VIEWS: 135 PAGES: 60 POSTED ON: 12/12/2009
Corporate Investment Decision Practices And the Hurdle Rate Premium Puzzle Iwan Meier and Vefa Tarhan1 September 22, 2006 ABSTRACT We survey a cross-section of 127 companies to gain insight on various dimensions of firms’ investment decisions. The questions posed by our survey address the hurdle rates firms use, calculations of project-related cashflows, and the interaction of cashflows and hurdle rates. Unlike previous studies which examine investment decisions by either using survey data or data obtained from financial tapes, we use both sets of data. This approach produced one of our primary findings that there is a hurdle rate premium puzzle, in that hurdle rates used by our sample of firms exceed their cost of capital, that we calculate using Compustat data, by a substantial magnitude. We investigate the determinants of this puzzle and find that it is related to factors that reflect financial flexibility considerations, managers’ confidence in the estimates of beta, financial health of firms, and the past performance of the industry they are in. Finally, our findings show that survey firms do not always appear to handle the cashflow dimension of their investment decisions in a consistent manner. 1 HEC Montréal, and Loyola University, Chicago, respectively. Corresponding author: Iwan Meier, HEC Montréal, 3000 chemin de la Côte-Sainte-Catherine, Montréal (Québec) H3T 2A7, Canada. Tel. +1-514-340-3198; e-mail address: iwan.meier@hec.ca. We thank Michael Fishman, Deborah Lucas, Robert McDonald, Mitchell Petersen, Artur Raviv, Ernst Schaumburg, Paul Spindt, and Timothy Thompson. We especially thank Ravi Jagannathan for his insightful comments, his continuous encouragement, and for providing resources through the Financial Institutions and Market Research Center. We also thank the focus group participants William A. Colaianni, James J. Cowhey, Pavel A. Dorosevich, Gaea Gomez, Art Mollenhauer, and Don Porth for their input to improve the design of the survey. This research is sponsored by the Zell Center for Risk Research and we are indebted to its Director and Dean Emeritus of the Kellogg School of Management Donald Jacobs for his support and advice. Introduction This paper discusses the findings we obtain from a survey that covers a comprehensive list of discount rate and cashflow related issues firms face in making their investment decisions. The survey was completed by the CFOs of 127 companies in October 2003. Most respondents reveal their identity which enables us to match their answers with CRSP and Compustat data. Some studies examine corporate decisions by testing models that use data available in financial data bases such as CRSP and Compustat. The advantage of such data is that it contains a substantial number of observations, which increases the level of confidence regarding the accuracy of the results. Additionally, the data in question is determined objectively. However, this objectivity may come at a cost. Financial models of investment decisions posit how managers should behave based on model specifications. Ideally, the predictions contained in these models need to be tested with data that captures the behavior of the managers. However, the data obtained from financial tapes does not fully reflect how managers behave since these data points represent the realizations of financial variables. These realizations are determined not only by the behavior of managers, but also by the behavior of other economic agents as well as by the parameters of the exogenous environment. For example, the realized level of sales of a firm does not only reflect how successful managers were in their past investment decisions, but it also reflects factors such as the current state of the economy, consumer demand, investments of competitors, etc. Thus, while the theories of corporate investment decisions specify how managers should make their investment decisions, the predictions of the models are tested with data that is produced only partially by managers’ behavior. Tests that use data contained in financial data bases may reveal whether a particular investment decision was a success or a failure ex-post, but they do not necessarily fully reflect whether the success or failure of the project is the result of the procedures managers followed ex-ante, since exogenous factors also play a role in the outcome. 1 The strength of the survey data based testing methodology is that the predicted behavior of agents can be captured directly. While data generated by surveys is subjective, this subjectivity is not necessarily undesirable, provided that the survey questions are well-designed. Since financial models make predictions about the behavior of economic agents, the hope of survey studies is that the survey questions will capture how managers do behave. On the other hand, tests based on survey data also have some weaknesses. First, surveys typically do not produce anywhere as many data points as the financial data bases. Additionally, if survey questions are not phrased correctly, tests based on survey responses could be misleading. Using both, survey data and data available from financial data bases, enables us to compare how managers do behave with how they should behave in making their investment decisions. Such comparisons could prove to be very revealing. In fact, it is on the basis of such a comparison that we are able to document one of the important findings of this study: There appears to be a hurdle rate premium puzzle. We find that hurdle rates we obtain from our survey data exceed, on average, the rate that managers should use (their actual WACC that we compute from financial data bases) by a substantial magnitude. Previous surveys on how firms make their investment decisions, such as Graham and Harvey (2001), Bruner, Eades, Harris, and Higgins (1998), Poterba and Summers (1995), Trahan and Gitman (1995), and Bierman (1993), primarily focus on documenting the popularity of the different capital budgeting techniques used by firms. Some of these studies also examine the discount rates firms use in evaluating their investment projects. The primary findings of the earlier surveys are as follows: First, over time, firms have shown an increasing tendency to rely on discounted cash flow (DCF) methods to evaluate projects. Second, most firms apparently use their weighted average cost of capital (WACC) as the discount rate in evaluating their projects. Third, it seems to be the case that in computing their discount 2 rates, firms typically infer the cost of equity from the capital asset pricing model (CAPM). Figure 1 displays the increased usage of these models and techniques over time. While our survey respondents by and large confirm these findings, we supplement the traditional investment decision issues addressed by past surveys by examining additional topics that are important to the investment decisions of firms. As a result, this study makes two major contributions to the literature on investment decisions. First, as mentioned above, we document the existence of the WACC-based hurdle rate premium puzzle. Furthermore, we explain this puzzle with reasonable success. Second, we examine the practices in calculating cashflows, as well as the interaction between cashflow and discount rates, topics typically not addressed by earlier studies. For example, we examine whether or not firms compute their cashflows correctly, and whether they are consistent in matching project cashflows with the appropriate discount rates (i.e., we examine what discount rates are used when cashflows are levered versus when they are unlevered). Our survey also contains questions to determine whether or not firms are consistent in accounting for inflation in the cashflow and the hurdle rate components of the projects they consider. Additionally, we probe survey participants about other cashflow specific topics such as how they account for sunk costs, and cannibalization of the sales of their existing products when making decisions about the introduction of new products. Finally, we examine whether firms correctly analyze their cross-border projects by using consistent foreign currency/domestic currency denominated cashflow/hurdle rate combinations. Our important findings on hurdle rates are as follows: First, the hurdle rates used by firms in our survey appear to exceed their WACC calculated using CRSP and Compustat data by 5.3% to 7.5%, depending on the equity premium assumptions we use. Second, when we examine the determinants of this puzzle, we find that desire for financial flexibility in high growth environments, firms’ financial health, the past performance of the industry that the firm is in, and managers’ confidence regarding their estimated betas go a long way towards explaining this puzzle. Based on our interpretation of the statistically significant 3 variables that contribute towards explaining the puzzle in question, a case can be made that it is understandable, perhaps even reasonable, that managers use hurdle rates in excess of their WACC. However, since the (upwards) adjustment managers make to the textbook definition of WACC is subjective in nature, it is not possible to judge whether or not the magnitude of the hurdle rate premium is appropriate. Third, our findings show that self-reported hurdle rates are related to firms’ systematic risk, implying that they use CAPM. However, our tests indicate that apparently unsystematic risk also plays some (albeit lesser) role in the determination of their discount rates. In addition to risk variables, we find that self-reported hurdle rates appear to be related to the same variables that are significant in explaining the hurdle rate premium puzzle. Finally, the results also show that some firms appear not to adjust their hurdle rates with sufficient frequency, and that some firms use firm-wide hurdle rates even when they have multiple divisions. Needless to say, both of these hurdle rate missteps have the potential to create underinvestment or overinvestment problems. Our important findings on the cashflow component of investment decisions are as follows: First, it appears that survey firms correctly incorporate inflation in evaluating projects. Second, it seems that they also handle the complex problem of domestic/foreign currency denominated cashflow/hurdle rate combinations associated with cross-border projects correctly. Third, on the negative side, firms in our survey have somewhat of a mixed record on the computation and use of cashflows, and the cashflowdiscount rate interaction issues. While in general they calculate both levered and unlevered cashflows correctly, they do not always use the correct cashflow-discount rate combinations. Moreover, not all firms appear to correctly account for sunk costs and erosion in sales of existing products related to new product introductions. We also briefly discuss capital budgeting methods and confirm the findings of earlier studies in that our survey firms rely on discounted cashflow (DCF) methods by a wide margin. 4 The plan for the remainder of the paper is as follows: We discuss survey design and sample characteristics in Section I. We present our findings on self-reported hurdle rates in Section II. Section III presents our calculations of WACC for our survey firms. Section IV documents the hurdle rate premium puzzle. To gain insight about this puzzle, in Section V we estimate bivariate regressions on self-reported hurdle rates, and also on the hurdle rate premium where we use two equity premium scenarios. We also examine the determinants of the hurdle rate premium in a multivariate regression framework where the explanatory variables are the statistically significant coefficients obtained from the estimation of the bivariate hurdle rate premium models. In Section VI, we present and discuss our findings about the cashflow dimension of the investment decisions and also about the interaction between cashflows and hurdle rates. In Section VII, we resume our discussion of self-reported hurdle rates. In particular, we examine whether or not firms in our survey adjust their hurdle rates over time as market conditions change, and to what extent multi-segment firms use company-wide versus divisional hurdle rates. A brief discussion of capital budgeting methods used by our survey participants follows in Section VIII. Finally, we present our conclusions in Section IX. Due to space constraints we are unable to present the full list of our findings. At times we discuss some of our results briefly and refer to “results not displayed here” or “unreported results.” Additionally, for the same reason we were not able to include some of the analysis that we conducted. The details of results we briefly discuss in the paper but do not display in tables, and additional analysis we conduct but do not discuss in the paper can be accessed at www.author. (hardcopy available upon request) I. Survey Design and Sample Characteristics A. Questionnaire In designing the survey we carefully followed the advice of experts in the fields of psychology and marketing. We designed the questions in such a way that we minimize the use of buzz words and names 5 of models that are taught in a typical MBA course. For example, as we conjecture that there might be a wedge between cost of capital and the discount rate, we avoid the term “cost of capital” in our questionnaire. Instead, the survey asks questions on their hurdle rates. Similarly, we tried to avoid using terminology such as “levered” and “unlevered” cashflows, but rather provided them with the definitions of the two types of cashflows from which to choose. It is a well documented observation in psychology, known as the social desirability hypothesis (see e.g. Singer and Presser (1989)), that respondents to surveys tend to try to please the sender of the survey by providing the answers they think the sender expects. Therefore, we did not want to prompt them by asking questions that contain technical buzz words. The input from numerous finance academics helped to further improve the content of the questions. In order to test the survey with practitioners, we invited six CFOs from the Chicago area to a focus group meeting on May 26, 2003. After filling out the survey, we discussed each question to assure that the wording was not ambiguous.2 The survey was sent out together with a cover letter from the Dean Emeritus of the Kellogg School of Management, Donald Jacobs, along with a postage-paid return envelope on September 12, 2003, to a total of 4,600 CFOs of U.S. companies listed in the Compustat name file. We asked the participants to return the questionnaire within ten days. At the beginning of October we sent a follow-up mailing. Most respondents reveal their identity. Almost all surveys are filled out completely and there is no decline in the number of responses towards the end of the four-page questionnaire. We have some evidence that the surveys were actually filled out by CFOs as we received a number of e-mails requesting an advance copy of the survey results and these mails came directly from the CFOs. In addition, many respondents provide elaborate comments to open questions. When comparing the self-reported sales 2 We also followed the guidelines on questionnaire design and focus group meetings provided by Gillman (2000) and Morgan (1988). 