Morales(2004). We also discuss the conditions that must be satisfied for this approach to work. We give both a numerical example that is based on a simulation study and an example using a real-life auto insurance data set. Various properties of this risk model are also discussed and compared with the existing literature.

]]>