Testing the Robustness of. Long-Term Under-Performance of. UK Initial Public Offerings

Transcription

1 Testing the Robustness of Long-Term Under-Performance of UK Initial Public Offerings by Susanne Espenlaub* Alan Gregory** and Ian Tonks*** 22 July, 1998 * Manchester School of Accounting and Finance, University of Manchester; **Department of Management, University of Exeter; and *** Department of Economics, University of Bristol.

2 Abstract We re-examine the evidence on the long-term returns of IPOs in the UK using a new dataset of firms over the period , in which we compare abnormal returns under a number of alternative benchmarks. Previous work has identified that IPOs underperform a market index, and the purpose of this paper is to examine the robustness of this finding in relation to a number of alternative benchmark portfolios. We find that there are negative abnormal returns to an IPO such that a one pound investment is worth less than 85 pence after three years. This finding is similar across benchmarks. 1

3 I INTRODUCTION The long-term share price performance of Initial Public Offerings (IPOs) have recently become the focus of attention. The seminal article by Ibbotson (1975) reported a negative relation between initial returns at the IPO and long-term share price performance. Ibbotson found that although initial returns were not erased in the aftermarket, average returns for one month holding periods were positive in the first year after the IPO, negative during the following three years, and again positive in the fifth year. Ritter (1991) analysed the performance of US IPOs issued between and found that for a three-year holding period, IPOs underperformed a control sample of matching firms. He concluded that IPOs make bad medium- to long-term investments, since a dollar invested in an IPO was only worth 83 cents three years later. In the UK, Levis (1993) identified the same extent of IPO under-performance over the longer-term. Summarising a wealth of other international IPO evidence, Loughran, Ritter and Rydqvist (1994) report that market adjusted 3-year abnormal performance following an IPO is always small and mostly negative in all countries, with the exception of Japan. These results are dramatic, and imply that investing in recent IPOs is a poor investment. But before we accept this example of an apparent market inefficiency, it seems reasonable to examine the robustness of these findings. An obvious difficulty with an event study research model is the decision of an appropriate benchmark to use. There is considerable evidence that the choice of benchmark can have an important impact on the scale of abnormal returns from event studies [e.g., Dimson and Marsh (1986), Gregory, Matatko, Tonks and Purkis (1994), Fama and French (1996)], and this problem is particularly acute given the current state of asset pricing theory [Fama and French (1992)]. A further issue noted by Loughran et al (1994) in relation to the international (non-us) evidence is that the data samples used are typically very small. In this paper we re-examine the evidence on the long-term returns of 588 IPOs in the UK using a new dataset of firms which came to the market over the period Moreover, the UK stock market provides a unique opportunity to examine the robustness of the findings on the performance of US IPOs within the setting of another market-based financial system, in which stock markets are supposed to play a crucial role in providing company finance [see e.g., Table 1 in Roell (1995)]. 2

4 We compare abnormal returns under a number of alternative benchmarks. We compute abnormal returns up to three years after the offering, so the accumulation period in this study is over the period Our study employs the basic capital asset pricing model (CAPM), the simple size-adjusted model of Dimson and Marsh (1986), a CAPM-type model extended for size effects, the Fama and French (1996) three-factor model, and Ibbotson s (1975) Returns Across Securities and Times (RATS) approach. II PREVIOUS LITERATURE Levis (1993) investigates the long-run performance of a sample of 712 UK IPOs issued during using share-price data from 1980 until the end of Levis recognises the importance of the size effect for UK stocks and reports long-run abnormal returns based on three alternative benchmarks: the Financial Times Actuaries All Share (FTA) Index, the Hoare Govett Small Companies (HGSC) Index and a specially constructed all-share equally-weighted index. Levis confirms Ritter s (1991) finding of statistically significant long-run IPO underperformance, although he notes that average under-performance in his UK sample appears to be less excessive than that in Ritter s US sample. While Ritter reports under-performance of up to 29% over the first three-years after the IPO, Levis finds under-performance of between 8% and 23% depending on the market benchmark. There are important differences between Levis (1993) and the present study. First, unlike this study, Levis makes no explicit adjustments for (systematic) risk or other factors such as size, adopting instead a zero-one model to calculate abnormal returns. Second, the sample period over which Levis measures long-term performance ranges from , a period during which small-capitalisation stock outperformed larger stocks. By contrast, the period under study here spans , a period during which there was much greater time-series variation in the size effect as smaller stocks overperformed during the initial part, but underperformed during the latter part of the period. It is well documented that although beta does have a role in explaining returns, so does firm size [Fama and French (1992)]. Size effects have been taken into account in empirical studies 3