6 figures with the numbers retrieved from Compustat, a reassuring 92.3% of the respondents checked exactly the correct sales range.3 B. Sample Description Figure 2 describes the characteristics of the 127 firms in our sample. Panel A shows the breakdown by industry. Similarly to previous surveys (e.g. Graham and Harvey (2001); Poterba and Summers (1995)), most firms in our survey come from the manufacturing sector (41.7% of the sample).4 Firms in the technology and energy/transportation sectors constitute 13.4% and 10.2%, respectively, of the sample. We excluded firms in the financial sector from the survey.5 Firm size measured by sales (based on the self-reported numbers of the survey) is below $100 million for 35.2% of the companies (see Panel B). 31.2% of the responding firms report sales in excess of $1 billion. Other characteristics of survey firms are as follows: The majority of the firms (72.0%) have multiple product lines. Fourteen respondents (11.3%) are privately owned firms. The equity stake of senior management in the firm is 5% or less for half of the respondents (53.3%), and 1% or less for 13.1% of the 107 respondents to this question.6 Appendix A.1 reports details on the profiles of the responding CFOs. 3 125 firms checked a sales range and for 98 firms we can match the answer with data from Compustat. Out of the six firms that did not exactly check the box consistent with the Compustat sales figure for 2003, an additional three firms were just slightly off (on average by 3.2%). Thus, 96.9% of the answers to this question on the last page of the questionnaire are almost exactly right. 4 In a number of surveys the fraction of manufacturing firms is even more pronounced. For example, in Gitman and Mercurio (1982) this ratio is 93.8%, while in Gitman and Forrester (1977) it is 74%. 5 Financial firms account for 15% of the respondents in Graham and Harvey (2001). Specifically, we exclude all finance and insurance companies with the major SIC code in the ranges 6000-6499, 6700-6799, as well as health, education, social services, and museums (7200+). We also drop radio and TV broadcasting, cable, and other pay TV services, as these firms might be driven by non-commercial interests, e.g. religious radio stations (4840-4949). 6 The number of quantitative responses to this question is 107. The other respondents either checked “Don’t know” (15 responses) or left it blank (5). Five answers to this question were either <1%, <5%, or <8%. In these cases we take the exact values 1%, 5%, and 8%. Similarly, the two responses 50%+ and >1% are replaced by 50% and 1%. 7 II. Self-reported Hurdle Rates In this section, we first discuss the summary statistics on self-reported hurdle rates (Section A). In Section B, we examine what the survey participants claim their hurdle rates represent (i.e., whether is it their weighted average cost of capital (WACC), cost of levered equity, etc.). In Section C, we describe how we convert cost of equity based hurdle rates to their WACC equivalents. After this conversion all self-reported hurdle rates reflect survey participants’ perception of their WACC. Later on, in Section VII, we resume our discussion of hurdle rates by examining the frequency of hurdle rate changes by our survey firms and in the case of multi-divisional firms, we investigate to what extent they use firm-wide versus divisional hurdle rates. A. Summary Statistics on Hurdle Rates In the survey, we ask the participants for the nominal hurdle rate that they have used for a typical project during the two years preceding the survey date.7 Table I displays summary statistics on self-reported hurdle rates for the sample of firms. The results show that the mean hurdle rate in our sample is 14.1% in nominal terms (the median is 14.0%).8 None of the numbers is less than 5% and the maximum is 40%. Furthermore, the skewness coefficient of 1.7 indicates that the distribution is fairly symmetric, and the kurtosis coefficient of 9.6 confirms that the distribution is centered around the mean and median. Adjusting for the average inflation of 2.2% during the past two years (January 2001 to December 2003)9 produces an average real hurdle rate of 11.6%, which is close to the 12.2% real hurdle rate reported in the survey conducted by Poterba and Summers (1995).10 7 The number of responses to this question is 119. Seven of the 8 non-responding firms do not use a discounted cash flow (DCF) technique as their primary capital budgeting technique. 8 If a range is provided instead of a single number, then we take the average (6 respondents). For 5 firms we infer the typical hurdle rate as the average of the lowest and highest rate. One observation is reported in real terms and we add the average inflation from 2001-2003 of 2.2%. 9 Inflation rates are based on the Consumer Price Index (CPI-U) compiled by the Bureau of Labor Statistics. 10 While earlier surveys such as Gitman and Forrester (1977), and Gitman and Mercurio (1982), also report high rates (14.1%, and 14.3%, respectively), the nominal rates that they report are not high in real terms, considering that during the time of the latter two surveys nominal rates were high due to high inflationary forces in effect. 8 The hurdle rates obtained from both Poterba and Summers (1995) and our survey appear to be high. They surmise that the discount rates they obtain from their survey appear to be higher than even the cost of equity of the survey firms. In this study, we define the hurdle rate premium as the difference between the WACC-based hurdle rates used by survey firms and their computed WACC (Section III). We then compute the hurdle rate premium (Section IV) and show that its size is substantial (about one half to one third of the self-reported hurdle rates). After documenting its existence we then conduct tests to identify its determinants (Section V). B. What Do Hurdle Rates Represent? Of the 117 firms that responded to the question on what their hurdle rate represents, the vast majority of the CFOs (71.8%) claim that the hurdle rate they use is their weighted average cost of capital (WACC). Apparently, in the case of 7 firms (6.0%), the hurdle rate represents their cost of levered equity, while for 9 firms (7.7%) it reflects their unlevered cost of equity. For 17 firms (14.5%), the hurdle rate falls into the “other” category. The bar chart in Figure 3 illustrates these fractions. The widespread use of WACC is consistent with the findings of Bruner, Eades, Harris, and Higgins (1998) and Bierman (1993) who find that even larger fractions of firms use WACC. Thus, it seems to be the case that similar to the increased use of discounted cash flow (DCF) techniques and CAPM, the use of WACC for capital budgeting has also increased over time. This positive trend is shown in Figure 1. For example, in a survey conducted 30 years ago, Petty, Scott, and Bird (1975) document that only 30% of the firms in Fortune 500 that responded to their survey use WACC. Similarly, a survey by Schall, Sundem and Geijsbeek (1978) finds that 46% of the firms use WACC. C. Converting Non-WACC Self-reported Hurdle Rates to WACC-based Self-reported Hurdle Rates In the cases where survey participants indicate that they use either levered or unlevered cost of equity as their hurdle rate, we transform cost of equity based hurdle rates to their WACC equivalents. If they indicate that the hurdle rate information they provide represents their cost of levered equity, we simply 9 plug in this rate for the cost of equity component of WACC and average it with their after-tax cost of debt using market value weights. If they indicate that their hurdle rate represents their cost of unlevered equity, we check if they had any long-term debt. For firms that do not have any debt, their unlevered cost of equity is obviously the same as their WACC. In cases where firms had debt in their balance sheet, we lever up their cost of unlevered equity to obtain their cost of levered equity, and then compute their WACC using their after-tax cost of debt and market value weights. These procedures enable us to obtain WACC-based self-reported hurdle rates for 101 firms.11 III. WACC Computations for the Survey Firms In this section we compute the weighted average cost of capital (WACC) for the survey firms using CRSP and Compustat data. For firms where before-tax cost of debt data is missing we make assumptions to fill the missing data. We also explain how we construct the tax rate data for the survey firms. We start our WACC computations for the survey firms in Section A by first calculating their cost of levered equity. In Section B, we discuss how we compute before-tax cost of debt, tax rates, and the weights for debt and equity. A. Computing Cost of Levered Equity Using CRSP and Compustat Data Needless to say, an important component of WACC is the cost of levered equity. In recent years, as documented by Bruner, Eades, Harris, and Higgins (1998), and Graham and Harvey (2001), the dominant model that firms use in calculating their cost of equity capital has been the CAPM.12 Graham and Harvey (2001) report that 73.5% of the firms in their survey use CAPM “always” or “almost always”. Their findings also show that only a small number of firms implements more complex, multi-factor models. To 11 Taking the unadjusted, self-reported hurdle rates for the total of 119 respondents, the mean hurdle rate is 14.8% (median 15.0%) with a standard deviation of 5.3%. The 25th and 75th percentiles are 12.0% and 16.0%, respectively. 12 The increased use of CAPM over time is depicted in Figure 1. In Gitman and Mercurio (1982) only 21.5% of the firms indicate that they use the CAPM to assess cost of equity capital. The low fraction of firms using CAPM (29.8%) in the survey by Trahan and Gitman (1995), which was conducted in 1992, is likely due to the wording of the question on how firms consider risk in capital budgeting decisions, and the range of choices. 10 determine whether or not our survey firms use CAPM, we did not ask the question directly in order to avoid the potential contamination of the data, as predicted by the social desirability hypothesis. Instead, we gave them various choices about what they consider to be important factors that would affect their hurdle rates. Some of the choices in the question involved CAPM related considerations.13 The responses imply that the use of CAPM is widespread in our sample of firms. For example, 68.6% of our survey participants check the following statement as being important or very important; “market risk of a project, defined as the sensitivity of project returns to economic conditions.” Similarly, a very high proportion of the respondents argued that “interest rate changes” and “changes in stock market returns” would play important roles in their decision to change their hurdle rates. In calculating the survey firms’ cost of equity, we chose proxies for the risk-free rate and the equity premium that are on the high side (we also repeat our tests using an equity premium obtained from a survey of CFOs). The choice of high values for these two CAPM inputs is made in order to make sure that the hurdle premium we find is not generated by “too low” values for these inputs. The mean life of a typical project for firms in our survey sample is 6.8 years. Partially for this reason, we use the 10-year Treasury bond rate, which was 4.3% at the time of our survey, as a proxy for the risk-free rate.14 For the equity premium we first use 6.6% which represents the difference in the arithmetic average of the return on the S&P 500 index and the long-term Treasury bond rate covering the period 1926-2003 (obtained from Ibbotson (2004)).15 This value is in line with the textbook suggestion of Brealey and Myers (2000, p.160) that using a premium above the T-bill rate (instead of the yield on 10-year T-bonds) in the range of 6 to 8.5% seems reasonable. Furthermore, the equity premium we use is similar to the 7% 13 For example, rather than using the word beta, we asked them whether their hurdle rates were related to “… the sensitivity of project returns to economic conditions”, and whether they would change hurdle rates if “interest rates change” and if “tax-rates change”, etc. 14 This choice seems to be justified for other reasons as well. In their survey of 27 highly regarded corporations, Bruner, Eades, Harris, and Higgins (1998) find that more than 70% use a 10-year or longer-term Treasury rate. They report that only 4% of the firms in their survey use the 90-day T-bill rate. 15 Using the arithmetic mean instead of the geometric mean also results in a higher estimate of the equity premium. 11 figure used by 226 financial economists in the Welch (2000) survey. However, some studies use a much lower equity premium. For example, using a forward-looking approach, Blanchard (1993), Wadhwani (1999), Jagannathan, McGratten, and Scherbina (2001), and Fama and French (2002) suggest that the equity premium should be in the 3-4% range. The lower risk-premium figures used in recent studies are based on data that shows that the equity premium has declined considerably during the nineties. The 217 respondents to the Duke/CFO Magazine survey in December 2003 by Graham and Harvey (2006) expect, on average, a premium of the S&P 500 index over the 10-year T-bond yield of 3.83% (median 3.60%). For this reason we repeat our analysis using a lower equity premium of 3.6%. Since this figure is obtained from a survey of CFOs that was conducted about two months after we mailed our questionnaire, it is likely to be close to the actual equity premium used in practice at the time of our survey. Beta coefficients for individual firms are often difficult to estimate. We obtain beta from the slope coefficient of the market model that we estimate. Since beta coefficients tend to be not very robust, we corroborate our results by calculating betas using various alternative estimations. These estimates involve different combinations of data frequencies (daily, weekly, and monthly), length of time period for estimating the model, and weighting schemes with industry betas. These estimation procedures are explained in Appendix A.2. Table II displays the summary statistics for beta coefficients estimated by using the methods described in Appendix A.2. Pair-wise simple correlations of estimated beta coefficients (not reported here) are relatively high, ranging from 0.55 to 0.90, with the exception of industry beta coefficients, Method (6), for which the correlations with the other beta estimates range from 0.25 to 0.39.16 Nevertheless, given that there is some dispersion in the estimated beta coefficients, calculating WACC and hurdle rate premiums on the basis of different beta estimates may be informative since it provides a robustness check for the hurdle rate premium we document. 16 The same conclusions are valid for the Spearman rank order correlations. 