How Asset Irreversibility Influences the Investment-Uncertainty Relationship

Transcription

1 How Asset Irreversibility Influences the Investment-Uncertainty Relationship Catharina Klepsch 1 Institute for Capital Markets and Corporate Finance, Ludwig-Maximilians-University Munich, Germany Abstract This paper examines how uncertainty affects firms investments for varying degrees of asset irreversibility (i.e. the wedge between purchase price and liquidation value of an asset). To identify more or less irreversible capital goods, we exploit unique survey data on German manufacturing firms over the sample period 2004 to 2012 in which managers provide information on investments purpose (capacity expansion, replacement, restructuring, rationalization, and other). Our results indicate that only investments into the most irreversible capital goods (capacity expansion)) will decrease if uncertainty rises. I also find support for other channels, such as the financial friction or the market power channel, to explain the investment-uncertainty relationship. Keywords: Firms investment behavior, Uncertainty, Irreversibility, Investment types 1. Introduction Several empirical papers find a negative impact of uncertainty on investments (Leahy and Whited, 1996; Guiso and Parigi, 1999; Bloom et al., 2007), which serves as evidence for the real options theory of Bernanke (1983), McDonald and Siegel (1986), Pindyck (1988) and Dixit and Pindyck (1994). Real options theory hereby indicates that under the presence of asymmetric adjustment costs (asset irreversibility), firms reduce investments if demand uncertainty increases. The reason is that irreversibility of capital creates a so-called option value of waiting which is higher during increased economic uncertainty. During times of increased uncertainty, firms will thus prefer waiting for new information, as this prevents firms from investing in projects which seem profitable ex-ante but turn out to be unprofitable ex-post (Koetse et al., 2006). To examine whether irreversibility truly explains this negative investment-uncertainty relation, several studies directly measure irreversibility at the firm-level (Guiso and Parigi, 1999; Chirinko and Schaller, 2009; Guariglia et al., 2012) or at the industry-level (Kim and Kung, 2016) and find that the negative effect of uncertainty on firms total investment outlays are even stronger for firms and industries with a higher degree of asset irreversibility. 1 adress: klepsch@bwl.lmu.de; adress: Schackstr. 4, Munich, Germany.

2 One shortcoming of most of these prior studies is that researchers examine the effects of uncertainty on total investments without taking into account that studying total investment outlays is only appropriate if firms capital goods are homogenous and do not differ in their degree of irreversibility. 2 In reality, however, capital goods are heterogeneous and differ in their degree of specificity, liquidity and adjustment costs (Guiso and Parigi, 1999). Accordingly, the true effects of uncertainty might be concealed, if irreversibility constraints are only binding for some types of capital goods. In this study, we thus explicitly take into account that the degree of asset irreversibility does not only vary across firms or industries (firm/industry-specific irreversibility), but also within firms and therefore across firms assets (asset-specific irreversibility). We therefore examine how varying degrees of asset-specific irreversibility influence the investment-uncertainty relationship. We empirically test how uncertainty affects our investment categories using panel data on private manufacturing firms in Germany over the sample period 2004 to 2012 provided by the Business & Economics Data Center. Our panel data set combines unique survey data from the ifo Investment Survey with financial statement data from either Amadeus or Bisnode. To measure asset-specific irreversibility, we exploit survey data in which managers provide information on the purpose for which investments are acquired. In detail, managers can choose between five investment types/purposes: capacity expansion, rationalization, restructuring, replacement, and other. We assign these five investment types to more and less irreversible investments by taking into account the characteristics and aims specific to each investment type. We argue that capacity expansion investments have the highest degree of irreversibility, as capacity expansions often include the set up of a new production site, which entail high administration and planning costs, and thus, high sunk costs. In contrast, non-capacity investments such as replacement, rationalization and restructuring should entail a low degree of irreversibility, as such investments often only maintain a firms capital stock without affecting firms capacities. Accordingly, planning efforts and thus sunk costs should be rather low, making a small wedge between purchase and resale prices more likely. We measure uncertainty using managers subjective expectations about future sales available from the ifo survey to construct a measure of disagreement dispersion for two-digit industry codes (Fuss and Vermeulen, 2008; Bachmann et al., 2013; Philipp and Röhe, 2016). We find that uncertainty negatively influences firms total investment plans as well as investment plans for capacity expansions. We cannot find an effect of uncertainty on non-capacity investment plans such as replacement, restructuring or rationalization. Our results indicate that capacity expansions are more difficult to revert and thus negatively correspond to increases in uncertainty. We conduct several tests controlling for the robustness of our results. Our tests indicate that our 2 Exceptions are Guiso and Parigi (1999) and Driver et al. (2006). Guiso and Parigi (1999) differentiate between investments in structures, machineries and equipments, and vehicles to account for heterogeneity across investments. The authors find a negative and significant effect of uncertainty on all three types. In contrast, Driver et al. (2006) differentiate between investments in machineries and buildings and find that machineries (which are more irreversible) are more strongly influenced by uncertainty shocks. 2

