Pierpont and the Capital Market

Trevor Sykes, alias Pierpont, began writing as a business journalist in 1972. In weekly columns for the Australian Financial Review, Sykes wrote poignant commentaries on Australian companies, usually in a critical yet humorous style. Pierpont's column was distinguished firstly by his independence and secondly by his large following.

We consider issues similar to those addressed by Foster (1979, 1987), who studied the share price reactions of U.S. companies when Abraham Briloff criticized them in his Barron's column for their financial reporting practices. While Pierpont had much in common with Briloff, they can be contrasted in important ways. As a full-time journalist (Briloff was an academic), Pierpont was prolific. He traversed more territory, raising questions on corporate governance, company announcements and corporate regulation and expressing negative, neutral and positive views about the companies concerned. And Pierpont's entertaining style garnered a large following.

We extend previous work in this area by examining the market reaction to the regular columns of a ‘star journalist-cum-analyst’, Pierpont, which were published weekly for many years. We further differentiate our work from earlier studies (such as the work by Foster on Briloff, 1979, 1987) by employing an independent, double-coding procedure that allows us to identify reliably and more finely partition the trading recommendations contained in each column. For example, where all of Briloff's comments were negative, we classify Pierpont's columns according to whether they are focused or passing references, and whether they translate into positive, negative or neutral recommendations.

The core issue we address is whether Pierpont's large following translated into unusual share market activity. Put another way, was Pierpont's column ‘just a good read’ or was there more to it?

The Press in the Information Process

Research on the relation between the press and commerce has been limited (Zingales, 2000; Miller, 2006), and the literature that does exist has reached conflicting conclusions. One strand suggests the press tends towards sensationalism, and lacks depth (Miller, 2006). Jensen (1979), for example, considered the press to be a form of entertainment, with articles being written to appeal to the lowest common denominator. Core et al. (2008) found press coverage of executive compensation was consistent with this view, concluding it concentrated on large ex post stock gains rather than the ex ante compensation expense to the company. They concluded press coverage had no impact on the compensation behaviour of firms and by inference no influence on the underlying value of the firms being covered. In contrast, Dyck and Zingales (2002a) suggested the press could encourage financial bubbles by adopting a company's spin in return for private information, rather than providing in-depth original analysis. Collectively, these studies question the utility of the press's role as an information intermediary, suggesting it may even play a negative role.

Others have concluded the press is an important source of information, and in doing so it influences the investing public and the actions of firms that are mentioned (Dyck and Zingales, 2002b; Djankov et al., 2003; Stromberg, 2004; Miller, 2006). Some studies have found that the press enhances the information environment by packaging and rebroadcasting information already available (Huberman and Regev, 2001; Dyck and Zingales, 2003; Dyck et al., 2008). As Miller (2006, p. 1005) put it, ‘press coverage per se influences the public's response to information and the actions of firms being covered’.

Miller (2006) explored these ideas, concluding that articles based on original analysis provide new information to the market while stock prices do not react significantly to the rebroadcast of ‘old’ information. Authors who publish repeatedly are more likely to provide original analysis with strong information content, developing their own brand, and readers are likely to perceive regular columns as more informative. Furthermore, the business-orientated press is more likely to undertake original analysis while non-business periodicals are engaged primarily in rebroadcasting.

Miller's (2006) findings do not rule out the possibility that investors respond to the increased salience of a sensationalized story as well as to the information contained within the story. Tetlock (2007) found salience can move stock prices even in the absence of news. In a related line, da Silva Rosa and Durand (2008) found that the likelihood of a stock appearing in a portfolio in a stock picking competition is much higher if it was featured in the national press shortly before the close of the competition.

Several North American studies of the role of newspaper columnists have focused on the ‘Heard on the Street’ column in the Wall Street Journal (e.g., Lloyd Davies and Canes, 1978; Liu et al., 1990; Beneish, 1991). ‘Heard on the Street’ reports revisions in analysts' recommendations (and cites the analysts' reasons), as opposed to disseminating new private information (Beneish, 1991). Buy (sell) recommendations have been found to be associated with positive (negative) abnormal stock price changes on the day of the publication and with smaller, but statistically significant, price changes over the two days before publication. Trading volume has also been found to be higher. Liu et al. (1990) reported a symmetric stock price reaction to buy versus sell recommendations and that the impact of single-company recommendations was greater than multi-company recommendations.

Pound and Zeckhauser (1990) used the ‘Heard on the Street’ column to identify firms rumoured to be the target of a public takeover offer. By doing so, they could gauge the effect of rumour on stock prices before, during and after the rumour surfaced in the press. They found the market to be efficient to the extent that no excess return could have been realized on average through an investment strategy of buying (or selling) rumoured takeover targets once the rumours were published. However, stock prices were more volatile on the publication date.