12 Excluding the beta estimates obtained from weekly data and daily data without lags due to potential noise problems at these frequencies, the range of beta means obtained from models (1), (2), (5), (6), and (7) is 0.93 to 1.03, with an arithmetic average of 0.95.17 Based on this data, it appears that our survey firms are fairly representative of the market. In the empirical tests we conduct, we report results where the beta is estimated from the market model that uses monthly returns over a five year period, i.e. Method (1). This choice is based on the fact that this particular estimation method is most commonly used in empirical studies that use CAPM and forms the baseline model for most service beta providers. Results not reported here show that for the survey firms, the estimated cost of equity using the models mentioned above ranges from 10.42% to 11.07%, with a mean of 10.59%, under the assumption of 6.6% equity premium. The mean of the cost of levered equity based on the beta estimate from Method (1) is 10.42% (the median is 9.76%). When the lower equity premium of 3.6% is used in the computations the range is 7.64% to 7.99%, with a mean of 7.73% and the mean value for Method (1) is 7.64%. B. After-tax Cost of Debt, Debt/Equity Weights and the WACC Computations As discussed earlier, a very high percentage of the respondents (71.8%) indicates that the hurdle rate they report in the survey represents their WACC. We compute the WACC of these firms by using Compustat data. We discuss below the details of these calculations and the assumptions we use. We are able to obtain computed WACC for 83 firms for which we have matching Compustat and CRSP data. In Section IV we compare the self-reported WACC-based hurdle rates with computed WACC for 70 firms for which we have matching data for both, self-reported and computed WACC. In our computations, we make some assumptions regarding the cost of debt and tax rate components of WACC. For the before-tax cost of debt we use the survey participants’ answers to our question regarding what the interest rate on their senior debt is. We do not know whether their answers refer to the coupon rate or the yield to maturity of their senior bonds. Thus, for firms that have not issued debt recently, it is 17 We include Method (5) since, under this procedure, the four lags alleviate the problem of stale prices. 13 possible that their answers do not reflect the marginal cost of debt if they report coupon rates.18 However, given the secular decline of interest rates that started in the late 1990s and continued during the early 2000s, this should work against finding a hurdle rate premium. 88 respondents supply data on their before-tax cost of debt. Using Compustat data, we check whether firms that left the interest rate question blank had any debt. Out of the 39 non-responding firms we can match Compustat data for 28, and 16 of these firms did indeed have not any debt. Obviously, in the case of these firms, their cost of unlevered equity represents their WACC. However, our check also reveals that 12 firms had debt even though they left the interest rate question blank.19 For these firms we use their Z-scores to assign interest rates. If a firm’s Z-score is greater than 3, a score that indicates a very low probability of default (8 firms), we assign the 10-year Treasury bond rate in effect at the time of the survey plus 1 percent (5.3%). For the 2 firms with Z-scores of less than 1.81 (financially unhealthy firms), we assign the 10-year Treasury rate plus 4 percent (8.3%). Firms that have Z-scores in the interval between 1.81 and 3 (2 firms) are assigned a before-tax cost of debt of 6.3%. Given the narrow default risk spreads at the time of the survey, the assumptions we employ likely exaggerate cost of debt of the 12 firms in question and thus, if anything, the difference between our assumption and the actual spreads at the time of the survey introduces a bias against finding hurdle rate premiums. Finally, some firms report a rate below the 10-year Treasury rate (4.3% at the time of the survey). In the case of these firms, we add a spread of 0.5% to the Treasury rate. Therefore, all our WACC calculations assume cost of debt of at least 4.8%. 18 To rectify these and other potential problems in cost of debt figures, we attempted to use Compustat to obtain the bond ratings of survey firms and use the applicable rates at the time of the survey. However, we were unable to obtain sufficient bond rating information on long-term debt for differing maturities and seniorities. As a result, we ended up using the survey participants’ responses as the before-tax cost of debt. 19 Out of these 12 firms, 2 have less than 1% debt (as a fraction of market value of assets) and another 6 less than 5%. 14 We calculate a firm’s tax rate by dividing total income taxes (Compustat item #16) by income before taxes (#170). When item #16 or #170 is negative (tax credits and negative profits, respectively), we set the tax rate to zero. The tax rate we obtain in this manner, of course, reflects a firm’s average and not marginal tax rate. However, we were unable to obtain a sufficient number of observations on marginal tax rates. Finally, we cap the tax rate at 34 percent. To compute the weight of debt, we divide total debt (the sum of Compustat items #9 and #34), by total debt plus market value of equity (the product of number of shares (#199) and end of year stock price (#25), plus the book value of preferred stock (#130)). For the weight of equity we use (1 – weight of debt). The weights used in calculating WACC should be based on target rather than on current capital structure. Due to lack of data on target capital structures, we assume that the current capital structure is also the target capital structure. 30.1% of the firms in our sample indicate that they do not plan to change their capital structure during the next three years, while 24.4% were planning on having higher leverage, and 45.5% had the intention of using less debt in the future. While we do not know the exact planned debt-to-equity ratios, these figures suggest that using the current capital structure weights as proxies for target rates may not bias computed WACC. The summary statistics of computed WACC using the beta estimation methods (1), (2), (5), and (6) are reported in Table III for the two equity premium scenarios (Panels A and B). Nominal, computed WACC means under the 6.6% (3.6%) equity premium assumption range from 9.30% to 9.55% (7.18-7.31%). These nominal mean rates correspond to real rates of 6.95% to 7.19% (4.87-5.00%). The arithmetic average of the four different costs of capital calculations is 9.45 (7.26) in nominal terms and 7.09 (4.95) in real terms. Skewness and kurtosis statistics for the various WACC calculations indicate that, by and large, the distributions of WACC are symmetric and that the observations are centered on means and medians. 15 IV. Documenting the Existence of the Hurdle Rate Premium Puzzle The summary statistics on the hurdle rate premium (self-reported WACC – computed WACC) under the 6.6% and 3.6% equity premium assumptions, using the same four beta estimation methods as before, are displayed in Table III. These results show that the WACC used by our survey firms exceeds the WACC that we compute using CRSP and Compustat data. Under the 6.6% (3.6%) equity premium scenario, the hurdle premium ranges from 5.11 to 5.28% (7.33-7.45%).20 For Method (1), columns 2 and 6 show that the mean (median) hurdle rate premium is 5.28% (5.23%), and 7.45% (6.90%) for equity premiums of 6.6% and 3.6%, respectively. The magnitudes of these premium figures are substantial. Given that the mean self-reported hurdle rate is 14.1%, the mean hurdle rate premium is more than half of the mean hurdle rate used by managers when the equity premium is assumed to be 3.6%. The mean premium under the 6.6% equity premium accounts for more than a third of the mean value of the self-reported hurdle rate. Results not reported here show that even if for each individual firm the maximum computed WACC figure is used, the hurdle premium is still large (3.04% and 6.22% for equity premiums of 6.6% and 3.6%, respectively). Furthermore, any doubt about the presence of the hurdle rate premium is dispelled after the cost of levered equity computed with the maximum of the beta estimates (displayed in Table II, Method (8)) for each firm is compared with self-reported WACC. The differences in the means are 1.59% and 5.60% for the higher and lower equity premium assumptions. Since cost of equity represents the upper bound of what the hurdle rate should be, the fact that self-reported WACC even exceeds the more expensive of its two sources of capital for the maximum beta estimate provides strong evidence that hurdle rates used in practice are significantly higher than those rates predicted by theory. Figure 4 plots the distribution of the hurdle rate premium. 20 For all methods (1) to (7) from Table II, the ranges are 4.68-6.06% when the equity premium is assumed to be 6.6% and 7.12-7.84% when the calculations are based on an equity premium of 3.6%. 16 V. Investigation of the Hurdle Rate Premium Puzzle In this section, we empirically examine the variables that could explain this hurdle rate premium puzzle. Since the hurdle rate premium is the difference between self-reported and computed WACC, we investigate the hurdle rate premium puzzle by running bivariate regressions on both, the self-reported hurdle rate and the hurdle rate premium. These results are discussed in Section A. We again report two sets of results corresponding to our equity premium assumptions of 6.6% and 3.6%. In Section B we estimate multivariate regressions where the explanatory variables are chosen on the basis of statistical significance at the 5% level in the estimation of the bivariate models. The multivariate regressions enable us to see to what extent we are able to explain the hurdle rate premium puzzle and whether the bivariate regression variables retain their sign and statistical significance in the multivariate setting. In Section C we discuss some scenarios that may be consistent with this puzzle. A. Determinants of the Hurdle Rate Premium Puzzle: Bivariate Regressions In this section we examine the determinants of the hurdle rates (self-reported WACC) and the two hurdle rate premiums by estimating bivariate models. These results are displayed in Table IV. The explanatory variables we use fall into five general categories: measures of risk (beta, standard deviation of stock returns, measures of systematic and unsystematic risk), variables that are designed to capture firms’ growth opportunities (market-to-book ratio, cash-to-assets ratio, debt-to-assets ratio, average industry returns during the 5 years prior to the survey), firm size (log of assets and log of sales), variables related to firms’ financial health (Altman’s Z-Score and the current ratio), and variables that reflect the survey participants’ views regarding whether or not their firms are financially or managerially constrained. Additionally, we use the R-squares obtained from the estimation of the market model. The reason for including this variable in the analysis is that, while we find strong evidence that firms in our sample use CAPM in determining their cost of equity, it is possible that managers make subjective adjustments to the CAPM-based cost of equity when they do not have confidence in the beta estimates obtained from the 17 market model. We expect this to be the case especially when the market model regressions have low Rsquares. As can be seen from the columns 2-4 of Table IV, self-reported WACC is positively related to calculated WACC, beta, total risk (standard deviation of monthly stock returns during the 5 years prior to the survey date), and to both systematic and unsystematic risk measures. All coefficients are significant at the 5 percent level. The positive correlation of systematic risk with self-reported WACC implies that firms use CAPM in setting their hurdle rates. The positive correlation of total risk and hurdle rates could be driven by the systematic risk component of total risk. Alternatively, unsystematic risk may have incremental explanatory power indicating that both types of risk may play a significant role in the determination of hurdle rates.21 To judge the relative importance of systematic versus unsystematic risk we need to measure them in comparable terms. Since beta is the index of systematic risk, while standard deviation of stock returns captures the level of total risk, the estimated coefficients for these two variables make their comparison difficult. To make the comparison meaningful, we measure systematic risk as i m and unsystematic risk as 2 i2 i2 m . The results indicate that, independent of how it is measured, systematic risk is an important determinant of hurdle rates. The results also show that, even though managers appear to use CAPM, to some extent they apparently incorporate unsystematic risk in their hurdle rates.22 Judging by the relative size of the coefficients (the estimated systematic risk coefficient is twice as big as the unsystematic risk coefficient), managers appear to consider systematic risk to be much more important. To further examine the relative importance we estimate a regression of hurdle rates on both, systematic and unsystematic risk. Results 21 Poterba and Summers (1995) report results indicating that for their sample of survey firms, neither systematic nor unsystematic risk is correlated with self-reported hurdle rates. 22 There is some evidence reported in the literature that unsystematic risk plays a role in the determination in the required rate of return of investors. See for example, Goyal and Santa-Clara (2003), Barberis and Huang (2001), and Malkiel and Xu (2001). 18 not displayed here show that when systematic risk is accounted for, the t-statistic for the unsystematic risk coefficient drops from 3.16 in the bivariate regression to 2.02. However, the adjusted R-square increases from 0.09 in the bivariate systematic risk regressions to 0.14 when both risk measures are included. This suggests that unsystematic risk still has some incremental explanatory power. The relative importance of systematic risk over unsystematic risk in the determination of hurdle rates is also supported by survey results. As we discuss in Section VII, when the participants are asked to judge the importance of systematic and unsystematic risk in the determination of their hurdle rates, the mean values for these answers indicate that CFOs consider systematic risk to be more important than unsystematic risk (mean scores of 0.86 versus 0.68 as shown in Table VI). Column 6 of Table IV shows that the systematic risk variables and the computed WACC are negatively and significantly (except for unsystematic risk) correlated with the hurdle rate premium when the equity premium is assumed to be 6.6%. However, this negative relationship may be an artifact of how the hurdle rate premium is defined. For example, for a given level of the self-reported hurdle rate, as beta increases, computed WACC also increases. The increase in computed WACC would in turn cause the stub portion of the self-reported hurdle rate (the hurdle rate premium) to become smaller. Thus, beta and the hurdle rate premium could be negatively correlated by construction. Furthermore, when the equity premium is 6.6%, a one unit increase in beta is expected to make the stub portion of the self-reported hurdle rate much smaller than when the equity premium of 3.6% is used. The results in Table IV confirm this is indeed the case. The absolute values of the estimated coefficients for the risk variables are smaller under the lower equity premium scenario and they also are statistically insignificant. The general point is that whether the risk variables are significant or not in explaining self-reported hurdle rates, the negative relationships between risk-related explanatory variables and the hurdle rate premium may be driven by the mechanics of the definition of the hurdle rate premium. 