Working Paper: CEPR ID: DP86
Authors: Paul Fisher; Mark Salmon
Abstract: Most model builders continue to treat their models as deterministic when forecasting, despite the fact that these models are composed of equations which are stochastic in nature. Deterministic solution methods ignore the stochastic information on the model structure and in addition produce biased forecasts in non-linear models. It is therefore important to investigate whether a given model is significantly non-linear. After commenting on the poor simulation methodology employed in a number of earlier studies, we find significant non-linear effects in two large macro models of the United Kingdom economy. This is confirmed by two tests that we propose for assessing the importance of non-linearity in such models.
Keywords: nonlinearity; large macroeconometric models; stochastic simulation
JEL Codes: 211; 212
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| traditional deterministic methods (C69) | biased and inefficient estimates of conditional expectation (C51) |
| efficient stochastic simulation methods (C15) | unbiased forecasts and proper policy analysis (E17) |
| treatment of nonlinearity (C32) | accuracy of forecasts (C53) |
| ignoring nonlinearity (C51) | biased forecasts (C53) |
| significant nonlinear effects (C29) | biased forecasts and inappropriate policy conclusions (H68) |