18 64 20 8 6 4 2 0 Size Matched First SEOs Second Third Year Year Fourth Year Fifth Year Year 18 1 12 10 8 7 StMle Matched First SEOs Second Year Third Year Fourth Year Fifth Year Year Figure 1-Post-issue returns for firms conducting seasoned equity offerings(SEOs)in 1970-2000.The average annual return for each of the five years after issuing is shown for firms conducting SEOs and(top panel)size- matched nonissuing firms,and (bottom panel)style-matched nonissuing firms.The style matches are based upon size(market cap)and book-to-market matching.For each of the five post-issue years,the average annual return is calculated as an equally weighted average of the CRSP-listed issuers that are present at the beginning of the year. All matching firms have been CRSP-listed for at least five years at the time of the SEO with which they are matched,and have not conducted an SEO during this time.If an SEO is delisted before five years from the issue date,its annual return during the year of delisting is calculated by splicing in the CRSP value-weighted market return for the remainder of that year.Additional details are described in Table 1,where the numbers in this figure are reported.Returns are computed using CRSP returns ending on December 31,2001. 11
11 First Year Second Year Third Year Fourth Year Fifth Year SEOs Size Matched 0 2 4 6 8 10 12 14 16 18 Annual Percentage Returns First Year Second Year Third Year Fourth Year Fifth Year SEOs Style Matched 0 2 4 6 8 10 12 14 16 18 Annual Percentage Returns Figure 1—Post-issue returns for firms conducting seasoned equity offerings (SEOs) in 1970-2000. The average annual return for each of the five years after issuing is shown for firms conducting SEOs and (top panel) sizematched nonissuing firms, and (bottom panel) style-matched nonissuing firms. The style matches are based upon size (market cap) and book-to-market matching. For each of the five post-issue years, the average annual return is calculated as an equally weighted average of the CRSP-listed issuers that are present at the beginning of the year. All matching firms have been CRSP-listed for at least five years at the time of the SEO with which they are matched, and have not conducted an SEO during this time. If an SEO is delisted before five years from the issue date, its annual return during the year of delisting is calculated by splicing in the CRSP value-weighted market return for the remainder of that year. Additional details are described in Table 1, where the numbers in this figure are reported. Returns are computed using CRSP returns ending on December 31, 2001
Using a size benchmark,however,introduces a confounding effect.Issuing firms tend to be growth firms,and nonissuers tend to be value firms.Thus,in addition to comparing issuers with nonissuers,growth firms are being compared with value firms.To remove this confounding effect,Table 1 also reports the average annual returns on issuing firms and nonissuers matched by both size and book-to-market(style"matching).In so doing,some issuers are lost because of missing book value information.Table 1 shows that when issuers are compared to style- matched nonissuers,the underperformance narrows slightly to 3.4%per year in the five years after issuing.Statistical significance levels are not reported in Table 1,because the large degree of overlap in post-issue returns among the sample greatly decreases the number of independent observations. Inspection of Table I shows that issuers do not underperform in the first six months after issuing.This is probably due to a combination of momentum effects and a desire to avoid litigation by making sure that earnings numbers meet analyst forecasts in the first two quarters after issuing(negative earnings surprises are rare immediately following an SEO(Korajczyk, Lucas,and McDonald(1991))).In the roughly two years after this six month honeymoon, however,there is very substantial underperformance,as firm performance fails to live up to optimistic expectations.By year five,however,the abnormal returns are close to zero, suggesting that the underperformance does not persist forever. Most of the empirical literature concerning the long-run performance of SEOs has used two procedures:buy-and-hold returns and 3-factor regressions.The results of studies using buy- and-hold returns with a style benchmark are reported in Table 2.3 3 Jegadeesh(2000)and Brav,Geczy,and Gompers(2000)also adjust for momentum,in addition to size and book- to-market,with qualitatively unchanged results. 12
