دانشگاه اصفهان مدیریت دارایی و تامین مالی 2383-1189 14 3 2026 09 23 The Term Structure of Investor Sentiment and Stock Return ساختار زمانی احساسات سرمایه‌گذار و بازده سهام 99 124 29940 10.22108/amf.2025.145197.1987 FA مریم دولو گروه مدیریت مالی و بیمه، دانشکده مدیریت و حسابداری، دانشگاه شهید بهشتی، تهران، ایران 0000-0003-3321-1165 محمد رضا فقیهی حبیب آبادی گروه آمار، دانشکده علوم ریاضی، دانشگاه شهید بهشتی، تهران، ایران نازنین کوکانی گروه مدیریت مالی و بیمه، دانشکده مدیریت و حسابداری، دانشگاه شهید بهشتی، تهران، ایران Journal Article 2025 05 16 This study investigates the impact of investor sentiment on stock returns across short-term daily, weekly, and monthly horizons—namely the sentiment term structure. While prior literature establishes a link between sentiment and stock price volatility, empirical evidence on the role of the time horizon remains limited and conflicting. We test the effect of investor sentiment on excess stock returns at these three frequencies using panel regression, controlling for market excess returns, size, and value factors. Investor sentiment is measured through four indirect proxies, synthesized into composite indices using both Principal Component Analysis (PCA) and the Kalman filter. The results indicate that investor sentiment exerts a significant yet temporally decaying influence on stock returns; the magnitude of the effect diminishes from the daily to the weekly and monthly horizons, revealing a downward-sloping term structure. This finding is robust, as it holds for both the PCA- and Kalman filter-based sentiment indices. To our knowledge, this is the first study to systematically examine the differential impact of investor sentiment across these high-frequency horizons and the first to empirically validate its temporal structure using both of these methodological approaches.<br /><strong>Key words:</strong><strong> </strong>Kalman Filter, Principal Component Analysis, Stock Return, Term Structure of Investors Sentiment<br /><strong>JEL Classification:</strong> C22, C38, G12, G41<br /><strong>Introduction</strong><br />Classical finance theories are predicated on the assumption of rational investors, asserting that stock prices are determined by fundamental factors while largely neglecting the influence of investor sentiment. However, this paradigm is challenged by historical market anomalies such as the 1929 Great Depression, Black Monday in 1987, and the dot-com bubble of the 1990s. Corroborating these events, a growing body of empirical research confirms that investor sentiment significantly contributes to stock price volatility (Baker & Wurgler, 2007; Kim & Ha, 2010; Kumar & Lee, 2006; Frazzini & Lamont, 2008; Antoniou et al., 2013).<br />Despite this established connection, the majority of empirical studies examine sentiment effects within a single, static time horizon. A critical gap exists, as emerging evidence suggests that the impact of sentiment is not uniform but varies across different timeframes (Li, 2020; Kim & Ryu, 2021). The underlying rationale is that as new information emerges over weekly or monthly periods, initial emotional reactions among investors are corrected, and stock prices tend to revert toward their intrinsic values, highlighting the time-dependent nature of sentiment effects.<br />This study, therefore, investigates the impact of investor sentiment on stock returns across daily, weekly, and monthly horizons—a relationship we term the temporal structure of sentiment. Elucidating this dynamic is crucial for informing sound investment decisions, effective policy-making, and robust risk management practices.<br />To accurately measure this latent construct, we rely on indirect proxies derived from financial and economic variables. Recognizing that any single variable captures both sentiment and unrelated noise, we synthesize multiple indicators to construct a more efficient and robust measure of unobservable investor sentiment (Baker & Wurgler, 2006). Specifically, this study employs two distinct methodologies to create composite sentiment indices: Principal Component Analysis (PCA), which extracts the common variation from a set of proxies (Berger & Turtle, 2011; Huang et al., 2014; Kamath et al., 2024), and the Kalman filter, a state-space technique designed to process all available information from the variables while optimally minimizing noise and estimation errors.