5 in a variety of ways. Most simply, Dimson and Marsh (1986) used size decile control portfolios, where each company is assigned a decile membership based upon its market capitalisation at the beginning of each year. More recent studies (Fama, Booth and Sinquefield, 1993; Loughran and Ritter, 1995; Fama and French, 1996) have used a multifactor benchmark approach. In particular, Fama and French (1996) suggest that many apparent anomalies in efficient markets studies can be explained by the use of a three-factor model, where the factors are the excess returns on the market, the difference in returns between companies with high book-to-market (BMV) and low BMV ratios, and the difference in returns between large and small companies. III METHODOLOGY To analyse long-term performance after an IPO we apply the standard event-study methodology. For a particular benchmark, monthly abnormal returns are computed for up to 36 months after the IPO (excluding the month of new issue); the minimum criterion for inclusion was 12 monthly observations post-ipo. To avoid any downward bias in returns caused by Jensen s inequality when averaging returns across portfolios, raw returns are used throughout this paper. 1 Event studies around IPOs are faced with the problem that data is not available to obtain estimates of the benchmark parameters in a pre-event period [Ibbotson (1975)]. In fact this problem is generic to a number of event study situations, since the event itself may cause a change the underlying structural parameters. 2 Thompson (1985) discusses the use of conditional (pre-event period only) and unconditional (pre- and post-event period) parameter estimates, and argues that in the case where the benchmark and the event are correlated that unconditional estimation is preferred. In this paper we follow Agrawal, Jaffe and Mandelker (1992) who investigate the long-run returns following a merger, and estimate the model parameters and the excess returns jointly and use in-sample estimates of abnormal returns. 1 Barber and Lyon (1997), p Boehmer, Musimeci and Poulson (1991) investigate the power of abnormal event methods when the event induces changes to the variance of returns 4

6 Abnormal returns with respect to each of five benchmarks are computed, and are cumulated over time up to period T after the IPO, using both the Cumulative Average Abnormal Return (CAAR T ) measure CAAR T T 1 = ε N t=+ 1 and the buy-and-hold return or Abnormal Performance Index (API T ). i it API T T 1 = ( 1+ ε it ) 1 N i t=+ 1 where ε it is the abnormal return in month t after the IPO for firm i, and there are N firms in the sample. To calculate the abnormal return ε it, the first benchmark in Model 1 below, is the standard Capital Asset Pricing Model. The second is a simple size adjustment where the benchmark is the return on the relevant size-decile portfolio. The third is a multi-index model using the market index as one factor and the Hoare-Govett Index as the measure of smaller company performance. The fourth is another multi-index model where the factors are those specified in Fama and French (1993). The final benchmark is derived from a RATS model, which allows the estimate of beta to vary during the returns window,. In the case of those models where the parameters are directly estimated from a single regression (models 1, 3, 4 and 5) the abnormal performance post-ipo is estimated by deducting the expected return calculated using parameters from the regression equation Model 1: CAPM ε c it [ β ] = R R + $ ( R R ) (1) it ft i mt ft where R it is the return on company i in event month t, R mt is the return on the market in event month t, R ft is the treasury bill return in event month t, β i is the CAPM beta of company i, estimated by an OLS regression up to 36 months after the IPO. Model 2: Size control portfolio (SS): ss ε it = R R it st (2) 5

