Why do Firms Switch Underwriters



Industries, Analysts, and Price Momentum

LESLIE BONI and Kent L. Womack*

Preliminary Version 1.02, please do not cite without permission

Comments Welcome

* Boni is at the University of New Mexico and Womack is at the Tuck School of Business at Dartmouth. We especially acknowledge the expert research assistance of Bob Burnham and helpful comments by Bhaskaran Swaminathan and seminar participants at Babson, Colorado and Penn State. We thank Keith Ferry at for data on analyst recommendations. We also appreciate data provided by I/B/E/S and Ken French. Comments and questions should be sent to boni@unm.edu or kent.womack@dartmouth.edu. An updated version of the paper will be found at and .

Industries, Analysts, and Price Momentum

Abstract

This paper examines the competition among sell-side analysts, who are predominately industry specialists, to provide investment value through their recommendations within their followed industries. Earlier “unconditional” research using event studies and consensus levels suggested that modest abnormal returns before transactions costs were available to investors using recommendation information. Our industry-based analysis shows the potential for even larger and more consistent abnormal returns. We also distinguish between analysts’ value for intra-industry stock picking and the value of their information in identifying industry momentum or sector rotation strategies.

We find that long-short industry-based portfolios using buy-sell consensus levels are ineffective at garnering significant excess returns, but monthly recommendation changes (upgrades and downgrades) within industries provide significant market neutral one-month-ahead returns that are 1.3%, or 18.9% annualized. However, we show that the value of these analyst opinion changes dissipates rapidly, over about 2 to 3 months.

At an industry level, aggregated analyst information precedes excess industry returns in the subsequent months. This suggests considerable promise for the use of analyst recommendations in explaining a substantial portion of price momentum as an industry phenomenon and in implementing sector rotation strategies.

JEL Classification: G14, G24

Keywords: Sell-Side Research; Value of Brokerage Recommendations; Market Efficiency; Price Momentum; Sector Rotation.

Brokerage research (often called “sell-side research”) departments are predominantly organized by industry. Thus, with few exceptions, the Wall Street analysts that write reports, estimate earnings, help underwrite new issues, and issue buy and sell recommendations are considered industry specialists. Not surprisingly then, analysts are regularly compared and evaluated relative to their competitors in the same industry. The Institutional Investor All-Star rankings within each industry published each October are a prime example of the reputational rewards that accrue to the most highly regarded analysts in each industry. Krigman, Shaw, and Womack (2001) show that Institutional Investor All-Star reputations of industry-specific analysts are instrumental in garnering highly profitable investment-banking business for the star analyst’s firm.

This study examines recommendation upgrades and downgrades of sell-side security analysts that are one of the most prominent sources of the information flow that investors appear to use in revising their portfolios. And, they use them with good reason. Earlier research has shown that trading volume doubles and returns increase 1-3%, on average, on days when analysts from the major U.S. brokerage firms upgrade their recommendations on individual stocks. Abnormal volumes and price decreases are even larger when stocks are downgraded (for example, from “buy” to “hold”). Barber, Lehavy, McNichols, and Trueman (2001), Stickel (1995), and Womack (1996) provide estimates of the value changes in this increased activity: returns for stocks upgraded continue to increase (after risk and market adjustments) and stocks downgraded continue to decrease for a month or more after the recommendation change. But, it should be noted that this challenge to available-information market efficiency is not unique to analysts’ recommendations. Bernard and Thomas (1989) and Gleason and Lee (2002) for example, show that the market is slow to incorporate and incorrectly interprets the information contained in earnings estimate revisions as well. And, the literature on price momentum begun by Jegadeesh and Titman (1993) possibly suggests that the market incorporates “some information” into market prices slowly, with a lag of at least a few months. “Some information” may be, in part, the opinions of sell-side analysts.

The focus of this paper is the value investors may derive by observing the competition among industry analysts to provide timely and accurate opinions on the stocks they follow. For small capitalization companies, this competition may include only one, two, or three analysts; but, for heavily-traded, large capitalization stocks such as Microsoft and General Electric, the “herd” may be as many as 30 or more analysts. Higher reputation and presumably higher pay reward the winners of this stock-picking competition. Indeed, the All-American research poll conducted by Institutional Investor has a special recognition category for “stock picking ability”.