Star Analysts

Trevor Sykes, alias Pierpont, can be characterized as an ‘unaffiliated star analyst’.¹ Like Briloff, he was unconstrained by allegiance and was free to write as either a buy- or a sell-side analyst. Over the period 1968 to 1998 examined by Desai and Jain (2004), Briloff wrote thirty-one articles. Pierpont's column was published weekly, with 506 columns published from 1995 (when he re-joined the Australian Financial Review) until 2005, when he ceased writing regular columns.

When Foster (1979, 1987) examined the stock prices of companies commented upon by Briloff, he observed a negative stock market reaction around the time of the commentary. Specifically, Foster (1979) found companies criticized by Briloff on average fell by 8.6 per cent (relative to a benchmark; 8.1 per cent in Foster, 1987) on the first day after publication and there was no reversal of the loss over the next thirty days. Desai and Jain (2004) extended Foster's post-publication period and found the featured companies lost more than 20 per cent on average over the two years following Briloff's story.

Making stock recommendations is one of the most important functions that analysts perform (Schipper, 1991; Francis and Philbrick, 1993; Womack, 1996). Womack (1996) confirmed analysts' recommendations have value by reporting four-day cumulative abnormal returns surrounding recommendation changes. In another study, buy-siders surveyed by Institutional Investor ranked an analyst's stock picking ability as an important consideration in their overall voting decisions (Leone and Wu, 2002).

Star analysts have greater impact. The two main ranking systems of U.S. analysts are the Institutional Investor All-America Research Team and the Wall Street Journal's ‘Best on the Street Analysts’. Both pick the best analysts based on performance and the usefulness of their recommendations. While Bagnoli et al. (2008) found little overlap in the top-three analysts in the two ranking systems in any given year during the period 1998–2003, their results were consistent whichever ranking system they used.

Early research found no systematic difference among analysts in terms of the accuracy of their forecasts (e.g., Brown and Rozeff, 1980; O'Brien, 1990; Butler and Lang, 1991). Stickel (1992) controlled for the age of the forecasts and showed All-American analysts were more accurate. Consistent with Stickel (1992), Gleason and Lee (2003) found forecasts of ‘celebrity’ analysts elicited a stronger immediate price response, but the subsequent price drift was less pronounced. Additionally, the market paid more attention to analysts with celebrity status but was more likely to under-appreciate revisions made by others with comparable forecasting ability. This result suggests an analyst's reputation, and not just their forecasting ability, will influence stock prices. Further, Leone and Wu (2002) found evidence consistent with all-star analysts outperforming the rest. Specifically, they made more accurate earnings forecasts, better stock recommendations, and were less optimistically biased. Highly ranked analysts also took greater risks, in that their forecasts often deviated further from the consensus.

Although all-star analyst rankings change from year to year, suggesting some randomness in their performance, it is widely believed that security analysts do differ systematically in their forecasting skill. For example, the Wall Street Journal's 1996 All-Star Analysts Survey reported that:

Jacob et al. (1999) examined these claims, finding that forecast accuracy of incoming analysts did not generally improve with experience. Rather, forecast accuracy was influenced by analyst-related factors such as aptitude or company alignment and by environmental factors such as brokerage size and industry specialization.

In the Australian setting, Brown and Walter (1982) conducted an experiment with a group of investment professionals. They obtained the professionals' recommendations and tracked the subsequent investment performance of the stocks that were identified as future winners and losers. The professionals consistently outperformed the market, their success mostly being due to their ability to identify losers to sell rather than winners to buy.

Research Question

In the light of the evidence just discussed, our aim is to assess the influence of Pierpont's columns on stock prices. Like Foster (1979), we adopt the classical event study methodology, which is well suited to this task. In brief, we attempt to reject the null hypothesis (that stock returns are unrelated to Pierpont's commentary) in favour of the alternative hypothesis, that a company which was the subject of a positive (negative) comment would have experienced a positive (negative) abnormal return around the time of the comment's publication. Similarly, we predict no significant return if the company was the subject of a neutral comment. Formally:

RESEARCH DESIGN AND SAMPLE

Pierpont's column was published weekly in the Australian Financial Review (AFR). Sykes, alias Pierpont, re-joined the AFR midway through 1995, and the AFR continued to carry his column until the end of 2005. Pierpont considered himself to be an expert on mining companies, and his fictitious Blue Sky Mines was a regular feature. However, Pierpont did not mind venturing an opinion on high tech and other firms as well.² His Dubious Distinction awards have been an annual event, celebrating ‘those denizens of the financial world who managed to make the biggest spectacle of themselves’.³ Companies were known to pay close attention to what Pierpont had to say about them.⁴

Many of Pierpont's columns are posted on his personal website and available for download. We collected 474 articles from this source. A further thirty-two articles were collected from the Factiva database, giving a total of 506. Table 1 summarizes his columns published in the second half of 1995 and the companies commented upon. On average, Pierpont wrote forty-six columns per year while at the AFR (48.3 columns per year when the half year, 1995, is excluded).

Within those 506 articles, Pierpont referenced 971 companies.⁵ Many were financially distressed or were otherwise suspended from trading, or delisted. Others were mentioned by Pierpont before they were first traded. Consequently, stock return data were unavailable for 530 of the 971 cases, potentially leaving 441 for study. A further eight were excluded from some of the more detailed analysis due to missing data.