<br /> <br /><strong>Materials and Methods</strong><br />To test the first hypothesis concerning the impact of investor sentiment on excess stock returns across daily, weekly, and monthly horizons, a panel regression was employed, following Li (2020). Equations (1) through (3) were specified for this purpose and were estimated separately for the sentiment indices derived from Principal Component Analysis (PCA) and the Kalman filter.<br /> <br /><br /><br /><br /><br />(1)<br /><br /><br />                                                                        <br /><br /><br /><br /><br />(2)<br /><br /><br /><em> </em><br /><em> </em>                     <em> </em>   <br /><br /><br /><br /><br />(3)<br /><br /><br />   <br /> <br /><br /><br /><br /><br /> <br />where  is investor sentiment change, is the size factor,<em> </em>  is the book-to-market factor,  is risk-free rate, and is market return at time t and i refers to i^th firm stock.<br />To test the second hypothesis—that the influence of investor sentiment diminishes over longer time horizons—the sentiment coefficients (β) estimated from the panel regressions were annualized using Equation (4). These annualized coefficients were then systematically compared across the daily, weekly, and monthly frequencies.<br /><br /><br /><br /><br />(4)<br /><br /><br /> <br /><br /><br /><br /><br /> <br />Where denotes the annual sentiment coefficient,  represents the sentiment coefficient corresponding to each frequency, and the constant 250 indicates the assumed number of trading days in a year.<br />Stock return was calculated as the natural logarithm of the difference in adjusted closing prices, which account for dividends and stock splits. The market risk premium was defined as the difference between the market return and the risk-free rate, the latter proxied by Central Bank bond yields. The market return was computed as the log return of the Tehran Stock Exchange (TSE) index. The size (SMB) and the value factor (HML) factors were constructed following the methodology of Fama and French (1992). The composite sentiment index was constructed for daily, weekly, and monthly frequencies using four variables and two methods of PCA and Kalman filter. PCA extracts common components assumed to capture investor sentiment, while the Kalman filter processes all observed information to estimate sentiment while minimizing noise (Brown & Cliff, 2004; Li, 2020). Sentiment proxies are measured as below: a) Adjusted turnover rate – reflects changes in stock liquidity due to investor sentiment (Baker & Stein, 2004).<br /> <br /><br /><br /><br /><br />(5)<br /><br /><br /> <br /> <br /><br /><br /><br /><br />In which, R_it​ represents the return of stock <em>i</em> at time <em>t</em>, VOL_it​ denotes the trading volume (measured in number of shares) for firm <em>i</em> at time <em>t</em>, and Shares Outstanding refers to the total shares outstanding for firm <em>i</em> at time <em>t</em>. b) Buy–sell imbalance – captures net retail demand for a stock at a given time.<br /><br /><br /><br /><br />(6)<br /><br /><br /> <br /><br /><br /><br /><br /> <br />In which  denotes the purchase volume of stock i on day j during period t, and  represents the selling volume of stock i on day j during period t. c) Trading volume – higher traded value indicates elevated investor sentiment (Li & Yang, 2017).<br /> <br /><br /><br /><br /><br />(7)<br /><br /><br /> <br /><br /><br /><br /><br /> <br />In which represents the total number of shares of stock i traded during period t, and denotes the closing price of stock i at time t. d) Psychological line index – measures the proportion of positive trading days, indicating general market sentiment toward a stock.<br /> <br /><br /><br /><br /><br />(8)<br /><br /><br /> <br /><br /><br /><br /><br /> <br />In which  indicates the number of trading days during period t on which the closing price of stock i exceeds that of the previous day, and  represents the total number of trading days for stock i during period t.<br /><strong>Findings</strong><br />To assess the impact of investor sentiment on excess stock returns, we estimated panel regressions using a sequence of models: a single-factor model (investor sentiment), a two-factor model (adding the market excess return), and a multi-factor model (further incorporating the size and value factors). The results for the PCA-based sentiment index are reported in Table 1.