國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
10
3. Empirical Analysis
3.1 Data
We construct our dataset using data from Compustat database for firms’ accounting data, ExecuComp database for managers’ compensation data, CRSP database for stock price and return data, and SDC Merger and Acquisition database for M&A data. First, following Harford (2014), we include all the S&P 1500 firms but financial firms (4-digits SIC 6000-6999) and regulated utilities (4-digits SIC 4900-4999) in Compustat ranging from 1990 to 2016. Second we merge the Compustat data with ExecuComp to obtain CEO compensation data. After merging with Execucomp, our data range shorten from 1993 to 2016 for ExecuComp only has data available after 1992 and we lag most of our independent variable by 1 year. Third, we merge the data with CRSP with same range to compute one of our control variable, Stock return.
Finally, we merge with SDC for M&A data. Here include all U.S. domestic completed merger, acquisition of majority interest, and asset acquisition undertaken by S&P 1500 firms ranging from 1993 to 2016. After deleting all the missing observations, we have 22800 firm years and 4248 M&A deals from 1993 to 2016. The mean of overconfident CEOs in our sample is 0.3767.
We show our summary statistics in Table 2.
Table 3 shows the statistics of overconfident CEOs. First, Panel A exhibits the distribution of overconfident CEOs in each year. It is noticeable that the percentage of overconfident CEOs decline in 2001 and 2008-2008 periods, which induce that recessions might make overconfident CEOs be less overconfident. We further examine if the percentage of overconfident CEOs is lower during recessions than during non-recessions. The statistics are shown in Panel B. In line with our expectations, the difference in the percentage of overconfident CEO is lower during recessions than non-recession periods no matter what measurement of recession years we adapt. Panel C exhibit the difference in mean of overconfident CEOs before and after the 2001 Recession. From the statistics, we can’t tell that if the mean of overconfident CEOs differ before and after 2001. However, Panel D reports that the means of overconfident CEOs is lower after the 2008 Recession
3.2 CEO Overconfidence
In this paper, we follow the logic of Malmendier and Tate (2008) and use the method of Campbell, Gallmeyer, Johnson, Rutherford, and Stanley (2011) respectively to construct
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
11
managerial overconfidence measurement. Unlike outside investors who can diversify their portfolio easily, CEOs have a large portion of their asset and human capital invested in firm which leave them unable to diversify the idiosyncratic risk they face. Under this scenario, risk-averse CEOs should exercise their options as soon as possible when the moneyness of the option is high enough in order to avoid the under-diversification problem (Hall and Murphy, 2002). If a CEO do not act in this way, which is postponing their option exercise when they are eligible to exercise and the stock price is sufficiently high, we can infer him as overconfident for he overestimates his ability to push the stock price higher in the future to capture more benefit. Following this logic, we can identify overconfident CEOs by their option holding/exercising behavior.
Following Campbell et al. (2011), we define a CEO as High-optimism (overconfident) by his postpone of option-exercising. Instead of using actual exercise price to calculate the moneyness of the options a CEO holds or exercises, we derive the moneyness using estimated average exercise price. To estimate average exercise price, we first compute the realizable value per option by dividing total realizable option value of exercisable options (ExecuComp variable OPT_UNEX_EXER_EST_VAL) by number of exercisable options (OPT_UNEX_EXER _NUM). Then we subtract realizable value per option from stock price at the fiscal year end (Compustat variable PRCC_F) to obtain average exercise price. Finally, the moneyness of an option is calculated by dividing the per-option realizable value by estimated average exercise price. By Campbell et al. (2011)’s definition, a High-optimism CEO would hold exercisable option which is over 100% in the money at least twice in his tenure. On the other hand, we also need to define CEOs who are underconfidence or having normal degree of confidence to complement the overconfidence measure. Similar to High-optimism CEO, we classify a CEO as Low-optimism (underconfidence) when he at least twice in his tenure exercise options that is under 30% in the money and do not hold any option which is over 30% in the money. There’s a little difference when we compute the moneyness of exercised options. We first compute average exercise price for the exercised option as follow: PRCC_F-OPT_EXER_VAL/
OPT_EXER_NUM. Then we divide per-option realized value by average exercise price to derive moneyness of exercised options. Finally, we define a CEO who is either classified as High-optimism or Low-optimism as Moderate-optimism (CEO having normal degree of confidence). They hold or/and exercise options which are between 30% and 100% in the money.