We examine two particular questions that have relevance to investors trying to gain a risk-reward edge. First, we examine whether industry analysts are able to identify future winners and losers within their industry specializations. Since analysts typically follow a small number (usually from 5 to 30) stocks in one industry, it would be reasonable to assume that their stock picking ability, if any exists, should most likely be in ranking the stocks in their industry sector as winners and losers. Earlier research, which focused primarily on portfolios and event studies, were not well industry-diversified. Barber, Lehavy, McNichols, and Trueman (hereafter Barber, et.al., 2001) found excess returns before transactions costs by constructing unconstrained “buy” and “sell/unattractive” portfolios. Indeed, we find that such an approach, choosing portfolios based on the highest and lowest absolute consensus levels is quite industry-unbalanced and loads heavily on momentum. Stickel (1995) and Womack (1996) identified upgraded and downgrades stocks but did little industry-based analysis. Hence, earlier analyses did little to answer the question of industry “skill”.

While we do not exhaust all or even the most optimal strategies for maximizing future return, we find that a simple long-short strategy of buying upgraded firms and selling downgraded firms within each industry in any calendar month yields a substantial 1.3% per month in the next calendar month and cumulatively 2.4% over 3 months. These returns are well diversified, computed across all industry-months, and have no large common risk factor loadings. This is in contrast to earlier portfolio-based attempts that have significant risk factor loadings and industry imbalances.

The second investment question that we address is whether observing the collective recommendation changes or consensus levels by all analysts within an industry provides value in capturing sector rotation or industry momentum. As mentioned earlier, one of the largest challenges to market efficiency is the phenomenon of price momentum. Moskowitz and Grinblatt (1999) argue that much of the observed price momentum previously documented by Jegadeesh and Titman (1993) and others is industry based. Our data provide an indication of the momentum (correlation) in industry information that may be driving industry price momentum. We find that conditioning long-short portfolios on the basis of “hot” industries with recommendation momentum (whether the industry was net upgraded or downgraded in the formation month) earns 1.56% per month, or about 3% more annually than the “all industry” case in the previous paragraph. Our findings that future intra-industry and industry sector returns are predictable and significant when using the information content in brokerage upgrades and downgrades suggests a possible informational explanation for price momentum that has not been previously explored.

Our examination, of course, does not exhaust all the potential strategies that investors might employ using recommendation information, and it does not provide evidence that these returns are fully realizable after transactions costs. However, the magnitude of our returns, relative to earlier price momentum and post-earnings announcement drift studies, shows considerable promise of the potential for predicting future returns from analysts’ pronouncements. We conjecture that recommendations are a proxy for the primary source of information Jegadeesh and Titman (1993) claim as a fundamental explanation for price momentum.

The rest of the paper proceeds as follows. Section 1 provides the institutional background on analysts and their recommendations and Section 2 describes the data and sample selection methods used in the paper. Section 3 examines the question of analysts’ intra-industry stock-picking ability, and Section 4 analyzes analysts’ collective pronouncements as a mechanism for predicting industry/sector rotation. Section 5 discusses the implications of our findings and concludes.

1. Analyst Recommendation Activity: The Institutional Setting and Prior Literature

Recommendations are only one of the informational products that analysts provide for investors. Others include building pro forma valuation models, forecasting future earnings, cash flows and price targets. The analyst’s specific information snooping and dissemination tasks can be categorized as 1) gathering new information on the industry or individual stock from customers, suppliers, and firm managers; 2) analyzing these data and forming earnings estimates and recommendations; and 3) presenting recommendations and financial models to buy-side customers in presentations and written reports.

The analyst’s dissemination of information to investment customers is usually disseminated through a morning research conference call. Such conference calls are held at most brokerage firms about two hours before the stock market opens for trading in New York. Analysts and portfolio strategists speak about, interpret, and possibly change opinions on firms or sectors they follow. Both institutional and retail salespeople at the brokerage firm listen to this call, take notes, and ask questions. Alternatively, urgent communications may be made following a surprising quarterly earnings announcement or some type of other corporate announcement while the market is open for trading. In both cases, analysts notify the salespeople at the brokerage firm, who in turn call customers who they believe might care (and potentially transact) on the basis of the change. Once the sales force is notified, the analyst may directly call, fax, or send e-mail to the firm’s largest customers if the analyst knows of their interest in the particular stock or sector. The information is sometimes retransmitted via the Dow Jones News Service, Reuters, CNN, or other news sources, especially when the price response in the market is significant.

The type of announcement analyzed extensively here is a change of opinion rating level by an analyst on a stock. New “buy” recommendations are usually scrutinized by a research oversight committee or the legal department of the brokerage firm before release. Thus, a new added-to-buy recommendation may have been in the planning stage for several days or weeks before an announcement. Sudden changes in recommendations (especially, removals of “buy” recommendations) may occur in response to new and significant information by or about the company.