<br /><strong> </strong><br /><strong>Table (1) Results of the effect of investor sentiment (PCA) on excess returns</strong><br /><br /><br /><br /><br /><strong>Variable</strong><br /><strong>Frequency</strong><br /><br /><br /><strong>Daily</strong><br /><br /><br /><strong>Weekly</strong><br /><br /><br /><strong>Monthly</strong><br /><br /><br /><br /><br />a<br /><br /><br /><br /><br />Intercept<br /><br /><br />0/0027^***<br /><br /><br />0/0084^***<br /><br /><br />0/0258^***<br /><br /><br /><br /><br />Investor Sentiment<br /><br /><br />0/0099^***<br /><br /><br />0/0249^***<br /><br /><br />0/0548^***<br /><br /><br /><br /><br />Adjusted R-squared<br /><br /><br />0/1536<br /><br /><br />0/2292<br /><br /><br />0/2485<br /><br /><br /><br /><br />Annual Coefficient of Investor Sentiment<br /><br /><br />9/7420<br /><br /><br />2/2722<br /><br /><br />0/9019<br /><br /><br /><br /><br />b<br /><br /><br /><br /><br />Intercept<br /><br /><br />0/0014^***<br /><br /><br />0/0044^***<br /><br /><br />0/1070^***<br /><br /><br /><br /><br />Investor Sentiment<br /><br /><br />0/0093^***<br /><br /><br />0/0224^***<br /><br /><br />0/0484^***<br /><br /><br /><br /><br />Excess Market Return<br /><br /><br />0/7290^***<br /><br /><br />0/6420^***<br /><br /><br />0/6140^***<br /><br /><br /><br /><br />Adjusted R-squared<br /><br /><br />0/2614<br /><br /><br />0/3340<br /><br /><br />0/3628<br /><br /><br /><br /><br />Annual Coefficient of Investor Sentiment<br /><br /><br />8/3086<br /><br /><br />1/9088<br /><br /><br />0/7674<br /><br /><br /><br /><br /> <br /><br /><br />c<br /><br /><br /> <br /><br /><br /><br /><br />Intercept<br /><br /><br />0/0010^***<br /><br /><br />0/0037^***<br /><br /><br />0/0048^***<br /><br /><br /><br /><br />Investor Sentiment<br /><br /><br />0/0091^***<br /><br /><br />0/0223^***<br /><br /><br />0/0440^***<br /><br /><br /><br /><br />Excess Market Return<br /><br /><br />0/8994^***<br /><br /><br />0/6906^***<br /><br /><br />0/7949^***<br /><br /><br /><br /><br />Size Factor (SMB)<br /><br /><br />0/0486^***<br /><br /><br />0/2648^***<br /><br /><br />0/5311^***<br /><br /><br /><br /><br />Value Factor (HML)<br /><br /><br />0/0270^***<br /><br /><br />-0/0202^**<br /><br /><br />0/0054<br /><br /><br /><br /><br />Adjusted R-squared<br /><br /><br />0/2925<br /><br /><br />0/3492<br /><br /><br />0/4286<br /><br /><br /><br /><br />Annual Coefficient of Investor Sentiment<br /><br /><br />7/8744<br /><br /><br />1/8951<br /><br /><br />0/6801<br /><br /><br /><br /><br /> <br />The results indicate a positive and statistically significant effect of investor sentiment on excess stock returns across all examined frequencies—daily, weekly, and monthly. Notably, the adjusted R² values exhibit an increasing trend from daily to monthly horizons, suggesting that the explanatory power of sentiment strengthens over longer timeframes. This finding is consistent with the literature, including Andleeb (2023) and McClure et al. (2004), which posits that investor sentiment exerts a more pronounced influence on short-term investment decisions, with this effect gradually dissipating as the investment horizon extends. Crucially, the persistence of a significant sentiment effect in our multi-factor models, which control for other systematic risks, indicates that its impact on returns is distinct and not subsumed by established risk factors. To ensure robustness, the analysis was replicated using the Kalman filter-based sentiment index. As summarized in Table 2, the results remain qualitatively unchanged, thereby reinforcing the primary conclusions.