7 where R st is the return on the size control portfolio in event month t. In this model, the control portfolios are equally weighted average returns on a portfolio of all firms in the decile to which firm i belongs. Note that model 2 does not depend on any estimated parameters, neither conditional nor unconditional. Model 3: Value weighted multi-index model using the Hoare-Govett Index as the measure of smaller company performance: hg hg [ mt ft i ht mt ] hg ε = R R + β $ ( R R ) + γ $ ( R R ) (3) it it ft i where R ht is the return on the Hoare-Govett Smaller Companies index in the event month t. The motivation for using the Hoare-Govett Smaller Companies Index (HGSCI) is that this is a value-weighted index of the bottom 80% of companies by market capitalisation. Fama and French (1996) report that many efficient markets anomalies can be explained by taking into account size and book-to-market effects through the use of a three factor benchmark. Under this model, abnormal returns are calculated as follows: Model 4: Fama and French (1996) Value-weighted three factor model: ff ff [ ( mt ft ) i ( ) i ( )] ff ε ff = R R + β$ R R + γ$ SMB + δ$ HML (4) it it ft i where SMB is the value weighted return on small firms minus the value-weighted return on large firms, and HML is the value-weighted return on high BMV firms minus the valueweighted return on low BMV firms The SMB and HML portfolios in model (4) above are formed, as in Fama and French (1996) by sorting all companies in each year by BMV and market capitalisations; only companies for which both figures are available are included in the portfolios. Again as in Fama and French (1996), value weighted returns are calculated for the bottom 30% of companies by market capitalisation and the top 30% of companies by market capitalisation, and the top 50% of companies by BMV and the bottom 50% of companies by BMV. The differences between 6

8 these value-weighted returns form the small minus big (SMB) and high minus low BMV (HML) returns. Model 5: RATS Models 1, 3 and 4 compute abnormal returns as an in-sample forecast. To obtain an out-ofsample forecast we would first need to estimate parameter estimates from prior data. Ibbotson (1975) recognised that the parameter estimates for new issues could not be estimated in a preevent estimation period, since the stock was not quoted. Ibbotson pioneered the RATS method which allows the estimate of beta to vary during the returns window, and which we estimate in the modified form used in Agrawal et al (1992). The modification allows for size effects by subtracting the decile return from the realised return in each case: RATS ε = R [ R α$ β$ ( R R )] (5) it it st t t mt ft Whereas CAARs derived from (1, 3 and 4) above assume beta (and other coefficients) remain constant through time for each firm, the RATS model in the form of (5) implicitly assumes that the difference between decile and firm betas captured by β t is constant across firms in any time period. For each of the benchmarks described by (1) to (5) above, the CAAR T, and the API T are reported. The t-test statistics are derived from the Brown and Warner (1980, p ) Crude Dependence Adjustment test for the CAARs, so that cross section dependence is taken into account, t test = CAAR T * t t ( 1 ε ε 36 ) 2 / 35 t =+ 1 T t = + 1 where ε t = 1 ε N i and we compute a simple cross-sectional t-test for the significance of the Abnormal Performance Index (API T ). it 7

9 Recently Kothari and Warner (1997) and Barber and Lyon (1997), have argued that longhorizon tests are mis-specified, and that there is significant over-rejection of the null hypothesis of no positive abnormal performance by both CAAR and API methods. The central finding of this study is that IPOs significantly underperform, and this conclusion is obtained using both CAAR and API measures of abnormal performance. Given the simulation results in Kothari and Warner (1996) it is unlikely that the magnitude of these results can be explained by specification errors. IV DATA AND DESCRIPTIVE STATISTICS The sample consists of all 588 IPOs issued by non-financial UK companies during the period from 1985 to 1992 and reported by KPMG Peat Marwick s New Issues Statistics. IPOs of investment companies, building societies, privatisation issues and foreign-incorporated companies, (including companies incorporated in the Republic of Ireland) were excluded. 3 Long-term total returns, including both capital gains and dividend payments, were computed from monthly returns data collected from the London Share Price Database (LSPD) , for all IPOs on the London Stock Exchange during The cumulative abnormal return for a holding period of m months, is measured by the sum of the monthly average abnormal returns from the end of the first month of trading to the close of the mth month. Table 1 shows some initial statistics for our sample of 588 IPOs. Calculating abnormal returns by simple market-adjusted returns (a zero-one model), table 1 reports cumulative average abnormal returns by year of issue (in Panel A) and by size of issue (in Panel B). The column Obs gives the number of firms trading in a given post-ipo month T. Logarithmic stock and market returns are taken from the LSPD database and converted into discrete returns. The market return is the total return (using the FTA dividend yield adjustment) on the FT- Actuaries (FTA) All Share Index. Using this very simple model of abnormal performance Panel A identifies significant 36-month abnormal under-performance with IPOs issued in , negative but insignificant performance in 1992, and in the remaining years the positive abnormal performance is insignificant. The cumulative returns are plotted in figure 1, and it can be seen that although the average raw returns are positive they under-perform the 3 New issues which did not constitute true IPOs, such as share issues at the time of a relisting after a firm was temporarily suspended from trading on the stock market, were excluded. 8