12 Using a size benchmark, however, introduces a confounding effect. Issuing firms tend to be growth firms, and nonissuers tend to be value firms. Thus, in addition to comparing issuers with nonissuers, growth firms are being compared with value firms. To remove this confounding effect, Table 1 also reports the average annual returns on issuing firms and nonissuers matched by both size and book-to-market (“style” matching). In so doing, some issuers are lost because of missing book value information. Table 1 shows that when issuers are compared to stylematched nonissuers, the underperformance narrows slightly to 3.4% per year in the five years after issuing. Statistical significance levels are not reported in Table 1, because the large degree of overlap in post-issue returns among the sample greatly decreases the number of independent observations. Inspection of Table 1 shows that issuers do not underperform in the first six months after issuing. This is probably due to a combination of momentum effects and a desire to avoid litigation by making sure that earnings numbers meet analyst forecasts in the first two quarters after issuing (negative earnings surprises are rare immediately following an SEO (Korajczyk, Lucas, and McDonald (1991))). In the roughly two years after this six month honeymoon, however, there is very substantial underperformance, as firm performance fails to live up to optimistic expectations. By year five, however, the abnormal returns are close to zero, suggesting that the underperformance does not persist forever. Most of the empirical literature concerning the long-run performance of SEOs has used two procedures: buy-and-hold returns and 3-factor regressions. The results of studies using buyand-hold returns with a style benchmark are reported in Table 2.3 3 Jegadeesh (2000) and Brav, Geczy, and Gompers (2000) also adjust for momentum, in addition to size and bookto-market, with qualitatively unchanged results
Table 2 Evidence on the long-run performance of SEOs,measured using buy-and-hold returns Horizon. SampleSample Mean buy-and-hold return Annualized Study weighting size period SEOs Matching Difference U.S.Data Mitchell Stafford 3 years,EW 4.439 1961-1993 34.8%45.0% -2.7% Eckbo.Masulis Norli 5 years,EW 3,315 1964-1995 44.3% 67.5% -4.8% Jegadeesh 5 years,EW 2.992 1970-1993 59.4% 93.6% -4.9% Spiess Affleck-Graves 5 years,EW 1.247 1975-1989 55.7% 98.1% -6.1% Brav,Geczy Gompers 5 years,EW 3,775 1975-1992 57.6% 83.9% -3.9% Mitchell Stafford 3 years,VW 4.439 1961-1993 41.1% 45.3% -1.1% Eckbo.Masulis Norli 5 years,VW 3.315 1964-1995 51.6% 62.2% -2.2% Brav,Geczy Gompers 5 years,VW 3,775 1975-1992 72.5% 97.5% -3.4% Japanese Data Cai Loughran 5 years,EW 1.389 1971-1992 74.1% 103.2% -3.5% Sources and assumptions:The numbers reported in this table for matching firms are all based on size and book-to- market matching("style"benchmarks).EW is equally weighted,and VW is value weighted.Most authors use buy- and-hold returns on individual stocks for their benchmarks,although some,such as Brav,Geczy,and Gompers (2000),use portfolios and rebalance their benchmark monthly and then compound the monthly average returns.All of the benchmarks delete issuing firms from the universe of potential matches.For Mitchell and Stafford(2000, Table III)the compounded annual return difference of 2.7%assumes a 3.0 year mean holding period(they reinvest early delistings in an index).Their sample period is July 1961-December 1993 and they include utilities.The annualized difference is calculated as [RTR00%where R is the average gross buy-and-hold return on the issuing firms,Ro is the average gross compounded return on the benchmark,and inverseT is the reciprocal of the average holding period length.For Mitchell and Stafford,R is 1.348,and the annualized difference is