Company specific references were independently coded by two of the authors according to the perceived tone of the reference (i.e., positive, neutral or negative), its type (i.e., whether the company was a focus of the column or mentioned in passing), and whether the column referred to a single company. Codings were identical in 96.4 per cent of the 971 cases and the initial differences on the remaining thirty-five cases were readily resolved. For ease of exposition, positive comments are designated buy recommendations, neutral comments are holds, and negative comments are sells.

Of the 971 company references in total, 675 (58 per cent) were sell recommendations, 188 (28 per cent) were holds and 108 (14 per cent) were buys. When companies with no share price data are removed, 258 of the remaining references (70 per cent) were sells, 122 (19 per cent) were holds and 61 (11 per cent) were buys (see Table 2). Of the 971 references, 521 (54 per cent) were focused comments and the remainder (450, or 46 per cent) were passing references. And of the 441 references with returns data, 179 (46 per cent) were focused and 262 (59 per cent) were passing.

The 441 references with returns data related to 207 companies, with each company mentioned on average 2.13 times. The 120 companies with passing references averaged 2.18 mentions, while focused references mentioned 129 companies, on average 1.39 times each. Buy recommendations related to 52 companies (mentioned on average 1.17 times), hold recommendations to 61 companies (average 2.00 times), and sell recommendations to 142 companies (average 1.81 times). These categories are not mutually exclusive, since some companies were the subject of different recommendations on different dates.

We assigned firms to industries according to their general industrial classification (GICS) codes. The mining sector comprises energy (GICS 1) and materials (GICS 2), and IT comprises information technology (GICS 8) and telecommunications (GICS 9). Consistent with Pierpont's believed expertise, 151 references were to mining companies and 71 to IT companies (see Table 2, Panel B). Of these references, 41.72 per cent and 42.25 per cent are focused references to mining and IT companies respectively. Within the mining and IT sectors, 59 per cent and 64 per cent of the sell recommendations were focused references. Combined, mining and IT companies accounted for over half the sample, indicating Pierpont gravitated to industries where information asymmetries between insiders and outsiders are endemic. Companies in the financial sector (GICS 7) also formed a relatively high proportion of Pierpont's references (there were 99), although a lower proportion of them (36 per cent) were focused references. All other industries accounted for less than 10 per cent of the total number of companies mentioned.

On average, each Pierpont column referred to 1.92 companies, of which 1.33 were sell recommendations and 1.03 were focused (see Table 3, Panel A). When the Dubious Distinction columns are removed (they mentioned 12.13 companies on average, with no buy recommendations), the average drops to 1.76 companies, with 1.19 being sells (see Panel B). Panel C of Table 3 shows the average number of references made when columns with no companies mentioned are excluded (but with the Dubious Distinction columns included). When at least one company was mentioned, the average was 2.52 company references, with 1.75 being sells and 1.35 companies being the focus of Pierpont's attention.

Table 4 presents the frequencies of the number of companies mentioned in a Pierpont column. Panel A shows the total number of companies mentioned and the number of companies mentioned classified by recommendation. Overall, 121 columns commented on general macroeconomic and other business related issues and did not refer to any company specifically. The majority of columns that did refer to companies commented on a single company (159 columns). In three instances, the column referred to 14, 15 and 16 different companies; they all received a Dubious Distinction award.

When grouped by recommendation type, the frequency of companies mentioned per column is revealing. The vast majority of columns (435) did not recommend any buys, 59 columns made a single buy recommendation, and only 16 recommended two or more companies to buy. Hold recommendations followed a similar trend. But sell recommendations exhibited a strikingly different pattern: 183 columns made no sell recommendation, 172 recommended selling a single company, 84 recommended selling two companies, and 67 recommended selling more than two.

The negative tone of Pierpont's columns is seen in his many sell recommendations, especially among focused references. Of the 441 company references included in our ‘final’ sample, 152 are focused sells (Focus*Sell). This is highlighted further when the number of columns that contained a particular type of recommendation is examined (see Table 4). Of the 506 columns, 435 (86 per cent) contained no buy recommendation, 385 (76.1 per cent) contained no hold recommendation, but only 183 (36.2 per cent) did not contain a recommendation to sell. When the 121 columns that did not mention any companies listed on the ASX are removed, only 62 of the remaining 385 (16 per cent) did not mention a company to sell.

We also collected data on the word count for each column and the size of each company that rated a mention.⁶ Daily returns, needed to measure the stock's return around the publication date, were sourced from the Securities Industry Research Centre of Asia-Pacific (SIRCA) database. Returns were examined over ten trading days (typically, two calendar weeks) up to and including the publication date and for another twenty trading days thereafter.

RESULTS

Descriptive statistics are examined first, followed by univariate and multivariate analyses. No new statistical tests are provided and Foster's work on Briloff is the guide.