10 average market return, so that the cumulative average market-adjusted returns are negative. Figure 2 plots the 36-month buy-and-hold market adjusted returns for each individual security by month of issue. Panel B shows that the under-performance is common across issue sizes though the smallest new issues exhibit some over-performance, but it is statistically insignificant. In the next section we report the long run abnormal performance applying the more sophisticated benchmarks discussed in Section III. A particular difficulty encountered when trying to apply the Fama-French three factor model to UK returns is the lack of availability of book-to-market value (BMV) figures for many firms on Datastream. For every company which has returns and market capitalisation data available on the LBS tape, the SEDOL number was extracted and used to search for BMV ratios on Datastream in each year from 1980 to Unfortunately, over fifty per cent of the firms on the LBS tape for which market capitalisations were available for January 1990, did not have BMV ratios available on Datastream. This suggests that survivorship bias may be a problem when the three factor model is applied to UK data. Furthermore, the simple transposition of a US model based on BMV to the UK can be questioned given the very different accounting treatment of some balance sheet items in the UK compared to the US. Though Strong and Xu (1997) find that book-to-market equity is a significant variable in explaining the cross-section of UK expected returns. 4 Given the prominence of the three-factor model in the recent US literature, it is used to estimate abnormal returns here, although the results should be treated with some caution. 4 It is also questionable whether BMV has any real role in explaining the cross-section of US stock returns (Kothari et al, 1995, Jaganathan and Wang, 1996). 9

11 V RESULTS For 588 IPO firms issued during , cumulative average abnormal returns (CAARs) and abnormal performance indices (APIs) for the first to the 36th month of seasoning are shown in table 2 and 3, respectively. Average holding period returns by month of seasoning (in event time) and average 36-month holding period returns by issue month (in calendar time) are illustrated in figures 1 and 2, respectively. For companies that drop out of the sample before the end of 36 months, we average abnormal returns across the surviving firms. According to these figures, the long-term performance of the UK IPOs in this sample is even worse than reported by Levis (1993) for a sample of UK IPOs issued during For the CAPM case the reported CAAR for month 36 is percent, compared to the -11 per cent in Levis. Examining table 2, in the standard CAPM case, we find that for the first 15 months, monthly abnormal returns are slightly positive and average 1/20th of one percent. After month 15 we observe steadily declining CAARs, all the way out to month 36. Comparing these CAARs across different benchmarks the dominant result that comes out of these diagrams is that these long-term significantly negative abnormal returns are robust to alternative specifications of the benchmark portfolio. There is a remarkable consistency in the results across alternative benchmarks with four of the five models exhibiting very similar declining abnormal returns over 36 months; the exception being the Fama and French three-factor model, which yields an even more dramatic rejection of the null hypothesis and suggests abnormal performance of almost -30 per cent. Given our concerns mentioned in Section IV on the applicability of the Fama-French three-factor model to the UK stock market, we would hesitate to place too great a reliance of the exceptional size of the abnormal returns generated by the Fama-French benchmark. The results are not altered by allowing for size in the benchmark portfolio, either by the single Hoare-Govett factor or by assigning firms to the appropriate decile portfolio. The RATS model which allows for a beta coefficient which is the same for firms in each size decile, but varies over time, also produces the same negative pattern of abnormal returns. The CAPM results are slightly different from the others in that the excess returns up to the fifteenth month are slightly positive. In all the other four benchmarks the excess return are negative from the first month onwards. 10