In our paper, we define High-optimism CEO as overconfident CEO and both Low-optimism
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
12
CEO and Moderate-Optimism CEO as non-overconfident CEOs.
Before any classification of high-optimism or low-optimism, we assume all the CEO are moderate-optimism. In addition, as we mentioned that a CEO is classified as high-optimism or low-optimism only if he shows the corresponding behavior at least twice during his tenure, however, his classification starts from the first time he shows the corresponding behavior.
Moreover, if a CEO shows opposite behavior twice after his classification of high-optimism or low-optimism, we would change his classification to the opposite from the first time he show the opposite behavior. For example, if a CEO was first classified as high-optimism, he would be classified as low-optimism if he show the opposite behavior twice after his previous classification. Finally, we can’t classify CEO who don’t have option at all or have all their options out of money, so we exclude those CEOs from our sample.
3.3 Recession Years
In this paper, we use National Bureau of Economic Research (NBER)’s definition as our main measurement of recession years. Because the recession definition from NBER is a widely used measurement in past academic literatures (Burns and Mitchell, 1946; Harford, 1999;
Halling, Yu, and Zechner, 2016) and it only capture recessions with sufficient depth and influence, we adapt it as our main measurement. The recession described in NBER is the period when macroeconomic activity goes form peak to trough and expansion is the period when macroeconomic activity goes from trough to peak. Using, NBER’s data, we define 2001, 2008, and 2009 as recession years and denote the recession dummy variable as 1 in these year.
In addition, we have two alternative measurements of recession years as complement, which are definition from The Organization for Economic Cooperation and Development (OECD) and Chicago Fed National Activity Index (CFNAI) respectively. First, OECD United States business cycle are defined by the movement of their Composite Leading Indicators.
There are turning points of this indicators when indicators deviate from trend series. Recessions and expansions are defined as periods between turning points. Second, the definitions of recessions and expansions from CFNAI are periods when the 3-month moving average of index goes below-0.7 and periods when the 3-month moving average goes beyond +0.2 respectively.
We denote the recession dummy variable as 1 in the recession years defined by the fore-mentioned measurement. The recession years defined by the fore-fore-mentioned institutions are shown in Table 1.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
13
3.4 Probability of Bidding
To examine if overconfident CEOs are more likely than non-overconfident CEOs to undertake M&As during recessions or after recessions, we adapt Harford and Uysal (2014)’s probit model. The model is used to examine the likelihood of a firm to undertake M&A deals and is described in Equation (1) as follows:
𝑃𝑟𝑜𝑏(𝐵𝑖𝑑|𝑋𝑖.𝑡) = 𝛽0+ 𝛽1𝑂𝑣𝑒𝑟𝑐𝑜𝑛𝑓𝑖𝑑𝑒𝑛𝑐𝑒𝑖,𝑡-1+ 𝛽2𝑅𝑒𝑐𝑒𝑠𝑠𝑖𝑜𝑛𝑖,𝑡 +𝛽3𝑂𝑣𝑒𝑟𝑐𝑜𝑛𝑓𝑖𝑑𝑒𝑛𝑐𝑒𝑖,𝑡-1× 𝑅𝑒𝑐𝑒𝑠𝑠𝑖𝑜𝑛𝑖,𝑡+ 𝛽4𝑆𝑎𝑙𝑒𝑠𝑖,𝑡-1 +𝛽5𝐶𝑎𝑠ℎ ℎ𝑜𝑙𝑑𝑖𝑛𝑔𝑠/𝑇𝐴𝑖,𝑡-1+ 𝛽6𝑀𝑎𝑟𝑘𝑒𝑡-𝑡𝑜-𝑏𝑜𝑜𝑘𝑖,𝑡-1 +𝛽7𝑀𝑎𝑟𝑘𝑒𝑡 𝑙𝑒𝑣𝑒𝑟𝑎𝑔𝑒𝑖.𝑡-1+ 𝛽8𝐸𝐵𝐼𝑇𝐷𝐴/𝑇𝐴𝑖.𝑡-1
+𝛽9𝑆𝑡𝑜𝑐𝑘 𝑟𝑒𝑡𝑢𝑟𝑛𝑖,𝑡-1+ 𝛽10𝑀𝐴 𝑙𝑖𝑞𝑢𝑖𝑑𝑖𝑡𝑦𝑖.𝑡-1
+𝛽11𝐻𝐻𝐼𝑖,𝑡-1+ 𝛽12𝑅𝑎𝑡𝑒𝑑𝑡-1+ 𝛽13𝐴𝑔𝑒𝑡+ 𝛽14𝑀𝑎𝑙𝑒𝑡+ 𝜖𝑖,𝑡 (1) The dependent variable “Bid” denotes 1 if a firm makes a bid in a firm year or 0 if not.