3 key findings are not sensitive to the empirical specification, to the estimation method applied or to our measure of uncertainty. Beside examining the importance of asset-specific irreversibility, we also explore the effects of firm-specific irreversibility on the investment-uncertainty relation. Therefore, this is the first study showing how uncertainty affects investments through different channels of irreversibility (asset versus firm-specific irreversibility). Our results indicate that uncertainty affects investments through asset-specific and firm-specific irreversibility, as we find that uncertainty also affects non-capacity investments, if we account for firm-specific irreversibility. However, a high degree of firm-specific irreversibility does not amplify the effects of uncertainty on capacity investments. According to Caballero (1991), the relation between investment and uncertainty is only negative under imperfect competition. We therefore also examine how differences in market concentration influence the effects of uncertainty on firms investment types. Our findings indicate that firms in perfectly competitive markets (according to the Herfindahl Hirschmann Index) respond with a stronger decline in investments when uncertainty increases compared to firms in imperfectly competitive markets. Beside the irreversibility channel, which might depress investments if uncertainty increases, investments might also decrease during times of high uncertainty due to the financial friction channel (Ghosal and Loungani, 2000; Minton and Schrand, 1999; Gilchrist et al., 2014). The financial friction channel indicates that with raising uncertainty on key determinants of investments, such as cash flow and marginal or average q, a firms default probability raises which leads to higher external financing costs (von Kalckreuth, 2003). We thus examine how increased uncertainty affects investments if we control for financial frictions. We find that financially constrained firms reduce non-capacity investments during times of increased uncertainty, while financially unconstrained firms do not. Capacity investments are not affect by financial constraints. Our results indicate that the financial friction channel explains declines in investments only for the least irreversible investments. This paper contributes to the investment-uncertainty literature by examining the effects of uncertainty on capacity and non-capacity investments. We argue that capacity expansion and noncapacity expansion investments differ in their degree of irreversibility and thus should be differently influenced by an increase in uncertainty. To the best of our knowledge, we are the first to control for differences in irreversibility across firms capital goods by categorizing investments according to their purpose. Differentiating between capacity and non-capacity investments is important for several reasons. First, we show that non-capacity investments build a large part of total investments, as they account for more than two thirds of firms overall investment outlays. Accordingly, understanding how uncertainty affects non-capacity investments is of relevance due to the sheer size of these investments. In addition, knowing that increased uncertainty only depresses capacity expansion investments allows for a more targeted intervention by policy makers. Accordingly, policy makers should think about interventions which especially affect capacity expansion investments, as higher uncertainty does not affect replacement investments. Finally, we make researchers aware of asset-specific irreversibility 3

4 90 and emphasize that researchers which do not account for asset-specific irreversibility might only find a negative effect of uncertainty on total investments due to aggregation effects. The remainder of this study is organized as follows. In Section 2 we classify investments according to their purpose and derive our hypotheses. Section 3 presents the study s empirical design by describing our data, our uncertainty measure and the empirical specification. Section 4 contains our empirical results. In Section 5 we examine how firm-specific irreversibility, financing constraints and market concentration influence our results. We conclude in Section Contrasting investment types and hypotheses development When analyzing the effects of uncertainty on firms investment behavior, prior studies mainly examine firms total investments (Leahy and Whited, 1996; Bulan, 2005; Bloom et al., 2007) and find a negative impact of uncertainty if investments are irreversible. Studying total investments is appropriate if capital goods are homogenous and do not differ in their degree of irreversibility. In reality, however, capital goods are heterogeneous and differ in their degree of specificity, liquidity and adjustment costs (Guiso and Parigi, 1999). In addition, some capital goods can be more easily substituted, which indicates that irreversibility constraints on some assets can be overcome by investing in less irreversible assets. Accordingly, the true effects of uncertainty might be concealed if irreversibility constraints are only binding for some types of capital goods. In this study, we explicitly take into account that the degree of asset irreversibility does not only vary across firms (firm-specific asset irreversibility), as it is assumed by most prior studies (Chirinko and Schaller, 2009; Guariglia et al., 2012; Kim and Kung, 2016), but that irreversibility also varies across firms assets (asset-specific irreversibility). As asset-specific irreversibility is difficult to measure, we categorize capital goods into more or less irreversible investments by referring to the purpose for which the investments are acquired. In doing so, we implicitly assume that the purpose of investment serves as an indicator for investments degree of irreversibility. The rationale is that each investment project is characterized by certain properties which depend on the purpose of investment. These characteristics in turn determine the degree of irreversibility. To categorize investments according to the purpose for which they are acquired, we exploit information from the ifo Investment Survey which is available on an annual basis for the years 2004 to Following the guidelines of the European Commission, the ifo Investment Survey differentiates between five investment purposes: capacity expansion, rationalization, restructuring, replacement, and other. In the following, we briefly describe the characteristics of each investment type. Based on these characteristics, we then explain why investment types categorized by their purpose differ in their degree of irreversibility. Capacity expansion investments comprise all investments which increase firms existing capacities. The main purpose of capacity expansion is hence to expand production output in order to meet increased demand or to access new markets, for example. While capacity expansions can include the purchase of additional equipment and machineries, it most often entails the set up of a new 4

