Natural Gas Price Uncertainty, Real Options, Investment and Firm Value: Evidence from Northwest Europe

Natural Gas Price Uncertainty, Real Options, Investment and Firm Value:

Evidence from Northwest Europe

Author: Jan Geschiere (S1789074) Supervisor: Dr. Peter Smid

Program: Msc. Finance Faculty of Economics and Business

University of Groningen

Abstract

In this research, we investigate how uncertainty regarding natural gas prices influences firm-level investment, using panel datasets of non-financial firms from Germany and the U.K. Employing panel regression techniques to estimate several widely-used investment specifications, we find the relation between natural gas price volatility and investment to differ between the countries under investigation. The evidence points towards a positive relationship for Germany, whereas the evidence for the U.K. is consistent with an inverse u-shaped relationship. Our evidence is inconsistent with standard real options theory of irreversible investment, and is supportive of other theories of investment under uncertainty such as theories of convex marginal products of capital, strategic options theory and other, more general models where investment simultaneously kills and creates multiple options. Prudence is required when interpreting our results, however, as results from related research are shown to be sensitive to choice of model specification and estimation techniques.

1

1. Introduction

The investment decisions made by firms and the valuation of investment projects and growth opportunities are central topics in the fields of capital budgeting, corporate valuation and economics in general. Growth opportunities are a primary constituent of firm value (Kester, 1984) and a firm’s investment decisions are key determinants of its future capability to create value and generate satisfactory returns. At a higher level of aggregation, private sector investment is a key component of aggregate demand and an important driver of economic growth and future well-being. This research aims to improve our understanding of these topics by investigating the role of uncertainty regarding natural gas prices.

Our main purpose is to assess if there is a significant relationship between natural gas price uncertainty and investment in Europe. As a further extension, we investigate if these relationships are stronger for firms that operate in relatively energy-intensive industries. Whereas research on investment-uncertainty relationships is nothing new in the literature, the emphasis here lies on a particular form of uncertainty, namely uncertainty regarding natural gas prices. Being a primary energy source, natural gas is an important production input for firms, making uncertainty regarding its price a potential key driver of uncertainty surrounding corporate decision-making. While several earlier contributions in the literature have investigated the relationships between energy prices, investment and firm value, the majority of this literature focuses on crude oil prices and U.S. firms. This line of research has found evidence that both the level and volatility of energy prices can have a significant impact on decision-making. For example, Uri (1980) finds energy price levels to be of importance to American investment at the macro- and sector-level and Edelstein & Kilian (2007) find such a relationship to hold for the primary resources industry. Regarding energy price uncertainty, Henriques & Sadorsky (2011) find oil price volatility to be a significant determinant of U.S. firm-level investment and Yoon & Ratti (2011) find similar results for consumer fuel prices. Our aim here is to innovatively contribute to the literature by looking at natural gas prices in the European context.

2

the market for oil, is that there is no perfectly integrated global market for natural gas. In general, one could speak of three separate continental markets: Europe, North America, and Asia (Siliverstovs et al., 2005). This separation is caused by the relative difficulty of transporting gas, which is done primarily via fixed pipeline infrastructures. At the regional level, market integration in Europe is also not perfect. A common distinction here is between the more developed Northwestern-European markets, which are generally seen as being more integrated, and the less developed markets in Southern- and Eastern-Europe. This implies that we cannot speak of pan-European market integration (Stern & Rogers, 2011) (Neumann et al., 2006). An implication of the separation of continental markets is that it grants a powerful position to large, dominant suppliers in separate markets. Against the background of declining domestic gas production, the increasingly powerful position of large foreign suppliers is a primary concern to European policy-makers with profound geopolitical consequences, besides significant economic implications.

Recently, European markets are subject to profound changes and are moving from a situation with tightly controlled supply and demand to a more competitive landscape (Stern & Rogers, 2011), which is characterized by increasing liberalization and a changing pricing mechanism. European policies directed at increasing competition have led to an increasing integration of national markets (Robinson, 2007), as well as the break-up of incumbent, integrated state-monopolies into separate utility and network companies (Spanjer, 2009). The dominant pricing mechanism is moving from long-term contracts, in which prices are formed according to complex formulae that explicitly link gas prices to crude oil prices (Creti & Villeneuve, 2004), towards price formation on so called ‘hubs’ where gas can freely be traded. As Stern & Rogers (2011) formulate it, this implies a move towards a situation where gas prices are determined by the supply- and demand conditions of gas itself, rather than those for oil.

3

paper on the exposure of the European economy to developments on natural gas markets, Reymond (2007) underlines the increasing vulnerability to price volatility due to market liberalization.

This research takes a two-staged approach to investigating the investment-uncertainty relationship. Firstly, we investigate the direct relationship between investment and uncertainty, using several different modeling strategies that are common in the investment literature. Subsequently, a direct test of the uncertainty-firm value relationship is conducted to assess if there is a relationship between uncertainty and the valuation of a firm’s investment options. We seek to answer these questions by employing panel regression techniques, attempting to account for both firm- and time-specific effects, which are applied to firm-level panel data on British and German non-financial firms over the 1997-2012 and 1991-2012 period respectively

Evidence of a causal relationship between natural gas price uncertainty, investment and the option value of investment opportunities has several relevant implications. At the firm level, it reveals the potential for welfare-enhancing hedging policies and the potential benefits of reducing the dependency on natural gas through the development of less energy-intensive production methods. In the domain of policy making, adverse effects of price volatility are an important argument in the debates surrounding European natural gas market reform and the security of energy supply, including the recent debate on shale-gas development. With regard to macroeconomic policy, policy-makers striving to stimulate investment might be better off focusing on the stabilization of energy price levels besides aiming at directly influencing the height of price levels themselves. Outside the economic domain, this research can help us to assess the importance of energy supply as a geopolitical device by providing more insight in the harmful effects of disruptions of supply and the manipulation of price levels by large natural gas suppliers.

4

be discussed in the subsequent section. Finally, section 7 concludes and provides recommendations for further research.

2. Literature review and hypotheses

This section provides an overview of theoretical developments from the literature that are relevant to the investigation at hand and the empirical investigations that precede our research. The first subsection covers the most influential theories on uncertainty, investment, firm value and benchmark model specifications. Having developed these concepts, we move on to earlier empirical work that underpins our model specifications and methodology. Finally, we formulate testable hypotheses based on the theoretical concepts and empirical works covered in the review.

