國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
21
investors’ decisions on trading but this is not driven by risk aversion difference.
Barber and Odean (2001) document men trades with greater overconfidence than women but not risk aversion level.
4) Procrastination. Inertia on CEOs’ personal account my carry over to the corporate account of the firm in a reluctance to conduct equity issues so CEO should not habitually purchase company equity under my measure.
5) Tax consideration: Jin and Kothari (2008) find the tax burden is a major factor inducing CEOs to sell stocks. However, there is only security trading taxes levied by trading stocks in our sample Taiwan market reducing the tax problem.
Moreover, to purchase a stock would also be taxed over showing the confidence level of the CEOs.
IV. Sample description and summary statistics
4.1 Sample
The sample is selected from public traded Taiwanese companies listing on Taiwan Stock Exchange (TSE) and Over-the-Counter (OTC) market. Since Wright and Philip (1980, 1978) discovered that Asia subjects are more overconfident than western subjects and the success of Taiwan exporting processing business is well-known makes me choose to study the behavior of Taiwanese managers. To measure a persistent characteristic on CEOs, I first collect CEOs shareholding data of their own companies from 1990 to 2010 from the Taiwan Economic Journal (TEJ), a special complete Taiwan database, and compute the Net Stock Purchase every year every CEOs in every firms. TEJ also consist CEOs’ pledge activities and employee bonus as stock dividends distribution statements. As distributed stock dividends for
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
22
shareholders each year in each firm, they are computed by announced stock dividends per share multiplying the original shares got by CEOs. There are 11,802 total CEO-year observations over a maximum time period of 20 years for any individual CEO.
The financial information of the companies, including sales revenues, research and development expense (R&D), book assets, book common stocks, deferred taxes, earnings before interest and taxes and depreciation and amortizations, and cash dividends are from TEJ- Finance DB. The stock price, return, shares outstanding and institutional ownership are from TEJ- Equity DB. The personal data of CEO and basic information of companies are from TEJ- Company DB. The factor premium data of equity market are from TEJ- Multiple equity factor DB. Excluding the missing data for any of the above measures yields a sample of 10,258 CEO-year observations across 1,217 firms over a maximum of 18 years for any individual CEOs.
4.2 Statistic Summary
Table1 presents summary statistics for the confidence measure. Panel A of Table 1 contains the results where the units and the percentage of excessively overconfident and excessively diffident CEOs measured as top and bottom 20% cutoff of distribution of Net Stock Purchase is a CEO-firm-year and Panel B contains the results where the units and percentage of excessively overconfident and excessive diffident CEOs is a CEO-industry.
As shown in Panel A, the average of frequency of excessive overconfident CEOs in fiscal years is 25% measured by Net Stock Purchases (EO_raw), 26% measured by Net stock purchases after pledge (EO_pledge), 29% measured by Net stock purchases excluding stock dividends (EO_div) and 30% measured by Net stock purchases excluding stock dividends after pledge (EO_pledge& div). The average of frequency
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
23
of excessive diffident CEOs in fiscal years is 48% by Net stock purchases (ED_raw), 43% by Net stock purchases after pledge (ED_pledge), 33% by Net stock purchases excluding stock dividends (ED_div), and 28% by Net stock purchases excluding stock dividends after pledge (ED_pledge& div). Despite the fact that we use upper and lower quintile cutoffs to define excessive overconfidence and excessive diffidence respectively, the resulting sample distribution should not necessarily be 20% each because the quintile breakpoints are computed using CEO-year observations each year but the excessive overconfidence (diffidence) would remain forward unless they are be classified to opposite group. From the panel, excessive overconfidence persists more in Net stock purchases excluding stock dividends after pledge (EO_pledge& div) than in rest measures since the average frequency deviates largest from quintile cutoff, certifying that the purer trait of CEO’s overconfidence is measured out. However, the trait of CEO’s diffidence is blurred by liquidity-motivated sale bias when the measure considers stock dividends and pledge, resulting in excessive diffidence persists more in Net stock purchases. While compared the frequency of excessive diffidence to of excessive overconfidence, higher frequency of excessive diffidence may indicate the sample is biased to moderate overconfident group, making the results may also bias.