5 production unit. Accordingly, for the majority of capacity expansion investments, firms will also have to hire new employees who will work at the new plant. Restructuring investments cover all investments which modify firms existing production processes without affecting firms capacities. The main purpose is to lower firms production costs without changing capacity. For instance, a firm which acquires a new machine to produce a certain (new) output good and simultaneously reduces the output (capacity) of an existing machinery in order to offset the increased capacity from the new machine, is undertaking restructuring investments. Rationalization investments contain all investments which replace an existing, still usable machine by a new, more efficient machine. In general, these new machines meet the latest technological standards or have improved technical capabilities. The rationale behind replacing a still usable machine is to improve the cost structure of firms production. Replacing an obsolete machine by a modern version, but without changing firms capacities, is a typical rationalization investment. Investments which substitute input factor labor by input factor capital due to cost considerations can be also regarded as rationalization investments. Replacement investments include all investments which maintain firms capital stock without affecting firms capacities. Replacement investments aim to replace an old, used asset by a new asset of the same kind. In general, the replacing asset is of similar type but not of equal value as the replaced asset, because replacement investments usually increase the qualitative level of assets. For example, replacing an old machine after ten years of usage with a new machine of the same type increases the level of quality as the performance of the replaced machine is significantly lower than the performance of the new one. According to Bitros (1976), investments on capital maintenance cover all outlays that preserve firms operative state of capital stock. Similarly,? define replacement investments as routine maintenance expenditures to replace and repair broken machines. By other investments, the Ifo Investment Survey understands all investments in environmental protection, investments which improve working conditions, and investments in research and development. To categorize the five investment types into more and less irreversible investments, we first define asset irreversibility. According to Arrow (1968) and Abel and Eberly (1996), capital goods are irreversible if there is a wedge between the purchase and the resale price. This wedge arises because of transactions costs such as installation costs, which increase investments purchase price but which cannot be recovered on a resale. Moreover, costs of arranging, detaching or moving machinery determine the magnitude of such a wedge (Arrow, 1968). Hence, all sunk costs related to an investment will increase the wedge between purchase and resale price and make an investment more irreversible. Taking into account the notion of irreversibility and the porperties of capacity expansion investments, we argue that capacity expansion investments face the highest degree of irreversibility. The reason is that capacity expansion investments often include the set up of a new production site which entails high administration and planning costs and thus high sunk costs. In addition, the set up of a new plant makes the recruitment of new employees necessary which in turn makes 5

6 165 capacity expansions even more difficult to revert. The reason is that the dismissal of employees is impeded by labor protection guidelines which are especially pronounced in Germany. Given that our considerations are valid, we expect to find that capacity expansion investments decrease after an increase in uncertainty because capacity expansion investments face a high degree of irreversibility. The hypothesis to be tested is H1: Capacity expansion investments decrease after an increase in uncertainty In contrast, replacement investments should entail the lowest degree of irreversibility, as replacement investments only maintain a firms capital stock without affecting firms capacities. Accordingly, planning efforts and thus sunk costs are rather limited making a small wedge between purchase and resale prices more likely. Due to high levels of technological progress even within short time periods, it might often be impossible to ascertain differences between replacement, rationalization and restructuring investments. Since all three investments have in common to keep firms capacities unchanged and thus extensive planning efforts at a minimum, we summarize these three investments under the category non-capacity investments. We assume that non-capacity expansion investments are less influenced by an increase in uncertainty compared to capacity expansion investments, as non-capacity investments face a lower degree of irreversibility. The hypothesis to be tested is H2: Non-Capacity expansion investments are less affected by uncertainty than capacity investments For the category other investments, we do not have a clear prediction on the degree of irreversibility as several different investments are summarized in this category. If we find a negative and significant effect of uncertainty on other investments this might indicate that firms mainly subsume investments into research and development (r&d) under this category. The reason is that r&d investments often entail a large amount of labor costs (research personnel), which cannot be reverted if the project fails (Dixit and Pindyck, 1994; Czarnitzki and Toole, 2013). In contrast, if we do not find any impact of uncertainty on other investments, this might indicate that investments, which are not specific to the firm (and thus can be easily sold) or investments which do not entail high transaction cost, are included in this category. 3. Empirical implementation 190 To examine the impact of uncertainty on firms investment types, we examine unique panel data on large, private manufacturing firms in Germany over the sample period 2004 to In the following subsections, we provide a brief overview on the data, derive our measure of uncertainty, describe our empirical specification and provide summary statistics Data 195 The unique feature of our data set is that it combines survey data from the ifo Investment Survey, which is conducted by the ifo Institute, with accounting data from either Bureau van Dijk s 6