2.1. Investment behavior and the investment-uncertainty relationship

Explaining investment behavior, both at a firm- and macro-level, remains one of the pressing challenges in the fields of finance and economics. Despite the importance of understanding investment behavior for research areas ranging from capital budgeting and corporate finance to the study of macroeconomic fluctuations, the performance of empirical investment models remains mixed (Carruth et al., 2000). In the vast literature on investment behavior, there is no consensus on a single most appropriate model (Chirinko, 1993) and estimating different alternative models is common practice. Therefore, this review discusses several of the most commonly used specifications in the literature. The foundations of modern investment theory lie in classical models such as those developed by Clark (1917) and Chenery (1952). In these so called ‘accelerator models’ investment is linked to changes in demand that necessitate investment in additional capacity. Based on these accelerator models are neoclassical models of investment such as those discussed in the review by Jorgenson (1971). The starting point of these neoclassical models is the desired capital stock, proportional to demand levels, of a firm at which firm value is maximized.

5

as mentioned above fall in the category of implicit models, with dynamics often entering the investment equation implicitly through a distributed lag specification (for example lagged values of the capital stock, sales and investment). Early examples include Hall & Jorgenson (1967) and Eisner & Nadiri (1968). An implicit model that gained much popularity in the empirical literature is the error-correction model as introduced by Bean (1981). He modified existing distributed lag specifications by adding an error correction term, which is intended to proxy for the deviation of the capital stock from its desired level. The more the capital stock falls below its desired level, the higher investment will be.

Early critics of the implicit models treated above criticize the reliance on distributed lag specifications to capture investment dynamics, most importantly expectations and adjustment costs, see for example Eisner (1974) and Gould (1968). Returning to the framework of Chirinko (1993), explicit, or structural, models were developed to deal with these issues by explicitly incorporating dynamics in the model specification. In these models investment no longer depends on the optimal capital stock, but on the tradeoff between the (expected) marginal value of an additional unit of capital and its cost, including adjustment costs (Chirinko, 1993). Here, we discuss two particularly popular explicit models of investment behavior, known as ‘Q-models’ and ‘Euler-equation’ models.

6

measurement error of Q-ratios. Sources for such an error are stock market valuations that do not reflect fundamentals or managerial expectations (LeRoy, 1989; Blanchard et al., 1993) and bookkeeping practices used to measure the replacement costs of assets (Lewellen & Badrinath, 1997).

So-called ‘Euler equation’ models, first introduced by Abel (1980), are explicit models founded on the same principles as Q-models, the tradeoff between the marginal value and -costs of capital. The basic intuition behind Euler equation models is that firms set investment so that the costs of investing now equal those of postponing investment to the future. Hence, investment depends on the marginal product of capital, which reflects the opportunity costs of waiting to invest, and the expectation of future investment, which determines the expectation of future adjustment costs. Assuming rational expectations, the expectation of investment enters the model as the realized value of investment in the following period. The main criticism associated with Euler equation models is the limited amount of information that enters the specification (Chirinko, 1993).

7

liquidity and investment. Finally, conflicts of interest between shareholders and debt holders can lead firms to pass up on profitable investment due to the debt overhang problem as presented by Myers (1977), a problem that is most severe for financially constrained firms.

Having developed the basic theories on investment behavior, we now turn towards the investment-uncertainty relationship central to the research at hand. Key elements of the causal relationship between uncertainty and investment are the irreversibility of investment and the option approach to investment opportunities. An implicit assumption, crucial to the treatment of uncertainty, of standard investment models discussed so far is that of reversible investment, i.e. that a firm can easily disinvest without incurring significant costs. Critics of this assumption include, among others, Arrow (1968), Nickell (1974) and Pindyck (1990), who argue that firms are often confronted with significant losses associated with disinvestment. In his paper on corporate borrowing Myers (1977) presents the principle of perceiving growth options and investment projects as call options on real assets (analog to options on financial securities) and discusses the importance of these options for corporate borrowing, capital structure decisions and firm value.

8

The underlying relationship between uncertainty and option valuation is described in the literature on financial options. Most notable in this respect are the seminal article by Black and Scholes (1973) and its extensions as formulated by Merton (1973). The basic gist of the work by Black and Scholes is the well known Black and Scholes formula for option pricing. In this formula the value of a call option depends positively on the stock price, the time to the exercise date, the risk-free interest rate and the variance of the underlying stock’s return. For the purpose of real options valuation, we can translate these concepts from financial instruments to the expected value of the investment project, the ‘window of opportunity’ in which the project can be executed and the variance in the projects expected value. Hence, the real option value of delay increases with the uncertainty regarding project value.

9

payoff threshold required to invest because the option value of delay increases, but the probability of actually reaching this threshold level also increases due to the increased variability of payoffs. Sarkar (2000) then proposes that the positive effect of increased probability initially dominates the negative option-value effect, whereas the option-value effect becomes dominant at higher levels of volatility.

2.2. Uncertainty, real options and firm value

When we interpret a firm’s set of investment opportunities as a portfolio of call options on real assets (Myers, 1977), we can visualize total firm value as the sum of the value of assets in place and the value of the firm’s real options portfolio. This concept is further developed by, among others, Pindyck (1988) and Kester (1984), who argue that this set of options constitutes a significant portion of total firm value. The prediction that follows from our discussion on real options theory is that, ceteris paribus, an increase in uncertainty increases firm value through the value of the firm’s real options portfolio, implying a causal relationship between uncertainty and firm value. Naturally, there are different channels through which uncertainty influences firm value, besides changes in option values, which are to be accounted for when one wishes to test the uncertainty-firm value relationship.

10

reducing firm value (Stulz, 1990; Froot et. al, 1993). Finally, it is useful here to make a distinction between systematic and unsystematic risks, which could have differential effects on firm-value. For example, Fama & French (1993) argue that growth firms have higher betas and face higher systematic risks, pointing towards a positive relationship between systematic risk and firm-value as measured by Tobin’s Q.

2.3. Empirical strategies and results

Having developed the main theoretical concepts, we now turn to an investigation of previous empirical investigations that are relevant to our research. We discuss the empirical results and performance of the investment models discussed in the literature before moving on to a discussion of earlier tests of the uncertainty-firm value relationship. The empirical contributions discussed in this section serve, in turn, as the foundations of the methodology employed in this research.

Previous empirical work employing error-correction specifications includes Bond et al. (2003), Bloom (2006) and Guariglia (2008). Consistent with theoretical predictions, these authors find sales growth and lagged capital stock to exert a significant influence on investment. Importantly, the error-correction term is found to be significant and correctly signed, with a capital shortfall increasing investment whereas a capital surplus depresses investment rates. Moreover, it is found by Bond et al. (2003) and Bloom (2006) that capital stock and sales are cointegrated non-stationary series, whereas the error-correction term is non-stationary. These results are consistent with error-correcting behavior as they suggest long-run proportionality of capital stock and sales. Using a similar model, applied to firm-level investment data from the United States and France, Mairesse et al. (1999) find the error-correction specification to empirically outperform, in terms of fit, standard neoclassical accelerator models of investment while providing a more parsimonious specification.