Univariate Analysis

Univariate analysis of stock returns around the date of a Pierpont column is summarized in Table 6. It shows the average market-adjusted buy-and-hold log return by recommendation type from an investment made at the end of day −10 and held to the end of day +20, where day 0 is the day the column was published in the AFR, for the full sample of firms with return data.

Table 6.
MARKET-ADJUSTED AVERAGE BUY-AND-HOLD CUMULATIVE RETURN BY DAY AND RECOMMENDATION TYPE

Event day	Full sample (N= 441)			Information technology (N= 71)			Mining (N= 151)
	Buy	Hold	Sell	Buy	Hold	Sell	Buy	Hold	Sell
	(N= 61)	(N= 122)	(N= 258)	(N= 11)	(N= 16)	(N= 44)	(N= 27)	(N= 41)	(N= 83)
−9	0.0096*	0.0014	−0.0009	0.0162*	−0.0002	0.004	0.0139	0.0059**	0.0035
−8	0.0076	0.0006	−0.0006	0.0142	−0.0042	0.0116	0.017	0.0083**	−0.0019
−7	0.0068	−0.0019	−0.0062	0.0324	−0.0082	0.0024	0.0102	0.0065	−0.0124
−6	0.0145	−0.0024	−0.0112*	0.0718	−0.015	−0.0089	0.0076	0.0082	−0.0151*
−5	0.0123	−0.004	−0.0116**	0.0629	−0.0151	−0.0254***	0.0091	−0.0025	−0.006
−4	0.0175	−0.0039	−0.0218***	0.0685	−0.0197	−0.0523***	0.0178	−0.005	−0.0138
−3	0.0173	−0.0017	−0.0296***	0.0594	−0.011	−0.0634***	0.0198	−0.0044	−0.0209
−2	0.0167	−0.0031	−0.0281***	0.0718	−0.01	−0.0739***	0.0137	−0.0065	−0.0186
−1	0.0168	−0.0023	−0.0263***	0.0716	−0.0144	−0.0743***	0.0152	−0.0035	−0.0169
0	0.0214	−0.0015	−0.0284***	0.0881	−0.0168	−0.0765***	0.0205	0.0015	−0.0195
1	0.021	−0.0017	−0.0333***	0.0794	−0.0168	−0.069***	0.0241	0.0036	−0.0253*
2	0.0302*	−0.001	−0.0408***	0.0891	−0.0146	−0.0793***	0.0299	0.0009	−0.0401***
3	0.0283*	0.0017	−0.0408***	0.0825	−0.014	−0.0866***	0.0261	−0.0001	−0.0401***
4	0.0301*	−0.0005	−0.0446***	0.0846	−0.0184	−0.0966***	0.0257	−0.0025	−0.0375**
5	0.0311*	0.0017	−0.0455***	0.086	−0.0228	−0.1006***	0.0322	−0.0001	−0.0400**
6	0.0335*	0.0001	−0.0461***	0.1051	−0.0267	−0.095***	0.0323	0.0005	−0.0416**
7	0.0316*	−0.0016	−0.0489***	0.0987	−0.0354	−0.0951***	0.0301	−0.0033	−0.0520***
8	0.0281	−0.0008	−0.0505***	0.0951	−0.0432	−0.1027***	0.0277	0.0047	−0.0562***
9	0.0305*	−0.0017	−0.0538***	0.091	−0.0385	−0.1112***	0.0352	−0.0033	−0.0575***
10	0.0211	−0.0042	−0.0521***	0.0817	−0.0413	−0.1064***	0.021	−0.004	−0.0555***
11	0.0291	−0.0051	−0.0550***	0.0942	−0.0379	−0.1106***	0.0391	−0.0054	−0.0635***
12	0.0305	−0.007	−0.0526***	0.1078	−0.0344	−0.1127***	0.0348	−0.0086	−0.0637***
13	0.0312	−0.0075	−0.0526***	0.1176	−0.0369	−0.1194***	0.0297	−0.0102	−0.0625***
14	0.0307	−0.0069	−0.0527***	0.1263	−0.0352	−0.1234***	0.0173	−0.0124	−0.0566***
15	0.0329	−0.0033	−0.0550***	0.1328	−0.0238	−0.1251***	0.0221	−0.0104	−0.0604***
16	0.0362	−0.0029	−0.0577***	0.1452	−0.0315	−0.114***	0.0179	−0.0069	−0.0649***
17	0.0388	−0.0025	−0.0585***	0.1612	−0.029	−0.102***	0.0198	−0.0066	−0.0720***
18	0.0523	−0.0018	−0.0577***	0.1758	−0.0238	−0.1079***	0.0366	−0.0073	−0.0712***
19	0.0614*	−0.0046	−0.0602***	0.1939	−0.0263	−0.1067***	0.045	−0.0129	−0.0698***
20	0.0640**	−0.0056	−0.0555***	0.2086	−0.0263	−0.0912***	0.0499	−0.0147	−0.0609**

• *, ** and ***  denote a statistically significant difference from zero at the 10%, 5% and 1% levels, respectively. The event date, day 0, refers to the date on which the Pierpont column was published.