7 Amadeus database (Amadeus) or Bisnode database. We briefly describe both survey and accounting data sources in the following. For a detailed description we refer to Appendix B. The ifo Investment Survey has been conducted every fall (September/October) since 1989 on mainly private manufacturing firms in West Germany. The survey contains managers responses on firms investment behavior and the factors influencing firms investment spending. Since fall 2004, the survey asks managers to state planned investment spending for the entire upcoming year. As we refer to survey information to measure stated investment plans, our sample period begins in Our sample period is restricted to 2012 as relevant survey data is only available until 2012 due to disclosure considerations. Accounting data are provided by either Amadeus or Bisnode. The former is run by Bureau van Dijk Electronic Publishing GmbH and offers extensive business and financial statement data on mainly non-listed firms. The Bisnode accounting database is provided by Bisnode Deutschland GmbH, which is one of the leading databases for business and industry information in Germany. Both accounting databases consider German accounting standards and are available since If information for a firm is available from both databases, information from Bisnode is preferred due to its higher accuracy and quality (Seiler, 2012). Annual survey data are linked to the corresponding annual balance sheet data via firms registered name and address information using a record linkage approach. 3 As relevant survey data have been only available since 2004, our final sample period is limited to the years from 2004 to Following Almeida et al. (2004), we account for firm discontinuities caused by merger activities, reorganizations, and other corporate actions by removing all firm-year observations with total assets or sales growth exceeding 100%. To ensure comparability among firms legal forms, we only keep limited liability companies in the sample. We finally delete 80 public firms (270 firm-year observations) as we only focus on private firms. We winsorize all ratios at the 1% and 99% levels to control for outliers. Our final data set includes survey data on firms investment behavior as well as extensive financial statements data for private manufacturing firms in Germany over the time period from 2004 to Measuring uncertainty 225 Prior literature examining the effects of uncertainty on firms investment, either rely on stock market data, accounting data or on survey data to measure uncertainty. Leahy and Whited (1996), for instance, use the volatility in the daily stock returns and its covariance with a market portfolio to measure firm-specific uncertainty. 4 One serious concern with stock market based measures is that variations might be driven by bubbles unrelated to firms fundamentals (Carruth et al., 2000; 3 For a detailed description of the exact matching approach we refer to information provided on the web page of the EBDC. For theoretical details on the record linkage approach see, e.g., Newcombe et al. (1959); Fellegi and Sunter (1969). 4 The volatility of stock market returns is also used by several other studies (Kim and Kung, 2016; Bulan, 2005; Bond and Cummins, 2004; Bloom et al., 2007). 7

8 Bloom et al., 2007). A measure based on accounting data is used by von Kalckreuth (2003). He proxies uncertainty using the volatility of errors from models that forecast firms real sales. Although measures derived from forecasting equations are forward-looking, one drawback is that one has to assume that all firms use the same forecast model for prediction. The most direct method to measure uncertainty is to ask managers about their certainty regarding output price or demand expectations. For example, Guiso and Parigi (1999) study cross-sectional survey data on Italian manufacturing firms and use managers subjective probability distributions for future product demand to measure firm-level uncertainty. Similarly, Fuss and Vermeulen (2008) use survey data to derive a measure of uncertainty based on managers expectations on future demand and output price changes. In our study, we follow the approach of Bachmann et al. (2013) and Philipp and Röhe (2016) and construct a qualitative measure of uncertainty based on ex ante forecast disagreement. 5 To construct our measure, we focus on the following question included in the ifo Investment Survey: Q1: How do sales expectations influence firms investment activities for the entire upcoming year Q1 contains information on the influence of managers future sales expectations on firms future investment spending. Managers can choose from five possible response options: a strongly stimulating effect, a weakly stimulating effect, no effect, a weakly moderating effect and a strongly moderating effect. For simplification purposes, we summarize these five response options to three categories: stimulating, moderating and neutral category. However, all subsequent results remain rather unchanged if we consider all five categories and do not subsume strongly and weakly moderating (stimulating) to moderating (stimulating). To construct our ex-ante forecast disagreement measure, we use responses from Q1. More precisely, we follow Bachmann et al. (2013) and categorize F ract stim t as the weighted fraction of firms in the cross section which state stimulating responses at time t for the upcoming year (t+1). F ract mod t is analogously constructed for the moderating responses. We then calculate the dispersion of the survey forecasts for each two-digit industry sector according to the common statistical classification of economic activities in Germany (WZ 2008 industry codes). 6 Equation (1) shows the exact construction of our uncertainty measure. F DISP j,t = F ract stim j,t + F ract mod j,t (F ract stim j,t F ract mod j,t ) 2. (1) By using a cross-sectional dispersion measure to proxy for uncertainty, we explicitly assume that idiosyncratic shocks are drawn from similar distributions, so that fluctuations in dispersion are the result of fluctuations in uncertainty and not only changes in the cross-sectional composition (Bachmann et al., 2013). Although this is a strong assumption, there is meta-study evidence 5 Several other studies also used forecast disagreement to measure uncertainty, see, e.g., Driver et al. (2006), Bloom et al. (2007) and Fuss and Vermeulen (2008). 6 For a detailed description on the appropriateness of this measure to capture uncertainty see, e.g., Bachmann et al. (2013) and Fuss and Vermeulen (2008). 8