11

the associated coefficients are found to be rather small in magnitude, granting the models low explanatory power and fit. Moreover, all these studies include other variables, notably liquidity-related variables such as cash flows, leverage and the presence of collateral, which are found to be strongly significant in combination with Q. This is inconsistent with the notion that, under the assumption of efficient markets, one would expect such considerations to enter the model via the market valuation measured by Q. A potential explanation of these findings would be the before mentioned measurement error associated with Q. Particularly relevant studies are those of Leahy & Whited (1995) and Baum et al. (2008) who use Q-models to test the investment-uncertainty relationship using risk measures derived from stock prices. Baum et al. (2008) find firm-specific uncertainty measures, based on individual stocks, to exert a negative influence in combination with Q, whereas market-based risk measures, based on the market index, are found to exert a positive influence. In contrast, Leahy & Whited (1995) find these risk measures to lose significance when combined with Q. Finally, Cummins et al. (2006) attempt to tackle the criticism of measurement error by using a unique dataset of analyst forecasts to calculate an alternative measure of Tobin’s Q based not on actual market values but on firm-values calculated based on these forecasts. The use of this alternative measure significantly increases the fit of their model and the explanatory power of Q, supporting the criticism of Q on the basis of measurement error. All though uncertainty effects were not investigated in combination with this alternative measure, results could potentially differ when uncertainty is more appropriately priced into this measure of Q.

12

studies find that increasing adjustment costs play a significant role through a significant coefficient on squared past investment. Moreover, significant positive coefficients on the sales-to-capital ratio are found, consistent with predictions on the role of marginal productivity.

Regarding the role of liquidity and financing constraints, a pioneering contribution is that of Fazzari et al. (1987), who use dummies to separate dividend-paying and non dividend-paying firms in order to distinguish between financially constrained and non financially constrained firms. Fazzari et al. (1987) find a positive coefficient on cash flow that is both statistically and economically significant. Furthermore, they find that this coefficient is significantly greater for more financially constrained firms. In a similar vein, Hubbard et al. (1995) find that a standard Euler-equation model is appropriate for a subsample of financially unconstrained firms, measured by dividend payouts, whereas it performs poorly using subsamples of financially constrained firms, indicating that capital market imperfections play a role. Whited (1992) includes the leverage ratio and the ratio of interest payments to cash flow in a Euler-equation model and finds these measures of financial constraints to have a significantly negative relationship to investment. A similar result is found by Aivazian et al. (2005) who find leverage to have a significant, negative coefficient when integrated in a Q-model. Carpenter & Guariglia (2008) and Almeida & Campello (2007) find cash flows to have a significant positive coefficient in Q-models, while Love (2003) and Bond et al. (2003) present results where cash flow carries a positive, significant coefficient in Euler-equation models and error-correction specifications. As a general remark, nearly all of the empirical investigations discussed in this review find a significant, positive influence of cash flows on investment.

13

literature. Yoon & Ratti (2011) use an error-correction specification to test for the influence of fuel price uncertainty on U.S. firm level investment and interact fuel price uncertainty with changes in sales, they find energy price uncertainty to reduce the sensitivity of firm investment to changes in demand. An interesting work on macro-events, such as terrorism and political events, is Bloom (2009). The general finding of these studies is a negative relationship between uncertainty and investment. Studies that find evidence of a positive investment-uncertainty relationship are Mohn & Misund (2009), who find oil price volatility to have a significant positive influence on investment in the oil & gas industry, and Huizinga (1993), who distinguishes between input- and output price uncertainty and finds a positive investment-uncertainty relationship for output price uncertainty and a negative relationship for input price uncertainty. This is consistent with the findings of Koetse et al. (2009), who perform a meta-analysis of results from the literature on investment under uncertainty and find most studies to predict a negative investment-uncertainty relationship, all though a significant minority is found to point towards a positive relationship. Hence, the empirical literature is generally supportive of an irreversibility effect where uncertainty reduces investment with firms holding on to their option to delay. Apparently, this effect dominates any potential positive counter-effects that arise from the creation of a put-option on new investment projects or convex functions that raise the expected payoffs of projects (Carruth et. al, 2000). Results supportive of a non-monotonic relationship include Henriques and Sadorsky (2011), who report a significant positive effect of squared oil price volatility on investment, providing evidence for a u-shaped relationship as predicted by the strategic options literature. Bo & Lensink (2005), using an accelerator model of investment for a sample of firms from the Netherlands, find empirical evidence of an inverted u-shaped relationship between investment and uncertainty, where investment increases at low levels of uncertainty and decreases at higher levels.

14

on firm value, as measured by Tobin’s Q, via real options. As discussed in our review of theoretical concepts, there are other channels through which uncertainty potentially influences firm value that need to be accounted for when one wishes to test the uncertainty-option value relationship. Shin & Stulz (2000) find no evidence supporting an increase in equity values based on the options approach to equity developed by Merton (1972), as the effects of risk on firm-value do not differ for different levels of leverage. Rountree et al. (2008) find that cash flow variance has a highly significant negative relationship to firm value, both statistically and economically, consistent with basic predictions from financial theory that risk-averse investors apply a higher discount rate to risky cash flows. Evidence on the uncertainty-firm value relationship via channels of financial distress costs can be found in the literature on hedging practices, where hedging is found to be value-enhancing by, among others, Nelson et al, (2005) and Allayannis & Weston (2001).

15

expenditures and sales growth, which are all found to be positively related to firm-value. Szewczyk et al. (1996) conduct an event-study to establish that abnormal stock returns following announcements of increased R&D expenditures are largest for firms with a high Tobin’s Q. Their results are supportive of the investment opportunities hypothesis of R&D that R&D approximates growth opportunities.

2.4. Hypotheses

For the first stage of research, we assess the direct relationship between natural gas price uncertainty and investment. Based on the real options theory of irreversible investment, it is to be expected that there is a negative relationship between natural gas price volatility and investment. This expectation finds strong support in empirical contributions on the subject. Moreover, one would expect any effect of energy price uncertainty to be strongest for firms that operate in energy intensive industries, a result that is indeed found by Yoon & Ratti (2011). From this it follows that we can formulate the following hypotheses on the relationship between natural gas price uncertainty and investment:

Hypothesis I: There is a negative relationship between natural gas price uncertainty and investment. Hence, we expect a negative and significant coefficient on natural gas price volatility in any investment specification.