The control variables of the model includes natural logarithm of sales for Almazan, De Motta, Titman, and Uysal (2010) evidence that larger firms have higher probability of conducting M&As. Besides, we include EBITDA/TA for Roll (1986) and Harford (1999) show that firms with better operating performance are more likely to undertake M&As. Also, we include market-to-book ratio and stock return to control the higher likelihood to make a bid among firms with more investment opportunities. At the same time, market leverage is included in the regressions to separate the effects of leverage and having a rating. Moreover, we include cash holding for Harford (1999) evidences that firms with larger cash holdings are more acquisitive.
Since Schlingemann, Stulz, and Walkling. (2002) propose that higher liquidity of the market for corporate assets make it more easier for firms to make M&A deals, we include the variable
MA Liquidity. Industry concentration can also influence the probability of bidding for there are
fewer targets within industry which is more concentrated. We use Herfindahl Index to proxy for industry concentration. As Harford and Uysal (2014) evidence that firms with bond-rating are more likely to undertake M&As, we add the rating dummy, Rated, to control that effect. In addition to Harford and Uysal (2014)’s control variables, we also add Age for Yim (2013) shows that young CEOs are more likely to undertake M&As than older CEOs. We also add the variable, Male, which denotes 1 if the CEO is male for Huang and Kisgen (2013) evidence that‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
14
male CEOs are more acquisitive than female CEOs. Finally, the construction of overconfidence dummy and recession dummy are described in Section 3.1.1 and 3.1.2. All the control variable are lagged by 1 year to examine the causality except for Recession, Age, and Male. The details of all the variables are shown in Appendix.
To test the effect of the 2001 Recession and the 2008 Recession separately, we divide our sample into 6 subsamples, which cover 3 and 5 years before and after 2001/2008, 1993 to 2007, and 2002 to 2016.
3.5 Announcement Effect