Several earlier academic studies form our priors as to how investors and the collective market respond to recommendation changes. Womack (1996) identified significant price drift in short time periods following recommendation upgrades and downgrades by the largest US brokerage firms in the 1989-1991 time period. For upgrades to “buy” or “strong buy”, he found using an event-study methodology that the market reacted strongly to the news announcement of the upgrade and continued to drift in the predicted upward direction for another one to two months. For downgrades from a buy category and for outright sell recommendations, he documented the negative returns with somewhat higher intensity: larger immediate reactions and larger amounts of drift over somewhat longer periods, three to six months. Two issues related to market efficiency were addressed that we examine more carefully here. First, adjustments for standard Fama-French and momentum risk factors were used and did not significantly affect the magnitudes of the short-term (1 to 3 month) returns. Second, in Figure 1, Womack demonstrates the lesser reactions of larger companies relative to smaller ones. Price reactions and drift were almost twice as large in the 5 smallest market capitalization deciles as in the top two.

Barber, Lehavy, McNichols, and Trueman (2001), take an importantly different perspective. Using a different dataset (Zacks) of recommendation information, they address the issue of whether forming portfolios through calendar time can capture the returns suggested by event study results in Womack (1996) and Stickel (1995). Two features distinguish the Barber et.al. results. First, they form portfolios by choosing stocks below and above two time-invariant cutoff levels (to identify attractive and unattractive stocks). To form long portfolios, they choose all stocks with analyst consensus level ratings of 1.5 or less. For the short portfolio stocks, they include stocks with ratings higher than 3.0. Second, they rebalance these portfolios daily, as stocks move into and out of these cutoff ranges. Their results suggest significant long-short returns before transactions cost (slightly less than 1% per month) for the 1986 to 1996 period, but implied transactions costs eat up these “profits”.

Jegadeesh and Titman (1993) and others document price momentum over periods of 3 months to one year and suggest that the evidence is consistent with delayed reactions to firm specific information. Our paper offers causal confirmation, showing that one type of or proxy for firm-specific information is analysts’ valuation changes.

Hong, Lim, and Stein (2000) link the phenomenon of price momentum with analyst following. This gives support to the hypothesis of Hong and Stein (1999) that momentum is a symptom of investors’ collective underreaction to individual pieces of private information. They use analysts as a proxy for intensity of information dissemination, and, after controlling for size, find that price momentum is greater where analyst coverage is lower.

2. Data and Sample Selection

The data used in this study were originally developed by , a now defunct Internet startup company that had planned to develop a product for institutional and individual investors to rank analysts and their recommendations. Validea’s choice of recommendations data (after searching for the most timely and accurate source with historical data back to at least 1995) was IBES, which constructs two recommendations databases used here. The first is a monthly Summary History-Recommendation file that compiles a monthly snapshot of each company followed by sell-side analysts whose brokerage firms provides data to IBES. This database tracks at mid-calendar month (similar to the Summary History-EPS file) the number of analysts following the stock, the average consensus rating level on a 1 to 5 scale (where 1 is a “strong buy” and 5 is a “sell”) and its standard deviation for the stock, and the number of analysts upgrading and downgrading their opinion level in the month. The Detail History-Recommendation file provides a database entry for each recommendation change made by each analyst. Important variables include the date of the change, the analyst and the brokerage firm’s name, to what level the change was made. Table 1 shows the brokerage firm level characteristics of the 150,873 recommendation changes we analyze. There were 7,960 companies followed during at least one month of the January 1996 to June 2001 time frame that we examine.

[Table 1 about here.]

McNichols and O’Brien (1998) discuss the determinants of analyst coverage, specifically the question of what would motivate an analyst to initiate coverage of a stock. While there are of course several answers, one of the most obvious is the potential for generating trading commissions and investment banking fees from the followed company. Not surprisingly then, we find and document in Table 2 that there is a significant correlation between the number of analysts following a company and the market capitalization of that company. For example, we find in the tenth (largest) CRSP size decile the median number of analysts following a stock is 13. In the 6th decile, by contrast, the median number of analysts following is 3.[1]

[Table 2 about here.]

In addition to the data supplied by IBES, Validea merged all companies with the Morgan Stanley GICS Industry codes after their research concluding that Morgan Stanley’s industry-coding scheme was most closely aligned to the natural industry distinctions used on the sell-side. This divides the 7,960 companies in our dataset into 120 distinct industry groups. The average number of companies in each industry is 66 but the median is 32. “Banks” is the largest (and unusual) industry comprising 791 companies, which skews the industry distribution. There are 6 industries with fewer than 10 companies.