9 that managers are actually influenced by the outcome of the ifo survey (Abberger et al., 2009). Accordingly, managers might become themselves more uncertain if they notice a large disagreement among peer firms. Our measure of uncertainty is thus forward-looking, sector-specific and time varying. We hence do not have to make any assumptions on firms expectation formation process as in von Kalckreuth (2003), for example, to obtain a forward-looking uncertainty measure. Table 1 contains summary statistics of our uncertainty measure. FDISP takes values between 0.68 and 0.93, whereas higher values indicate a higher level of industry-specific uncertainty. Mean and median values are rather similar, taking a value of 0.80 and 0.83, respectively and the standard deviation of our dispersion measure is [Table 1 about here.] For illustration purposes, we also show in Figure 1 how our measure of forecast disagreement depends on the fraction of stimulating and moderating responses regarding the impact of sales expectations. From Figure 1, it becomes evident that our uncertainty measure takes the highest values if the number of firms responding a positive impact of sales expectations is similar to the number of firms responding a negative impact. Accordingly, unanimity within an industry regarding the influence of firms sales expectations indicates low uncertainty, while nonconformity within an industry will lead to high industry-specific uncertainty. Our uncertainty measure FDISP moves also counter-cyclical, as we observe higher levels of uncertainty during economic recessions (financial crisis in 2009) and lower levels during economic boom phases. 8 A negative co-movement of our uncertainty measure with the business cycle is consistent with the findings of prior studies (Bloom, 2009; Bachmann et al., 2013; Gilchrist et al., 2014). [Figure 1 about here.] Several prior studies use forecast disagreement as a measure of uncertainty. For instance, Bond and Cummins (2004) use disagreement among analysts profit forecasts to measure uncertainty. Driver et al. (2006) use cross-sectional dispersion of industry optimism about the general business situation. The measure closest to ours is the one of Fuss and Vermeulen (2008). One difference, however, is that Fuss and Vermeulen (2008) have monthly responses on demand and price expectations, while we have only annual data. Moreover, Fuss and Vermeulen (2008) use managers demand expectations to measure uncertainty, while we use managers sales expectations. Our measure thus considers both demand and price uncertainty, as sales and demand expectations will only differ if firms future output prices change. 7 To save space, we did not include summary statistics for each industry, however, results are available from the author upon request. 8 This pattern becomes even more evident if we look at the time period 1989 to This graph is available from the author upon request. 9

10 3.3. Model specification and estimation methodology To determine the influence of uncertainty on firms investment spending, we follow prior research and estimate the following reduced-form investment model: P lanned inv i,t = δ i + ϑ t + γ 1 Actual inv i,t 1 + βx i,t 1 + θ 1 F DISP j,t + ɛ i,t, (2) where the index i = 1,..., N is an index for individuals and the index t = 1,..., T indicates time. Firm fixed effects (η i ) are included to control for omitted variables that vary across firms, but are constant over time. By including time fixed effects (θ t ), we control for all variables that are constant across firms but evolve over time. ɛ denotes our error term. The dependent variable is P lanned inv i,t which is a firm s planned investment spending for year t stated in year t-1 scaled by beginning-of-year total assets. Plan data are obtained from the ifo Investment Survey, in which managers provide information on firms planned total investment spending in PPE in absolute values for the entire upcoming calendar year. We scale stated investment plans by firms beginning-of-year total assets. We use planned investment spending instead of realized investment spending as dependent variable as both planned investments and our perceived uncertainty measure coincide in terms of timing. Hence, we match stated investment plans for the upcoming year with perceived uncertainty for the upcoming year. In doing so, we follow Guiso and Parigi (1999), Fuss and Vermeulen (2008) and Bontempi et al. (2010). For ease of reference we mean firms total planned investments when talking about planned investments. To examine the effects of uncertainty on different investment types, we estimate Equation (2) for capacity expansion, non-capacity expansion and other investment plans. We construct firms investments into each investment category by using managers survey responses on the purpose of investments. In detail, the ifo Investment Survey asks managers in t-1 how investments for year t are allocated to capacity extensions, restructuring, rationalization, replacement and other investments, each as a percentage of total investment plans. We use this information and calculate Non-Capacity inv. as the sum of firms replacement, restructuring and rationalization investment plans stated in year t-1 for year t. The variable is constructed using managers stated fractions (in percent) into each of the just stated investment categories, multiplying this fraction with firms total planned investments (variable: Planned inv.) and adding up all three fractions. Analogously, we construct Capacity inv. and Other inv. We follow Guiso and Parigi (1999) and Fuss and Vermeulen (2008) and include firms realized investments in property, plant and equipment (PPE) in year t over beginning-of-year total assets (Actual inv i,t ) as control variable. Since data from cash flow statements are not available, we calculate realized investments in PPE as the change in tangible assets from year t-1 to t and add firms depreciation as of t. Untabulated results indicate that our estimations are qualitatively unchanged if we use realized investments stated from survey data. By considering realized investments of the year when plans are stated, we explicitly control for adjustment costs related to firms investments. 10