Hypothesis II: This negative effect is most pronounced for firms in energy-intensive industries. This effect is to be reflected in a negative and significant coefficient on the interaction between natural gas price volatility and8 an energy-intensive industry dummy in any investment specification.

16

to be controlled for. Consistent with our earlier predictions, we would expect a potential value-enhancing effect of natural gas price uncertainty through growth opportunities to be most pronounced in energy-intensive industries. This leads to the formulation of the following testable hypotheses:

Hypothesis III: There is a positive relationship between natural gas price uncertainty and the value of investment opportunities. Hence, we expect a positive and significant coefficient on the interaction between natural gas price uncertainty and growth opportunities in our firm-value specification.

Hypothesis IV: The positive relationship between natural gas price uncertainty and the value of investment opportunities is most pronounced for firms in energy intensive industries. This effect is to be reflected in a positive and significant coefficient on the interaction between natural gas price volatility, growth opportunities and an energy-intensive industry dummy in our firm-value specification.

3. Model specification

Having reviewed the theoretical concepts and earlier empirical contributions we now turn to specifying the models to be estimated in order to test our hypotheses. Firstly, we specify the models employed in order to test the investment-uncertainty relationship. Subsequently, the model employed to test the uncertainty-option value relationship is developed.

17

investment-cash flow relationship by including cash flows as an explanatory variable. The three baseline specifications are given below:

(1)

(2)

(3)

The dependent variable for each specification is capital expenditure I for firm i in year t, scaled by the firm’s stock of physical capital, property plant and equipment, K. S denotes firm output, as measured by sales, while s and k are the logarithms of sales and the capital stock respectively. CF denotes a firms operating cash flow and Q is the ratio of a firm’s market value to the replacement cost of its assets, i.e. Tobin’s Q. Finally, as in conventional notation, ∆ is the first difference operator, α0 is the intercept term and ε is the regression error term. A more detailed

description of all variables used is found in table 1 below.

Equation (1) gives the error-correction specification. In this specification scaled capital expenditures are regressed upon a lagged value of itself, as is common in implicit models in order to account for expectations and investment dynamics. Following Bond et al. (2003) and Guariglia (2008), we model the accelerator principle, investment as a response to changes in demand, with the inclusion of the logarithm of current and lagged changes in sales in order to approximate changes in demand. As in Guariglia (2008) and Bond et al. (2003) the error-correction term measures the difference between the logarithm of capital and sales lagged by two years and serves as an approximation of the deviation of the capital stock from its desired level in that period. Following the convention in the literature, liquidity considerations enter the specification via the inclusion of scaled cash flows.

18

(1994), the expectation of future investment enters the specification via the inclusion of the future realized value as an explanatory variable. The intuition behind the inclusion of future expectations regarding investment is that they capture the extent of adjustment costs associated with postponing investment, where higher future investment is expected to discourage delay of investment and increase current investment. A severe drawback of using the Euler equation specification here is that we lose the most recent year of available data in estimation. We follow Bond et al. (2003), Love (2003) and Bond & Meghir (1994) by including the ratio of sales to capital, reflecting the opportunity cost of postponing investment in terms of the marginal product of capital, and scaled cash flows in order to capture the sensitivity of investment to cash flow and liquidity.

Equation (3) presents a Q-model of investment, based on the baseline specifications of Carpenter & Guariglia (2008) and Baum et al. (2008). Once again, the investment rate is regressed upon a lagged value of itself to capture serial correlation and investment dynamics. The core of the model is the inclusion of Tobin’s Q, which approximates the market valuation of additional assets installed by the firm, and is expected to exert a positive influence on investment rates. Following Carpenter & Guargilia (2008) and Baum et al. (2008) we include scaled cash flows to account for the widely documented sensitivity of investment to cash flows in combination with Tobin’s Q.

19

macroeconomic developments, both at the national and international level (Asprem, 1989; Fifield et al., 2002; Kwon & Shin, 1999; Tsouma, 2009). We follow the notion in the literature that stock index returns are a leading indicator of economic performance by including a one-period lag. To test our hypotheses on energy-intensive industries, a dummy variable is included to account for a different effect of natural gas price volatility on investment in the most energy-intensive industries. Finally, we include an interaction with domestic assets as a fraction of total assets owned by the firm in order to account for the fact that multinational firms will run a significant portion of their operations in countries outside Europe, different conditions on natural gas markets will apply to these operations due to the separation of global gas markets

For sake of brevity, we drop the time and firm subscripts unless the variable is a lagged or future variable. The three model specifications now read:

(4)

(5)

(6)

In the specifications above σS denotes the volatility of a firm’s stock returns, intended to capture

the effect of general uncertainties, whereas σI is the volatility of stock index returns, intended to

capture macroeconomic uncertainty. σG is our primary variable of interest and represents the

volatility of natural gas prices. RI denotes the annual return on the national stock index, included to capture macroeconomic circumstances that would otherwise correlate with σG. DA is the

20

capturing any stand-alone effects of these variables in the interaction terms. In line with our hypotheses we would expect the coefficients on σG, the interaction between DA and σG, and the

interaction between EI and σG to be significant and of negative sign. If we find these coefficients

to be simultaneously negative and significant, this would imply that firms reduce their capital expenditures in response to natural gas price volatility, with the effect being most pronounced for firms that run the bulk of operations domestically and for energy intensive industries.

We now turn to the second stage of our research, which is to investigate the relationship between uncertainty and the value of real options. Following the empirical contributions discussed in the previous section, Tobin’s Q is used as an approximation of firm value. As discussed previously, there exists a vast body of empirical work on the determinants of Tobin’s Q. The goal of the specification developed here is to provide a parsimonious specification that is fit to test the effect of uncertainty on Tobin’s Q through the value of growth opportunities. We follow the works of López-Iturriaga & Rodríguez-Sanz (2001) and Yermack (1996) by using return on assets, as an approximation of firm profitability, the leverage ratio and firm size as control variables in determining Q. Building on the works of Lang & Stulz (1993), Morck et al. (1988), Kim & Lyn (1996) and Szewczyk et al. (1996) we include R&D expenditures as an approximation of growth opportunities. The baseline specification is given in equation 7 below:

(7)

As before Q denotes Tobin’s Q, whereas and denote the intercept and regression error term respectively. ROA denotes the return on assets, A represents total assets and enters the specification as a logarithm, L gives the leverage ratio. Finally, RD denotes R&D expenditures, which enter the specification scaled by total assets. A more detailed description of all variables used is found in table 1.

