Outperforming the Market: A Comparison of Star and Non-Star Analysts' Investment Strategies and Recommendations

The study examines whether analyst recommendations—particularly from star-ranked analysts—can produce abnormal returns and outperform non-star analysts, thereby informing market efficiency and signaling theory. The context is the NASDAQ market, with a focus on the tech-heavy NASDAQ 100, where dealer-based trading and modified cap-weighting allow diversification. The paper addresses two research questions: RQ1: Can investors use analyst recommendations as an investment strategy to earn above-average returns? RQ2: Do higher-reputation (star) analysts make more profitable recommendations than lower-reputation (non-star) analysts? The work is distinctive in using StarMine’s objective, algorithmic rankings (Top Stock Pickers and Top Earnings Estimators) rather than Institutional Investor Magazine’s survey-based rankings; it treats analysts ever designated as stars as having persistently superior skill without splitting pre- and post-star periods. Methodological novelties include testing regressor endogeneity (Durbin-Wu-Hausman), modeling investor sentiment’s role in volatility via a GARCH(1,1) framework using NASDAQ VXN as an exogenous variance regressor, and assessing the timing of sentiment-volatility interactions with frequency-domain causality. The study contributes by showing that investor sentiment significantly affects stock return volatility with short-lived effects and by validating the investment value of analyst recommendations, particularly in short portfolios.

The literature on analyst recommendations shows mixed evidence on investment value. Early work (Womack, 1996; Stickel, 1995) found significant price reactions and post-recommendation drifts (upgrades around 30 days; downgrades up to six months), with a positive link between reputation and short-term reactions. Later research often finds limited profitability after costs or short-lived reactions (Jegadeesh and Kim, 2006; Leone and Wu, 2007; Barber et al., 2010; Loh and Stulz, 2011; Su et al., 2019), though some evidence indicates persistent value for highly ranked analysts and AA status (Leone and Wu, 2007). Studies comparing star versus non-star performance (Emery and Li, 2009; Fang and Yasuda, 2014; Kucheev and Sorensson, 2019; Byun and Roland, 2020) suggest reputation correlates with better recommendations, yet concerns remain about bias and the role of brokerage resources. Ranking systems are either subjective (Institutional Investor, II) or objective (StarMine TSP/TEE, WSJ Best on the Street). Subjective II rankings can behave like popularity contests and may be biased by firm visibility and recognition, while objective systems use quantitative performance and forecast accuracy. Evidence suggests StarMine and WSJ are more informative for profitability than II (Kucheev and Sorensson, 2019). Recent findings note that star analysts’ outperformance can persist and be linked to better access to management (Barber et al., 2007), with stronger alphas for small/low-coverage stocks (Kadan et al., 2020). The present study aligns with this stream, employing CAPM, Fama-French 3-factor, and Carhart 4-factor models to compare risk-adjusted returns of star and non-star recommendations and extends literature by modeling investor sentiment’s role in volatility via GARCH and frequency-domain causality.

Data: Analyst recommendation changes and initiations (I/B/E/S Detailed Recommendations, standardized 1=Strong Buy to 5=Strong Sell) for NASDAQ 100 constituents from 2010–2020; daily holding period returns with dividends and corporate actions from Refinitiv Eikon; Fama-French daily factors (market, SMB, HML) and momentum (MOM); risk-free rate is 1-month T-bill. StarMine rankings (Top Stock Pickers, Top Earnings Estimators) are hand-collected for each year and merged to identify star status using analyst name, broker, and industry codes. Anonymous or unmatched entries are dropped. Reiterations at the same level are excluded; only changes in levels and initial coverage are kept. Stars are defined by combining TSP and TEE to expand the pool; a Star-1 subgroup denotes analysts ranked number one in a GICS industry.

Sample: 93 NASDAQ 100 firms on average; 2010–2020 yields 6587 observations (1934 initial negative and 2325 initial positive; plus level revisions). Stars issue about 2% of recommendations annually; non-stars about 98%.

Portfolio construction: Following Barber et al. (2006) and Fang and Yasuda (2014), construct calendar-time, value-weighted portfolios by analyst group. Long portfolio includes Strong Buy and Buy; Short portfolio includes Hold, Sell, Strong Sell. For each recommendation, invest $1 at close of the announcement day (or next business day) and hold 30 days; short positions are recorded analogously for negative/neutral recommendations. Positions are closed after 30 days. If multiple analysts recommend the same stock, it appears multiple times.

Return computation: Daily compounded returns for each recommendation are computed and used to form value-weighted daily portfolio returns across active recommendations on day t. This produces daily time series of portfolio returns (2010–2020).

Risk-adjusted performance: Estimate alphas via asset pricing models:

• CAPM: Excess portfolio return on market excess return.
• Fama-French 3-factor: Adds SMB and HML.
• Carhart 4-factor: Adds MOM to FF3. Average daily excess returns are scaled by 21 trading days for monthly figures. Alphas are computed for Stars, Non-Stars, and Star-1, and differences tested parametrically across groups and portfolios (Long, Short, Long–Short).

Endogeneity testing: Assess potential endogeneity of the momentum factor (MOM) using the Durbin-Wu-Hausman specification test, comparing OLS and IV-based estimators under the null that regressors are exogenous.