11 X i,t 1 reflects standard control variables used in the investment literature. In particular, we control for firms cash flows, investment opportunities and size. To account for size effects, we include the natural logarithm of firms total assets (Size). Firms operating cash flows (CF) are calculated as earning after tax plus depreciation, scaled by beginning-of-year total assets. Given that our sample consists of only private firms, we follow Asker et al. (2015), D Espallier and Guariglia (2012) and Bloom et al. (2007) and use sales growth as our measure for investment opportunities. 9 All control variables are measured as of t-1 to avoid endogeneity concerns and to account for the fact that only variables can influence investment plans that are known to the manager at the time when plans are stated. Our measure of uncertainty F DISP j,t captures the effects of uncertainty on planned investment spending. Alternatively, we could also examine the effects of perceived uncertainty at the time when investment plans are stated (F DISP t 1 ). Fuss and Vermeulen (2008), for instance, can show that uncertainty at the time when plans are stated also depresses firms investments. For our base results, however, we follow Guiso and Parigi (1999) and use uncertainty perceived for the upcoming year. Hence, investment plans are based on the same information available to the firm when expectations on future sales are reported. As our sample contains a large number of firms which plan to undertake zero investments into non-capacity expansion, capacity expansion and other (left censored at zero), we estimate Equation (2) using panel Tobit estimators. We apply Tobit estimators as censoring leads to biased coefficient estimates when using ordinary least squares or fixed effects estimators, for instance (Cameron and Trivedi, 2009). By applying Tobit models, we thus directly take into account that uncertainty might lead to periods of inaction (investments are delayed) which explains zero values of the dependent variable. However, in robustness tests, we find that our results are rather unchanged if we use standard panel fixed effects estimators Summary statistics Table 2 contains summary statistics on all relevant variables used for estimation. Panel A contains all investment related variables, while Panel B contains summary statistics on our control variables. The first two rows contain total realized investments (from accounting data) as well as stated investment plans. It becomes evident that on average, firms plan to invest about 6.0% of beginning-of-period total assets in PPE. Our figures on planned investments are comparable with the figures of Guiso and Parigi (1999) who find average investment plans of 5,99% when studying survey data on Italian manufacturing firms in Figures on realized investments are slightly smaller, as firms invest on average only 5.0%. 9 In untabulated robustness tests, we also use the two-digit level of the common statistical classification of economic activities in Germany (WZ 2008 industry codes) to calculate industry s sales growth as a proxy for investment opportunities. The results remain rather unchanged. 11

12 Panel A also contains summary statistics on stated shares into each investment type as well as summary statistics on investment type ratios. Ratios are calculated as stated share into the respective investment type times total investment plans. All share and ratio figures on investments by purpose refer to plan data. We find that the majority of planned investments are non-capacity investments as stated shares correspond to approx. 74%. Stated shares into capacity investments amount to 18%, and other investment shares are on average 8.2% of total investments. 10 Our figures emphasize the importance of non-capacity expansion investments and coincide with the results of prior studies examining replacement versus capacity expansion investments (Eisner, 1972). Similar results are found for investment type ratios. Non-capacity investments display the highest ratios again, as firms plan to invest into replacements about 4.1% of beginning-of-year total assets, while capacity expansion investments amount to 1.5% and other investments to 0.5% of total assets. From Panel B of Table 2 it becomes further evident that our sample contains mainly large firms, as the median firm has total assets of approx. EUR 30 million. Furthermore, our sample contains firms with negative cash flows and negative sales growth, as well. Overall, the summary statistics in Table 2 point to a certain firm heterogeneity within our data set. [Table 2 about here.] 4. Estimation results In the following section, we examine to what extent uncertainty influences total investment plans as well as capacity and non-capacity investment plans Results In Table 4, we estimate our full model specification as described by Equation 2 using Tobit estimations over the full sample period 2004 to In Column 1, we find that uncertainty has a negative and significant effect on firms total investment plans. A one standard deviation increase in uncertainty (0.116) decreases planned investment ratios by In terms of average level of planned investment ratio (0.061) this implies a drop in investments of about 3.23%. Moreover, we find a positive and significant effect of lagged cash flows on total investment plans. Accordingly, higher cash flows increase stated investment plans for the upcoming year. In Columns 2 to 4, we regress investment plans for each type on our measure of uncertainty. Column 2 contains the results using non-capacity expansion plans as dependent variable. We cannot find any impact of uncertainty on non-capacity expansion investments. The only factors positively influencing non-capacity plans is lagged realized investment spending and cash flows. Investment plans for replacement, restructuring or rationalization investments also do not depend on investment opportunities. This seems reasonable as replacement investments are often investments which have 10 In Table C.1, we show summary statistics for all five investment types. 12