19 However, there may also be a fundamental reason for the negative correlation between systematic risk and hurdle rate premiums. The results displayed in Table IV show that under both equity premium assumptions there is a negative correlation between the hurdle premium and the median R-squares obtained from the estimation of the market model for the firms in the same industry (two-digit SIC code) as the survey firms.23 The coefficients are statistically significant at the 1 percent level. Why would hurdle rate premiums be higher when the goodness of fit of the market model becomes poorer? A possible explanation of this result is that when the goodness of fit is poor, managers may be less confident that their estimated beta reflects their “true” beta. It seems reasonable that in such cases they would be reluctant to use CAPM in a mechanical manner, but instead make subjective adjustments to their CAPMbased cost of equity. Of course, given their lack of confidence in their estimated betas when the market model has poor goodness of fit, the adjustment to the CAPM-based cost of equity could be in the upwards or downwards direction depending on their intuition about whether their “true” beta is higher or lower than the beta estimate that the market model generates. This would suggest that CFOs would be just as likely to use a higher cost of equity than what the CAPM dictates, as they are to use a lower figure. Nevertheless, this does not seem to be the case. First, virtually all firms, including those where the market model has reasonably high R-squares, use hurdle rates that are higher than their computed cost of capital. Second, the subjective adjustment to the computed WACC in the upward direction increases in magnitude as the confidence regarding beta estimates declines, i.e., when R-squares of the market model become smaller. One possible explanation for the two stylized facts in question may be that risk aversion considerations account for this behavior. Managers may believe that unless the R-square of the market model is extremely high, to be on the safe side they should use a cost of equity figure that is higher than the CAPM-based cost of equity. It follows then that they may become even more risk-averse as their 23 The negative relationship holds for our survey firms as well. But, given the relatively small sample size, in order to eliminate the possibility that outliers may disproportionately affect the estimates and to obtain more robust results, we estimate the market model for individual firms in the industry (two-digit SIC code) to which the survey firms belonged and use the median R-squares from each industry. The survey firms belong to 29 two-digit SIC codes. For each SIC code we have between 7 and 899 firms with a mean of 156 firms. 20 level of confidence regarding their estimated beta declines and, as a result, may add higher premiums to their computed WACC. The above discussion assumes that the positive difference between self-reported WACC and computed WACC is driven by a positive discrepancy between the CAPM-based cost of equity and the cost of equity as perceived by managers. It can be argued that the hurdle rate premium may be due to differences between weights and cost of debt used in our computed WACC calculations, as compared to what managers use for these two variables in determining their self-reported hurdle rates. In particular, we use market value based weights, while it may be the case that managers use book values. Furthermore, while we use market interest rates, managers may be using the coupon rate of their bonds as their cost of debt. However, even if such differences exist, we consider them to be unlikely drivers of the hurdle rate premium. First, using book value weights would exaggerate the importance of debt in firms’ capital structure since for most firms the market value exceeds the book value of their equity. In fact, 90% of the survey firms have market-to-book equity ratios equal or greater than one. In other words, if managers are indeed using book weights instead of market weights, the hurdle rate premium we document is smaller than the “actual” hurdle rate premium. Second, the cost of debt component of WACC is also unlikely to account for the hurdle rate premium that we document. Firms tend to refinance their debt when interest rates decline by a sufficient amount. Thus, coupon rates and yield to maturities of bonds are unlikely to differ by a large magnitude given that the interest rates have declined prior to our survey date. Additionally, given the small (after-tax) magnitude of cost of debt relative to the cost of equity, it is unlikely that yield to maturity versus coupon rate differences would significantly affect the hurdle rate premium data that we construct. In fact, we have a small sample of seven firms which report that their hurdle rate represents their cost of equity. We 21 compute the CAPM-based cost of equity of these firms and compare it with their self-reported cost of equity. Admittedly, this is a very small sample, but the cost of equity premium (self-reported cost of equity – CAPM-based cost of equity) is positive and large in magnitude for each one of these firms. The mean value of their cost of equity premium is 9.59, the median is 7.12, and the minimum is 2.67 under the 6.6% equity premium scenario.24 Based on this limited data and more importantly, the conceptual points we discussed above, we conclude by a process of elimination that it is likely that the source of variation in hurdle rate premium comes from differences in the cost of equity used by managers compared to the CAPM-based cost of equity. What explains the negative correlation between the hurdle rate premium and beta (statistically significant at the 1% level when the equity premium is assumed to be 6.6%)? Above, we discussed that the inverse correlation between these two variables may be purely mechanical. However, we also raised the possibility that the negative correlation is driven by how confident managers are that their estimated beta reflects their “true” beta. To further investigate this possibility we estimate a bivariate model where the dependent variable is the median industry R-square of the market model and the explanatory variable is median industry beta. Results not displayed here show that the estimated coefficient for the beta variable is 0.026 (with a t-statistic of 4.09), the estimated constant coefficient is 0.043 (the t-statistic is 5.28), and the R-square of the regression is 0.38.25 Thus, when market model regressions have low beta estimates, they apparently also have low R-squares. In fact, as shown in Table IV, when systematic risk is used as an explanatory variable in the bivariate hurdle rate premium regressions, the estimated coefficient is negative and significant at the 1% level under the 6.6% equity premium assumption. The estimated coefficient continues to have a negative sign when the equity premium is assumed to be 3.6%, although is 24 For the second scenario, with an equity premium of 3.6%, the statistics are: mean 12.45, median 9.20, and minimum 4.20. 25 The results appear to be robust. In fact, when we estimate the same model for the median R-squares and the median beta of all two digit SIC code firms (SIC codes with at least 20 firms), and not just the industry that survey firms are in, the estimated coefficient for the median industry beta is 0.066 (t-statistic is 13.2) and the R-square of the regression is 0.70. 22 becomes statistically insignificant. However, when both, systematic risk and the median industry Rsquares, are used as explanatory variables and the equity premium is assumed to be 3.6%, systematic risk is no longer significant (the estimated coefficient is 0.00 with a t-value of 0.96), while the estimate for the R-square variable is -35.20 and it is statistically significant at the 1% level. When 6.6% is used as the risk premium, both low levels of systematic risk (-0.14, significant at the 1% level) and high R-squares variable (-38.49, significant at the 1% level) appear to account for high risk premiums independent of each other. This result suggests that even though betas and R-squares of the market model are positively correlated, the goodness of fit associated with the estimation of the market model may be more important than the estimated beta in explaining the hurdle rate premium since the goodness of fit variable is statistically significant in trivariate models under both equity premium scenarios, while beta is significant only when the equity premium is 6.6%. Table IV also displays the bivariate relationships between proxies for growth opportunities and the hurdle rate premium (and the self-reported hurdle rates). The growth opportunities proxies we use consist of market-to-book value of assets, cash-to-assets and debt-to-assets. The most frequently used growth opportunities proxy in the literature is the market-to-book ratio. As shown in Table IV, this variable is positively correlated with the hurdle rate premium under both equity premium assumptions. The level of significance is at the 5% level for the 6.6% equity premium assumption and at the 1% level under the 3.6% equity premium scenario. This result is consistent with the hypothesis that high growth firms behave in a more discriminating manner by using higher hurdle rates. We also consider cash-to-assets and debt-to-assets ratios as potential proxies for growth opportunities.26 While we do not know what portion of a firm’s cash is excess cash and what the borrowing capacity of a 26 Opler, Pinkowitz, Stulz, and Williamson (1999) find that firms with strong growth opportunities and riskier cashflows hold relatively high ratios of cash to total non-cash assets. Their finding supports our argument that such firms value financial flexibility. However, (excess) cash is only one component of financial flexibility. Firms with high growth opportunities are also likely to preserve their borrowing capacity. 23 given firm is, it is plausible that the levels of cash holdings are positively correlated with excess cash holdings. Similarly, we assume that low levels of debt are correlated with high levels of unused debt capacity. To the extent these assumptions are realistic, firms with high levels of cash and low amounts of debt are likely to be firms where financial flexibility is important. Given the strong forces of the market for corporate control and high levels of shareholder activism that exist in today’s corporate finance environment, firms that under-borrow and/or have high levels of excess cash typically suffer from low stock prices and are frequently forced to distribute their excess cash and borrow additional funds to take advantage of interest tax shields. The natural opt-out from this shareholder imposed pressure are firms with high growth opportunities. Investors tolerate such firms to operate below their “optimal capital structure” provided they have healthy growth opportunities, because they probably realize that financial flexibility enables these firms to avoid capital market access concerns and accept positive NPV projects in the future. Technology firms, which typically have substantial amounts of excess cash and very little debt, are the prime examples of firms that are not penalized by shareholders, in spite of their seemingly “under-levered” capital structure. The results in Table IV support our hypothesis that firms with higher cash-to-assets and lower debt-to-assets have higher hurdle rate premiums. The final growth opportunity proxy we use is the past performance of firms. For reasons discussed above, we use industry (two-digit SIC code) median monthly returns instead of individual firm returns to capture the past performance of survey firms’ past growth opportunities. To the extent managers subscribe to the notion that high growth opportunities experienced in the past are indicative of similarly rich growth opportunities in the near future (the median project life for our survey sample is 5.3 years), they may behave in a more discriminating manner in accepting projects relative to the firms which experienced mediocre performance during the recent past. Thus, the expected sign of the median past 5year industry return variable is positive in both the self-reported hurdle rate and the hurdle rate premium equations. The estimated coefficients in Table IV indicate that this indeed is the case. 24 We next examine two variables that potentially reflect a firm’s financial health; the current ratio and Altman’s Z-score. We use current ratio, which measures a firm’s liquidity, partially because this is the only statistically significant variable that Poterba and Summers (1995) find in explaining self-reported hurdle rates of their survey firms. While we find that the current ratio is not related to the hurdle rate variables, the Z-score which measures a firm’s overall financial health in the context of its default risk is positively correlated with the hurdle rate premium calculated under both equity premium scenarios. The Z-score is not considered to be linearly related to a firm’s default risk. Instead, firms with a Z-score below 1.81 are considered to be financially unhealthy, while Z-scores in the range of 1.81 and 3.0 are thought to represent firms that are partially healthy, and Z-scores above 3.0 are associated with financially healthy firms. For this reason, we use Z-scores to construct a variable with three categories. The categorical variable reflecting to what extent hurdle rates used by survey participants are adjusted for risk is also positively correlated with self-reported hurdle rates and hurdle rate premiums. Other variables that are not related to the hurdle rates and premiums in the bivariate regressions are firm size, whether it is measured as the log of sales or assets (we included this variable to see if larger firms have lower discount rates due to having easier access to capital markets), and whether or not the survey participants felt their firms faced managerial or capital constraints. In sum, Poterba and Summers (1995) find only the current ratio to be significant in explaining selfreported hurdle rates and comment that “a striking conclusion is that none of the traditional financial variables that may proxy for risk, like the firm’s stock market beta, correlates with hurdle rates.” Our results, on the other hand, show that risk related variables, the managers’ level of confidence regarding estimated betas, variables that reflect firms’ financial health, financial flexibility considerations that accompany rich growth prospects, and the past performance of the industry that the survey firms belong to are important determinants of the hurdle rates. As we discuss below, we find essentially the same set of variables to be statistically significant in the hurdle rate premium equations as well. 25 B. Determinants of the Hurdle Rate Premium Puzzle: Multivariate Regressions We consider the statistically significant variables of the bivariate regressions to be informative about why the hurdle rate premium exists. To see to what extent we are able to explain this puzzle, we first estimate multivariate regressions where the independent variables are the variables which have p-values of 0.05 or lower in the bivariate regressions. Then, we estimate hurdle rates using step-wise regressions. The multivariate regression approach is essential for two reasons. First, we want to see to what extent the independent variables we use explain the variation in the hurdle rate premiums. Second, even though the statistically significant variables have the “correct” sign in the bivariate models, we want to see whether or not they still have the expected sign, and also continue to be statistically significant in multivariate regressions. Table V displays the estimates of the multivariate regressions. Models (1) and (3) display the multivariate regression estimates when all the variables that are statistically significant (at the 5% level) in the bivariate regressions are used as explanatory variables. The difference between Models (1) and (3) is the equity premium assumption. Models (2) and (4) display the estimates obtained from the step-wise regression procedure that excludes variables that do not contribute towards an increased adjusted Rsquare. For the step-wise regression procedure, we set the significance level for removal from the model at 0.20. As the unsystematic component of total risk is not significant, and the two measures of systematic risk are perfectly correlated, we use i m as the systematic risk variable. 