To provide an indication of variation within and across sectors, Table 3 shows descriptive data for all industries along with three representative industries (“Systems Software”, “Restaurants”, and “Electric Utilities”) and one company in each of the industries. The average consensus level for all industries is close to a “Buy” (IBES rating level = 2) each month (with little variation). As shown in Figure 1, the average consensus level for these industries varies from 2.0 by less than + 0.5. As shown by the spreads between the consensus levels each month for the “best” company (lowest consensus level) and “worst” company (highest consensus level), analysts do appear to be willing to differentiate among companies within the industry they follow.[2] Figure 2 provides an example of correlation between price movement and consensus level both at the industry and company level for the Systems Software industry and Microsoft.

[Table 3, Figure 1, and Figure 2 about here.]

The key informational output collected by IBES and examined in this paper are recommendation changes from one rating level to another. In the 1996 to 2001 timeframe of our data, there are 83,634 recommendation changes. In addition to changes, where both an earlier level and the new current level are identified, we also observe 67,239 “first recommendations”, which are recommendations where we obtain the new current level, but do not know the previous rating level, either because it did not exist (if an analyst initiates coverage) or because the company could not be matched to an earlier rating in the database. Note that known reiterations (common in the communications of analysts to their clients) of prior ratings levels are not recorded or examined, unlike in the Zacks data examined by Barber, et.al. The total number of recommendation changes examined is therefore 150,873 over five and one-half years, as shown in Table 4, Panel A.

Although buy and strong buy recommendations grossly outnumber hold and sell recommendations, interestingly Table 4 shows that downgrades (55.5% of recommendation changes) actually exceed upgrades (44.5% of recommendation changes). This suggests a broader universe of stocks is available for portfolio strategies based on going long upgrades / short downgrades than for strategies based on consensus level cutoffs of going long buy while shorting underperform and sell categories.

[Table 4 about here.]

It is instructive to note the magnitude of the three-day market-adjusted returns for the various change categories shown in Table 4, Panel B. When a stock is moved to the strong buy (IBES Code “1”) category from “buy” or “hold”, the average market response is a size-decile-adjusted return of 3%. This is significantly higher than the 1.06% to 1.48% range reported by Barber, et.al. for the 1985 to 1996 time period in the Zacks data. We offer two possible explanations for the difference. First, the Zacks database, while including most of the significant US brokerage firms, omitted a few large ones. Further, Zacks collected the data second-hand, so that often the dates of the recommendation changes were the dates of written reports that may have been several days later than the actual recommendation event. Hence, the averages reported in Barber, et.al. may have included some “event” returns after the actual event dates. It is important to note that this issue does not likely affect the important Barber et.al. portfolio strategy results reported later in that paper.

Second, there is some evidence that market responses to new information may have been more intense in the more recent time frame analyzed here. In some ways, the late 1990s was recognized as the advent of the “day trader” due to convenient and inexpensive transactions services available via the Internet. For example, Busse and Green (2002) show that traders respond to televised analysts’ recommendations within a minute of their broadcast in the year 2000.

3. Analysts’ Industry Stock-Picking Expertise

Event-study returns in Table 4 show that the market responds in the direction that security analysts predict. Upgraded stocks go up and downgraded stocks go down in the three-day period centered on the recommendation change date in the IBES database. More importantly, earlier work by Womack (1996) and Stickel (1995) show that highlighted stocks continue to drift in the direction predicted by the analyst for 1 to 6 months. These analyses were summaries of all recommendation changes that may have been heavily weighted towards some well-followed industries. Research by Barber et al (2001, 2002) also suggest profitable strategies may be derived using consensus level criteria. Therefore, a fundamental question on analysts’ expertise remains: Can industry analysts identify future winners and losers within their industry specializations?

To analyze the intra-industry industry stock picking ability of analysts, we examine four calendar-month long-short portfolio strategies. Strategies A and B define portfolios based on consensus level criteria at month end, while Strategies C and D use recommendation changes within the “formation” month.

In Strategy A, two stocks (the “best” and the “worst”) in each of the 120 industries are chosen for a long-short portfolio based on the two extreme (most favorable and least favorable) average consensus rating levels for each stock at month end. If there are ties in the highest and lowest level in each industry, the portfolio is formed by equally weighting the returns of the two or more stocks involved.