Investor sentiment mechanism: Model the effect of investor sentiment on volatility using a GARCH(1,1) with NASDAQ VXN (volatility index) as an exogenous regressor in the variance equation to test whether sentiment significantly raises stock return volatility.

Frequency-domain causality: Apply the Breitung and Candelon (2006) frequency-domain Granger causality test in a bivariate VAR to examine causality between sentiment and volatility across frequencies, distinguishing short-run versus long-run effects and identifying cycle lengths (T = 2π/θ).

• All analyst groups (Stars and Non-Stars) generate statistically significant abnormal returns, indicating investment value in recommendations and challenging strict EMH in this context.
• Long portfolios: Significant alphas for both stars and non-stars; no statistically significant alpha differential between stars and non-stars. Example (daily; monthly in parentheses): Stars CAPM alpha 0.0146% (0.3066%); Non-Stars 0.0151% (0.3171%); differences insignificant across CAPM, FF3, Carhart.
• Short portfolios: Stars significantly outperform Non-Stars. Stars exhibit negative alphas (profitable when shorting) of about −0.0085% to −0.0086% daily (≈0.18% monthly), while Non-Stars show positive alphas (losses for shorting) of about 0.0178% daily (≈0.3738% monthly). The parametric test indicates a daily alpha differential of −0.0265% (≈−0.5565% monthly) favoring Stars; highly significant and consistent across models.
• Long–Short portfolios: Stars’ Carhart alpha ≈0.0229% daily (≈0.4809% monthly), significant at 1%. Non-Stars’ long–short alpha ≈−0.00270% daily (≈−0.0567% monthly), marginally significant.
• Star-1 subgroup: Highest and most significant alphas across portfolios; e.g., Carhart Long ≈0.0174% daily (≈0.3717% monthly); Short ≈−0.0092% daily (≈0.1932% monthly); Long–Short ≈0.0266% daily (≈0.5586% monthly). Differences vs Non-Stars are highly significant in short portfolios; long portfolio differentials remain insignificant.
• Endogeneity test: Durbin-Wu-Hausman indicates momentum factor exogeneity (Difference in J-stats = 0.0985, p = 0.7536), supporting OLS consistency and efficiency.
• GARCH(1,1) with VXN: ARCH and GARCH terms are positive and significant (RESID(−1)^2 = 0.3816, p < 0.0001; GARCH(−1) = 0.0823, p = 0.0006), sum < 1; VXN coefficient is positive and highly significant (0.000191, p < 0.0001), indicating investor sentiment significantly increases stock return volatility.
• Frequency-domain causality: Investor sentiment influences volatility across short and longer horizons; market volatility causes changes in investor sentiment predominantly over 5–10 days, with non-permanent effects.

The results address RQ1 by demonstrating that portfolios constructed from analyst recommendations earn significant abnormal returns, indicating actionable investment value, particularly in short strategies. Regarding RQ2, star analysts do not significantly outperform non-stars in long portfolios but do so decisively in short portfolios, suggesting that reputation manifests more in profitable sell/hold calls than in buy calls. This asymmetry may reflect positivity bias in recommendations and the greater difficulty of issuing negative opinions; star analysts’ better access to information and superior processing capabilities likely enhance the effectiveness of their negative/neutral calls. The presence of significant alphas implies short-term market inefficiencies or delayed information incorporation, though the lack of differential in long portfolios is compatible with EMH-like rapid price adjustment for positive signals in a highly visible tech-heavy index. The exogeneity of momentum supports the robustness of multi-factor alpha estimates. The sentiment mechanism analysis shows that higher investor sentiment raises volatility and that volatility changes feed back into sentiment within 5–10 days, revealing a dynamic, cyclical relationship among analyst signals, market sentiment, and volatility. Together, these findings validate the usefulness of objective star rankings (StarMine) for identifying analysts whose recommendations, particularly on the short side, can enhance performance, while highlighting the role of investor sentiment in amplifying short-term volatility dynamics.

The study provides empirical support that analyst recommendations, especially those from StarMine-identified star analysts, can be used to earn abnormal returns. Stars generate superior alphas overall and significantly outperform non-stars in short portfolios, validating the effectiveness of StarMine’s objective rankings. However, in long portfolios there is no significant difference between stars and non-stars, consistent with rapid information incorporation in prices. The sentiment analysis reveals that investor sentiment significantly and positively affects stock return volatility, and market volatility induces sentiment changes within 5–10 days without permanent effects. Practically, investors can benefit from incorporating analyst recommendations—using star status as a quality signal—and should exercise due diligence, especially when considering short strategies. The findings also suggest that while star analysts possess superior skills, intensifying competition and widespread analytical tools compress advantages over time, aligning with EMH expectations for sustained outperformance. Future research could broaden datasets (e.g., NYSE/S&P constituents), extend time horizons, and further explore sentiment–volatility dynamics across sectors and news types, building on GARCH and frequency-domain causality frameworks.

The analysis relies on specific data sources and scope: StarMine rankings and I/B/E/S recommendations for NASDAQ 100 firms from 2010–2020. The limited market and firm coverage (tech-heavy, large-cap index) may constrain generalizability. Rankings were compiled from specific datasets and years, and recommendation reiterations were excluded. Broadening data sources, including newer periods, other markets and indices (e.g., NYSE, S&P), and alternative ranking systems could enhance external validity and applicability.