Time-varying fiscal multipliers for South Africa : a large time-varying parameter vector autoregression approach

A critical requirement for efficient fiscal policy is a reliable understanding of its impact on the aggregate economy for different policy instruments and under different economic conditions. Indeed, there is strong evidence to suggest that fiscal multipliers vary with economic conditions, the components of government decision-making that are considered, and the identification strategy and modelling approach used. Previous studies on South Africa have typically used small-scale models or constant coefficient linear settings, which do not fully capture either the disaggregated components of spending and tax revenue or the time-varying nature of fiscal multipliers. In this paper we add to the critical evaluation of these limitations by using a large time-varying parameter vector autoregression approach estimated with Bayesian methods. We argue for an agnostic approach that studies the components of aggregate output in an economy which imposes as few restrictions and assumptions as possible. We model the impact of the government-controlled components of output on all other components and present a new way of reconciling these results to aggregate results in other studies. We find multipliers at the lower end of other findings in the literature on South Africa: our estimate of the average cumulative fiscal multiplier of government consumption on output is 0.155, while that of government investment is -0.118. Our approach also casts a cautionary light on both existing research and novel methods used to measure fiscal multipliers. As a result, convincing evidence that fiscal policy can be used actively for business cycle stabilization remains elusive.