Academic journal article Journal of Money, Credit & Banking

New and Old Models of Business Investment: A Comparison of Forecasting Performance

Academic journal article Journal of Money, Credit & Banking

New and Old Models of Business Investment: A Comparison of Forecasting Performance

Article excerpt

RECENT EMPIRICAL RESEARCH on investment has focused on the estimation of the stochastic first-order conditions, or Euler equations, from dynamic models derived under rational expectations. Because these models have an explicit structural interpretation, they are theoretically more appealing than traditional models of investment. However, the empirical performance of Euler-equation models has not been tested against the traditional models. This paper performs such a test by adding two Euler equations to the usual group of traditional investment models examined in previous studies--namely, the accelerator, neoclassical, modified neoclassical, and Q-theory models.(1)

The first of our two Euler equations is a "canonical" model that typifies the equations estimated in recent years.(2) Despite its popularity, this canonical model has a restrictive dynamic structure that is unlikely to capture the time lags inherent in the investment process. In contrast, our second Euler equation explicitly accounts for the lag between the start of an investment project and the later date at which the new capital begins to contribute to the firm's production. By embedding such "time-to-build" lags, which were emphasized by Kydland and Prescott (1982), this equation has a richer structure than most previous investment Euler equations.(3)

This paper focuses on the ability of the various models to forecast investment in equipment and in nonresidential structures. From a practical standpoint, such out-of-sample tests are needed to determine which models have the most value as forecasting tools. Moreover, beyond this practical objective, out-of-sample performance is a powerful test of model specification (see, for example, Hendry 1979). We conduct two sets of tests. The first set of tests examines the size, bias, and serial correlation of the models' one-step-ahead forecast errors, similar to the out-of-sample tests performed by Kopcke (1985) and Clark (1979). In addition, we compare the information content of model forecasts by regressing actual investment on predictions from pairs of models. Fair and Shiller (1990) have argued that such regressions provide a powerful test of alternative models.

To summarize the results, we find that the Euler equations produce extremely poor forecasts of investment for both equipment and nonresidential structures. The time-to-build version of the Euler equation outperforms the basic Euler equation in our tests, but the improvement is modest. All the Euler equations have mean squared forecast errors many times larger than those of the traditional models. Moreover, the Fair-Shiller tests suggest that, as a group, the traditional models for equipment dominate the Euler equations; for nonresidential structures, the Fair-Shiller tests show that neither the Euler equations nor the traditional models have any forecasting ability.

The paper is organized as follows. The next section describes the models in our horse race, while section 2 briefly discusses our data set. Section 3 presents full-sample estimates of each model, in order to gain some initial information on their relative fit. Section 4 documents that the Euler equations produce relatively poor forecasts, and section 5 attempts to explain why this is so, arguing that the standard assumptions that underlie these equations could be to blame. Section 6 concludes the paper and suggests areas for future research.


A. Two Investment Euler Equations

To derive our Euler equations of investment spending, we adopt several assumptions that are fairly standard in the literature:

* The firm's production function is Cobb-Douglas with constant returns to scale:


where [Y.sub.t] and [L.sub.t] are output and employment during period t, and [K.sub.t - 1] is the capital stock at the end of period t - 1. The marginal product of capital is


* Capital is a quasi-fixed factor subject to the usual quadratic adjustment costs, while employment is assumed to be a variable factor. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.