New Developments in Economic Scenario Generator Modelling – Dec. 7, 2023

Session Description

Join the Ontario Regional Centre of the Canadian Statistical Sciences Institute (CANSSI) at the Statistical Sciences Applied Research and Education Seminar (ARES) with Jean-François Bégin.

Over the last 40 years, various frameworks have been proposed to model economic and financial variables. These frameworks—called economic scenario generators—are comprehensive models that allow actuaries and risk managers to grasp the long-term uncertainty underlying financial market values and economic variables. Their primary aim is to generate a set of future scenarios covering a range of plausible outcomes. The main end-users of these frameworks are pension, life insurance, and banking practitioners. In this presentation, we explore two important modelling questions relating to economic scenario generators and their use: model uncertainty and model averaging.

Given today’s knowledge and technology, one could construct complicated frameworks to fit the data better. However, this process would lead to highly parametrized models, which goes against the idea of parsimony in statistics—the desire to explain phenomena using fewer parameters. The first part of this presentation investigates this tradeoff: would a more complex economic scenario generator perform better, or would a simple model accomplish the same performance? To answer this question, we propose a new complex generator that nests versions of well-known actuarial frameworks. We then assess whether complex models perform better than simple models using both in- and out-of-sample analyses.

Second, we investigate model averaging. This strategy has been used extensively in data-heavy domains such as weather forecasting and in fields where forecasts come from diverse methods and datasets, such as election polls. In our context, we answer whether we can create better, more reliable economic scenario generators by combining them. This part of the presentation considers a recently proposed Bayesian averaging technique and data from multiple countries.