RDP 8810: The Relationship Between Actual Investment and Survey-Based Expectations 2. Empirical Models of Business Investment

Two principal approaches to investment theories models have evolved in the literature – the stock-oriented approach and the flow-oriented approach. In the former, the neoclassical model of Jorgenson (1963) is most prominent. Investment is determined as a stock-adjustment process of the capital stock to its desired level. The desired level of the capital stock is that level which maximises the firm's expected present value as given by its production function, demand for output and user cost of capital.

Within the latter approach, the most frequently used investment specifications are the Keynesian accelerator model and neoclassical-based models found in Lucas (1967), Gould (1968) and Tobin's (1969) Q theory. The accelerator model relates investment spending to the rate of change in output. In the Q model investment is determined by equating the marginal market value of new capital goods with their marginal cost of replacement.

In fact it can be shown[1] that both the stock and flow theories are special cases of a more general model of investment behaviour. However, most empirical research has tended to adopt a partial approach and test alternative explanations of investment.

In an extensive review of the Australian literature Hawkins (1979) found that the accelerator model explained investment in Australia up until the early 1970s. Thereafter the relationship seemed to break down. Attempts to extend the basic accelerator model by introducing financial constraints which may limit the attainability of the desired capital stock[2] have met with some success but the results are not robust over extended time horizons. Favourable results for the accelerator model were found in the U.S.A. by Clark (1979), but Kopcke (1985) found that profits and user costs of capital were also important.

The failure of accelerator models to be grounded in microeconomic foundations further left the theory open to the criticism that the output term may be picking up a conglomerate of influences. In particular, the failure of such models to incorporate relative price effects was seen as restrictive. In contrast, as pointed out by Carmichael (1979), the demand functions derived from the neoclassical model can be interpreted as incorporating most explanations of investment.

However, the neoclassical model did not perform well in early empirical studies. Early demand oriented versions of the neoclassical model including Mackrell et al (1971), McLaren (1971) and Hawkins et al (1972) found implausibly low parameter estimates for output elasticities and in general had little predictive ability in explaining movements in business investment even prior to the early 1970s.

Further studies, including research related to the NIF and RBII Models,[3] incorporated supply factors (such as the cost of using capital) into the neoclassical model. Of the single equation NIF studies most reported both poor ex ante and ex post forecast ability reflected in low R2 values and large standard errors. There also appeared to be a tendency towards the persistence of serial correlation in the errors of these studies.

Simulations using the RBII model were more promising. In particular, the neoclassical capital equation of the model provided low error sizes. But more recent research by Kohli and Ryan (1986) found incorrect signs on parameter estimates for the user cost of capital. According to these studies, investment appeared more responsive to changes in output than to changes in relative prices.

In his survey of determinants of investment spending in the U.S.A., Kopcke (1985) found that while the neoclassical model performed reasonably well during the first half of the 1980s, no one model “can clearly supply the one true description for causes of capital spending”.

One reason suggested for the poor performance of neoclassical models is the difficulty of correctly measuring variables such as the user cost of capital. Related to this, the treatment of expectations, which are typically introduced adaptively, could explain the poor explanatory and predictive performance.

Recent studies which have attempted to incorporate a forward looking expectations process for investment include the neoclassical flow models based on Tobin's Q theory. While consistent with the neoclassical stocks principle of equating the expected value of new investment with the expected cost of capital at the margin, in Tobin Q models the expected value is determined by the current stock market valuation of the firm's capital assets. As such, it is assumed that the current market valuation incorporates all available information in generating expectations and so avoids the difficult problem of specifying an explicit expectations function.

In a review of overseas studies Chirinko (1986) concluded that empirical results with the Q model are less than satisfactory. Similarly, Rider (1987) found little support for a tax-adjusted Q theory in the Australian context. However, once an explicit constraint for the cost of adjusting the capital stock is incorporated into the Q theory then McKibbin and Siegloff (1987) found that the Q theory explained at least 10 per cent of the variation in investment with the remainder explained by current profits.

An alternative forward looking concept which is similar in concept to Tobin's Q is EPAC's measure of the incentive to invest. This is based on the expected rate of return on capital relative to the opportunity cost of capital.[4]

Where it differs from Tobin's Q is that it uses a long-run measure of expected returns and focuses on the cost of debt finance. In their review, Carmichael and Dews (1987) concluded that empirical results with the EPAC variable were quite credible. Importantly, the EPAC measure explained the lacklustre performance of investment over the past two years better than did Tobin's Q.

The approach in this paper differs to those studies mentioned above. Rather than attempting to test directly models of investment we are concerned with examining the relationship between expected and realised investment plans using “news” in the variables usually proposed for explaining investment.

Our study suggests that the behaviour of firms in setting forecasts of business investment is such that the differences between these projected investments from their actual level can be attributed to deviations in expectations of certain variables (formed at the time of the survey) from their actual outcomes. Aside from providing some systematic study as to how survey data on business investment intentions relate to actual outcomes, the analysis provides some explanation as to the factors affecting changes in investment plans over time.

Footnotes

See Abel (1980). [1]

See Stegman (1982) who incorporates a profitability constraint. [2]

For studies related to the NIF model see Smith (1974), Johnston (1975), Higgins et al (1976). RBII model related research using a neoclassical model of business investment includes Jonson et al (1977) and Kohli and Ryan (1986). [3]

See EPAC (1986). [4]