calculated as 1.348 to the 1/3 power minus 1.450 to the 1/3 power,and then converted to a percentage.Eckbo,Masulis,and Norli(2000,Table 3,excluding utilities)truncate a firm's return if and when it conducts a subsequent SEO.The compounded difference in returns of 4.8%assumes a 3.5 year mean holding period.For Brav,Geczy,and Gompers (2000,Table 3,Panel A),the compounded difference in annualized returns assumes a 4.5 year mean holding period. Their sample period is from 1975-1992,with returns ending on December 31,1995,and includes utilities.For Jegadeesh(2000,Table 2),a 4.5 year mean holding period is assumed,giving a compounded difference in returns of 4.8%per year for his sample period of 1977-1994,with returns truncated at Dec.31,1997.For Spiess and Affleck- Graves(1995,Table 3),a 4.5 year average holding period is assumed.They restrict their sample of SEOs to pure primary issues (offers where only the firm is selling shares),unlike the other studies,which include combination offers where both the firm and existing shareholders sell shares in the SEO.For Cai and Loughran(1998,Table 2. Panel A),where Tokyo Stock Exchange-listed firms are used,the annualized difference is computed assuming a 5.0 year holding period,since very few of the sample firms delist early. 13
13 Table 2 Evidence on the long-run performance of SEOs, measured using buy-and-hold returns Horizon, Sample Sample Mean buy-and-hold return Annualized Study weighting size period SEOs Matching Difference U.S. Data Mitchell & Stafford 3 years, EW 4,439 1961-1993 34.8% 45.0% -2.7% Eckbo, Masulis & Norli 5 years, EW 3,315 1964-1995 44.3% 67.5% -4.8% Jegadeesh 5 years, EW 2,992 1970-1993 59.4% 93.6% -4.9% Spiess & Affleck-Graves 5 years, EW 1,247 1975-1989 55.7% 98.1% -6.1% Brav, Geczy & Gompers 5 years, EW 3,775 1975-1992 57.6% 83.9% -3.9% Mitchell & Stafford 3 years, VW 4,439 1961-1993 41.1% 45.3% -1.1% Eckbo, Masulis & Norli 5 years, VW 3,315 1964-1995 51.6% 62.2% -2.2% Brav, Geczy & Gompers 5 years, VW 3,775 1975-1992 72.5% 97.5% -3.4% Japanese Data Cai & Loughran 5 years, EW 1,389 1971-1992 74.1% 103.2% -3.5% Sources and assumptions: The numbers reported in this table for matching firms are all based on size and book-tomarket matching (“style” benchmarks). EW is equally weighted, and VW is value weighted. Most authors use buyand-hold returns on individual stocks for their benchmarks, although some, such as Brav, Geczy, and Gompers (2000), use portfolios and rebalance their benchmark monthly and then compound the monthly average returns. All of the benchmarks delete issuing firms from the universe of potential matches. For Mitchell and Stafford (2000, Table III) the compounded annual return difference of 2.7% assumes a 3.0 year mean holding period (they reinvest early delistings in an index). Their sample period is July 1961-December 1993 and they include utilities. The annualized difference is calculated as [Ri inverseT – Rb inverseT]×100% where Ri is the average gross buy-and-hold return on the issuing firms, Rb is the average gross compounded return on the benchmark, and inverseT is the reciprocal of the average holding period length. For Mitchell and Stafford, Ri is 1.348, and the annualized difference is calculated as 1.348 to the 1/3 power minus 1.450 to the 1/3 power, and then converted to a percentage. Eckbo, Masulis, and Norli (2000, Table 3, excluding utilities) truncate a firm’s return if and when it conducts a subsequent SEO. The compounded difference in returns of 4.8% assumes a 3.5 year mean holding period. For Brav, Geczy, and Gompers (2000, Table 3, Panel A), the compounded difference in annualized returns assumes a 4.5 year mean holding period. Their sample period is from 1975-1992, with returns ending on December 31, 1995, and includes utilities. For Jegadeesh (2000, Table 2), a 4.5 year mean holding period is assumed, giving a compounded difference in returns of 4.8% per year for his sample period of 1977-1994, with returns truncated at Dec. 31, 1997. For Spiess and AffleckGraves (1995, Table 3), a 4.5 year average holding period is assumed. They restrict their sample of SEOs to pure primary issues (offers where only the firm is selling shares), unlike the other studies, which include combination offers where both the firm and existing shareholders sell shares in the SEO. For Cai and Loughran (1998, Table 2, Panel A), where Tokyo Stock Exchange-listed firms are used, the annualized difference is computed assuming a 5.0 year holding period, since very few of the sample firms delist early