21

intensive firms, which we expect to have valuable growth opportunities. This allows us to isolate the value-effect of uncertainty via growth opportunities from other firm-value effects of uncertainty. As before, macro-economic circumstances are accounted for by including stock index returns and -volatility. In order to test our hypotheses regarding energy intensive industries, we once again include a dummy variable for energy-intensive industries, both separate and in interaction. As before, the fraction of domestic assets is included to account for operations in different natural gas markets. Each variable used in interaction is also included separately to prevent capturing stand-alone influences in the interaction. The extended model is specified in equation 8 below. Once more, we drop all firm-year subscripts unless a variable is lagged.

(8)

In line with our hypotheses on the real option value of investment opportunities we would expect positive and significant coefficients on the interactions between general uncertainty, natural gas price uncertainty and our R&D intensity dummy. As we expect this effect to be stronger for firms with domestic assets, we would also expect a positive coefficient on the three-way interaction between natural gas price volatility, domestic assets and R&D intensity. Finally, in line with our hypothesis on sector-specific effects we would expect a positive and significant coefficient on the three-way interaction between the energy intensive industry dummy, natural gas price volatility and R&D intensity. Such findings would imply that the market valuation of a firms growth opportunities increases with both general and natural gas price uncertainty, and that the latter relation is strongest for firms with a large proportion of domestic assets that operate in energy intensive industries.

4. Data

22

Thomson Reuters Datastream, as well as our data on stock indices intended to proxy for macroeconomic circumstances. Data on natural gas prices comes from different sources. Monthly BAFA price data, the average monthly price of gas imported to Germany, comes from publicly available data published by the German Federal Office of Economics and Export Control. Daily hub-price data is not publicly available, but here we employ daily hub-prices on the British NBP hub, kindly made available by Royal Dutch Shell.

Whereas the British natural gas market has been characterized by liberalization and hub-based price formation since the late 1990s, natural gas hubs in continental Europe have only recently developed into sufficiently developed, mature and liquid trading platforms (Stern & Rogers, 2011; Petrovich, 2013). For our research, this implies that hub prices are only suitable for an analysis of the investment-uncertainty relationship from the late 2000s onwards. As noted earlier, the dominant gas prices in the period preceding the development of hub-based price formation were gas prices formed according to long-term contracts. The so-called ‘BAFA price’, a monthly series of the average monthly price of gas imported into Germany, collected by the German federal government, is the only transparent and publicly available approximation of natural gas contract prices. For the British market, prices formed on the National Balancing Point (NBP), the primary natural gas hub of the U.K., are widely regarded as the prevailing market price for natural gas.

23

Table 1: Description of variables

Detailed description of all variables employed in estimation. Datastream variable codes, if applicable, are given in brackets.

Variable Definition

I Capital expenditures: funds used to acquire fixed assets other than those obtained in acquisitions (WC04601).

K Capital stock: net property, plant & equipment (PPE), defined as gross PPE minus accumulated reserves for depreciation and amortization (WC02501).

S Net sales: gross sales and other operating revenue minus discounts, returns and allowances (WC01001).

CF Operating cash flow: represent the sum of cash receipts from operations, working capital changes and extraordinary items (WC04860).

Q Tobin’s Q: the ratio of market value to the replacement cost of assets, defined as the sum of equity market value (WC08001) and total debt book value (WC03255) divided by book value of total assets (WC02999).

σs Annual volatility of a firm’s stock price, measured as the percentage variation of

the annual high and low prices from the annual mean (WC08806).

σG Natural gas price volatility. This is defined as the annualized standard deviation of

logarithmic returns on natural gas. For the German sample this is calculated based on monthly return series of BAFA prices, for the U.K. sample this is based on daily return series of NBP hub prices.

ROA Return on Assets, calculated as net income divided by total assets, multiplied by 100 to provide a percentage (WC08326).

A Total book value of assets on the firm’s balance sheet (WC02999).

L Leverage ratio, calculated as the total book value of debt (WC03255) divided over the book value of total assets (WC02999).

RD Total R&D expenses, including all direct and indirect R&D cost with potential commercial applications (WC01201).

RDI A dummy for R&D intensive firms, which equals 1 when a firm is in the top 3 deciles of the ratio of R&D expenditures (WC01201) over total assets (WC02999) and 0 otherwise.

DA A firms fraction of domestic assets, calculated as 100 minus the percentage of international assets on the balance sheet, divided by 100 (WC08736).

EI Dummy variable taking a value of 1 when a firm operates in an energy-intensive industry and 0 otherwise, industries considered energy-intensive are defined as such by the ODYSSEE MURE project (2012). A more detailed description of sectors can be found in table A1 in the appendix.

σI Annualized standard deviation of logarithmic weekly total returns of the relevant

national stock index. This is the HDAX index for our German sample [PRIMHDX(RI)] and the FTSE All Stocks index for the British sample [FTALLSH(RI)]

24

Our German sample consists of 827 non-financial firms over the 1989-2012 period and is used in combination with the available time series on BAFA-prices. The time period is chosen so that it matches the available time-period of BAFA prices, 1991-2012, while allowing for the use of lagged variables dating two years back. Moreover, two different samples are used for our two stages of research, the investment-uncertainty and uncertainty-firm value relationship respectively. This is due to the fact that Datastream coverage for Germany is not complete. Given that both model specifications require different variables, the common samples differ. The key-variable of interest, natural gas price volatility, is measured by the annualized standard deviation of logarithmic monthly BAFA price returns. Tables 2 and 3 below provide descriptive statistics on the samples used for investment- and firm-value estimation respectively. Table A1 in the appendix presents a detailed description of sectors, according to the Thomson Reuters Business Classification. Table B1 and B2 in the appendix provide sample correlation matrices.

Our second sample is a U.K. sample of 1120 non-financial firms over the 1995-2012 period. Company data is used in combination with data from the NBP hub over the 1997-2012 period. The time-period for company data is chosen so as to have the option to use lagged values dating 2 years back, while retaining our data on natural gas prices. As the natural gas market of the U.K. is relatively developed, the NBP hub is widely regarded as a good benchmark for the prevailing market price in the U.K. (Petrovich, 2013; Stern & Rogers, 2011). Once more, two different samples are used for the investment estimation and the firm value estimation to account for the incomplete coverage of Datastream. Tables 4 and 5 provide descriptive statistics on the samples used for investment- and firm value estimation respectively. Table A1 in the appendix presents a detailed breakdown by sector of all observations in our dataset. Tables B3 and B4 in the appendix present sample correlation matrices. When comparing the German and U.K. samples, it is noteworthy that all data for the British sample is given in Pound Sterling, whereas for the German sample these data are denoted in Euro.