27 Estimates obtained from Model (1) indicate that none of the signs of the significant coefficients of the bivariate regression estimates are reversed. However, the estimated coefficient for the market-to-book ratio of assets becomes statistically insignificant, perhaps because this growth opportunities proxy is correlated with the other two growth proxies (median 5-year monthly returns and cash-to-assets ratio). 27 The estimate for the systematic risk variable we use has a somewhat higher t-statistic (in absolute value) than the estimate for beta in the bivariate models (-3.79 versus -3.56). 26 The step-wise regression estimates are displayed in Model (2). The adjusted R-square increases from 0.41 to 0.44 as a result of the step-wise regression procedure. Thus, we are able to explain 44% of the variation in the hurdle rate premium. Unlike Model (1), the regression in Model (3) does not include systematic risk because in the 3.6% equity premium scenario this variable is not statistically significant in the bivariate regression, on the other hand the variables debt-to-assets and risk adjustment to hurdle rates are added. Neither of these variables is statistically significant in Model (3) and the step-wise procedure excludes both of these variables again. The resulting adjusted R-square increases from 0.39 to 0.40. The end result is that essentially the same set of variables contributes towards explaining the hurdle rate premium puzzle under both equity premium assumptions. Additionally, neither the explanatory power of the regressions nor the sign and the significant coefficients appear to depend on the equity premium used. Thus, the results appear to be robust. Furthermore, adjusted R-squares in the range of 0.40 to 0.44 indicate that the explanatory variables we use go a long way towards explaining the hurdle rate premium puzzle. Nevertheless, given the important implications of the possible hurdle rate premium for individual firms and the economy as a whole (potential under-investment problems), the presence of the hurdle rate premium as well as its determinants should be the subject of future research. The need for further research is reinforced by the fact that the hurdle rate premium we report may not be sample-specific. Poterba and Summers (1995) comment that “the real discount rate implied by the responses is well above the historical real return on either debt or equity in U.S. financial markets.” While they do not compute the hurdle rate premium, their conjecture supports the existence of the hurdle rate premium. We document the magnitude of this premium and also find that it makes up a substantial portion (one half to one third) of the hurdle rate firms use in practice. While it appears that firms use a discount rate which is higher than the textbook version of WACC, it may be a mistake to conclude that hurdle rates used in practice firms are “incorrect”. In fact, our analysis 27 indicates that 40 to 44% of the variation in the hurdle rate premiums can be explained. Given the relatively high adjusted R-squares, combined with the fact that the estimated coefficients have the expected sign, a case can be made that using hurdle rates that are higher than what is dictated by the CAPM may be reasonable. C. Other Possible Explanations of the Hurdle Premium Puzzle In addition to the statistically significant variables discussed above, it is possible that the hurdle premium puzzle is related to managerial or capital market access constraints. Firms that have a significant number of profitable projects may not undertake all positive NPV projects due to managerial constraints, as discussed in Jagannathan and Meier (2002). Under such a scenario, firms may use high hurdle rates as a rationing device. Firms may also use higher hurdle rates due to capital market constraints. There is also an extended literature on capital constraints. While some studies, such as Fazzari, Hubbard, and Petersen (1988) and Rauh (2006) find that some firms face capital market access constraints, others, such as Kaplan and Zingales (1997), Cleary (1999), and Pulvino and Tarhan (2006), report findings that show firms are not hindered from raising funds in the external markets. Our evidence on the importance of managerial and capital constraints on the hurdle premium puzzle is mixed. The bivariate regression estimates indicate that these variables are not statistically significant. On the other hand, some managers express a degree of concern about the presence of managerial and capital market constraints. For example, 60.8% of the respondents “strongly agree” or “agree” with the following statement: “We cannot take all profitable projects due to limited resources in the form of qualified management and manpower.” Similarly, 43.2% of the survey participants expressed agreement or strong agreement with “there are some good projects we cannot take due to limited access to capital markets”. Additionally, there was agreement or strong agreement on the part of 51.2% of the managers about “we invest more in projects in years when the firm has more operating cash flows.” 28 VI. Cashflow Related Practices and Interactions between Cashflows and Hurdle Rates Surveys about investment decisions typically focus on questions about capital budgeting methods, hurdle rate related issues, such as whether or not firms use CAPM, and ignore the practice of calculating cashflows. One of the contributions of this paper is that we fill this gap by including questions on cashflows in our survey questionnaire. We consider cashflow-related practices to be important since all DCF techniques use cashflows as well as discount rates as inputs. Even if a firm is using a correct capital budgeting method, and computing its discount rate “correctly”, it could still undertake in value-destroying investments by making errors in the computation of project cashflows and/or by using incorrect cashflow/hurdle rate combinations. In Section A, we discuss our survey results on cashflows. In Section B we turn our attention to the interaction between cashflows and hurdle rates. A. Calculation of Cashflows, Sunk Costs, and Cannibalization of Existing Product Sales Table VII shows that 45.5% of the firms compute cashflows as: Earnings before interest and after taxes (EBIAT) + depreciation – capital expenditures – net change in working capital (i.e., unlevered cashflows) when evaluating projects. Levered cashflows, which are defined as net income + depreciation-capital expenditures – change in net working capital, appear to be the next popular cashflow measure used (25.2% of the firms use it). 16.3% of the firms apply an incorrect “unlevered cashflow” definition by not subtracting fixed and current assets investments. In sum, about 71% of the firms employ correct definitions of either levered or unlevered cashflows while the remaining 29% of the survey firms use cashflows that are defined incorrectly. Obviously, firms could use either levered or unlevered cashflows in evaluating projects, provided that they use the correct combinations of cashflows and discount rates. In Section B, we examine to what extent firms are successful in determining the correct cashflow/discount rate combinations. 29 Table VIII displays how the survey sample handles sunk costs and the loss of sales in existing products when new products are introduced (cannibalization or erosion). We address the sunk costs issue with the following question: “In valuing projects, do you incorporate into the cashflows the money you spent before making accept/reject decisions?” Surprisingly, 52.4% of the respondents answer this question affirmatively. When the survey participants are asked whether or not they subtract expected losses in the sales of existing products in evaluating the introduction of new products, 81.3% of the respondents unequivocally say “yes”, while only two respondents (1.8%) qualify their answer by checking the option that they would do so only if their competitors were unlikely to introduce products similar to the new product they are considering. 16 firms (14.3%) indicate that they would never adjust projected sales of new products for the erosion they will cause in the sales of existing products. Given the highly competitive nature of U.S. industries, it is surprising that 81.3% of the firms indicate that they would forecast sales for new products as if there were economic, technological, or legal barriers to entry.28 B. Interactions between Cashflows and Hurdle Rates As we discussed above, 71% of the firms calculate either levered or unlevered cashflows correctly. In this section, we first examine whether or not these firms match the cashflows they use with the appropriate discount rate. The intersection of two survey questions on cashflows and discount rates is displayed in Table IX. While 71.3% of the firms that responded to both questions use WACC as their hurdle rate, and 44.4% of the firms use unlevered cashflows, only 34.8% of the firms choose the correct combination of matching unlevered cashflows and WACC in evaluating investment projects. 19.1% of the respondents apparently make the mistake of discounting levered cashflows at their WACC. 28 Due to patent protection in certain industries such as the pharmaceutical and technology sectors, firms would be justified in accounting for sales erosion of existing products in introducing new products for the duration of their patents. However, given that 42% of our survey firms are in the manufacturing sector and an additional 10% are engaged in energy/transportation industries, the fact that 84% of the survey firms behave as if competitors would be unable to introduce similar products seems difficult to justify. 30 Furthermore, while 25.2% of the firms use levered cashflows, apparently only one of those firms makes the correct decision of discounting levered cashflows at the levered cost of equity. We next investigate whether firms incorporate expected inflation consistently in project cashflows and hurdle rates. The results are displayed in Table X. In the questionnaire, we deliberately did not pair nominal/real hurdle rates and cashflows in the same question. Instead, we asked whether they use nominal/real discount rates and nominal/real cashflows in two separate questions. In fact the two questions were in different sections (and different pages) of the questionnaire. We did this to minimize the possibility that they may stumble on the correct answer on grounds that the correct answer must be the one where the measurement of discount rates and cashflows need to be consistent. The table shows that 41.3% of the respondents use nominal cashflows and 49.6% use nominal hurdle rates. However, the table also shows that 29.8% of the respondents correctly match nominal hurdle rates with nominal cashflows. Similarly, while 58.7% of the respondents rely on real cashflows and 50.4% employ real hurdle rates, the real hurdle rate/real cashflow combinations represent 38.4% of the answers. Overall, our survey firms seem to be knowledgeable about the treatment of inflation since 68.2% of the sample incorporates inflation into their analysis correctly. When survey participants are asked whether they incorporate risk by adjusting the hurdle rate upwards or by adjusting project cashflows downwards, 56.7% of the respondents indicate that they make the adjustment in the hurdle rate (“important” or “very important”) while 60.6% of the firms make the adjustment in project cashflows.29 The survey also contains three questions on cross-border investments. One question was about the risk of domestic projects compared with similar foreign projects. 50.9% of the respondents consider foreign 29 25.2% of the respondents consider adjustments to both, hurdle rates and cashflows, to be “important” or “very important.” However, only 6.3% selected “very important” for both types of adjustments. 31 projects to be riskier than similar domestic projects. It appears that firms account for this incremental risk differently: 35.1% of the respondents would use higher hurdle rates than they use in similar domestic projects, while 15.8% of the firms prefer to deal with the higher risk of foreign projects by using more conservative cashflow projections. However, 40.4% of the firms indicate that they do not consider foreign projects to be riskier than similar domestic projects. The survey results also show that 90.4% of the firms handle currency denomination of cashflow/hurdle cross-border investments correctly: 50.0% of the firms indicated that they would evaluate both foreign project cashflows and hurdle rates in dollar terms, while 40.4% would consider both the cashflow and hurdle rate components of foreign investments in foreign currency-denominated terms. VII. Self-Reported Hurdle Rates Revisited In Section II we discussed the summary statistics on self-reported hurdle rates and the participants’ views on what the discount rate represents. In this section we discuss how frequently firms change their hurdle rates and whether or not multi-divisional firms use firm-wide versus divisional hurdle rates. Additionally, we examine how the survey firms handle strategic projects. A. Frequency of Hurdle Rate Changes Firms could under- or over-invest if they do not adjust their hurdle rates as market conditions change. In an environment where cost of debt and equity are declining, if firms do not lower their hurdle rates accordingly, they would run the risk of under-investing. Similarly, in an environment of increasing cost of capital, firms that do not adjust their hurdle rates upwards would suffer from over-investing. 