Hold references, as predicted, do not exhibit any significant return over the period examined. Buy references are predicted to yield positive returns on publication. The results confirm this prediction, with days +2 to +7 yielding positive and significant cumulative returns. Additionally, the cumulative return to days +9, +19 and +20 are positive and significant. Day −9 also exhibits a positive and significant return, which cannot be explained by publication of a comment by Pierpont. However, this result is only for a single day and is not prolonged.

In contrast, sell recommendations show a more consistent pattern. They are predicted to yield negative returns over the period of interest, which is indeed what the data reveal. From day −6 to day +20, the cumulative return is negative and significantly so. A possible explanation for the poor performance before the article is published is that Pierpont sometimes chose to comment on topical matters already known, at least in part, by the market. When Pierpont reinterpreted the data, his ‘take’ was informative, so that stock prices on average fell further.

Our results are consistent with Foster's (1979, 1987) studies of the market reaction to Briloff's articles, which were strongly associated with negative returns, and there was no reversal for at least the next thirty days. Also of interest is the comparison of the magnitude of cumulative returns of the buy and sell portfolios. While the cumulative return on sell recommendations is reliably negative over a longer period, the equivalent return on buy recommendations is on average slightly larger in magnitude (0.0640 compared to −0.0555). Figure 1, Panel A, graphs the average cumulative returns from the end of day −10 to the end of day +20, with the vertical line indicating the announcement date.

Table 6 also shows the average buy-and-hold cumulative market-adjusted returns for the IT and mining sectors. For IT and for mining, and consistent with the full sample (which includes both), Pierpont's hold recommendations were not associated with a significant market-adjusted return. For buy recommendations in the IT sector, the returns, while positive, were not statistically different from zero. This might well be due to a lack of statistical power as only eleven IT firms (15.49 per cent of cases from that sector) received a buy recommendation. Interestingly, and highlighted in Panel B of Figure 1, from day +11 through to the end of the holding period there is a strong drift upwards in the average cumulative return, which reaches 20.86 per cent by the end of day +20. On the sell side, the results mirror those of the full sample, with significantly negative cumulative returns from day −5 onwards. Again, Pierpont may have decided to investigate some companies based on topical news and events, which contributed to the return before Pierpont's column was published. Subsequently, the negative return persisted. By the end of day +20, the negative cumulative return on sell recommendations in the IT sector was larger in magnitude than for the full sample and for all other industry groups. While Pierpont could be considered a mining specialist, his expertise clearly extended to the IT sector as well.

With respect to the mining sector, the returns associated with buy recommendations, while positive, were not reliably so. Again, this could reflect a lack of statistical power due to the small sample size. Also, as highlighted in Figure 1, Panel C, Pierpont's hold recommendations on average were not associated with any clear pattern in returns. On the sell side, the results for the mining sector exhibit a different pattern from the full sample and the IT sector. Although the cumulative return on sell recommendations in the mining sector was on average negative from day −8 onwards, it was significantly different from zero only from the day after the column was published (day +1). The pattern is consistent with Pierpont being regarded as an expert in mining stocks, and the market adopting his views. Also, compared to the magnitude of the average return on the full sample of sell recommendations, the mining sample experienced a larger abnormal loss (0.0609), although the average loss was not as large as the average loss on IT companies.

Overall, the evidence on market-adjusted returns around the time the Pierpont columns were published is clear. No significant return is documented for firms receiving a hold recommendation. For sell recommendations, the returns were significantly negative and persistent, while for the buys they were positive, although not reliably so.

As part of our univariate testing, we also examined trading volume around the date of the Pierpont column. To do this, we collected data on the total volume of trades in dollar terms.⁸ The average dollar volume per day within the window of day −9 to day +20 was used to normalize each day's volume.⁹ From prior literature we know that trading volume varies by day of the week. Consequently, we obtained the total daily dollar volume of all ASX-traded shares from 1 July 1995 to 31 January 2006, which is the period we are investigating. We then calculated the average value of trading for each day of the week and standardized the daily average, to yield the unconditional expectation of each day's trading volume.¹⁰ The unconditional expectation was then used to test whether the normalized volume observed around the time a stock was mentioned in one of Pierpont's columns could have been simply a chance occurrence.

Untabulated results demonstrated that trading volume on the day a stock was mentioned in a Pierpont column (day 0) and on the next day as well (day +1) was significantly greater than would otherwise have been expected. Specifically, on the event date (day 0), trading volume was 8.9 per cent greater than expected for the average stock on Fridays and the margin is statistically significant (one-tailed p-value 0.07).¹¹ When we broke the sample down by recommendation type, we found this result was driven largely by sell recommendations, for which the trading volume was 16.8 per cent higher than expected. On day +1 average dollar volume was 6.9 per cent greater than expected for the full sample (one-tailed p-value 0.05). When the sample was broken down by recommendation type, we found buy recommendations experienced trading volume 11.4 per cent higher and sell recommendations 8.8 per cent higher. Thus the trading volume evidence is consistent with the univariate evidence from the market-adjusted daily returns: Pierpont was indeed seen to be a skilled investment analyst.