13 to be undertaken to maintain firms capital stock and to ensure production. Accordingly, these investments should not depend on investment opportunities [Table 4 about here.] In Column 3, we find a depressing effect of uncertainty on capacity expansion plans. A one standard deviation increase in uncertainty (0.116) reduces capacity expansion investment plan ratio by This corresponds to a 20% decrease of average capacity expansions. Moreover, we find a positive and highly significant effect of past realized investments and cash flows on capacity expansion plans. Similar results are found by Guiso and Parigi (1999); Fuss and Vermeulen (2008); Bontempi et al. (2010). In addition, we find a highly significant effect of investment opportunities for stated investment plans on capacity expansions. Hence, higher investment opportunities increases firms capacity expansions plans. From Column 4 of Table 4 we find that other investment plans seem to be only influenced by firm size. In the last row of Table 4, we further present the number of left-censored observations (zero investments). Obviously, the number of firms stating zero non-capacity expansion investments is rather low (30) compared to the firms with zero capacity expansion plans (1092) or other plans (1529). The reason might be that non-capacity investment plans mainly contain investment plans into replacements, which most firms have to undertake every year in order to maintain a firm s capital stock. We conclude that although replacement investments represent a large share of total investments, they are not sensitive to increased uncertainty. One explanation might be, that replacement investments and thus (non-capacity expansion) investments do not face a binding irreversibility constraint and thus do not respond to increased uncertainty. Accordingly, if irreversibility explains the effects of uncertainty on investments, the negative coefficient for capacity expansion investments might indicate that capacity expansion investments face a higher degree of asset irreversibility than non-capacity expansion investments. The results in Table 4 thus indicate that we cannot reject our hypothesis 1 (capacity expansion investments face a high degree of asset irreversibility). Hence, studies which only examine total investments and do not take into account that uncertainty only affects capacity expansion outlays might lead to flawed results Robustness tests 425 We perform several tests to confirm the robustness of our results. For instance, we run all regressions and include only a subset of control variables or no control variables (only include uncertainty variable and respective interaction terms). Our results (untabulated) are qualitatively unchanged. We further examine whether our results are sensitive to our aggregation method. We therefore separately regress replacement, rationalization and restructuring investments on our measure of uncertainty and control variables. Untabulated results indicate that our main results remain 13

14 qualitatively the same, as we cannot find any significant effect of uncertainty on one of these variables. We also test whether summarizing only rationalization and restructuring investments into one category influences our results. Both investment types have the common purpose to reduce firms production costs and thus to enhance efficiency. Again, we cannot find any effects of uncertainty. Furthermore, we examine whether our main results depend on how we measure uncertainty. We therefore exploit only the largest and thus most likely exogenous variations in our uncertainty measure. More precisely, we construct a dummy variable which takes the value of one if the increase in our uncertainty measure belongs to the bottom quartile of the annual FDISP distribution, and zero else. We argue that this dummy variable captures only the largest and hence most unexpected increases in industry-specific uncertainty. When re-estimating Table 4 using this uncertainty measure, we still find a negative and significant effect on capacity expansion investments (untabulated). 11 However, the coefficient estimate is clearly smaller. The results indicate that investments are not only influenced by the largest increases in uncertainty, but that already moderate uncertainty levels (and increases) influence firms investment behavior. We also analyze whether measuring uncertainty based on disagreement regarding firms earning expectations changes our main results. In contrast to sales expectations, profit expectations capture shocks to both demand side and supply side factors (Abel and Eberly, 1994). Table 5 contains our results when re-estimating Table 4, but using an uncertainty measure based on profit expectations. We find a negative and significant effect of uncertainty on total planned investments and on capacity expansion plans, again. The effects on capacity expansions are slightly smaller compared to the base results. We conclude that our results are not sensitive to our measure of uncertainty. [Table 5 about here.] We finally apply different panel estimators to test whether our results are sensitive to the estimator used. Untabulated results indicate that our results are qualitatively the same if we apply pooled cross-sectional tobit or ordinary least squares estimators. The application of seemingly unrelated regression estimators, which account for correlations between the error terms of the three investment types, yields also qualitative similar results. In Table C.2, we also apply standard fixed effects estimators which control for time invariant firm fixed effects. The results are comparable to the base results shown in Table 4, as we find a negative and highly significant effect of uncertainty on firms total investment plans and on capacity expansion plans. The estimated coefficients on our uncertainty variable are, however, slightly lower. A one standard deviation increase in uncertainty reduces total investment plans by 2.78% and capacity expansion plans by 10.8%. 11 All untabulated robustness tests can be obtained from the author upon request. 14