52.5% of the firms in our sample indicate that they did not change their hurdle rates during the three years preceding the survey date. Apparently, one fourth of the firms in our survey have adjusted their hurdle rates once (24.6%), or made more than one adjustment (23.0%). This finding is in line with the survey results of Brigham (1975), Gitman and Mercurio (1982), and Bruner, Eades, Harris, and Higgins (1998) 32 who provide evidence that a relatively low portion of U.S. firms schedule reviews of their hurdle rates with some regularity.30 The finding that one half of our survey firms have not changed their hurdle rates for at least three years is not comforting. Given that during the pre-survey time period S&P returns were significantly negative (i.e., cost of equity increased), it is surprising that firms did not increase their hurdle rates.31 According to a 1999 article in The Economist, corporate practices on this issue are worse in Europe. The article argues that “while U.S. firms often review their hurdle rates, in continental Europe they do so sometimes, and in Britain, rarely”.32 While U.S. firms may fare better in adjusting their hurdle rates relative to European firms, given the serious consequences of under-/over-investment problems generated by the non-revision of hurdle rates in line with changes in market conditions, the potential damage inflicted by over half of the firms in our survey is likely to be significant. In our survey we also ask what significant factors would lead firms to change their hurdle rates. Table VI summarizes how respondents answer this question. On a scale of -2 (not important), to 2 (very important), cost of capital factors in the form of interest rate changes (considered to be very important or important by 79.3% of the survey respondents), and changes in the expected risk premium (also 79.3% of the respondents) are ranked highest in terms of their influence in triggering hurdle rate changes. On the other hand, variables reflecting the state of the economy, such as cyclical changes in the economy and political uncertainty are not considered to be very important. These responses imply that firms correctly assess that sensitivity of a firm’s fortunes to the cyclical phases of the economy, and its sensitivity to changes in the political environment are already factored into its beta. Thus, as expected, only changes in 30 Brigham (1975) finds that 39% of the firms in his sample change hurdle rates more frequently than once a year. Gitman and Mercurio (1982) report that 24.3% of their survey firms change their cost of capital annually, 24.3% less frequently than annually, and 21.5% change their hurdle rates “whenever environmental conditions warrant reevaluation”. Bruner, Eades, Harris, and Higgins (1998) report that 37% of their survey firms re-estimate cost of capital on an annual basis. 31 The annual S&P returns (with dividends) from 2000 to 2002 are -9.1%, -11.9%, and -22.1%. 32 See “How high a hurdle?” The Economist, May 8, 1999. 33 the risk-free rate and risk premium components of CAPM cause firms to adjust their cost of capital. This result not only corroborates the finding that the vast majority of CFOs use the CAPM (for example Graham and Harvey (2001) for U.S. firms; Bounen, de Jong, and Koedijk (2004) for European firms), but it also shows that firms use CAPM correctly. B. Multi-Segment Firms and Use of Firm-Wide versus Divisional Hurdle Rates We ask firms with multiple divisions or business segments to indicate how often they use a companywide hurdle rate on a scale from -2 (never) to 2 (always). A substantial portion of the respondents (80.3%) give an answer of “always” or “almost always” to this question, with a mean value of 1.24. It appears that multi-divisional firms by and large do not use proxy firms to determine divisional hurdle rates. 25.0% of the respondents indicate that they always or almost always use the hurdle rate of other firms in the particular industry that their division operates (the mean value is -0.45).33 We then refined our question further by asking them if they adjusted proxy firm data to their divisions by taking into account differences in leverage, tax rates, and costs of debt. Only 15.8% of the respondents always or almost always follow such a procedure (the mean value is -0.82). Earlier studies report similar findings. Gitman and Mercurio (1982) summarize previous studies and conclude that one third to half of the companies do not adjust hurdle rates for different projects. 34 Findings from our survey and earlier surveys show that corporate practice in this respect has not changed much during last 20 years or so (as can be seen again in Figure 1). In fact, there is other (anecdotal) 33 The reported percentages for these alternatives to company-wide hurdle rates are potentially biased upwardly, as firms that did not respond are likely not using those methods. 34 In Bierman’s (1993) sample, 61.8% of the firms report that they use their firm-wide hurdle rate for all divisions. Apparently, only 26.5% of the firms report that they use project-risk adjusted hurdle rates. He also finds that only 35% of the firms would ever consider using divisional rates, and just 6% of the CFOs rank divisional cost of capital to be an important issue. Bruner, Eades, Harris, and Higgings (1998) find that 41% of the firms do not make adjustment to company’s cost of capital to reflect the risk of individual investment opportunities, 33% only sometimes. 34 evidence which shows that the use of company-wide hurdle rates by multi-divisional firms is not uncommon even in the case of well-known multi-national firms.35 The distortions caused by the use of firm-wide hurdle rates for divisional projects are well known. Such behavior results in over-investment in risky divisions and under-investment in low risk divisions. In addition to destroying shareholder wealth, not using divisional hurdle rates also increases the risk profile of firms unintentionally since these firms would allocate capital away from low risk divisions and towards riskier divisions. The summary statistics on self-reported hurdle rates for the 52 firms that have multiple divisions and use company-wide hurdle rates “always” or “almost always” are very similar to those reported in Table I for the 101 firms for which we have the Compustat data to construct their computed-WACC. In fact the means of the two samples are identical and the medians differ by only 0.3%. C. Strategic Projects Survey participants find the strategic versus non-strategic nature of the projects, and the size of the projects to be somewhat important factors in the determination of hurdle rates (on a scale of -2 to 2, the mean scores are 0.70 and 0.68, respectively). When survey participants are asked how they would compare strategic projects and non-strategic projects, their responses indicate that, other things being equal, 59.3% of the respondents consider strategic projects to be more valuable than non-strategic projects. We also ask them how they would evaluate strategic projects. Their response is as follows: 24.6% of the participants express the view that they would 35 “[…] Siemens, a German industrial giant, last year started assigning a different hurdle rate to each of its 16 businesses, ranging from household appliances to medical equipment and semiconductors,” Economist, May 8, 1999 excerpt from the article “How high a hurdle?” 35 use a lower hurdle rate to account for the option value associated with strategic projects, while 34.7% of the firms answer that they capture the additional benefits in question by valuing the future projects made possible by the project in question separately and add this amount to the value of the stand-alone value of strategic projects. 40.7% of the respondents, on the other hand, argue that they do not treat strategic projects any differently. VIII. Capital Budgeting Methods We ask the survey participants to select their first and second choices for the capital budgeting methods they use. Figure 5 summarizes their responses. The options represent the discounted cashflow (DCF) methods (NPV, IRR, profitability index, and APV), as well as non-DCF methods such as payback and average rate of return. The survey results indicate that 87.5% of the respondents rank the DCF methods as either their first or second choice. The IRR technique was the first choice for 42.1% of the firms, followed closely by NPV (36.5%). These results, displayed in Figure 1, confirm the increased use of DCF-based capital budgeting methods over time. The figure shows that different surveys conducted over time indicate that DCF capital budgeting methods increased from less than 20% around 1960 to almost 100% for large firms in 2000. Payback period and discounted payback period are the only methods that are more often second rather than a first choice. About one third of the firms (30.6%) select payback, either discounted or not, as the second method. Typically, the payback period ranges from 1.9 to 4.3 years (means for the lower and upper bound). The average range of the payback periods is substantially shorter than the average project life of 6.8 years. 31.9% of the CFOs explicitly state that they do not use (discounted) payback at all. All firms with a typical project life of beyond 5 years (50.8% of all respondents in our sample) use DCF methods as their first or second choice, whereas firms with a typical project life of 5 years or less apply DCF significantly less, namely in 75.8% of the cases. 36 Next, we examine whether the use of DCF-based capital budgeting method is linked to the size of the firm. We pool the methods NPV, IRR, and adjusted present value or APV (3 firms) into a single category (DCF techniques), and the other methods into a second group. Table XI displays the fraction of firms that use DCF techniques as their primary capital budgeting method by different firm size groups. About two thirds (63.6%) of the small firms (defined as firms with maximum sales of $100 million) in our sample, use DCF techniques. For firms with sales below $500 million, the DCF-usage percentage increases to 70.4%. In the case of companies that report sales above the $1 billion threshold, the fraction jumps to 94.9%. Finally, all of the 15 respondents with sales in excess of $5 billion use DCF methods. These results show that while even smaller firms rely more on DCF methods than non-DCF methods by a significant margin, use of DCF methods increases with firm size.36 In sum, while there are differences based on average project life, firm size, and the sectors in which survey firms operate, it is clear that in today’s world the vast majority of firms rely on DCF methods in making their investment decisions. IX. Conclusion In this study we examine how firms make their investment decisions in practice. Since we know the identity of the sample firms we combine their survey responses with publicly available data. This enables us to document a hurdle rate premium puzzle. We show that the self-reported WACC that financial managers use to discount cash flows exceeds computed WACC by 5.3 to 7.5%, depending on the equity premium assumption used. When we examine this puzzle, we find that financial flexibility considerations associated with high growth opportunities, past performance of the industry that survey firms belong to, manager confidence in their estimated betas, and firms’ current financial health are important 36 We also find that the manufacturing firms in our sample rely less on DCF methods as their primary capital budgeting method than non-manufacturing firms (69.8% vs. 83.8%). We do not find any difference in DCF usage on the basis of firms’ leverage. Apparently, it is also the case that neither the age nor the tenure of CFOs is related to their firms’ use of DCF methods. 37 determinants of the puzzle in question. The multivariate regressions we estimate to explain this puzzle have adjusted R-squares of 0.40 and 0.44. Furthermore, based on our interpretation of the estimated coefficients of the explanatory variables, managers’ apparent habit of using hurdle rates that exceed their computed WACC may be reasonable. While the existence of the hurdle rate premium appears to be sensible, whether or not the magnitude of the premium is appropriate should be the subject of future research. Furthermore, given the significant implications of the hurdle rate premium puzzle that we document, the existence of this puzzle, as well as its determinants, also deserve further examination. Our other key findings involve the cashflow component of investment decisions. While our results show that firms by and large compute levered and unlevered cashflows correctly, they are not as successful in matching cashflows and discount rates correctly. Additionally, a significant percentage of survey firms incorporate inflation into their analysis in a consistent manner, and also successfully determine domestic currency/foreign currency denomination of cashflows and discount rates in cross-border investments. However, our survey firms seem to not do as good a job in accounting for some other cashflow related issues such as sunk costs and sales erosion that new product introductions generate. The survey firms also have somewhat of a checkered record when it comes to changing their hurdle rates when market conditions change, and also in using firm-wide versus divisional WACC. Finally, our findings regarding the capital budgeting methods used by firms confirm the results of earlier studies in that 88% of our survey firms use DCF-based capital budgeting methods and that use of DCF methods increases with firm size. In fact while only 64% of firms which have sales of $100 million or less use DCF methods, this ratio is 100% for firms with sales that exceed $5 billion. 38 Appendices A.1 Age, Experience, and Education of the Respondents This section summarizes the characteristics of the responding CFOs. Nearly half of the CFOs (44.6%) are between 40 and 49 years old. Seventy-eight percent fall into the age group 40-59. Experience in the job is evenly distributed across the three categories “less than 5 years”, “5-9 years”, and “10 years or more.” Two-thirds of the CFOs (65.5%) graduated from an MBA program and an additional 12.9% hold a nonMBA masters degree or a higher degree. The degree alone does not necessarily reflect the education of the CFO, as the typical MBA curriculum has changed over the years and the quality of the programs differ. To control for the former effect, we ask the survey respondents for the year they graduated from their last school. On average, twenty years have passed since a CFO completed his last degree (the median year is 1982). A.2 Beta Coefficient Estimation Procedures To check the robustness of beta coefficients, we estimate the market model using various procedures. The standard practice is to estimate the market model by running a regression of stock returns against the returns on a market index like the S&P 500. The accuracy of the estimation results depends on a sufficient number of observations. Two obvious ways to increase the number of observations are to use a longer time period or to measure the returns at a higher frequency. The disadvantage of the former approach is that it runs the risk of including historical data that may no longer be representative of the firm’s current and future sensitivity to macro-economic fluctuations. Daily data, on the other hand, tends to be noisy, especially for infrequently traded stocks. For this reason, we estimate the market model using various time period and frequency combinations. We also use some additional specifications including lagged market returns and beta coefficients of comparable firms. 