Mitchell and Stafford have the lowest abnormal performance,which is presumably attributable to their sample being relatively intensive in utilities and SEOs listed on the New York Stock Exchange (NYSE)from the 1960s and early 1970s.Mitchell and Stafford (2000) report that the only issuers that underperform are small value firms.By contrast,Jegadeesh (2000)reports that it is growth firms among the issuers that have the worst subsequent performance,and that large firms as well as small growth stocks underperform. As discussed by Barber and Lyon(1997),Kothari and Warner(1997),Lyon,Barber,and Tsai (1999),Brav (2000),and others,unbiased statistical significance levels are difficult to compute using buy-and-hold returns.Consequently,starting with Loughran and Ritter (1995), the long-run returns literature has commonly used 3-factor time-series regressions,introduced by Fama and French(1993),of the form rpt-ra a b(rmt-ra)+SSMBt VVMGt ept where rpt-ra is the excess return over the risk-free rate on a portfolio in time period t,rmt-ra is the realization of the market risk premium in period t,SMBt is the return on a portfolio of Small stocks Minus the return on a portfolio of Big stocks in period t,and VMGt is the return on a portfolio of Value stocks Minus the return on a portfolio of Growth stocks in period t.Value and growth are measured using book to market ratios,and VMG is denoted HML in the literature (High book-to-market(value)Minus Low book-to-market (growth)stocks).The intercepts from these regressions are interpreted as abnormal returns.In Table 3,the intercepts (and t-statistics) reported in various studies of abnormal performance following SEOs are listed. 14
14 Mitchell and Stafford have the lowest abnormal performance, which is presumably attributable to their sample being relatively intensive in utilities and SEOs listed on the New York Stock Exchange (NYSE) from the 1960s and early 1970s. Mitchell and Stafford (2000) report that the only issuers that underperform are small value firms. By contrast, Jegadeesh (2000) reports that it is growth firms among the issuers that have the worst subsequent performance, and that large firms as well as small growth stocks underperform. As discussed by Barber and Lyon (1997), Kothari and Warner (1997), Lyon, Barber, and Tsai (1999), Brav (2000), and others, unbiased statistical significance levels are difficult to compute using buy-and-hold returns. Consequently, starting with Loughran and Ritter (1995), the long-run returns literature has commonly used 3-factor time-series regressions, introduced by Fama and French (1993), of the form rpt - rft = a + b(rmt - rft) + sSMBt + vVMGt + ept where rpt - rft is the excess return over the risk-free rate on a portfolio in time period t, rmt - rft is the realization of the market risk premium in period t, SMBt is the return on a portfolio of Small stocks Minus the return on a portfolio of Big stocks in period t, and VMGt is the return on a portfolio of Value stocks Minus the return on a portfolio of Growth stocks in period t. Value and growth are measured using book to market ratios, and VMG is denoted HML in the literature (High book-to-market (value) Minus Low book-to-market (growth) stocks). The intercepts from these regressions are interpreted as abnormal returns. In Table 3, the intercepts (and t-statistics) reported in various studies of abnormal performance following SEOs are listed
