Shifting sentiments in firm investment: an application to the oil industry

Abstract

Recent developments in the oil and gas industry suggest that investment behaviour is not necessarily changeless over time. We propose a micro-econometric procedure to investigate the stability of investment behaviour at the firm level. Applying system Generalized Method of Moments (GMM) on a panel data set for 253 oil and gas companies over 14 years, we estimate accelerator models of investment with error-correction. Robust econometric evidence indicates a structural break in oil and gas investment in 1998. The process of capital formation over the last few years is more flexible than before, with significant and material changes in the role of explanatory factors like cash flow and uncertainty.

Acknowledgements

This study has benefited from comments and suggestions from Frank Asche, Petter Osmundsen, Knut Einar Rosendahl, seminar participants at the University of Stavanger, Statistics Norway, Norges Bank (Central Bank of Norway) and conference participants at the 2008 FIBE conference at the Norwegian School of Economics and Business Administration Bergen. The usual disclaimer applies.

Notes

¹ See Bond and van Reenen (2007) for a general overview.

² Based on the result from preliminary estimation, the shift is restricted to the error-correction term (e_it ) and the vector of control variables (x_it ). Econometric tests are not supportive of a structural shift for the dynamic part of the model (i_{it −1}, Δy_it , Δy_{it −1}).

³ A range of control variables has been tested, including various oil price variables, result variables and cash-flow variables. The best econometric results were obtained in a model with the described cash-flow indicator as the financial variable. Joint significance of model parameters of this version outperformed alternative specifications. We therefore assume that all relevant financial information, including oil price variation, is captured by the cash-flow variable retained in the preferred model version.

⁴ The key differentiating factor of the production technology among oil and gas companies is the reserve concept. The stock of oil and gas reserves represents a crucial input in this production process. But oil and gas reserves are not readily available in well-functioning input markets, like the case is for most other traditional inputs. Rather, oil and gas companies have to invest in very risky exploration activities to support and grow the base of oil and gas reserves. Thus, our reserve replacement variable captures the impact on total investment from companies’ efforts to sustain production activity over the longer term.

⁵ Founded in 1948, IHS Herold. is an independent research firm that specialises in the analysis of companies, transactions and trends in the global energy industry (http://www.herold.com/).

⁶ A factor of 3 is only slightly higher than the maximum year-to-year change in oil price in our sample. A lower factor may therefore exclude observations not affected by a major restructuring, where annual sales growth is simply induced by oil price changes.

⁷ See Terregrossa (1997) for a comparative analysis of various depreciation patterns and investment demand.

⁸Our proxy for the price of investment goods is the implicit price deflator for nonresidential gross private domestic investment (structures, equipment and software) from the US national accounts.

⁹Bond et al. (2003) also assume a constant rate of depreciation of 8% for manufacturing companies. For comparability, we assume the same rate of depreciation.

¹⁰ A range of options is available for uncertainty indicators, and no consensus is yet obtained for the appropriate way to proxy uncertainty in empirical models of investment. Forward-looking volatility forecasting models (Engle, 1982) are common in early empirical studies of investment and uncertainty. However, this strategy also introduces model uncertainty in two stages. Furthermore, Autoregressive Conditional Heteroscedasticity (ARCH) and Generalized Autoregressive Conditional Heteroscedasticity (GARCH) models estimated on high-frequency data will usually imply a low persistence of shocks for our purpose. The majority of recent work therefore relies more directly on observed uncertainty measures (Carruth et al., 2000).

¹¹ This assertion is supported by statistical inference from preliminary estimations, where the structural shift was allowed also for the dynamic part of the model. These full-fledged models produced more insignificant parameter estimates, and their overall quality diagnostics proved them inferior to the presented estimated models.

¹² The Hansen J statistic is the minimized value of the two-step GMM criterion function (Hansen, 1982).

¹³ Based on the far-right column of Table 2 (Model 3), the sum of the two estimated volatility coefficients is −0.554 + 1.128 = 0.574, with p < 0.01.

¹⁴ Boyle and Guthrie (2003) also suggest that financial frictions may increase the risk of future funding shortfalls. This will reduce the general value of waiting, and may, therefore, increase investment beyond the optimal level. Moreover, an additional source of a positive investment/uncertainty relationship is introduced. Based on this idea, an interaction term between cash flow and uncertainty has also been tested in our model. However, this combined variable did not produce significant parameter estimates, and could, therefore, not be included in the preferred model.