25

more recent years for which data availability is generally best. Secondly, the minimum and maximum values of each variable tell us that there are a few large outliers that appear to be driving the mean, given the sizeable differences between mean and median values for several variables. This indicates that we need to account for outliers when performing our estimations. Moreover, the average firm-year observation in our sample invests 30.8% of its physical capital stock each year and is valued only slightly above the value of its assets in place. Furthermore, it is noteworthy that the average observation in the two samples has 73.2% or 68.6% of its total assets in domestic assets. Finally, 11.7% of the observations in the investment sample belong to the most energy-intensive industries, compared to 14.6% for the firm-value sample. Looking at the correlation matrices in the appendix we can learn that multicollinearity is not likely to be a large issue for most explanatory variables. Striking observations are the strong correlations between capital, sales and cash flows. All though the correlation between sales and cash flows is hardly surprising, the correlation between sales and capital is supportive of the proportionality of capital and sales that underpins error-correction models. Finally, we can observe a rather strong correlation between Tobin’s Q and R&D expenditures, supportive of the notion that R&D expenditures reflect growth opportunities.

26

before in the German sample, strong correlations between sales and capital, as well as between Tobin’s Q and R&D expenditures.

5. Results

Having developed our models, hypotheses and datasets in the preceding sections we now turn to the estimation of our model specifications and inferences regarding our hypotheses. We first look at the results found when estimating our investment specifications in equations (4), (5) and (6), after which we look at the results from our firm-value specification in equation (8).

27

Table 2: Descriptive statistics, investment specification, Germany 1991-2012

Descriptive statistics based on our German sample of 827 non-financial firms over the 1991-2012 period. All descriptive statistics are based on the common sample, i.e. the observations for which all variables are available in the dataset, of the variables used in direct estimation of investment ratios. All variable definitions are as in table 1.

Table 3: Descriptive statistics, firm value specification, Germany 1991-2012

Variable Q ROA(%) A (x1000) L RD/A σS σG DA EI σI RI

Mean 1.118 4.166 10006985 0.206 0.041 32.413 0.100 0.686 0.146 0.220 0.060 Median 0.851 5.205 524629 0.182 0.023 30.440 0.094 0.727 0.000 0.176 0.187 Minimum 0.152 -262.370 2157 0.000 0.000 7.240 0.058 -0.629 0.000 0.093 -0.578 Maximum 16.261 90.530 301729000 1.052 2.320 75.490 0.217 1.000 1.000 0.401 0.342 Std. Dev. 1.129 13.567 28604935 0.166 0.081 11.688 0.043 0.250 0.353 0.086 0.284 Observations 1314 1314 1314 1314 1314 1314 1314 1314 1314 1314 1314

Descriptive statistics based on our German sample of 827 non-financial firms over the 1991-2012 period. All descriptive statistics are based on the common sample, i.e. the observations for which all variables are available in the dataset, of the variables used in the estimation of firm value. All variable definitions are as in table 1.

28

Table 4: Descriptive statistics, investment specification, United Kingdom 1997-2012

Variable I/K K (x1000) S(x1000) Q CF/K σS σG DA EI σI RI Mean 0.485 791297 1727975 1.471 -5.456 34.195 0.569 0.755 0.157 0.173 0.056 Median 0.185 21560 107671 0.985 0.267 31.460 0.582 0.880 0.000 0.172 0.128 Minimum 0.000 1 -34300 0.081 -5255.333 4.000 0.244 -7.233 0.000 0.084 -0.404 Maximum 487.475 106000000 293000000 96.754 1336.222 84.170 0.967 2.983 1.000 0.354 0.306 Std. Dev. 8.560 4539005 10764221 2.372 98.248 13.433 0.200 0.328 0.364 0.067 0.193 Observations 5980 5980 5980 5980 5980 5980 5980 5980 5980 5980 5980

Descriptive statistics based on our U.K. sample of 1120 non-financial firms over the 1997-2012 period. All descriptive statistics are based on the common sample, i.e. the observations for which all variables are available in the dataset, of the variables used in direct estimation of investment ratios. All variable definitions are as in table 1.

Table 5: Descriptive statistics, firm value specification, United Kingdom 1997-2012

Variable Q ROA(%) A (x1000) L RD/A σS σG DA EI σI RI

Mean 1.678 -1.620 3996949 0.187 0.062 34.935 0.563 0.716 0.136 0.171 0.058 Median 1.115 5.425 131950 0.154 0.019 31.840 0.555 0.766 0.000 0.165 0.128 Minimum 0.104 -416.380 69 0.000 0.000 8.610 0.244 -0.519 0.000 0.084 -0.404 Maximum 45.931 255.280 219000000 3.143 3.584 83.970 0.967 2.983 1.000 0.354 0.306 Std. Dev. 2.328 30.898 17374854 0.216 0.146 13.778 0.207 0.293 0.343 0.067 0.191 Observations 2434 2434 2434 2434 2434 2434 2434 2434 2434 2434 2434

29

As we saw in the data section, both our data samples contain some severe outliers. Outliers of this magnitude would severely penalize any OLS estimation. Hence, we follow Bond et al. (2003) by excluding all observations where the investment ratio exceeds unity, or goes below zero. Moreover, we bind the ratio of domestic assets to lie between 0 and 1. Furthermore, Tobin’s Q, cash flow ratios and sales ratios are cut at their 0.025th and 0.975th percentiles, removing the observations beyond these percentiles, which is a common practice in the literature. All models are estimated using White heteroskedasticity robust standard errors, as the White test, not reported here, indicates that all our estimations face heteroskedasticity. Finally, the Jarque-Berra test, not reported here, indicates non-normality for the residuals of all our estimations. However, given the rather large amount of observations employed in estimation, we can rely on the central limit theorem to assume that non-normality of our residuals will be virtually inconsequential for inference.

Table 6 presents the results for our German sample. Models (1), (3) and (5) report the pooled OLS estimations of the error-correction, Euler-equation, and Q-model specifications of investment respectively. All variables are reported as in the original model specifications, where C denotes the intercept term. Models (2), (4) and (6) report the results for the error-correction, Euler-equation and Q-model respectively, accounting for firm fixed effects.

30

Alternatively, it could be that equity risk increases with growth opportunities as argued by Cao et al. (2008), and that our current controls do not capture all firm-specific growth opportunities.