39 (1) Regressing five years of monthly stock returns on returns of the S&P 500 from January 1999 to December 2003. This is the baseline calculation for most service beta providers (Bloomberg, Ibbotson, Merrill Lynch, Reuters, or Standard & Poor’s).37 We require a minimum of 20 observations, which excludes one firm from our sample. ri ,t i(1) i(1) rS & P ,t i(,1) t (1) (2) Estimating the Bloomberg adjusted beta. Bloomberg uses five years of monthly data for the “raw beta”, using price appreciation and ignoring dividends. Bloomberg then calculates an “adjusted beta” as i( 2) 0.66 i(1) 0.33 1 (2) Instead of raw returns based on price changes, we use returns that are corrected for dividends. (3) Beta estimation as in (1), with two years of weekly data. (4) Same as in (1), using two years of daily data. (5) To reduce the non-trading bias when using daily data we also estimate the characteristic line regression with lagged coefficients, as suggested by Scholes and Williams (1977) and Dimson (1979). We use the concurrent value of the S&P 500 index and four lags, corresponding to one trading week. ri ,t i(5) ∑0 i(,5j) rS & P ,t j 4 37 j i(,5) t (3) Value Line takes the NYSE composite index (Value Line) as the market index. Using the value-weighted or equally-weighted market index of the three major U.S. stock exchanges NYSE, AMEX, and Nasdaq (available at the Center for Research in Security Prices, CRSP) in place of the S&P 500 does not yield substantially different results. 40 The beta coefficient we obtain from this procedure is the sum of the five estimated beta coefficients in the above equation: β i(5) = ∑ 4 j =0 β i(,5j) (4) (6) In order to mitigate the problem that for some firms the R-squares from the market model are low, we calculate the industry beta at the two-digit SIC code level. We then use this equally weighted average beta of all firms with the same two-digit SIC code and at least 20 monthly returns leading up to December 2003 to represent the beta coefficient of the individual firm. (7) A weighted average of the beta estimates based on monthly data for the individual firm in model (1) and the industry beta from (6). We compute the weights based on the R-squares of the two models. This procedure can be written as: (7) i Rindustry Ri2 2 i(1) 2 i( 6) 2 2 Ri Rindustry Ri Rindustry 2 (5) (8) For each individual firm we take the maximum of all beta coefficients from models (1) to (7). 41 References Barberis, Nicholas, and Ming Huang, 2001, Mental accounting, loss aversion, and individual stock returns, Journal of Finance 56(4), 1247-1292. Bierman, Harold J., 1993, Capital budgeting in 1992: A survey, Financial Management 22(3), 24. Blanchard, Olivier J. 1993, Movements in the equity premium, Brookings Papers on Economic Activity 2, 75-118. Brealey, R.A., Myers, S.C. ,2000. Principles of Corporate Finance 6th ed. (Irwin/McGraw-Hill, New York, NY). Brigham, Eugene F., 1975, Hurdle rates for screening capital expenditure proposals, Financial Management 4, 17-26. Brounen, Dirk, de Jong, Abe, and Kees Koedijk, 2004, Corporate finance in Europe: confronting theory with practice, Financial Management 33(4), 71-101. Bruner, Robert F., Eades, Kenneth M., Harris, Robert S., and Robert C. Higgins, 1998, Best practices in estimating the cost of capital: survey and synthesis, Financial Practice and Education 8, 13-28. Cleary, Sean, 1999, The relationship between firm investment and financial status, Journal of Finance 54(2), 673-692. Dimson, Elroy, 1979, Risk measurement when shares are subject to infrequent trading, Journal of Financial Economics 7, 197-226. Fama, Eugen F., and Kenneth R. French, 2002, The equity premium, Journal of Finance 57(2), 637-659. Fazzari, Steven M., Hubbard, R. Glenn, and Bruce C. Petersen, 1988, Financing constraints and corporate investment, Brookings Papers on Economic Activity 1, 141-195. Fremgen, J. M., 1973, Capital budgeting practices: A survey, Management Accounting 54(11), 19-25. Gillham, Bill, 2000, Developing a questionnaire (Continuum, New York, NY). Gitman, Lawrence J., and John R. Forrester Jr., 1977, A survey of capital budgeting techniques used by major U.S. firms, Financial Management 6, 66-71. 42 Gitman, Lawrence J., and Vincent A. Mercurio, 1982, Cost of capital techniques used by major U.S. firms: Survey and analysis of Fortune’s 1000, Financial Management 11, 21-29. Goyal, Amit K., and Pedro Santa-Clara, 2003, Idiosyncratic risk matters! Journal of Finance 58(3), 9751007. Graham, John R., and Campbell R. Harvey, 2001, The theory and practice of corporate finance: evidence from the field, Journal of Financial Economics 60, 187-243. Graham, John R., and Campbell R. Harvey, 2006, The Equity Risk Premium in January 2006: Evidence from the Global CFO Outlook Survey, working paper available at SSRN. Ibbotson, 2004. Yearbook (Ibbotson Associates, Chicago, IL). Istvan, Donald F., 1961, The economic evaluation of capital expenditures, Journal of Business 36, 45-51. Jagannathan, Ravi, McGratten, Ellen R., and Anna Scherbina, 2001, The declining U.S. equity premium, Quarterly Review, Federal Reserve Bank of Minneapolis. Jagannathan, Ravi, and Iwan Meier, 2002, Do we need CAPM for capital budgeting? Financial Management 31(4), 55-77. Kaplan, Steven N., and Luigi Zingales, 1997, Do investment-cash flow sensitivities provide useful measures of financing constraints? Quarterly Journal of Economics 112, 169-215. Klammer, Thomas P., 1972, Empirical evidence of the adoption of sophisticated capital budgeting techniques, Journal of Business 45(3), 387-397. Klammer, Thomas P., and Michael C. Walker, 1984, The continuing increase in the use of sophisticated capital budgeting techniques, California Management Review 27(1), 137-148. Malkiel, Burton G., and Yexiao Xu, 2002, Idiosyncratic risk and security returns, working paper, SOM, The University of Texas at Dallas. Mao, James C.T., 1970, Survey of capital budgeting: Theory and practice, Journal of Finance 25(2), 349360. Moore, James S., and Alan K. Reichert, 1983, An analysis of the financial management techniques currently employed by large US corporations, Journal of Business, Finance, and Accounting 10, 623-645. 43 Morgan, David L. ,1988. Focus groups as qualitative research (Sage Publications, Newbury Park, CA). Opler, Tim, Pinkowitz, Lee, Stulz, René, and Rohan Williamson, 1999, The determinants and implications of corporate cash holdings, Journal of Financial Economics 52, 3-46. Petry, Glenn H., 1975, Effective use of capital budgeting tools, Business Horizons (October), 57-65. Petty, William J., Scott, David F., and Monroe M. Bird, 1975, The capital expenditure decision-making process of large corporations, The Engineering Economist 20(3), 159-172. Poterba, James M., and Lawrence H. Summers., 1995, A CEO survey of U.S. companies’ time horizon and hurdle rates, Sloan Management Review (Fall), 43-53. Pulvino, Todd, and Vefa Tarhan, 2006, Cash flow sensitivities with constraints, working paper, Northwestern University. Rauh, Joshua, 2006, Investment and financing constraints: Evidence from the funding of corporate pension plans, Journal of Finance 61(1), 33-72. Schall, Lawrence D., Sundem,Gary L., and William R. Geijsbeek Jr., 1978, Survey and analysis of capital budgeting methods, Journal of Finance 38(1), 281-287. Scholes, Myron, and Joseph Williams, 1977, Estimating betas from nonsynchronous data, Journal of Financial Economics 5, 309-327. Singer, Eleanor, and Stanley Presser, 1989. Survey research methods: A reader (University of Chicago Press, Chicago, IL). Trahan, Emery A., and Lawrence J. Gitman, 1995, Bridging the theory-practice gap in corporate finance: A survey of chief financial officers, Quarterly Review of Economics and Finance 35(1), 73-87. Wadhwani, Sushil B.,1999, The US stock market and the global economic crisis, National Institute Economic Review, 86-105. Welch, Ivo, 2000, Views of financial economists on the equity premium and on professional controversies, Journal of Business 73 (4), 501-537. 44 Table I: Summary statistics for self-reported hurdle rates. The table shows summary statistics of the self-reported hurdle rates. The hurdle rates represent the nominal rate that the company has used for a typical project during the previous two years. Self-reported hurdle rates that represent the cost of levered or unlevered equity are converted to their WACC equivalents. This conversion procedure is explained in Section II.C. Out of the 119 respondents to this question, 103 use either WACC, cost of equity, or cost of unlevered equity, and 16 answers fall into the category “other” and are dropped. For two out of the 17 firms that use either cost of equity or unlevered cost of equity we cannot match the debt-equity ratio from Compustat to calculate the WACC equivalent. Therefore, we report the hurdle rates for the remaining 101 firms. Mean Median Minimum Maximum Std. dev. 25th percentile 75th percentile Skewness Kurtosis N 14.1 14.0 5.0 40.0 4.9 10.8 15.0 1.7 9.6 101 45 Table II: Summary statistics for beta estimates. The table contains summary statistics for seven beta estimates from historical market data, and the results when for each individual firm the maximum of all beta estimates is taken. The historical beta coefficients are calculated from regressions of dividend-adjusted stock returns, ri ,t , on S&P 500 index returns, rS & P,t , using various frequencies, time windows, and shrinkage procedures. (1) Historical beta using five years of monthly data: ri ,t = α i(1) + β i(1) rS & P ,t + ε i(,1t) . (2) The beta coefficient β i(1) adjusted towards the overall market value of one: β i( 2) = 0.66 × β i(1) + 0.33 × 1 . (3) Same regression as in (1) using two years of weekly data. (4) Same regression as in (1) using two years of daily data. (5) Regression of two years of daily returns on the concurrent and four lags of the S&P 500 index; ri ,t = α i(5) + beta then equals β i(5) = ∑ 4 j =0 β i(,5j) rS & P ,t j + ε i(,5) . The sum t ∑ 4 j =0 β i(,5j) . (6) The industry beta computed as the mean of the beta coefficients from regressions as in (1) for all peer firms within the same two-digit SIC category. The individual regressions are based on five years of monthly data. (7) A weighted average of the individual firm beta from (1) and the industry beta in (6). The R-square from the regression in (1) for the individual 2 firm i, Ri2 , and the average R-squares across the same regressions for all peer firms within the same two-digit SIC category, Rindustry , define (8) For each firm i the maximum of the seven beta coefficients (1) to (7) is taken. Ri2 2 Rindustry the weights: i(7 ) Ri2 (1) i 2 Rindustry Ri2 2 Rindustry i(6) . 46 Period Frequency Details Method # Mean Median Minimum Maximum Std. dev. 25th quantile 75th quantile Skewness Kurtosis N 5 years Monthly (1) 0.93 0.83 -0.27 3.12 0.72 0.42 1.18 1.15 4.30 92 5 years Monthly Adjusted towards one (2) 0.94 0.88 0.16 2.39 0.47 0.61 1.11 1.15 4.30 92 2 years Weekly 2 years Daily 2 years Daily Sum beta (3) 0.73 0.62 -0.32 2.14 0.50 0.45 1.01 0.73 3.57 93 (4) 0.69 0.64 -0.09 2.16 0.47 0.34 0.95 0.81 3.45 93 (5) 0.94 0.87 0.00 3.14 0.56 0.63 1.19 1.03 4.96 93 5 years Monthly Industry average (6) 0.93 0.83 0.13 1.94 0.48 0.67 0.92 0.96 3.19 94 5 years Monthly (1) and (6) weighted (7) 1.03 0.83 0.02 2.76 0.56 0.67 1.37 0.97 3.63 92 Various Various Maximum of (1) to (7) (8) 1.33 1.12 0.37 3.14 0.62 0.87 1.87 0.90 3.35 94 47 Table III: Summary statistics for the weighted average cost of capital (WACC) and the hurdle premium = self-reported hurdle rate WACC. The table shows summary statistics for the weighted average cost of capital (WACC) and the hurdle premium for the sample firms that reveal their identity and where we can match with CRSP and Compustat data. We define the hurdle premium as the difference between the self-reported hurdle rate and computed WACC. In calculating cost of equity from the Capital Asset Pricing Model (CAPM) we compare two scenarios for the equity premium (Panels A and B): The historical average excess return of large stocks over long-term bonds from January 1926 to December 2003 of 6.6%, and the median CFO forecast in December 2003 as reported by Graham and Harvey (2006). The risk-free rate in the CAPM is set to 4.3%, the rate for 10-year Treasury bonds at the time of the survey at the end of October 2003. For each equity premium scenario we tabulate statistics for WACC and the hurdle premium using four different methods to estimate beta coefficients. The specific regressions are detailed in the caption of Table II and explained in Appendix A.2. Panel A: Equity premium 6.6%. WACC 5 years 2 years Monthly Daily Adjusted Sum beta towards one (2) (5) 9.43 9.55 8.51 8.82 5.09 4.26 17.98 19.97 2.82 3.07 7.52 7.69 10.48 11.74 1.21 0.85 4.16 3.74 83 84 Hurdle premium 5 years 2 years Monthly Daily Adjusted Sum beta towards one (2) (5) 5.19 5.12 5.21 4.48 -2.98 -1.96 23.90 26.27 4.47 4.66 2.11 1.71 7.32 6.76 1.09 1.65 6.24 7.73 70 71 Period Frequency Details Method # Mean Median Minimum Maximum Std. dev. 25th quantile 75th quantile Skewness Kurtosis N 5 years Monthly (1) 9.30 8.16 2.68 21.96 4.08 6.78 10.82 1.23 4.34 83 5 years Monthly Industry beta (6) 9.50 9.03 5.13 17.06 2.94 7.76 10.20 1.05 3.58 85 5 years Monthly (1) 5.28 5.23 -6.96 21.07 4.88 2.55 8.04 0.27 4.30 70 5 years Monthly Industry beta (6) 5.11 4.95 -5.76 29.88 5.61 2.14 7.96 1.16 6.96 72 48 Panel B: Equity premium 3.6%. WACC 5 years 2 years Monthly Daily Adjusted Sum beta towards one (2) (5) 7.25 7.31 6.84 6.86 4.66 4.20 12.04 12.87 1.66 1.78 6.12 6.16 8.05 8.50 1.12 0.76 3.82 3.32 83 84 Hurdle premium 5 years 2 years Monthly Daily Adjusted Sum beta towards one (2) (5) 7.40 7.33 6.52 6.72 0.65 0.84 29.25 30.54 4.57 4.65 4.09 4.26 9.18 9.01 1.80 2.06 8.96 10.23 70 71 Period Frequency Details Method # Mean Median Minimum Maximum Std. dev. 25th quantile 75th quantile Skewness Kurtosis N 5 years Monthly (1) 7.18 6.57 3.38 14.21 2.32 5.68 8.31 1.08 3.85 83 5 years Monthly Industry beta (6) 7.28 6.97 4.64 11.87 1.69 6.15 7.66 0.96 3.29 85 5 years Monthly (1) 7.45 6.90 0.51 27.71 4.58 4.20 9.66 1.43 7.11 70 5 years Monthly Industry beta (6) 7.34 6.70 -0.57 32.51 5.19 3.95 9.35 1.81 9.22 72 49 Table IV: Bivariate regressions of the self-reported hurdle rate and the hurdle premium on selected financial variables. We run the regression below with the self-reported hurdle rate (columns 2-4) or the hurdle premium (columns 5-10) as the dependent variable y. We define the hurdle premium as the self-reported hurdle rate minus the computed weighted average cost of capital (WACC). The results for the hurdle premium regressions are reported for two scenarios for the equity premium: The equity premium is set to the historical average excess return of large stocks over long-term bonds from January 1926 to December 2003 of 6.6% (columns 5-7) or the CFO consensus forecast in December 2003 of 3.6% taken from Graham and Harvey (2006) (columns 8-10). yi a b (Financial variable) i ei The table shows the set of explanatory variables in the first column, the estimated coefficients a and b along with the t-statistics in parenthesis below, and the R-squares. The variable systematic risk is defined as i m , where i is the beta regression for firm i and m the standard deviation of the monthly returns on the S&P 500 over the past five years, and unsystematic risk is defined correspondingly as i2 i2 i2 using 5 years of monthly data, where i is the standard deviation of firm i over the same time horizon. The observations above/below the mean +/- two standard deviations are dropped for the two ratios market/book assets (2 observations with very small book value) and current ratio (4 observations). The variable “past average 5-year industry return” measures the median return over the past five years for all firms in CRSP with the same two-digit SIC code. The binary variables “capital constraints”, “managerial constraints”, and “adjust for optimistic cash flows” are 1 for firms answering that these are important or very important, and 0 otherwise. Similarly, the binary variable “risk adjustments to hurdle rates” is 1 if firms always or almost always adjust hurdle rates for risk (scores 1 and 2 on a scale from -2 to 2) and 0 otherwise. “Firm recently changed hurdle rate” is 1 if the firm has adjusted the hurdle rate during the past three years, and 0 otherwise. “Average industry R-square” is the median R-square of the market model regressions for all firms within the same two-digit SIC category, using 5 years of monthly data. Significantly different from zero at the 1% level ***; at the 5% level **; at the 10% level *. Financial variable Self-reported hurdle rate Equity premium 6.6% Constant. Coeff. R2 9.85 0.52 0.18 (7.08)*** (3.81)*** . 