The Equity Risk Premium in September 2005: Evidence from the Global CFO Outlook Survey John R. Graham Fuqua School of Business, Duke University, Durham, NC 27708, USA Campbell R. Harvey* Fuqua School of Business, Duke University, Durham, NC 27708, USA National Bureau of Economic Research, Cambridge, MA 02912, USA ABSTRACT We analyze the results of the September 2005 survey of U.S. Chief Financial Officers (CFOs). We present expectations of the equity risk premium measured over a 10-year horizon relative to a 10-year U.S. Treasury bond. This multi-year survey has been conducted every quarter from June 2000 to September 2005. Each quarter the survey also provides measures of cross-sectional disagreement about the risk premium, skewness, and a measure of individual uncertainty. The individual uncertainty is deduced from the 80% confidence interval that each respondent provides for his or her risk premium assessment. We also present evidence on the determinants of the long-run risk premium. Our analysis suggests there is a positive correlation between the ex ante risk premium and real interest rates as reflected in Treasury Inflation Indexed Notes. The level of the risk premium also appears to track market volatility as reflected in the VIX index. JEL Classification: G11, G31, G12, G14 Keywords: Cost of capital, equity premium, long-term market returns, long-term equity returns, expected excess returns, disagreement, individual uncertainty, skewness, asymmetry, survey methods, risk and reward, TIPs, VIX ______________________________________________________________________________ *Corresponding author, Telephone: +1 919.660.7768, Fax: +1 919.660.8030, E-mail: [email protected]. We appreciate the comments of an anonymous referee. We appreciate the research assistance of Hai Huang and Runeet Kishore. Version September 19, 2005. This is part of a series of papers that will be posted to FEN each quarter.Graham-Harvey: Equity risk premium in September 2005 1 1. Introduction We analyze the results of the September 2005 survey of Chief Financial Officers (CFOs) conducted by Duke University and CFO Magazine. In particular, we poll CFOs about their long-term expected return on the S&P 500. Given the current 10-year T-bond yield, we provide estimates of the equity risk premium and show how the premium changes through time. We also provide information on the disagreement over the risk premium as well as average confidence intervals. 2. Method 2.1 Design The quarterly survey of CFOs was initiated in the third quarter of 1996.1 Every quarter, Duke University polls financial officers with a short survey on important topical issues (Graham and Harvey, 2005). The usual response rate for the quarterly survey is 5%-8%. Starting in June of 2000, a question on expected stock market returns was added to the survey. Fig. 1 summarizes the results from the risk premium question.2 While the survey asks for both the one-year and ten-year expected returns, we focus on the ten-year expected returns herein, as a proxy for the market risk premium. The executives have the job title of CFO, Chief Accounting Officer, Treasurer, Assistant Treasurer, Controller, Assistant Controller, or Vice President (VP), Senior VP or Executive VP of Finance. Given that the overwhelming majority of survey respondents hold the CFO title, for simplicity we refer to the entire group as CFOs. 2.2 Delivery and response In the early years of the survey, the surveys were faxed to executives. The delivery mechanism was changed to the Internet starting with the December 4, 2001 survey. Among other things, we now collect the respondents’ IP addresses (though not their identity or company) and are able 1 The surveys from 1996Q3-2004Q2 were partnered with a well-known national organization of financial executives. The 2004Q3-2004Q4 surveys were solely Duke University surveys, which used Duke mailing lists (previous survey respondents who volunteered their email addresses) and purchased email lists. The surveys from 2005Q1 to present are partnered with CFO Magazine. The sample includes both the Duke mailing lists and the CFO Magazine subscribers that meet the criteria for policy-making positions. 2 In the text and in Fig. 1, we refer to the most recent survey as 2005Q4. The table refers to the exact closing date of the survey, which was August 29, 2005.Graham-Harvey: Equity risk premium in September 2005 2 examine consistency of responses across different surveys. Respondents are given four business days to fill out the survey. Usually, two-thirds of the surveys are returned within two business days. The response rate of 5-8% could potentially lead to a non-response bias. There are four reasons why we are not overly concerned with the response rate. First, our response rate is within the range that is documented in many other survey studies. Second, Graham and Harvey (2001) conduct a standard test for non-response biases (which involves comparing the results of those that fill out the survey early to the ones that fill it out late) and find no evidence of bias. Third, Brav, Graham, Harvey and Michaely (2005) conduct a captured sample survey at a national conference in addition to an Internet survey. The captured survey responses (to which over two-thirds participated) are qualitatively identical to those for the Internet survey (to which 8% responded), indicating that non-response bias does not significantly affect their results. Fourth, Brav et al. contrast survey responses to archival data from Compustat and find archival evidence for the universe of Compustat firms that is consistent with the responses from the survey sample. 2.3 Data integrity In each quarter, we trim the top two and bottom two risk premium observations. Given that we have, on average, more than 200 responses each quarter, this implies a less than 1% trim in each of the tails. In addition, of the over 5,300 survey observations, there was only a single observation (in the June 2000 survey) that we consider not credible. The trimmed and untrimmed data are very similar with the exception of the June 2000 survey. There are two other steps that we take. First, for the purpose of some of our statistics, we require that the expected risk premium forecast be no more than the best-case scenario and no less than the worst-case scenario. If the ordering is violated, then the observation is deleted. Second, there are two instances in which