Sensitivity analysis was performed to ensure the validity of our results. First, we analysed whether there was any anticipation effect associated with Pierpont's columns. We did this by calculating pairwise correlations between the pre-event returns over the window [−9, −2] (i.e., from the last sale price on day −10 to that on day −2) and the event and post-event period returns [−1, +4] and [−1, +20]. Untabulated results indicated no significant correlation between these returns. Additionally, when we analysed focus sells and focus buys with minimal prior [−9, −2] returns, our results remained qualitatively the same.

We also investigated whether a mentioned company made any announcement to the ASX within the period from two to six days before the Pierpont column was published, in case company announcements could have confounded our results. Of the 441 companies mentioned by Pierpont, 148 had an announcement in this period. When we performed a non-parametric resampling test to compare the average return for firms that made an announcement with those that did not, we could not reject the null hypothesis of no difference between the average returns of the two groups.

Finally, we broke the sample into different time periods: 1995–99, 2000–02 and 2003–05. We found the portfolio comprising all sell recommendations did not perform as badly in 1995–99 as in the later two periods, while focus sells performed significantly worse in 2003–05 than in either of the earlier periods. Although there was some evidence that sell recommendations in the second sub-period performed worse than in the first, the difference in their average performance was not statistically significant.

Multivariate Analysis

Multivariate analysis was conducted to provide additional insight into the sources of the abnormal returns. We did this by regressing cumulative market-adjusted returns, measured over three overlapping holding periods, on several combinations of firm size plus a set of categorical variables reflecting industry (mining, information technology, other), the nature of Pierpont's ‘recommendation’ (buy, sell, hold), and whether the article referenced one or more companies. The full structural model is specified in equation (1):

()

Table 7 reports the results from estimating, by ordinary least squares, equation (1) and several of its reduced forms. Panel A is for the full sample, Panel B the subsample of sell recommendations and Panel C the subsample of focus sell recommendations. Panel D is for the full sample of ‘action’ recommendations (i.e., it excludes holds, which in a statistical sense add noise to the tests). Within Table 7, models 1, 2 and 3 exclude Single*Sell from the model and models 4, 5 and 6 include it. The three dependent variables are Mkt_Ret4, Mkt_Ret20 and Mkt_Retpost.

Table 7.
ORDINARY LEAST SQUARES REGRESSION OF CUMULATIVE MARKET-ADJUSTED RETURN ON RECOMMENDATION TYPE, FIRM SIZE AND INDUSTRY