To investigate the market reaction to the M&A deals announced by overconfident CEOs during or after recessions, we follow Harford and Uysal (2014)’s model to run the regression of cumulative abnormal return (CAR) of the acquirers. The predicted return is derived from market model. We use the firm’s return and benchmark return from 205 days before event date to 6 days before event date to compute the beta for each firm for each time. The benchmark return is the value-weighted market return including dividend in CRSP database. The length of our CAR window is 5 days, ranging from 2 days before the event date to 2 days after the event date. The model is described in Equation (2) as follows:
𝐶𝐴𝑅𝑖,𝑡 = 𝛽0+ 𝛽1𝑂𝑣𝑒𝑟𝑐𝑜𝑛𝑓𝑖𝑑𝑒𝑛𝑐𝑒𝑖,𝑡-1+ 𝛽2𝑅𝑒𝑐𝑒𝑠𝑠𝑖𝑜𝑛𝑖,𝑡 +𝛽3𝑂𝑣𝑒𝑟𝑐𝑜𝑛𝑓𝑖𝑑𝑒𝑛𝑐𝑒𝑖,𝑡-1× 𝑅𝑒𝑐𝑒𝑠𝑠𝑖𝑜𝑛𝑖,𝑡+ 𝛽4𝑆𝑎𝑙𝑒𝑠𝑖,𝑡-1
+𝛽5𝐶𝑎𝑠ℎ ℎ𝑜𝑙𝑑𝑖𝑛𝑔𝑠/𝑇𝐴𝑖,𝑡-1+ 𝛽6𝑀𝑎𝑟𝑘𝑒𝑡-𝑡𝑜-𝑏𝑜𝑜𝑘𝑖,𝑡-1 +𝛽7𝑀𝑎𝑟𝑘𝑒𝑡 𝑙𝑒𝑣𝑒𝑟𝑎𝑔𝑒𝑖.𝑡-1+ 𝛽8𝐸𝐵𝐼𝑇𝐷𝐴/𝑇𝐴𝑖.𝑡-1
+𝛽9𝑆𝑡𝑜𝑐𝑘 𝑟𝑒𝑡𝑢𝑟𝑛𝑖,𝑡-1+ 𝛽10𝑀𝐴 𝑙𝑖𝑞𝑢𝑖𝑑𝑖𝑡𝑦𝑖.𝑡-1+ 𝛽11𝐻𝐻𝐼𝑖,𝑡-1 +𝛽12𝑅𝑎𝑡𝑒𝑑𝑡-1+ 𝛽13𝑅𝑒𝑙𝑎𝑡𝑖𝑣𝑒 𝑠𝑖𝑧𝑒𝑖,𝑡 + 𝛽14𝑃𝑢𝑏𝑙𝑖𝑐 𝑡𝑎𝑟𝑔𝑒𝑡𝑖,𝑡
+𝛽15𝑃𝑟𝑖𝑣𝑎𝑡𝑒 𝑡𝑎𝑟𝑔𝑒𝑡𝑖.𝑡+ 𝛽16𝑊𝑖𝑡ℎ𝑖𝑛-𝑖𝑛𝑑𝑢𝑠𝑡𝑟𝑦-𝑎𝑐𝑞𝑢𝑖𝑠𝑖𝑡𝑖𝑜𝑛𝑖.𝑡 +𝛽17𝐴𝑙𝑙 𝐶𝑎𝑠ℎ𝑖.𝑡 + 𝛽18𝐻𝑜𝑠𝑡𝑖𝑙𝑒𝑖.𝑡+ 𝛽19𝐶𝑜𝑚𝑝𝑒𝑡𝑒𝑑𝑖,𝑡
+𝛽20𝐴𝑔𝑒𝑡+ 𝛽21𝑀𝑎𝑙𝑒𝑡+ 𝜖𝑖,𝑡 (2) There are several factors evidenced by previous literatures that would affect M&As announcement effect. Those factors include acquirer’s size, profitability, market-to-book ratio,
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
15
and stock return. Besides, M&A deal’s characteristics such as method of payment, the target is private or public, if there are multiple bidder, if the deal made by acquirer is hostile or friendly, and deal size relative to acquirer would also affect the market reaction to the announcement of the deal. Furthermore, industry characteristics such as M&A liquidity and concentration of the market play an important role in determining the announcement effect.
Besides the original control variable in Harford and Uysal (2014)’s model, we add Age for Yim (2013) shows that M&As undertaken by young CEOs are more value-decreasing than those undertaken by older CEOs. Moreover, we add the variable, Male, which denotes 1 if the CEO is male. Because Huang and Kisgen (2013) evidence that the announcement effect of M&As undertaken by male CEOs are lower than M&As undertaken by female CEOs. Finally, we add our overconfidence dummy and recession dummy to the model, which are described in Section 3.1.1 and 3.1.2. All the control variable are lagged by 1 year to examine the causality except for Recession, Age, and Male. The details of all the variables are shown in Appendix.
To test the effect of the 2001 Recession and the 2008 Recession separately, we divide our sample into 6 subsamples, which cover 3 and 5 years before and after 2001/2008, 1993 to 2007, and 2002 to 2016.