12 Turning to table 3, the API measures tell a similar story of negative abnormal returns after the IPO which is again robust across benchmarks. Again the CAPM model is something of an exception since it reports positive abnormal returns up to month 23 before there is a dramatic decline in the last 13 months. For the other benchmarks declining abnormal returns are reported from the first month onwards. Again for the API measure, the Fama and French 3- factor benchmark documents the largest negative abnormal return. According to the results of the Kothari and Warner (1997) simulations, returns accumulated using the CAR approach, over-rejects the null hypothesis over 36 month windows for positive abnormal performance [Kothari and Warner (1997), Table 1] in 26% to 35.2% of the samples at 5% significance level. Whereas negative abnormal performance is observed in only 2.4% to 8.4% of samples. Further buy and hold (API) abnormal returns are found to over-state the significance of longer term positive abnormal performance in up to 90% of the samples [Kothari and Warner (1997), Table 3], and these API results are even more asymmetric with a tendency to under-state the significance of longer term negative abnormal performance. Barber and Lyon (1997) recommend that buy-and-hold returns should be used in preference to CAARs 5, but Kothari and Warner find that the distributions of APIs are significantly skewed to the right, though CAARs are only slightly negatively skewed. As we reported in Table 4 it is indeed the case that for all of our benchmarks the distribution of APIs is highly negatively skewed. Median APIs and the number of positive and negative APIs are also shown in Table 4. In all cases, the median APIs are lower (more negative) than the mean APIs, and a simple sign test shows the number of negative APIs to be significantly greater than the number of positive APIs at the 1% level 6. 5 Barber and Lyon (1997) mean true buy-and-hold returns, whereas Kothari and Warner (1997) use the term to describe the API. 6 The reported z-statistic is based upon a null hypothesis of an equal number of positive and negative APIs. 11

13 VI CONCLUSIONS In this paper we have re-examined the evidence on the long-term returns of IPOs in the UK using a new dataset of firms over the period , in which we compare abnormal returns under a number of alternative benchmarks. These benchmarks allow for the standard CAPM, size effects, the Fama-French three-factor model which includes returns to book-to-market portfolios as a factor, and the modified RATS procedure. We find that on average there are negative abnormal returns to an IPO such that typically a one pound investment after the IPO is worth less than 85 pence after three years. This finding is remarkably similar across four of the five alternative benchmarks. The Fama-French three factor model documents a more dramatic decline in the performance of IPOs but the lack of book-to-market data available on Datastream for a large sample of firms suggests that survivorship bias may be a problem when the three factor model is applied to UK data.. Recently the simulation results reported in Kothari and Warner (1997) have questioned the significance of positive long-run abnormal performance. However given that the central finding of this study is that IPOs under-perform and that this conclusion is obtained using both CAARs and buy and hold returns, the significance of the results is likely to be under-stated rather than over-stated. It is unlikely that the magnitude of the results can be explained away by specification errors. 12

14 References Agrawal, A., Jaffe, J.F. and Mandelker, G.N. (1992), 'The post-merger performance of acquiring firms; a re-examination of an anomaly', Journal of Finance, Barber, B.M. and J.D. Lyon (1997) Detecting long-run abnormal stock returns: the empirical power and specification of test statistics, Journal of Financial Economics, vol. 43, Brown, S. and Warner, J.B. (1980), Measuring security price performance, Journal of Financial Economics 8, Dimson, E. and Marsh, P. (1986), 'Event study methodologies and the size effect', Journal of Financial Economics, 17, Dimson, E. and Marsh, P. (1995), Hoare Govett Smaller Companies Index 1995, Hoare Govett, London. Fama, E.F., Booth, D. and Sinquefield, R. (1993), Differences in Risks and Returns of NYSE and NASD Stocks, Financial Analysts Journal, 49, Fama, E.F. and French, K.R. (1992), 'The cross-section of expected stock returns', Journal of Finance, 47 no.2, Fama, E.F. and French, K.R. (1996), Multifactor Explanations of Asset Pricing Anomalies, Journal of Finance, 50, Gregory, A., Matatko, J., Tonks, I. and Purkis, R. (1994), UK Directors Trading: The Impact of Dealings in Smaller Firms, Economic Journal, January 1994, Ibbotson, R.G. (1975), Price Performance of Common Stock New Issues, Journal of Financial Economics, 3, Kothari, S.P. and Warner, J.B. (1997), Measuring Long-Horizon Security Price Performance, Journal of Financial Economics, vol. 43, Levis, M (1993), The long-run performance of initial public offerings: the UK experience , Financial Management, vol. 22, Loughran, T., J.R. Ritter and K. Rydqvist (1994), Initial public offerings: International insights, Pacific-Basin Finance Journal, vol. 2, Loughran, T. and Ritter, J.R., The New Issues Puzzle, Journal of Finance, March, Ritter, J. (1991) The long-run performance of initial public offerings, Journal of Finance, vol. 46, Roell, A. (1995) The Decision to Go Public: An Overview, LSE Financial Market Group Discussion Paper no. 225 Strong, N. and X.G. Xu (1997) Explaining the cross-section of UK expected stock returns, British Accounting Review, vol. 29, no. 1, Thompson, R. (1985), Conditioning the returns-generating process of firm-specific events: a discussion of event study methods, Journal of Financial and Quantitative Analysis, vol. 20,