We compare this best-worst industry consensus level strategy to a consensus level strategy (which we refer to as Strategy B) approximating the Barber, Lehavy, McNichols, and Trueman (2001) strategy (which does not diversify across industries) for the 1996-2001 time frame of our data. In their strategy, they choose for the long side of the portfolio all stocks with a consensus rating of 1.5 or lower (predominantly “strong buy” stocks) and then short all stocks with a level higher than 3.0 (“unattractive” or “sell” stocks). While they rebalance their portfolio daily, our version of their strategy forms the portfolios using their cutoffs at the end of the month.

To examine portfolio strategies based on recommendation changes, we define an aggregate change measure for stocks and also for industry sectors. For each stock each month, we define an aggregate recommendation change measure (“AgChange”), which is the sum of the number of analyst recommendations which are upgrades plus the number of first recommendations that fall into the IBES buy or strong buy category less the number of downgrades and the first recommendations that are to IBES category hold, underperforform, or sell. Using this simple measure, we can refer to stocks with AgChange greater (less) than zero as “net upgraded” (“net downgraded”) stocks each month.

For Strategy C, we construct industry-by-industry portfolios that are long (short) the “best” (“worst”) stock based on most positive (negative) value for AgChange measure.[3] Strategy D is a more evenly diversified version of Strategy C. In Strategy D, we construct industry-by-industry portfolios that are long all net upgraded stocks and short net downgraded stocks during the month based on the AgChange measure.

Table 5 shows the results of forming industry-based calendar-time portfolios for each strategy across all 66 months for all industries in our sample. In Panel A, we focus on returns to a long-short portfolio when stocks across all market capitalizations are available to investors. Panels B, C, and D provide constrained results, showing the effects of liquidity constraints proxied for by limiting portfolio inclusion to high market cap and analyst coverage.

3.1 Portfolio Strategies based on Consensus Rating Levels

Table 5 shows that the mean return from using the consensus level Strategy A is 0.59% in the calendar month following portfolio formation. However, after controlling for the three risk factors identified by Fama and French (1993), the excess return is 0.41%, and when momentum is also factored out, the return is only 0.11% per month, not economically or statistically different from zero. An initial conclusion is that future returns are not very predictable by conditioning on extreme rating levels.

[Table 5 about here.]

Table 5 documents that Strategy B, which is also industry unconstrained, is not well-diversified. The mean return for the post-formation month is higher than in Strategy A, at 0.92% but, after adjusting for the three Fama-French factors, becomes 0.76% and after adjusting for momentum, becomes 0.05%, insignificantly different from zero. Barber et.al. report significant monthly excess returns (intercepts) of 0.99% for Fama French and 0.75% for the four-factor model in the earlier 1986 to 1996 time frame. One conclusion that might be drawn from these differences in our time frame versus that of Barber et.al. (1986-1995) is the more significant influence of momentum in the late 1990s, a topic that we address more fully in Section 4.

3.2 Portfolio Strategies based on Individual Rating Changes within Industries

Note that Strategy C reduces the number of industry-month observations by about 22% relative to consensus level Strategy A (from 7,716 to 6,027) because in 1,689 industry-months, the industry sector does not have both a net upgraded and net downgraded stock. We document that this strategy, based on recommendation changes in the calendar formation month, provides substantially higher returns in the first post-formation month. The market neutral portfolio return is 1.38% per month (18.9% annualized) before transactions costs. Controlling for Fama French and momentum factors through calendar time regressions, the intercept remains 1.12% per month, highly significant both economically and statistically.

The numbers for Strategy D are quite similar to those in Strategy C, the raw return for the long-short portfolio being 1.30% per month and 1.08% after controlling for the Fama French factors and momentum. Thus, we conclude that the significant returns subsequent to the recommendation change information in a month are not caused by a few outliers each month, but are quite robust to broader portfolios incorporating all net upgrade and net downgrade information.

The results of the comparison between Strategies A and B, based on consensus levels, and Strategies C and D, based on recommendation changes, are quite striking and highlight the dissipating nature of the value of analysts’ information. Consensus levels usually do not change considerably from month-to-month. Knowledge that analysts are currently rating a stock a “strong buy” does not suggest measurable outperformance in the future, unless the rating has been recently upgraded to “strong buy”.[4]

3.3 The Value of Brokerage Information and the Limits of Market Efficiency

Table 5, Panel A allows analysts to choose the most and least attractive stocks in each industry, unconstrained by liquidity and market capitalization considerations. Hence, the documented strategy is likely to be unrealizable by large institutional investors, to the extent that small firms’ returns contribute significantly to the results. Thus, in Panels B, C, and D, we constrain the allowable stocks available for portfolio formation in three ways.