3.6 Long-term Performance
We follow Duchin and Schmidt (2013)’s model to investigate if the long-run performance following M&A undertaking by overconfident CEOs during recessions are worse. The measurement for long-run performance is the 24 month buy-and-hold abnormal return (BHAR) after the announcement of the deal. The benchmark portfolios follow Lyon et al. (1999)’s method, which use all NYSE/AMEX/Nasdaq’s firms to construct 70 portfolios based on 14 size and 5 book-to-market ratio. The model is described in Equation (3) as follows:
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
16
𝐵𝐻𝐴𝑅𝑖,𝑡 = 𝛽0+ 𝛽1𝑂𝑣𝑒𝑟𝑐𝑜𝑛𝑓𝑖𝑑𝑒𝑛𝑐𝑒𝑖,𝑡-1+ 𝛽2𝑅𝑒𝑐𝑒𝑠𝑠𝑖𝑜𝑛𝑖,𝑡 +𝛽3𝑂𝑣𝑒𝑟𝑐𝑜𝑛𝑓𝑖𝑑𝑒𝑛𝑐𝑒𝑖,𝑡-1× 𝑅𝑒𝑐𝑒𝑠𝑠𝑖𝑜𝑛𝑖,𝑡+ 𝛽4𝑆𝑖𝑧𝑒𝑖,𝑡-1
+𝛽5𝑅𝑒𝑙𝑎𝑡𝑖𝑣𝑒 𝑑𝑒𝑎𝑙 𝑠𝑖𝑧𝑒𝑖,𝑡-1+ 𝛽6𝑃𝑟𝑖𝑐𝑒 𝑟𝑢𝑛-𝑢𝑝𝑖,𝑡-1
+𝛽7𝐼𝑛𝑑𝑢𝑠𝑡𝑟𝑦 𝑇𝑜𝑏𝑖𝑛′𝑠 𝑄𝑖.𝑡-1+ 𝛽8𝐼𝑛𝑑𝑢𝑠𝑡𝑟𝑦 𝑙𝑒𝑣𝑒𝑟𝑎𝑔𝑒𝑖.𝑡-1 +𝛽9𝐹𝑟𝑒𝑒 𝑐𝑎𝑠ℎ 𝑓𝑙𝑜𝑤𝑠𝑖,𝑡-1+ 𝛽10%𝑑𝑒𝑎𝑙 𝑣𝑎𝑙𝑢𝑒𝑖.𝑡
+𝛽11%𝑁𝑖,𝑡+ 𝛽12𝐷𝑖𝑣𝑒𝑟𝑠𝑖𝑓𝑦𝑖𝑛𝑔𝑖,𝑡+ 𝛽13𝑃𝑢𝑏𝑙𝑖𝑐 𝑡𝑎𝑟𝑔𝑒𝑡𝑖,𝑡 +𝛽14𝐴𝑙𝑙 𝑐𝑎𝑠ℎ𝑖,𝑡 + 𝛽15𝑃𝑢𝑏𝑖𝑐 𝑡𝑎𝑟𝑔𝑒𝑡 × 𝐴𝑙𝑙 𝑐𝑎𝑠ℎ𝑖.𝑡
+𝛽16𝐻𝑖𝑔ℎ 𝑣𝑎𝑙𝑢𝑎𝑡𝑖𝑜𝑛𝑖.𝑡 + 𝛽17𝐴𝑔𝑒𝑖.𝑡+ 𝜖𝑖,𝑡 (3) There are some factors, which would affect firm’s long-run performance after the announcement of M&A. First, we control acquirer’s size in firm level as well as their Tobin’s Q and leverage in industry level. Second, as evidenced by Bouwman, Fuller, and Nain (2009), who show the lower long-term performance follows acquisition during periods of high market valuation, we add the valuation-related variables, High-valuation and Price run-up, respectively. Also, we control the effect of excess cash flow on long-term stock performance after M&As by deriving excess cash flow from Dittmar and Mahrt-Smith (2007)’s model In addition, we control several deal characteristics which would affect M&As’ long-run performance including relative deal size, private/public target, all-cash financing, and diversifying deal. Moreover, we include CEO’s age for Yim (2010) find that M&As undertaken by young CEOs worth less. As Duchin and Schmidt (2013) show that M&As long-term performance following M&As undertaken during merger waves is lower. We then add number and value of M&A deals of an industry in prior 12 month before the announcement of the deal relative to the total number and value of M&A deals in the sample periods to control the effect of merger wave. Finally, we add our overconfidence and recession dummies. All the control variable are lagged by 1 year to examine the causality except for Recession, Age, and Male.
The details of all the variables are shown in Appendix.
To test the effect of the 2001 Recession and the 2008 Recession separately, we divide our sample into 6 subsamples, which cover 3 and 5 years before and after 2001/2008, 1993 to 2007, and 2002 to 2016.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
17