15 Table 1 Abnormal Returns for UK Initial Public Offerings in The column Obs gives the number of firms trading in a given post-ipo month T. Logarithmic stocks and market returns are taken from the LSPD database and converted into discrete returns. The LSPD stock return for firm i in month t, R it, for t = 1,..., 36, is the total monthly return measured from the last trading day of calendar month t-1 to the last trading day of calendar month t, where month zero is the calendar month during which the stock is first traded. Abnormal returns are measured as market-adjusted returns (a zero-one model), where the market return is the total return (using the FTA dividend yield adjustment) on the FT-Actuaries (FTA) All Share Index: ε it = R it R mt The cumulative average abnormal return for the Tth post-ipo month, CAR 1, T, for T = 1, 12, 24, 36, are calculated as T N 1 CAR1, T = ARt where ARt = εit t = 1 N i= 1 t-statistics for the CARs are CART t = where var is the variance of the AR t series within a given category for t = 1,..., 36. T var Panel A: Long-Run Returns by Year of IPO Post IPO Month T = 1 Post IPO Month T = 12 Post IPO Month T = 24 Post IPO Month T = 36 IPO Year Obs CAR 1, 1 (%) t-stat Obs CAR 1, 12 (%) t-stat Obs CAR 1, 24 (%) t-stat Obs CAR 1, 36 (%) t-stat

16 Table 1 (continued) Panel B: Long-Run Returns by IPO-Firm Size Decile Firm size is in terms of market capitalisation at issue (IPO) price measured in millions at constant 1995 prices (MCAP) Post IPO Month T = 1 Post IPO Month T = 12 Post IPO Month T = 24 Post IPO Month T = 36 Size Decile MCAP ( m) Obs CAR 1, 1 t-stat Obs CAR 1, 12 t-stat Obs CAR 1, 24 t-stat Obs CAR 1, 36 (%) t-stat

17 TABLE 2 Cumulative Average Abnormal Returns For Alternative Benchmark Models. Figures are percentages; models are the Capital Asset Pricing Model (1), the Dimson-Marsh (1986) simple size adjustment model (2), a multi-index model using the return on the HGSCI minus the return on the FTASI (3), the Fama and French (1996) three factor model (4), and the RATS model (5). The t-statistics are computed according to the crude dependence adjustment method of Brown and Warner (1980, 1985). Month CAPM Results SS Results HG Results FF Results RATS Results CAR (%) t-test CAR (%) t-test CAR (%) t-test CAR (%) t-test CAR (%) t-test t (-0.15) (-2.31) (-1.24) (-2.83) (-2.19) t (0.05) (-2.65) (-1.37) (-3.09) (-2.34) t (0.64) (-2.09) (-0.93) (-2.99) (-1.84) t (0.60) (-2.40) (-1.21) (-3.59) (-2.15) t (0.48) (-2.38) (-1.36) (-4.01) (-2.16) t (0.41) (-2.65) (-1.66) (-4.59) (-2.39) t (0.57) (-2.94) (-1.65) (-5.05) (-2.63) t (0.07) (-3.82) (-2.56) (-5.80) (-3.40) t (0.29) (-3.78) (-2.48) (-5.60) (-3.16) t (0.50) (-3.80) (-2.46) (-5.70) (-3.11) t (0.27) (-4.33) (-2.81) (-6.08) (-3.60) t (0.32) (-4.29) (-2.88) (-6.54) (-3.59) t (0.50) (-4.38) (-2.73) (-6.78) (-3.68) t (0.55) (-4.37) (-2.53) (-6.80) (-3.70) t (0.31) (-4.76) (-2.84) (-7.34) (-4.02) t (-0.04) (-5.11) (-3.15) (-7.69) (-4.27) t (-0.38) (-5.06) (-3.27) (-8.04) (-4.21) t (-0.73) (-5.39) (-3.62) (-8.55) (-4.38) t (-0.74) (-5.54) (-3.59) (-8.62) (-4.39) t (-0.75) (-5.73) (-3.70) (-8.99) (-4.52) t (-0.71) (-5.58) (-3.61) (-9.04) (-4.37) t (-0.77) (-5.39) (-3.61) (-9.03) (-4.22) t (-0.76) (-5.10) (-3.50) (-8.96) (-3.96) t (-1.16) (-5.42) (-3.84) (-9.49) (-4.18) t (-1.57) (-5.68) (-3.86) (-9.85) (-4.37) t (-1.71) (-5.82) (-4.01) (-10.07) (-4.49) t (-1.90) (-5.79) (-4.12) (-10.46) (-4.49) t (-2.26) (-5.90) (-4.35) (-10.76) (-4.52) t (-2.53) (-5.90) (-4.43) (-11.00) (-4.51) t (-2.69) (-5.63) (-4.19) (-11.04) (-4.26) t (-3.05) (-5.70) (-4.20) (-11.23) (-4.26) t (-3.48) (-5.99) (-4.46) (-11.62) (-4.46) t (-3.71) (-6.29) (-4.69) (-11.85) (-4.64) t (-3.51) (-6.05) (-4.17) (-11.39) (-4.43) t (-3.67) (-6.21) (-4.24) (-11.62) (-4.54) t (-3.84) (-6.01) (-4.00) (-11.62) (-4.40) 16