In Panel B, we eliminate from consideration all stocks in the bottom half of the CRSP size deciles. Therefore, the remaining stocks (deciles 6 through 10) have market capitalizations of approximately $111 million and above in year-end 2000 dollars. Consistent with Panel A, strategies A and B, based on levels, do not provide much look-ahead value, but strategies C and D, based on recommendation upgrades and downgrades in the formation month, are valuable but are of smaller magnitude than the unconstrained case in Panel A. The intercepts for the long-short portfolios, adjusted for Fama and French factors and momentum, are 0.88% and 0.79% respectively and have T-statistics above 3.5.

Similarly, in Panel C we report portfolios that are only populated by the largest CRSP decile (10). This panel represents the ability of analysts to identify mispricing in the largest and most liquid stocks. Again, we find that the returns on the strategies based on consensus ratings levels are insignificantly different from zero. In strategies C and D (based on ratings changes), we find that post-one-month returns are still positive but marginally significant. Panel D shows a 0.45% per month return for Upgraded minus Downgraded stocks.

In Panel D, we show a different slice, based on the stocks most followed by sell-side analysts. Panel D limits portfolio choice to stocks with at least 15 analysts following them. Table 2 shows that this constraint corresponds to the largest 9% of stocks in the IBES universe. The returns for the changes-based strategies are reported as 0.28% and 0.43%, and are not significant at conventional levels.

Therefore, the exercise documented in Panels C and D begins to show the limits of analysts’ value and the power of competition in the largest, most-followed stocks to eliminate ostensible informational inefficiencies. Naturally, we emphasize that these strategies are not exhaustive and are conservative to one degree in that they wait, on average, a half-month to exploit apparently valuable and dissipating information. But, we also point out that the returns documented are before transactions costs.

3.4 Robustness Checks on Analyst Information Portfolios

The previous section begins to show the limits of analysts’ value and the efficiency of pricing in the most highly followed stocks. On the other hand, Table 6 demonstrates the robustness of the documented long short (Upgraded minus Downgraded) portfolio strategy in two ways. First, we observe the decomposition of the long-short portfolio into the long (Upgraded) component and the short (Downgraded) component. Not surprisingly, the calendar month returns in the portfolio formation month are very substantial, 2.97% for Upgraded stocks (in excess of an equal-weighted benchmark) and minus 4.05% for Downgraded stocks, consistent with the 3-day event returns documented in Table 4. In the first month after the recommendation information, the returns to the Upgraded portfolio are 0.64% while the returns to the Downgraded Portfolio are –0.66%. Thus, the net portfolio return of 1.30%, reported in Table 5, Panel A. If we extend the analysis to the three months post formation, the cumulative return is 2.39%, and to 6 months, 2.88%, both significantly different from zero and economically significant. Hence, we observe the majority of the market’s response to analyst information in the 1st three months after recommendation changes. For the Downgraded portfolios, we observe the continuation of underperformance even further out than the post-3 month window, consistent with the asymmetry of response to positive and negative news documented in Womack (1996). Equally interesting (and comforting) are the results of all stocks that are not upgraded or downgraded in the portfolio formation month (the “No Change” portfolio). The market-adjusted returns are not significantly different from zero in all post-event months.

[Table 6 about here.]

In Table 6, Panel B, we show the consistency of the results year-by-year. The post-one-month overall average of 1.30% ranges from 1.07% to 1.66% in the 6 calendar-year periods. Likewise in the cumulative post-event three-month period, the results are consistent in 5 out of six periods with the year 2000 being the only year with results less than 2.2% (+0.98% for year 2000).

The results industry-by-industry are remarkably consistent as well. Post-event one-month and three-month post-formation-month returns are positive for 94 and 93 respectively of the 117 industries for which long-short portfolios can be constructed.[5] Our conclusion from the results in Section 3 is that analysts appear able to distinguish temporarily under- versus over-valued stocks within their followed industries, at least in the short term ranging from 1-3 months.

4. The Value of Analyst Industry Information for Predicting Sector Rotation

The second investment question that we examine is whether observing the collective recommendation changes or consensus levels by all analysts within an industry provides any value in explaining sector rotation or industry momentum. Moskowitz and Grinblatt (1999) demonstrate that the puzzling anomaly of individual stock momentum is substantially driven by relative industry momentum. In this section, we examine whether industry momentum may have its roots in delayed response to analyst information.

