An evaluation and comparison of econometric methods of forecasting business investment : Summer's Q, Neoclassical, ARIMA, and VAR models
Date
1993
Authors
Parker, James Anthony
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Aggregate investment is a large and important component of GDP. The current level of investment not only determines current GDP, but it also plays a large part in determining future productivity gains and competitiveness as well. Thus, a correct model of the investment process is vital for both policy makers and forecasters.
Two distinct models of aggregate investment compete in the literature, the Neoclassical model and the Q model. The more traditional Neoclassical approach is currently used by most practitioners even though it has been rejected by most theorist (Lucas, 1976) in favour of the Q approach. The principle theoretical advantage of the Q model over the Neoclassical model is that it incorporates forward looking expectations. In theory, the backward looking expectations of the Neoclassical model leave it particularly vulnerable to the Lucas Critique.
This thesis compares the forecasting accuracy of a tax adjusted version of the Q model (Summer's Q), with two versions of the Neoclassical model, using quarterly Canadian data. In addition to the theory based models, a pair of time series models, an ARIMA model and a Vector Auto Regression (VAR) model are examined to assess the contribution of economic theory to the prediction of investment. Accuracy is measured with both Theil' s U and Mean Absolute Percent Error (MAPE). The quantitative significance of the Lucas Critique is assessed with the CUSUM test which tests for model stability. The Neoclassical and Q models are also tested for model specification using a co-integration test and the Ramsey test.
A comparison of the theory based models with the time series models reveals that the VAR and ARIMA models outperform both theory based models in the prediction of investment in non-residential construction. This result casts doubt on the value of economic theory as an explanation of the investment process. Further, both Neoclassical models, and Summer's Q model, fail the co-integration test for investment in machinery and equipment.
A comparison between the two theory based models shows that in general, Summer's Q is superior to the Neoclassical models in both forecasting accuracy and model stability. Further, the superiority of the Q model may be due directly to the quantitative significance of the Lucas Critique. Evidence is provided by the CUSUM test which shows instability of the Neoclassical models during the forecast period. The backward looking expectations of the Neoclassical models leave them particularly vulnerable in times of changing expectations whereas the Q model incorporates changing expectations immediately through the valuation of the stock market. Improvement in the modelling of expectations may lead to improvement in forecasting accuracy and model stability of the Neoclassical models.
Description
Keywords
UN SDG 8: Decent Work and Economic Growth