15 5. Testing for alternative channels 465 Although most prior studies identify the irreversibility channel as main reason for a negative effect of uncertainty on firms investments, several studies also examine the role of market competition and financial market frictions on the investment-uncertainty relation. In the following, we therefore examine the impact of these two alternative channels on our results. In addition, we measure asset irreversibility at the firm level, as standard in the literature, and examine whether the effects of asset-specific irreversibility remain if we control for firm-specific asset irreversibility Irreversibility channel Beside examining the degree of irreversibility at the asset level, we also explore asset irreversibility at the firm level. We thus construct a measure for firm-specific asset irreversibility following Chirinko and Schaller (2009) and Guariglia et al. (2012). Both studies measure firms degree of asset irreversibility by taking into account firms depreciation rate. The authors argue that irreversibility arises if firms cannot easily sell previously acquired capital goods. Consequently, the only way of reducing unwanted capital stock will be through depreciation. A firm is thus more likely to face asset irreversibility if its depreciation rate is below the median depreciation rate of the industry in which a firm operates. Our irreversibility dummy takes the value of one if the depreciation rate is below the median depreciation rate of two-digit industry, and zero else. To measure the effects of firm-specific irreversibility, we include an interaction term of uncertainty and our irreversibility dummy in our base specification. The coefficient on the interaction term shows us the additional effect of uncertainty for firms with high firm-specific asset irreversibility. 12 According to real options theory, firms with high asset irreversibility should be stronger affected by uncertainty. Accordingly, firms with high asset irreversibility (at the firm level), which plan to do capacity expansion investments (high asset irreversibility at the asset level), should be strongest affected by increased uncertainty. Table 6 contains the results. [Table 6 about here.] From Column 1 of Table 6 we find that total investment plans are still negatively influenced by an increase in uncertainty. However, we cannot find that firms with a high degree of firm-specific asset irreversibility are more strongly affected by increased uncertainty, as the interaction term between our uncertainty measure and our irreversibility dummy is insignificant. Similarly, we cannot find any additional effect of increased uncertainty for firms with a high degree of asset irreversibility which plan to undertake capacity expansion investments (Column 3). We find, however, a negative and highly significant effect of our interaction term in Column 2. Hence, increased uncertainty reduces planned investments into non-capacity investments if firms face a high degree of asset 12 In untabulated results we also included our irreversibility dummy as main variable as well as its interaction with uncertainty. However, we did not find any significant effects of the irreversibility dummy. 15

16 irreversibility. The results indicate that uncertainty affects investments through irreversibility at the asset and at the firm level. Hence, while an increase in uncertainty will lower total investments and especially firms capacity expansion investments it will not affect non-capacity investments. Thus uncertainty will depress especially investment types which are more difficult to revert (capacity expansion investments). However, uncertainty also affects non-capacity investments, if firms face in general a high degree of asset irreversibility. We are the first to show how uncertainty affects investments through different channels of irreversibility (asset-specific versus firm-specific asset irreversibility) Financial friction channel Beside the irreversibility channel, which might depress investments if uncertainty increases, investments might also decrease during times of high uncertainty due to the so-called financial friction channel (Ghosal and Loungani, 2000; Minton and Schrand, 1999; Gilchrist et al., 2014). The intuition is that with raising uncertainty on key determinants of investments, such as cash flow and q, a firm s default probability raises which leads to higher external financing costs and thus increases the wedge between internal and external sources of funds (von Kalckreuth, 2003). Similarly, Kim and Kung (2016) argue that an increase in uncertainty might negatively influence the collateral values of firms assets, especially for assets with low alternative usage in other firms or industries. A drop of collateral value then decreases debt capacity which might explain lower investments during times of increased uncertainty. The impact of uncertainty on investments thus varies with access to external financing. We hence examine whether financial market frictions can explain the negative relation between investments and uncertainty. To examine if the influence of uncertainty on investments varies with access to external financing, we partition firms into groups of constrained and unconstrained firms. We therefore use managers selfperception on financing status obtained from the ifo Investment Survey. More precisely, managers are asked every fall of a year to respond on a five-point scale how a firm s current year s investments are influenced by a firm s sales status, financing options, earnings expectations, technical developments, economic-political conditions or other factors. Regarding the five-point scale, managers can choose between a strong animating effect, a weak animating effect, no effect, a weak moderating effect, and a strong moderating effect. Using this information, we construct a dummy variable which takes the value of one if managers claim a strong or a weak moderating effect of financing options on current year s investments (financially constrained firms). It takes a value of zero if managers state no effect or either a weak or strong animating effect (financially unconstrained firms). We examine the importance of financial frictions by interacting our financial constraints dummy with our measure of uncertainty and include it in our base model. Table 7 contains the results. We find that once controlling for firms financial status, we cannot find any effect of uncertainty on firms total investments (Column 1), neither for our uncertainty measure nor for the interaction term. However, we still find a negative and significant effect of uncertainty on capacity expansion investment plans (Column 3). The interaction between uncertainty and our financial constraints 16