Panel A: Full sample
	Model 1		Model 2		Model 3		Model 4		Model 5		Model 6
	Mkt_Ret4		Mkt_Ret20		Mkt_Retpost		Mkt_Ret4		Mkt_Ret20		Mkt_Retpost
Intercept	−0.0603	0.5359	0.2041	0.3516	0.0482	0.8007	−0.0603	0.5348	0.2041	0.3493	0.0482	0.8004
Buy	0.0478	0.7310	−0.0517	0.8685	0.0594	0.8275	0.0478	0.7304	−0.0517	0.8678	0.0594	0.8272
Sell	−0.1201	0.2611	−0.6617	0.0061	−0.2993	0.1535	−0.1002	0.3498	−0.6049	0.0121	−0.2663	0.2059
Size	0.0029	0.5044	−0.0090	0.3612	−0.0020	0.8152	0.0029	0.5033	−0.0090	0.3588	−0.0020	0.8149
Size*Buy	−0.0016	0.8011	0.0023	0.8740	−0.0022	0.8596	−0.0016	0.8006	0.0023	0.8733	−0.0022	0.8593
Size*Sell	0.0052	0.2843	0.0293	0.0075	0.0130	0.1711	0.0045	0.3589	0.0272	0.0129	0.0118	0.2160
Mining	−0.0018	0.9145	−0.0154	0.6901	−0.0141	0.6767	−0.0018	0.9143	−0.0154	0.6887	−0.0141	0.6763
Mining*Buy	0.0018	0.9536	0.0385	0.5882	0.0201	0.7465	0.0018	0.9534	0.0385	0.5864	0.0201	0.7461
Mining*Sell	0.0055	0.7959	0.0152	0.7494	0.0061	0.8838	0.0037	0.8621	0.0101	0.8321	0.0031	0.9409
IT	−0.0156	0.5197	−0.0227	0.6772	−0.0138	0.7721	−0.0156	0.5186	−0.0227	0.6757	−0.0138	0.7718
IT*Buy	0.0124	0.7563	0.2253	0.0122	0.1352	0.0843	0.0124	0.7556	0.2253	0.0118	0.1352	0.0838
IT*Sell	0.0212	0.4646	0.0093	0.8863	0.0438	0.4407	0.0172	0.5521	−0.0020	0.9755	0.0372	0.5127
Single*Sell							−0.0292	0.0748	−0.0833	0.0234	−0.0484	0.1320
N	433		433		433		433		433		433
Adj R2	0.0296		0.0847		0.0270		0.0346		0.0937		0.0299
Panel B: Sell recommendations
	Mkt_Ret4		Mkt_Ret20		Mkt_Retpost		Mkt_Ret4		Mkt_Ret20		Mkt_Retpost
Intercept	−0.1804	0.0004	−0.4577	<0.0001	−0.2511	0.0104	−0.1605	0.0022	−0.4008	0.0003	−0.2181	0.0305
Size	0.0081	0.0008	0.0203	0.0001	0.0110	0.0181	0.0074	0.0028	0.0182	0.0005	0.0098	0.0388
Mining	0.0036	0.7988	−0.0002	0.9948	−0.0080	0.7713	0.0018	0.8980	−0.0054	0.8583	−0.0110	0.6899
IT	0.0055	0.7601	−0.0134	0.7266	0.0300	0.3919	0.0016	0.9306	−0.0247	0.5208	0.0234	0.5069
Single*Sell							−0.0292	0.1218	−0.0833	0.0360	−0.0484	0.1835
N	250		250		250		250		250		250
Adj R2	0.0342		0.0574		0.0138		0.0397		0.0704		0.0169
Panel C: Focus sell recommendations
	Mkt_Ret4		Mkt_Ret20		Mkt_Retpost		Mkt_Ret4		Mkt_Ret20		Mkt_Retpost
Intercept	−0.2406	0.0018	−0.6104	0.0002	−0.3079	0.0376	−0.2349	0.0029	−0.5527	0.0010	−0.2695	0.0745
Size	0.0110	0.0033	0.0277	0.0006	0.0138	0.0553	0.0108	0.0044	0.0257	0.0015	0.0124	0.0867
Mining	0.0109	0.6216	0.0146	0.7585	−0.0090	0.8346	0.0103	0.6450	0.0079	0.8671	−0.0134	0.7555
IT	0.0007	0.9802	−0.0282	0.6213	0.0263	0.6100	−0.0005	0.9865	−0.0395	0.4898	0.0188	0.7172
Single*Sell							−0.0086	0.7345	−0.0869	0.1077	−0.0578	0.2377
N	148		148		148		148		148		148
Adj R2	0.0437		0.0753		0.0108		0.0378		0.0855		0.0136
Panel D: Full sample (excluding holds)
	Model 1		Model 2		Model 3		Model 4		Model 5		Model 6
	Mkt_Ret4		Mkt_Ret20		Mkt_Retpost		Mkt_Ret4		Mkt_Ret20		Mkt_Retpost
Intercept	−0.0125	0.9078	0.1523	0.5414	0.1076	0.6250	−0.0125	0.9075	0.1523	0.5393	0.1076	0.6245
Sell	−0.1679	0.1564	−0.6100	0.0259	−0.3587	0.1369	−0.1480	0.2126	−0.5531	0.0433	−0.3257	0.1784
Size	0.0013	0.7954	−0.0067	0.5657	−0.0042	0.6825	0.0013	0.7949	−0.0067	0.5637	−0.0042	0.6821
Size*Sell	0.0068	0.2214	0.0270	0.0357	0.0153	0.1783	0.0061	0.2759	0.0249	0.0523	0.0140	0.2167
Mining	0.0000	0.9998	0.0231	0.7296	0.0060	0.9192	0.0000	0.9998	0.0231	0.7283	0.0060	0.9190
Mining*Sell	0.0036	0.9092	−0.0233	0.7521	−0.0140	0.8298	0.0018	0.9541	−0.0284	0.6982	−0.0170	0.7940
IT	−0.0033	0.9247	0.2026	0.0113	0.1214	0.0845	−0.0033	0.9245	0.2026	0.0108	0.1214	0.0841
IT*Sell	0.0088	0.8193	−0.2160	0.0155	−0.0914	0.2445	0.0049	0.8997	−0.2273	0.0107	−0.0979	0.2127
Single*Sell							−0.0292	0.1025	−0.0833	0.0427	−0.0484	0.1836
N	311		311		311		311		311		311
Adj R2	0.0344		0.0922		0.0326		0.0397		0.1015		0.0351

• Coefficients and their associated two-tailed p-values are shown in the table. Mkt_Ret4 is the cumulative market-adjusted buy-and-hold return (CAR) over days −1 to +4 (inclusive); Mkt_Ret20 is the CAR over days −10 to +20; Mkt_Retpost is the CAR over days 0 to 20; Size is the natural logarithm of MVE; Buy and Sell are dummy variables indicating whether a company receives a buy or sell recommendation; Mining and IT are dummy variables indicating whether a firm is in the mining (GICS 1, 2) and IT (GICS 8, 9) industries respectively; and Single is a dummy variable indicating whether that company is the sole company referenced in an article.