Looking at our variable of interest, natural gas price volatility, we obtain mixed results. Whereas the standalone coefficient on natural gas price volatility (σG) turns out significantly negative,

consistent with our hypotheses, we see that natural gas price volatility in interaction with domestic assets show a significantly positive relationship, as well as in interaction with the energy-intensive industry dummy. Based on our knowledge of imperfectly integrated natural gas markets, we would expect any causal relationship between investment and natural gas price volatility to be most likely for a firm’s domestic operations rather than for its foreign operations, which are not affected by domestic gas markets. The magnitude of the coefficient on the interaction between σG and the domestic asset ratio is large enough to completely cancel out

the negative effect of standalone σG for firms that operate purely domestically (i.e. where DA

equals 1). Combined with the highly significant and positive coefficient on the interaction with the energy intensity dummy, the evidence in table 6 is mostly supportive of a positive relationship between natural gas price volatility and investment for domestic and energy intensive firms, contrary to our hypotheses.

We now turn to the results for the U.K. sample, presented in table 7. Again, models (1), (3) and (5) present the results for standard pooled OLS estimation, whereas models (2), (4) and (6) report the results of our firm fixed effects estimations. The findings for the control variables are rather similar to that of our German sample, with all variables from the baseline specifications being significant and of expected sign, with the exception of cash flows and net sales ratios. As before, all coefficients of our control variables except for that of stock price volatility do not change in sign or significance with the inclusion of fixed effects. The results on natural gas price volatility oppose our results from Germany. We observe a positive coefficient on σG, which is not

31

Based on the strategic options theory of Folta & O’Brien (2004), we would expect a u-shaped relationship between investment and uncertainty, as is found by Henriques & Sadorsky (2011) to apply to oil price volatility in the U.S. Ignoring the presence of such a non-linear effect could influence our results and the sign of our coefficients. Therefore, we perform robustness checks for both our samples, including quadratic terms of natural gas price volatility to account for the possibility of a u-shaped relationship. Moreover, the primary challenge for our empirical research is the potential correlation of natural gas price volatility with other year-specific effects. To test the robustness of our results to the inclusion of a different approximation of year-specific macro effects we replace lagged stock index returns in our estimations with GDP growth, drawn from publicly available IMF data sources. Tables 8 and 9 below provide these robustness checks for our German and U.K. samples respectively, where all estimations account for firm fixed effects.

32

Table 6: Regression results, investment specifications, Germany 1991-2012

Regression results for the investment specifications of equation (4), (5) and (6), estimated using our sample of 827 non-financial German firms over the 1991-2012 period. Models (1), (3) and (5) present the results for pooled OLS estimation for the error-correction, Euler and Q-models respectively, whereas models (2), (4) and (6) present the estimation results including firm fixed effects. All models are estimated using White, heteroskedasticity robust, standard errors. The dependent variable for each model is (I/K). To deal with outliers, investment ratios and domestic asset ratios are truncated to lie between 0 and 1, Q, (NS/K) and (CF/K) are cut at their 0.025th and 0.975th percentiles to remove extreme observations. *, ** and *** denote significance at the 1%, 5% and 10% level respectively. All variable definitions are as in table 1.

33

Table 7: Regression results, investment specifications, United Kingdom 1997-2012

Regression results for the investment specifications of equation (4), (5) and (6), estimated using our sample of 1120 non-financial U.K. firms over the 1997-2012 period. Models (1), (3) and (5) present the results for pooled OLS estimation for the error-correction, Euler and Q-models respectively, whereas models (2), (4) and (6) present the estimation results including firm fixed effects. All models are estimated using White, heteroskedasticity robust, standard errors. The dependent variable for each model is (I/K). To deal with outliers, investment ratios and domestic asset ratios are truncated to lie between 0 and 1, Q, (NS/K) and (CF/K) are cut at their 0.025th and 0.975th percentiles to remove extreme observations. *, ** and *** denote significance at the 1%, 5% and 10% level respectively. All variable definitions are as in table 1.

34

Table 8: Robustness checks, Germany 1991 - 2012

Robustness checks for the investment specifications of equation (4), (5) and (6), estimated using our sample of 827 non-financial German firms over the 1991-2012 period. Models (1), (3) and (5) present the results for fixed effects estimation for the error-correction, Euler and Q-models respectively, accounting for a potential u-shaped investment-uncertainty relationship. Models (2), (4) and (6) present the fixed effects estimation results accounting for a different proxy of year-specific effects. All models are estimated using White, heteroskedasticity robust, standard errors. The dependent variable for each model is (I/K). To deal with outliers, investment ratios and domestic asset ratios are truncated to lie between 0 and 1, Q, (NS/K) and (CF/K) are cut at their 0.025th and 0.975th percentiles to remove extreme observations. *, ** and *** denote significance at the 1%, 5% and 10% level respectively. All variable definitions are as in table 1.

35

Table 9: Robustness checks, United Kingdom 1997 - 2012

Robustness checks for the investment specifications of equation (4), (5) and (6), estimated using our sample of 1120 non-financial U.K. firms over the 1997-2012 period. Models (1), (3) and (5) present the results for fixed effects estimation for the error-correction, Euler and Q-models respectively, accounting for a potential u-shaped investment-uncertainty relationship. Models (2), (4) and (6) present the fixed effects estimation results accounting for a different proxy of year-specific effects. All models are estimated using White, heteroskedasticity robust, standard errors. The dependent variable for each model is (I/K). To deal with outliers, investment ratios and domestic asset ratios are truncated to lie between 0 and 1, Q, (NS/K) and (CF/K) are cut at their 0.025th and 0.975th percentiles to remove extreme observations. *, ** and *** denote significance at the 1%, 5% and 10% level respectively. All variable definitions are as in table 1.

37

In general, the evidence from our estimations is rather mixed. We can observe that our findings on the relationship between natural gas price uncertainty and investment are rather different for our two samples, with evidence from Germany pointing mostly towards a linear, positive relationship and the evidence from the U.K. pointing towards an inverse u-shaped relationship. Furthermore, the relationships are not always completely robust to the choice of specification. Also, the impact of the inclusion of a different proxy for macroeconomic circumstances is far less consequential for the U.K. sample relative to the German sample. A potential explanation could be that the FTSE All Share index is a broader index than the HDAX index, and that the importance of listed firms is greater for the U.K. in general, making index returns a better approximation of macroeconomic circumstances.