12.78 2.12 0.09 (13.91)*** (2.64)** . 11.38 0.06 0.13 (10.18)*** (3.36)*** . 12.76 0.12 0.09 (13.89)*** (2.67)*** . 11.67 0.06 0.12 (10.51)*** (3.16)*** . Hurdle premium Equity premium 6.6% Constant Coeff. R2 9.85 -0.48 0.16 (7.08)*** (-3.56)*** . 8.04 -2.94 0.18 (8.95)*** (-3.82)*** . 7.33 -0.04 0.05 (6.09)*** (-1.94)* . 8.03 -0.17 0.17 (8.92)*** (-3.79)*** . 6.79 -0.03 0.03 (5.72)*** (-1.46) . Equity premium 3.6% Constant Coeff. R2 8.70 -0.17 0.01 (4.73)*** (-0.71) . 8.07 -0.67 0.01 (8.74)*** (-0.84) . 7.30 0.00 0.00 (6.30)*** (0.14) . 8.05 -0.04 0.01 (8.71)*** (-0.81) . 7.08 0.01 0.00 (6.27)*** (0.38) . WACC Beta Standard deviation Systematic risk Unsystematic risk 50 Market/book assets Cash/assets Debt/assets Average 5-year industry return Ln(assets) Ln(sales) Z-score Current ratio Capital constraints Managerial constraints Risk adjustments to hurdle rates Adjust for optimistic cash flows Firm recently changed hurdle rate Average industry R-squares 13.47 0.54 (14.96)*** (1.32) 12.47 20.48 (19.41)*** (4.53)*** 15.47 -4.95 (21.06)*** (-2.21)** 11.15 2.21 (7.42)*** (2.54)** 14.56 -0.03 (8.53)*** (-0.13) 16.02 -0.27 (11.84)*** (-1.25) 10.85 1.44 (4.29)*** (1.43) 12.12 0.73 (15.36)*** (2.31)** 14.15 -0.02 (28.47)*** (-0.06) 14.13 0.04 (26.66)*** (0.08) 13.05 1.33 (16.14)*** (2.12)** 14.46 -0.63 (22.81)*** (-0.61) 14.97 -1.61 (20.67)*** (-1.62) 16.45 -23.30 (14.00)*** (-1.68)* 0.02 . 0.21 . 0.06 . 0.08 . 0.00 . 0.02 . 0.03 . 0.07 . 0.00 . 0.00 . 0.05 . 0.00 . 0.03 . 0.04 . 3.76 1.04 (3.74)*** (2.03)** 4.20 10.99 (5.59)*** (2.19)** 6.01 -3.41 (7.61)*** (-1.37) 2.19 1.95 (1.40) (2.13)** 5.55 -0.04 (2.94)*** (-0.15) 6.48 -0.20 (3.74)*** (-0.73) -1.67 2.82 (-0.61) (2.62)** 3.96 0.34 (3.90)*** (0.85) 5.30 -0.04 (8.76)*** (-0.10) 5.25 0.19 (8.51)*** (0.39) 4.23 1.37 (4.38)*** (1.91)* 5.14 0.41 (6.69)*** (0.33) 5.33 -0.04 (6.07)*** (-0.04) 8.98 -49.31 (7.85)*** (-3.66)*** 0.06 . 0.07 . 0.03 . 0.06 . 0.00 . 0.01 . 0.09 . 0.01 . 0.00 . 0.00 . 0.06 . 0.00 . 0.00 . 0.16 . 5.39 (5.69)*** 5.90 (8.80)*** 8.75 (12.27)*** 3.75 (2.62)** 8.02 (4.55)*** 9.64 (6.00)*** 0.29 (0.11) 5.24 (6.39)*** 7.51 (13.28)*** 7.39 (12.83)*** 6.23 (6.74)*** 7.60 (10.56)*** 7.50 (9.00)*** 10.40 (9.46)*** 1.32 (2.72)*** 15.70 (3.51)*** -6.08 (-2.69)*** 2.33 (2.77)*** -0.09 (-0.34) -0.37 (-1.45) 2.92 (2.92)*** 0.71 (2.19)** 0.13 (0.32) 0.29 (0.61) 1.48 (2.17)** -0.30 (-0.26) -0.08 (-0.07) -39.37 (-3.04)*** 0.10 . 0.15 . 0.10 . 0.10 . 0.00 . 0.03 . 0.11 . 0.07 . 0.00 . 0.01 . 0.07 . 0.00 . 0.00 . 0.12 . 51 Table V: Explaining the hurdle premium. The table shows the results for regressions of the hurdle premium on various financial variables. The estimated coefficients and the corresponding t-statistics in parenthesis are tabulated for two scenarios for the equity premium: The historical average excess return of the S&P index over the 10-year T-bond rate of 6.6% (from January 1926 to December 2003) and the CFO consensus forecast in December 2003 of 3.6% taken from Graham and Harvey (2006). For each equity premium scenario the hurdle premium is first regressed on all significant variables (at the 5% level) in the bivariate regressions from the previous Table IV. From the three risk variables essentially measuring systematic risk (WACC, beta, and systematic risk) only systematic risk is included. In the adjacent Models (2) and (4) are the coefficients for those variables that remain after using stepwise regression. The significance level for removal from the model is set to 0.20. Equity premium 6.6% (1) (2) t-statistic Coeff. t-statistic (-3.40)*** -0.16 (-3.92)*** (1.12) . . (2.36)** 10.73 (2.50)** . (2.05)** 1.73 (2.33)** (2.52)** 0.40 (2.73)*** . . . . . . (-2.58)** -35.14 (-3.03)*** (2.51)** 5.25 (3.25)*** 69 0.44 Equity premium 3.6% (3) (4) t-statistic Coeff. t-statistic . . . (0.71) 0.68 (1.64) (1.32) 10.14 (2.60)** (-0.03) . . (2.37)** 1.81 (2.52)** (1.88)* 0.51 (3.41)*** (-0.23) (0.42) . . (-2.99)*** -32.69 (-2.97)*** (1.75)* 3.13 (1.95)* 68 0.42 Model # Systematic risk Market/book assets Cash/assets Debt/assets Average 5-year industry return Z-score Current ratio Risk adjustments to hurdle rates Average industry R-squares Constant N Adjusted R-square Coeff. -0.15 0.48 10.23 . 1.53 0.41 . . -30.54 4.31 68 0.41 Coeff. . 0.36 6.76 -0.07 1.87 0.38 -0.10 0.24 -34.71 4.10 56 0.21 52 Table VI: Reasons to change hurdle rates and important factors to assess hurdle rates. The table analyzes which factors induce CFOs to change hurdle rate(s) and which factors are considered to be important to determine the level of the hurdle rate(s). The table shows the fraction of the responding firms that consider the factor in question as important or very important, the mean of all answers on a scale from -2 (not important) to 2 (very important), and the number of respondents. The bar charts illustrate the mean scores. Not important Very important -2 -1 0 1 2 Important 1. If you were to change your hurdle rate(s), how important would the following factors be? a) Interest rate changes. b) Cyclical changes in the economy. c) Cyclical changes in the industry(ies) you operate in. d) Changes in political uncertainty. e) Changes in the expected risk premium. f) Changes in the corporate tax rates. 2. a) b) c) d) e) f) How important are the following factors in determining the hurdle rate you use? Whether it is a short-lived or long-lived project. Whether it is a strategic or non-strategic project. Whether it is a revenue expansion or a cost reduction project. Whether it is a replacement project or a new investment. Whether it is a domestic project or a foreign project. Whether the project in question requires significantly more funds than the typical project your firm takes. Mean N 79.3% 35.3% 53.8% 27.1% 79.3% 40.8% 40.0% 59.5% 44.6% 46.2% 35.1% 65.8% 1.08 -0.03 0.36 -0.21 1.02 0.10 0.08 0.70 0.26 0.30 0.01 0.68 121 116 117 118 121 120 120 121 121 119 114 120 3. How important are the following risk factors in determining the hurdle rate? a) Market risk of a project, defined as the sensitivity of the project returns to economic conditions. b) Project risk that is unique to the firm and unrelated to the state of the economy. 68.6% 60.9% 0.86 0.68 118 115 53 Table VII: Calculation of cash flows. Summary of the answers to the question how firms calculate cash flows when evaluation projects. The questionnaire provided five alternatives to choose from, a) to e), and allowed for an open end answer under “other.” Tabulated are the absolute number and the fraction of firms employing a given method. A total of 123 CFOs answered this question. In evaluating projects the cash flows you use are calculated as a) earnings before interest and after taxes (EBIAT) + depreciation. b) earnings before interest and after taxes (EBIAT) + depreciation – capital expenditures – net change in working capital. c) earnings. d) earnings + depreciation. e) earnings + depreciation – capital expenditures – net change in working capital. f) Other. # of firms 20 56 7 6 31 3 Fraction 16.3% 45.5% 5.7% 4.9% 25.2% 2.4% Table VIII: Sunk cost and cannibalization. The table shows the number of firms and the corresponding percentage answering to two survey questions regarding sunk costs (total of 124 respondents) and cannibalization (112 respondents). 1. In valuing projects, do you incorporate into the cash flows the money you spent before the period when you make the accept/reject decision? a) Yes. b) No 2. If a new product will cause a decline in the sales of an existing product (erosion, cannibalization), do you subtract the erosion from the estimated sales figures of the new project? a) Yes. b) Yes, but only if competitors are likely to introduce a product similar to the new product. c) Yes, but only if the competitors are unlikely to introduce a similar product. d) No. # of firms 65 59 Fraction 52.4% 47.6% 91 3 2 16 81.3% 2.7% 1.8% 14.3% 54 Table IX: Consistency between hurdle rates and cash flow calculations. The rows in the cross-tabulation indicate what the self-reported hurdle rate represents and the columns denote five different ways to calculate cash flows, a) to e), plus the “other” category. Each cell displays the percentage of all 113 respondents for a given combination. The definitions of the cash flow calculations are: a) Earnings before interest and after taxes (EBIAT) + depreciation b) Earnings before interest and after taxes (EBIAT) + depreciation – capital expenditures – net change in working capital c) Earnings d) Earnings + depreciation e) Earnings + depreciation – capital expenditures – net change in working capital Hurdle rate WACC Equity levered Equity unlevered Other Total a) 11.3 0.9 1.7 2.6 16.5 b) 34.8 2.6 1.7 5.2 44.4 Cash flow calculation c) d) 2.6 3.5 0.9 0.0 0.9 0.9 1.7 0.9 6.1 5.2 e) 19.1 0.9 1.7 3.5 25.2 Other 0.0 0.9 0.9 0.9 2.6 Total 71.3 6.1 7.8 14.8 100.0 Table X: Consistency of nominal and real terms in hurdle rates and cash flows. The rows in the cross-tabulation show whether the firm uses a nominal or real hurdle rate, and the columns indicate whether cash flows are calculated in nominal or real terms. The cells contain the percentage of firms, out a total of 121 respondents to the two separate survey questions, that use the respective combination. Hurdle rate Nominal Real Total Cash flows Nominal Real 29.8 19.8 11.6 38.4 41.3 58.7 Total 49.6 50.4 100.0 Table XI: Fraction of firms using discounted cash flow techniques or payback. The discounted cash flow techniques (DCF) include net present value, adjusted present value, internal rate of return, and the profitability index. The table shows the absolute number of firms and the corresponding percentage using DCF techniques with increasing firm size. Firm size is measured by self-reported sales per year. A total of 126 firms responded to this question. Sales < $100 million < $500 million > $500 million > $1 billion > $5 billion # of respondents 44 71 54 39 15 DCF techniques first choice 28 63.6% 50 70.4% 48 88.9% 37 94.9% 15 100.0% Payback first choice 11 15 9 4 1 25.0% 19.7% 16.7% 10.3% 6.7% Payback first or second choice 21 47.7% 29 40.9% 19 35.2% 13 33.3% 2 13.3% 55 Figure 1: Adoption of discounted cash flow (DCF) methods, WACC, CAPM, and company-wide hurdle rates over time. The figure provides an overview of the survey literature on capital budgeting decisions of U.S. firms in practice. The studies are listed in chronological order below the horizontal time axis. The graph summarizes their findings regarding the percentage of firms that a) use discounted cash flow (DCF) methods, including net present value (NPV), adjusted present value (APV), internal rate of return (IRR), and the profitability index (PI); b) use the weighted average cost of capital (WACC) to discount cash flows, c) employ the Capital Asset Pricing Model (CAPM) to compute cost of equity, and d) use a company-wide hurdle rate. DCF methods 100% WACC CAPM Company-wide hurdle rate 80% 60% 40% 20% 0% 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 Istvan (1961) Schall et al. (1978) Bruner et al. (1998) Mao (1970) Klammer (1972) Gitman and Mercurio (1982) Graham and Harvey (2001) Fremgen (1973) Moore and Reichert (1983) Meier and Tarhan (2006) Brigham (1975) Klammer and Walker (1984) Bierman (1993) Petry (1975) Petty et al. (1975) Trahan and Gitman (1995) Gitman and Forrester (1977) Poterba and Summers (1995) 56 Figure 2: Company characteristics. The two panels summarize the self-reported industry affiliation and sales figures and show the percentage of firms within each category. A: Industry Retail, wholesale Mining, construction Manufacturing Communication, media Technology Services Transportation, energy Other 0 10 20 30 % of firms 40 40 B: Sales % of firms 0 10 20 30 <$100m $500-999m >$5b $100-499m $1-5b Figure 3: What the hurdle rate represents. A total of 117 firms responded to the question what the firm’s hurdle rate represents. The eleven firms that explicitly indicate that they add a premium to the weighted average cost of capital (WACC) as their hurdle rate are included in the category WACC. WACC Cost of levered equity Cost of unlevered equity Other 0 20 40 60 % of firms 80 57 Figure 4: Hurdle premium using two scenarios for the equity premium. We define the hurdle premium as the difference between the self-reported hurdle rate and the weighted average cost of capital (WACC) that we compute from CRSP and Compustat. We use the Capital Asset Pricing Model (CAPM) to infer the cost of equity with 4.3% as the risk-free rate (10-year Treasury bond rate in October 2003) and two scenarios for the equity premium (Panels A and B): A historical equity premium of 6.6% (return on large stocks minus the return on long-term government bonds from 19262003), and the median CFO forecast for the premium of the S&P 500 index over the 10-year Treasury bond yield in December 2003, reported by Graham and Harvey (2006). Beta coefficients are inferred from a market model regression of the firm’s monthly returns on returns on the S&P 500 over the past five years. A: Equity premium 6.6% 20 20 B: Equity premium 3.6% 15 % of firms % of firms 10 5 0 -10 0 % 10 20 0 5 10 15 0 10 % 20 30 58 Figure 5: Capital budgeting methods. The bar chart illustrates the ranking of the two preferred capital budgeting methods. For the eleven respondents that do not provide a ranking, a rank of one is assigned to all methods checked. Therefore, the total of first choices (149) exceeds the number of 126 respondents to this question. The number of second choices (85) is substantially lower as some firms rely on a single technique. The comments to the choice “other” include economic value added (EVA) and multiples. Internal rate of return (IRR) Net present value (NPV) Payback period Return on invested capital Discounted payback period Other Adjusted present value (APV) Profitability index Average rate of return 0 20 40 # of firms 1st choice 60 2nd choice 80 59