18 TABLE 3 Abnormal Performance Index For Alternative Benchmark Model: Figures are percentages; models are the Capital Asset Pricing Model (1), the Dimson-Marsh (1986) simple size adjustment model (2), a multi-index model using the return on the HGSCI minus the return on the FTASI (3), the Fama and French (1996) three factor model (4) and the RATS model (5). The t-statistics are computed according to a simple cross-section t-test. Month CAPM Results SS Results HG Results FF Results: API t-test API t-test API t-test API t-test t (-0.26) (-2.30) (-1.32) (-3.07) t (0.05) (-2.80) (-1.56) (-3.26) t (1.08) (-1.75) (-0.69) (-2.40) t (1.04) (-2.01) (-1.01) (-2.95) t (1.09) (-1.72) (-0.84) (-3.00) t (1.01) (-1.77) (-1.08) (-3.46) t (1.47) (-1.54) (-0.70) (-3.22) t (1.06) (-2.08) (-1.24) (-3.40) t (1.35) (-1.94) (-1.21) (-3.22) t (1.69) (-1.93) (-1.19) (-3.23) t (1.72) (-1.94) (-1.12) (-3.01) t (1.71) (-2.12) (-1.39) (-3.68) t (2.01) (-2.19) (-1.14) (-3.85) t (2.16) (-2.07) (-0.93) (-3.75) t (2.10) (-2.23) (-1.08) (-3.98) t (1.78) (-2.51) (-1.22) (-4.26) t (1.61) (-2.34) (-1.16) (-4.35) t (1.20) (-2.41) (-1.29) (-4.47) t (1.22) (-2.45) (-1.22) (-4.51) t (1.21) (-2.59) (-1.49) (-4.89) t (1.25) (-2.38) (-1.36) (-4.91) t (1.36) (-2.08) (-1.16) (-4.76) t (1.37) (-1.65) (-0.92) (-4.53) t (0.99) (-1.69) (-1.08) (-4.76) t (0.59) (-1.79) (-1.11) (-4.93) t (0.36) (-1.83) (-1.09) (-5.03) t (0.16) (-1.85) (-1.11) (-5.24) t (-0.36) (-2.03) (-1.36) (-5.75) t (-0.65) (-1.95) (-1.24) (-5.88) t (-0.68) (-1.65) (-0.95) (-5.82) t (-0.95) (-1.55) (-0.82) (-5.65) t (-1.37) (-1.72) (-0.92) (-5.81) t (-1.55) (-1.80) (-1.06) (-6.02) t (-1.63) (-1.74) (-0.84) (-5.94) t (-1.81) (-1.82) (-0.83) (-6.10) t (-1.94) (-1.62) (-0.62) (-6.08) 17

19 Table 4 Distribution of API Returns CAPM Results SS Results HG Results FF Results to to to to mnth mnth mnth mnth Mean API Median API No +ve API No -ve API Sign test