A Variance-Based Sensitivity Analysis of a Goodwin-Keen type Economic Model

Sensitivity analysis is a well-known tool of the trade in a number of scientific fields, but is not yet widespread in economics, in spite of its central usefulness in evaluating a model robustness. Furthermore, the discipline makes scant use of dynamical modeling, to the notable exceptions of the post-keynesian and ecological economics sub-fields, where a rigorous and versatile macroeconomic dynamical modeling framework, dubbed stock-flow consistent (or SFC) modeling, is gaining popularity. The purpose of the present paper is to present a two-tiers form of variance-based sensitivity analysis, focusing first on parameter groups before looking at individual parameters themselves in the context of macroeconomic dynamics. Such an approach offers a powerful way to tackle models with moderate to large numbers of parameters in a hierarchical fashion, helping researchers to make sense of the results of a sensitivity analysis and of its insight into their model dynamics. We deploy this method on a recent model (IDEE) in the Goodwin-Keen family, as the Goodwin-Keen approach to nonlinear macroeconomic dynamics has gained a lot of momentum in the last two decades. IDEE is a model of intermediate complexity; as many similar models, its macroeconomic core harbors highly nonlinear interrelationships among key economic variables such as employment rate, wage share, debt ratio, and inflation rate. Our findings highlight the paramount influence of parameter groups dictating shareholders' incomes in shaping value distribution within the model. Interestingly, while inflation has historically been considered pivotal in prior studies, our analysis suggests that it plays a relatively minor role in trajectories converging toward a Solow-type attractor---except insofar as it influences bifurcating dynamics. Finally, and perhaps most importantly, the sensitivity analysis performed allows us to show that the model results are robust with respect to expected variations and uncertainties in its parameters' values. We believe that the numerical methods presented in this paper can help to understand and improve numerical economic models, and eventually to improve their overall soundness and robustness.