Table 3 Equally weighted and value-weighted intercepts from 3-factor regressions on U.S.SEOs rpt-ra =a+b(rmt-ra)+SSMBt+VVMGt+ept Sample size EW VW Mitchell Stafford,including utilities 4,911 -0.33 -0.03 (July 1961-December 1993) (-5.19) (-0.44) Mitchell Stafford,excluding utilities 3,842 -0.37 0.06 (July 1961-December 1993) (-5.58) (0.77) Eckbo,Masulis Norli,excluding utilities 1,704 -0.12 -0.17 (March 1964-December 1997),Amex/NYSE only (-0.65) (-1.12) Eckbo,Masulis Norli,excluding utilities 2,147 -0.42 -0.12 (February 1974-December 1997),Nasdaq only (-2.37) (-0.19) Jegadeesh 2,992 -0.45 -0.33 (January 1975-December 1995) (-5.07) (-2.84) Loughran Ritter 6,461 -0.47 -0.32 (January 1973-December 1996) -5.42) (-3.00) Loughran Ritter,purged 6,461 -0.61 -0.35 (January 1973-December 1996) (-6.08) (-3.38) Brav,Geczy Gompers 4,526 -0.37 -0.14 (January 1976-December 1995) (-4.81) (-1.36) Brav,Geczy Gompers,purged 4,526 -0.40 -0.17 (January 1976-December 1995) (-4.65) (-1.63) T-statistics are in parentheses.A coefficient of-0.33 represents underperformance of 33 basis points per month,or -3.96 percent per year before compounding.Sources:Mitchell and Stafford(2000,Table 9);Eckbo,Masulis,and Norli(2000,Table 10,Panel C);Loughran and Ritter(2000,Table 7);and Brav,Geczy,and Gompers(2000,Table 6).Mitchell and Stafford use issuing firms from 1958-1993 and keep a firm in the portfolio for five years after issuing.Monthly returns from July 1961 through December 1993 are used in their regressions.Brav,Geczy,and Gompers use issuing firms from 1975-1992,and keep a firm in the portfolio for five years after issuing.Jegadeesh (2000,Table IV)uses SEOs from 1970 to 1994,and keeps a firm in the portfolio for five years after issuing.The VW results that are reported for Jegadeesh are his EW "large firm"results.Loughran and Ritter use issuing firms from 1970-1996,and keep a firm in the portfolio for three years after issuing.Loughran and Ritter exclude utilities from their issuer portfolio.The purged and unpurged numbers in Brav-Geczy-Gompers and Loughran-Ritter refer to whether the size and book-to-market factors are constructed exclusive(purged)or inclusive (unpurged)of firms that have issued equity within the prior five years. 15
15 Table 3 Equally weighted and value-weighted intercepts from 3-factor regressions on U.S. SEOs rpt - rft = a + b(rmt - rft) + sSMBt + vVMGt + ept Sample size EW VW Mitchell & Stafford, including utilities 4,911 -0.33 -0.03 (July 1961-December 1993) (-5.19) (-0.44) Mitchell & Stafford, excluding utilities 3,842 -0.37 0.06 (July 1961-December 1993) (-5.58) (0.77) Eckbo, Masulis & Norli, excluding utilities 1,704 -0.12 -0.17 (March 1964-December 1997), Amex/NYSE only (-0.65) (-1.12) Eckbo, Masulis & Norli, excluding utilities 2,147 -0.42 -0.12 (February 1974-December 1997), Nasdaq only (-2.37) (-0.19) Jegadeesh 2,992 -0.45 -0.33 (January 1975-December 1995) (-5.07) (-2.84) Loughran & Ritter 6,461 -0.47 -0.32 (January 1973-December 1996) (-5.42) (-3.00) Loughran & Ritter, purged 6,461 -0.61 -0.35 (January 1973-December 1996) (-6.08) (-3.38) Brav, Geczy & Gompers 4,526 -0.37 -0.14 (January 1976-December 1995) (-4.81) (-1.36) Brav, Geczy & Gompers, purged 4,526 -0.40 -0.17 (January 1976-December 1995) (-4.65) (-1.63) T-statistics are in parentheses. A coefficient of –0.33 represents underperformance of 33 basis points per month, or –3.96 percent per year before compounding. Sources: Mitchell and Stafford (2000, Table 9); Eckbo, Masulis, and Norli (2000, Table 10, Panel C); Loughran and Ritter (2000, Table 7); and Brav, Geczy, and Gompers (2000, Table 6). Mitchell and Stafford use issuing firms from 1958-1993 and keep a firm in the portfolio for five years after issuing. Monthly returns from July 1961 through December 1993 are used in their regressions. Brav, Geczy, and Gompers use issuing firms from 1975-1992, and keep a firm in the portfolio for five years after issuing. Jegadeesh (2000, Table IV) uses SEOs from 1970 to 1994, and keeps a firm in the portfolio for five years after issuing. The VW results that are reported for Jegadeesh are his EW “large firm” results. Loughran and Ritter use issuing firms from 1970-1996, and keep a firm in the portfolio for three years after issuing. Loughran and Ritter exclude utilities from their issuer portfolio. The purged and unpurged numbers in Brav-Geczy-Gompers and Loughran-Ritter refer to whether the size and book-to-market factors are constructed exclusive (purged) or inclusive (unpurged) of firms that have issued equity within the prior five years