38

Table 10: Robustness checks, large firm sample, United Kingdom 1997-2012

39

Looking at the results for our German sample in table 11, model (1), we see that return on assets and the ratio of R&D expenses are significant and of expected sign, whereas leverage and size are insignificant in the explanation of Tobin’s Q. Contrary to our expectations, firm-value is positively related to stock price volatility in our German sample, whereas natural gas price volatility exerts no significant influence on its own. With regard to our main variables of interest, the interactions between uncertainty and our dummy for R&D intensive firms, we find little significant results besides a significantly negative coefficient on the interaction between natural gas price volatility, R&D intensity and the energy-intensive industry dummy. This would indicate that natural gas price uncertainty reduces the value of growth opportunities for energy intensive firms. We are to be cautious with the interpretation of this result however, as the number of observations on energy intensive, R&D intensive firms is rather small, consisting of 44 firm-year observations, and runs a serious risk of being biased. In general, the fact that we have no significant findings for any of the other interactions is not supportive of the hypothesis that uncertainty increases the value of growth opportunities. This is inconsistent with the general findings from our investment specifications estimated earlier, where we find uncertainty to boost investment.

40

uncertainty increases the value of investment opportunities, which can, in turn, increase investment.

Table 11: Regression results, firm-value estimations

Germany United Kingdom

(1) (2) C _{0.798*** 2.156***} ROA _{0.029*** -0.008**} log(A) _{0.005 -0.033*} L _{-0.133 -0.156} RD/A _{4.773*** 2.210***} σS 0.006*** -0.015*** σG -0.108 0.658*** DA _{-0.224 -0.001} (DA * σG) _{1.937* -0.664*} EI _{0.223 0.303} (EI * σG) -0.789 -0.554* (σs * RDI) 0.003 0.011** (σG * RDI) -0.886 -0.900 (EI * σG * RDI) -6.285*** 1.168 (DA * σG * RDI) -1.988 1.071* σI -1.157*** -1.271*** RIt-1 0.033 0.526*** Adj. R2 0.209 0.132 Incl. Obs. 920 1567

41

6. Discussion and implications

In line with the predictions of real options theory under irreversible investment, we expected a significantly negative relationship to exist between natural gas price volatility and investment, consistent with the earlier findings of Henriques & Sadorsky (2011) and Yoon & Ratti (2011) on energy price volatility. The evidence from our estimations is, however, rather mixed. Our results differ markedly between the two country samples employed, and these differences are found to be robust to accounting for the different time period of estimation as well as size effects. We now move on to a further discussion and interpretation of the results at hand.

42

of a value-enhancing effect of uncertainty on the value of growth opportunities turns out indecisive, with no evidence found on a significant relationship besides a small subset of R&D intensive, energy-intensive firms.

For our U.K. sample, the results of our initial specifications point toward a positive relationship between natural gas price volatility and investment, and a negative relationship for firms that operate in energy-intensive industries. The fraction of domestic assets, contrary to our expectations, does not appear to influence the relation. These results are not robust, however, to the inclusion of a squared volatility term. When accounting for a potential u-shaped relationship, the results for our standalone relationship points towards a u-shaped relationship as predicted by Folta & O’Brien (2004). In interaction with the domestic asset ratio and the energy-intensity dummy, however, we find evidence of an inverse u-shaped relationship, where investment increases at lower levels of natural gas price volatility and decreases only at higher levels. Finally, the results for our U.K. sample prove to be not entirely robust to the inclusion of a different macroeconomic proxy, although with less dramatic changes as with the German sample, indicating that stock-index based measures may be more appropriate macroeconomic proxies for the U.K. As the coefficients on international and domestic firms strongly oppose each other, this could mean that we face the same issues as before with accounting for other year-specific effects. When disregarding the standalone effects of σG, and taking into account the

u-shaped relation, the results indicate that for a purely domestic firm an increase in natural gas return volatility from 0 to 10 percentage points leads, on average, to an increase in investment ratios of 2.72 or 3.01 percentage points, dependent on the specification. The inflection point of the u-shaped relation for a purely domestic firm lies at a value of σG of 55.99 or 60.10

percentage points, dependent on the specification. For energy-intensive firms, the same increase in volatility leads to an additional 0.85 or 1.73 percentage points increase in the investment ratio, with the inflection point lying at 39.11 or 56.21 percentage points. Again, these results appear economically significant, all though potential counteracting effects of σG as

43

in natural gas return volatility leads to an increase in Tobin’s Q of 0.11, which is a magnitude of high practical significance. Given our evidence of a positive relationship between natural gas price volatility and investment from our investment specifications, these results point towards a value-increase of growth opportunities themselves, encouraging execution of projects, rather than a value-increase of the associated options to delay, which would discourage investment.

44

Koetse et al. (2009) that a sizeable minority of the literature finds a positive investment-uncertainty relationship. Our results from the U.K. that suggest an inverse u-shaped relationship are consistent with the empirical evidence of Bo & Lensink (2005).

With regard to the differences between our samples, there are a few potential explanations for these differences that can be derived from the structural differences between natural gas markets in Germany and the U.K. As volatility for BAFA prices is generally not as high as that of hub-based prices (Stern & Rogers, 2011), it might very well be that BAFA prices are not sufficiently volatile to reveal an inverted u-shaped relationship as for our U.K. sample, as volatility never reaches a sufficient level to exert a negative influence. Moreover, as forwarded by Haase (2009), the U.K. has faced particularly high volatility on its natural gas markets during our period of investigation, due to its shift from being a exporter towards being a net-importer and due to a shortage of storage capacity to mediate price shocks. Another distinct possibility, mentioned by Stern & Rogers (2011), is that in more recent years, contract prices, for which BAFA prices are considered an approximation, do not properly represent the prices that apply to large users of natural gas, such as many of the listed firms under investigation here. The intuition behind this is that following recent gas market liberalization, natural gas utilities cannot pass on the contract prices they pay for gas, as customers gained access to hub-based prices (Stern & Rogers, 2011).

45

As for the short run, it would be too simplistic to conclude that our evidence of a positive investment-uncertainty relationship indicates that natural gas price uncertainty, at least at intermediate levels, is socially desirable. As established earlier, increased uncertainty generally reduces the value of assets in place, making the total effect on welfare ambiguous. In this respect, it is worth mentioning that our research has not considered any potential welfare effects of uncertainty for economic actors other than firms, such as consumers and governments. Moreover, it is not a clear-cut conclusion that an increase in investment leads to a welfare-enhancement per se. In the strategic options literature, investment serves as an entry-deterrent, which implies that investment induced by strategic options influences market structure. Whether the implications for market structure positively or negatively influence social welfare depends, in turn, on market characteristics and assumptions (Thijssen et al., 2006). Moreover, the exercise of strategic options potentially induces suboptimal simultaneous investment (Smit & Ankum, 1993).

7. Summary and conclusions