In panel B, Taiwanese listed companies are divided into 26 industry groups based on Taiwan Stock Exchange’s classification and found out that excessively overconfident CEOs are higher in Electronics industry: 14% in Electronic Components, 13% in Semiconductors, and 11% in PC Electronics and excessively diffident CEOs are higher in Traditional industry: 9% in Textile and 7% in Construction. As we know, riskier industries such as electronics need overconfident CEOs to take more risk in investment thus attract overconfident CEOs to do the job.
Table 2 shows the summary statistics of sample firm and CEO characteristic on firm profitability as industry-adjusted Q, ROA, and ROE. Panel A shows the mean,
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
24
median, standard deviation, minimum, and maximum of full sample, Panel B shows that of excessively overconfident and excessively diffident CEOs sample measured by Net stock purchases, Panel C shows that of those measured by Net stock purchases after pledge sample, and Panel D shows that of measured by Net stock purchases excluding stock dividends after pledge. Considering the different characteristic existing in different level of excessive overconfidence and excessive diffidence, Rows 1 show the statistics of excessive overconfident and diffident CEOs by top and bottom 40% end of the distribution of Net stock purchases and rows 2 show those by top and bottom 10% end. As shown in Panel A, B and C, the means and medians of Q, ROA, and ROE in excessively overconfident CEO sample are statistically over than those in excessively diffident CEO sample and full sample, so as Sales, R&D, and CEO ownership and Firm age is smaller, manifesting that excessively overconfident CEOs are centered in young companies help value creation consistent with Mueller’s predictions but excessively diffident CEOs reduce the profitability same as Goel and Thakor’s prediction.
In addition, profitability measure decreases when approaching to the extreme end top 10% of the continuum, certifying the prediction that more extremely overconfident CEO reduce firm profitability, but the situation does not exist in excessively diffident CEO firms. However, the means and medians of Q, ROA, and ROE in excessive overconfident CEO firms in Panel D are lower than those of Q, ROA, and ROE in full sample and in excessively diffident CEO sample, showing that excessively overconfident CEOs might harm the company after using stricter measure on overconfidence classification. Nevertheless the results of excessively diffident CEO sample in Panel D are affected by liquidity selling problems.
Table 3 presents the summary statistics of sample firm and CEO characteristic on stock performance and related control variables. Panel A, B, C and D of Table 3 are
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
25
defined as same as Table 2. The total stock return of excessively overconfident CEO firms is worse than of full sample in all confidence measure method indicating that excessively overconfident CEO may let the company value go down but the effect seems lessen when approaching to the extreme cutoff. Contrarily, the total return of excessively diffident CEO firms is higher than of full sample maybe due that the excessively diffident CEOs are not really diffident as the measure is bias by liquidity-motivated sale activities. However, the trend is going down when approaching to the extreme cutoff, showing the extremely diffident CEOs make the value downward moving. In addition, excessively overconfident CEOs are centered in growth-oriented companies with lower book-to-market value, higher market value, and younger firms.
To understand the difference among my four measures of confidence level:
Confidence including stock dividends (confidence raw), Confidence including stock dividends after pledge (confidence pledge), Confidence excluding stock dividends (confidence div), and Confidence excluding stock dividends after pledge (confidence pledge& div), a correlation of four measures by top and bottom 20% cutoff is described in Table 4. As shown in table, the difference among four measures enlarged after considering stock dividends. Moreover, higher correlation among excessive overconfidence measures than excessive diffidence manifests CEOs have different consideration between purchases and sales of firm stocks.
Table 5 reports the correlation of excessive overconfidence, excessive diffidence and profitability measure and other related control variables. As shown in the tables, industry-adjusted Q, ROA, and ROE have certain positive correlation as same as finance theories. Confidence measure no matter excessive overconfidence or excessive diffidence does not have significant correlation with other variables.
Excessive overconfidence measured including stock dividends no matter before or
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
26
after pledge are positive to the profitability but excessive diffidence negative while that measured considered stock dividends are reverse consistent the results from statistics summary. Sales and R&D are positive correlated with Q, ROA and ROE and firm age are negative. Table 6 also reports the correlation of excessive overconfidence, excessive diffidence and stock performance and other related control variables.
Between the excessive overconfidence and excessive diffident, the total return is negative correlated with excessive overconfidence but positive with excessive diffidence no matter before or after pledge but excessive overconfidence and excessive diffidence are all negative correlated to total return after excluding stock dividends, moving to my prediction.