In the earlier section, we used the Aggregate Change Measure for each stock to form portfolios of net Upgraded and Downgraded stocks within each industry (each industry-month observation is a within-industry portfolio). In this section, we aggregate (sum) the Aggregate Change Measure for all stocks in each industry each month to calculate portfolios returns of Upgraded industries minus Downgrades industries in each month. Industries are differentiated by whether the sum of the Aggregate Change Measure for all stocks in the industry was positive (hence, labeled an In-Favor industry) or negative (and called an Out-of-Favor industry). Then, these strategies calculate an equal-weighted return for long positions in In-Favor industries, financed by short positions in Out-of-Favor industries in each month. In essence, do industries where there are, in aggregate, more recommendation upgrades than downgrades outperform industries where downgrades dominate upgrades over the next month?

In Table 7, we document two distinct versions of these portfolios. Having shown in Section 3 that analysts are ostensibly good month-ahead forecasters within their own industries (Table 5, Strategies C and D), the next logical question is whether investors selecting the subset of In-favor industries and avoiding Out-of-Favor industries can earn even higher net returns. We label this as Strategy E in Table 7. The mean return increases from 1.30% to 1.59% in the post-formation month. This implies that a one-month-ahead strategy using aggregate analyst industry recommendation information adds about 3.5% annually to an already large naïve all-industry within-industry strategy return.

[Table 7 about here.]

In Strategy F, we record the self-financing portfolio return when we equal weight all companies within In-Favor industries on the long side and equal weight all companies in Out-of-Favor industries on the short side. Note that this strategy does not take advantage of the within-industry expertise documented earlier (since stocks are not differentiated within each industry). In sum, it is one simple measure of one-month-ahead industry momentum attributable to analyst information. The mean return is 0.43% in the post-formation month.

In Table 8, we take a different tack at understanding industry momentum and analysts’ contribution to it. Table 7 suggested that (at least) month-ahead returns are predictable from the aggregate sum of analysts’ upgrades and downgrades. Earlier work has suggested that price momentum is most robust at intervals of about 6 months. We now examine the determinants of industry returns through a series of regressions.

[Table 8 about here.]

We take the industry excess return for 1 month (alternatively, 6 months) as the dependent variable to be explained. We define industry excess return as the equal weighted return on an industry sector minus the equal weighted return on the market portfolio in the calendar month. Each observation is an industry-month. Independent variables that we use to explain the industry excess return one month (and six months) ahead are:

1) AgChangeCurrent, the sum of the AgChange measures (as defined in Section 3) for all companies in the industry in the formation month divided by the sum of the analyst-coverage (i.e., number of analysts) for the companies in the sector less the mean of this variable for all the industry sectors for that month,

2) AgChangePrior, same as 1) above but for the month prior to formation,

3) ConsensusLevelCurrent, the consensus level of the industry at the end of the formation month minus the average consensus level for the all the stocks in our data for that month,

4) IndustryExcessReturnCurrent, the equal-weighted return of the industry minus the return of an equally weighted average of all the companies in our dataset that have CRSP returns for the formation month,

5) IndustryExcessReturnPrior, the equal-weighted return of the industry minus the return of an equally weighted average of all the companies in our dataset that have CRSP returns for the month prior to the formation month,

6) IndustryAverageSizeDecile, the average CRSP decile for stocks in the industry for the month minus the mean decile of all stocks in our data that month, and

7) IndustryAverageAnalysts, the number of analysts following stocks in the industry in the month minus the mean analyst coverage of all stocks in our data that month.

If aggregate industry analyst information has value in predicting future returns, the coefficient for 1) and possibly 2) will be positive. If a favorable current average rating level for the industry is predictive, the coefficient on 3) will be negative (since a higher rating number is a less favorable rating. Variables 4) through 6) are used as control variables for industry price momentum and market capitalization, and variable 7) shows the impact of above average analyst following in an industry, controlling for the other variables.

The results in Table 8 show that the industry recommendation AgChangeCurrent measure is quite significant in explaining the one and six-month ahead returns. For the longer period only, the AgChangePrior month is also significant. The ConsensusLevelCurrent measure also loads positively, but only for the longer period, meaning that returns are higher when the industry consensus level was less favorable (a higher rating number) controlling for the other factors. IndustryExcessReturnCurrent and IndustryExcessReturnPrior are also significant and positive, suggesting substantial industry price momentum, even at short horizons of one to three months. Finally, industries with lower average market capitalization have higher returns and, interestingly, industries with higher analyst coverage, all else equal, have higher returns.