Panel A contains estimates of the structural model fitted to the full sample. When the dependent variable is the market-adjusted return over the narrowest window (Mkt_Ret4), no variable has significant explanatory power although jointly they explain 3 per cent (see Adj R² of model 1). When a dummy variable (Single*Sell) is added to the equation to capture a sell recommendation when the firm was the only one mentioned in an article (model 4), the results from model 1 hold, but Single*Sell is also negative and statistically significant (two-tailed p-value 0.0748). From this analysis, and from the descriptive statistics discussed above, it appears that, when Pierpont focused his attention on a single company and recommended it be sold, its stock price fell quickly by 3 per cent more. This result is consistent with Liu et al. (1990), who found single-company recommendations have greater impact.

When the widest return window is examined (Mkt_Ret20), the regression models have greater explanatory power. Size is statistically significant when combined with sell recommendations (larger firms suffer smaller losses). When Single*Sell is excluded from the model, companies in the IT sector that received a buy recommendation from Pierpont (IT*buy) appear to have gained in price by more than firms in other industries.¹² The implication is that, when Pierpont made a positive comment about a firm within his area of expertise, the market paid particular attention. When Single*Sell is included, its coefficient is reliably negative (p-value 0.0234). The results for Mkt_Retpost (models 3 and 6) are consistent with those for Mkt_Ret20 (models 2 and 5), although the significance levels are lower.

Panel B contains estimates for the subsample of cases where the firm received a sell recommendation. Across all models, over both the narrower and wider return windows, the results are consistent. Firm size is positive and significant (p-values of 0.0008 and 0.0001 for Mkt_Ret4 and Mkt_Ret20 respectively excluding Single*Sell, and p-values of 0.0028 and 0.0005 for Mkt_Ret4 and Mkt_Ret20 respectively including it). The Single*Sell dummy has a negative coefficient (p-values of 0.1218 and 0.0360 for Mkt_Ret4 and Mkt_Ret20 respectively), implying when a firm was the sole reference in a column and Pierpont recommended it be sold, it fell significantly further. Again, as in the full sample, firms in the mining and IT sectors favoured by Pierpont did not experience significantly different returns once we control for firm size.¹³ As in Panel A, the results in Panel B for Mkt_Retpost are generally weaker than the results for Mkt_Ret20.

Panel C contains results for the subsample of firms that received a sell recommendation that was a focus reference. As in Panel B, Size is consistently positive (p-values of 0.0033 and 0.0006 for Mkt_Ret4 and Mkt_Ret20 respectively excluding Single*Sell, and p-values of 0.0044 and 0.0015 for Mkt_Ret4 and Mkt_Ret20 respectively including it). For the narrower return window (Mkt_Ret4), firms that were the sole reference in a column have a negative coefficient but it is not significant (p-value 0.7345). For the wider return window, the Single*Sell variable loads negatively and is marginally significant (p-value 0.1077). For Mkt_Retpost, the only significant variable is Size, although the coefficient on Single*Sell has the predicted sign.

Panel D contains regression estimates when holds are excluded and the sample is confined to cases of positive (buy) and negative (sell) recommendations. The models in Panel D are reduced form, in that they exclude the dummy variable Buy and its interactions. When abnormal returns are measured over the widest window (Mkt_Ret20), the coefficients of the IT sector along with the Sell dummy and its interactions with Size and IT are all significant; when Single*Sell is also added to the structural equation its coefficient is significant and negative. Once again, the results for the narrowest return window (Mkt_Ret4) are generally not statistically reliable, apart from Single*Sell, which has a negative coefficient (one-tailed p-value = 0.0512); and the results for Mkt_Retpost are consistent with but statistically weaker than the results for Mkt_Ret20.

Additional sensitivity analysis was conducted by including a dummy variable for companies that received one of Pierpont's Dubious Distinction awards and a continuous variable for the word count in each article. Further, other event windows were examined. The results were robust to these alternative specifications.

CONCLUSION

We examined share market movements around the time a company was mentioned in a Pierpont column in the Australian Financial Review. We did this by examining market-adjusted buy-and-hold returns and trading volume over thirty trading days beginning with the return on the ninth day before the column was published. While our results bear similarities to those of Foster (1979, 1987) and both calibrate and bear witness to the claim that star analysts play a non-trivial role in share price discovery, our study is differentiated first by the length of history and frequency of Pierpont's columns, and second by our use of an independent double-coding procedure that allows us to reliably identify and more finely partition Pierpont's trading recommendations.

Results confirm arguably what many investors must have believed: Pierpont was a skilled investment analyst. Coverage by Pierpont was associated with abnormal returns over these thirty days that averaged 6.4 per cent for positive coverage and −5.5 per cent for negative coverage. The results were clearer when IT and mining firms were examined separately. Over the same thirty day window, IT firms experienced a 20.86 per cent return on positive coverage, and lost 9.12 per cent on negative coverage, while the corresponding averages for mining firms were 4.99 per cent (gain) and 6.09 per cent (loss) respectively.

From multivariate analysis we found larger firms tended to lose less when they were the subject of a sell recommendation, while firms that were the sole reference in a Pierpont column tended to lose more. Firms in the IT sector that received a buy recommendation did better than others over the thirty day period.