5. Interpretation and Conclusions

We document in this study that analyst information, processed at the industry level, is quite useful in explaining future returns. First, our evidence suggests that analysts are indeed able to distinguish between future outperforming and underperforming stocks within their followed industry, since industry portfolios formed in a variety of ways earn consistent returns that are significantly higher than comparable price momentum strategies. Using information obtained over the course of one calendar month, long-short industry portfolios earn an average of 1.3% in the next month. This return is a conservative estimate of the actual effect, since the value of analyst information dissipates rapidly and our calculations approximate a half-month delay in forming portfolios after the information is available.

Our second contribution is to show that aggregated analyst information within an industry is valuable for explaining relatively under- and out-performing industries for at least the next six months. This suggests a direct link and partial explanation for the puzzle of price momentum. Our evidence is consistent with the claim of Moskowitz and Grinblatt (1999) that price momentum is significantly an industry phenomenon. In fact, we may be able to show, with future work, that analyst information is an appropriate and accurate proxy for the firm-specific information that Jegadeesh and Titman (1993) claim investors react to with a delay.

Bibliography

Barber, Brad, Reuven Lehavy, Maureen McNichols, and Brett Trueman. 2001. Can Investors Profit from the Prophets? Security Analyst Recommendations and Stock Returns. The Journal of Finance LVI (No 2):531-563.

Barber, Brad, Reuven Lehavy, Maureen McNichols, and Brett Trueman. 2002, Prophets And Losses: Reassessing the Returns To Analysts' Stock Recommendations, forthcoming, Financial Analysts Journal

Bernard, Victor, and J. Thomas, 1989, Post-earnings Announcement Drift: Delayed price response on risk premium? Journal of Accounting Research, 27, 1-48.

Busse, Jeffrey A., and T. Clifton Green, 2002, Market efficiency in real time, Journal of Financial Economics, 65: 415-437.

Fama, Eugene, and Kenneth French, 1993, Common Risk Factors in the Returns on Stocks and Bonds, Journal of Financial Economics, 33, 3-56.

Gleason, Cristi, and Charles M.C. Lee, 2002, Analyst Forecast Revisions and Market Price Discovery, working paper, Cornell University.

Grundy, Bruce D., and J. Spencer Martin, 2001, Understanding the Nature of the Risks and The Source of the Rewards to Momentum Investing, Review of Financial Studies 14, 29-79.

Hong, Harrison, and Jeremy C. Stein, 1999, A Unified Theory of Underreaction, Momentum Trading, and Overreaction in Asset Markets, Journal of Finance 54, 2143-2184.

Hong, Harrison, Terence Lim, and Jeremy C. Stein, 2000, Bad News Travels Slowly: Size, Analyst Coverage, and the Profitability of Momentum Strategies, Journal of Finance, LV, 265-295.

Jegadeesh, Narasimhan, and Sheridan Titman, 1993, Returns to buying winners and selling losers: Implications for stock market efficiency, Journal of Finance 48, 65-91.

Jegadeesh, Narasimhan, Joonghyuk Kim, Susan D. Krische, and Charles M.C. Lee, 2002, Analyzing the Analysts: When Do Recommendations Add Value?, working paper, Cornell University.

Krigman, Laurie, Wayne Shaw, and Kent L. Womack. 1999. Why Do Firms Switch Underwriters?, Journal of Financial Economics 60: 245-284.

McNichols, Maureen F., and Patricia C. O’Brien, 1998, Self-Selection and Analyst Coverage, Journal of Accounting Research 35, 167-199.

Moskowitz, Tobias J., and Mark Grinblatt, 1999, Do Industries Explain Momentum?, Journal of Finance 54, 1249-1290.

Stickel, Scott E. 1995. The Anatomy of the Performance of Buy and Sell Recommendations. Financial Analysts Journal Sept-Oct 1995:25-39.

Womack, Kent L. 1996. Do Brokerage Analysts' Recommendations Have Investment Value? Journal of Finance L1 (no.1):137-167.

-----------------------

[1] Analyst coverage is defined as the number of analysts that provide earnings forecasts for a stock each month per the IBES Summary History-EPS file.

[2] We update the IBES summary month consensus level numbers, which are usually mid-month based, to calendar-end basis using the daily recommendation data.

[3] When there are ties among stocks in the maximum number of upgrades or downgrades in a month, which is quite often, we equal weight all qualifying companies’ returns.

[4] By forming portfolios at month-end rather than immediately after recommendation changes obviously biases (lowers) our reported results since recommendation changes occur approximately uniformly distributed within a calendar month. Hence our approach is conservative, in that it measures a first post-event return approximately a half-month after the information event.

[5] For 3 of the 120 industries, in no month are there both a net-upgraded stock and a net-downgraded stock.

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download