<br /> <br /><strong>Table 2: Effect of Investor Sentiment (Kalman Filter) on Excess Return</strong><br /><br /><br /><br /><br /><strong>Variable</strong><br /><strong>Frequency</strong><br /><br /><br /><strong>Daily</strong><br /><br /><br /><strong>Weekly</strong><br /><br /><br /><strong>Monthly</strong><br /><br /><br /><br /><br />a<br /><br /><br /><br /><br />Intercept<br /><br /><br />0/0007^***<br /><br /><br />0/0033^***<br /><br /><br />0/0155^***<br /><br /><br /><br /><br />Investor Sentiment<br /><br /><br />0/0023^***<br /><br /><br />0/0092^***<br /><br /><br />0/0236^***<br /><br /><br /><br /><br />Adjusted R-squared<br /><br /><br />0/0531<br /><br /><br />0/1483<br /><br /><br />0/2119<br /><br /><br /><br /><br />Annual Coefficient of Investor Sentiment<br /><br /><br />0/7396<br /><br /><br />0/5549<br /><br /><br />0/3245<br /><br /><br /><br /><br />b<br /><br /><br /><br /><br />Intercept<br /><br /><br />0/0002^***<br /><br /><br />0/0007^***<br /><br /><br />0/0033^***<br /><br /><br /><br /><br />Investor Sentiment<br /><br /><br />0/0023^***<br /><br /><br />0/0087^***<br /><br /><br />0/0218^***<br /><br /><br /><br /><br />Excess Market Return<br /><br /><br />0/5878^***<br /><br /><br />0/5929^***<br /><br /><br />0/6084^***<br /><br /><br /><br /><br />Adjusted R-squared<br /><br /><br />0/1429<br /><br /><br />0/2550<br /><br /><br />0/3565<br /><br /><br /><br /><br />Annual Coefficient of Investor Sentiment<br /><br /><br />0/7396<br /><br /><br />0/5182<br /><br /><br />0/2967<br /><br /><br /><br /><br /> <br /><br /><br />c<br /><br /><br /> <br /><br /><br /><br /><br />Intercept<br /><br /><br />0/0000<br /><br /><br />0/0003^*<br /><br /><br />-0/0010<br /><br /><br /><br /><br />Investor Sentiment<br /><br /><br />0/0023^***<br /><br /><br />0/0085^***<br /><br /><br />0/0205^***<br /><br /><br /><br /><br />Excess Market Return<br /><br /><br />0/7609^***<br /><br /><br />0/6395^***<br /><br /><br />0/7630^***<br /><br /><br /><br /><br />Size Factor (SMB)<br /><br /><br />0/4387^***<br /><br /><br />0/2389^***<br /><br /><br />0/4806^***<br /><br /><br /><br /><br />Value Factor (HML)<br /><br /><br />0/0315^***<br /><br /><br />-0/0084<br /><br /><br />0/0051<br /><br /><br /><br /><br />Adjusted R-squared<br /><br /><br />0/1770<br /><br /><br />0/2709<br /><br /><br />0/4152<br /><br /><br /><br /><br />Annual Coefficient of Investor Sentiment<br /><br /><br />0/7396<br /><br /><br />0/5037<br /><br /><br />0/2770<br /><br /><br /><br /><br />Consistent with the primary results, the Kalman filter-based sentiment index also demonstrates a positive and statistically significant influence on stock return across all observed frequencies. However, a key finding emerges when examining the annualized sentiment coefficients: their magnitude displays a monotonic decline as the investment horizon extends from daily to monthly. This pattern indicates a decaying term structure for investor sentiment, where its pricing effect attenuates over longer periods. This result robustly confirms that the impact of investor sentiment on excess returns follows a consistently downward-sloping term structure.<br /><strong> </strong><br /><strong>Discussion and conclusion</strong><br />This study provides robust evidence of a direct and significant impact of investor sentiment on excess stock return across daily, weekly, and monthly investment horizons. These findings align with a growing body of literature that underscores the importance of sentiment in short-term price formation (e.g., Andleeb, 2023; Seok et al., 2018; Ryu et al., 2016; Dai, 2025).<br />A central finding is the decaying influence of investor sentiment as the observation horizon extends from daily to monthly data, revealing a consistently downward-sloping term structure. This pattern suggests that irrational behavioral factors disproportionately drive short-term investment decisions, whereas their influence wanes over longer periods. The results corroborate the findings of Yang & Gao (2014) and extend the work of Li (2020) to the context of the Tehran Stock Exchange, demonstrating that short-term waves of investor optimism or pessimism are ultimately transient. Consequently, investor sentiment appears to be a key driver of short-term asset mispricing, generating excess returns that are subsequently corrected as prices converge toward their fundamental values over the long run.<br />Crucially, this downward term structure is not an artifact of measurement, as it holds consistently for sentiment indices constructed using both Principal Component Analysis and the Kalman filter. This methodological robustness strongly confirms the time-dependent nature of sentiment effects, as theorized by Fu (2024). Furthermore, the persistent explanatory power of sentiment even after controlling for market, size, and value factors—a consistency also noted by Brown & Cliff (2004)—indicates that the informational content of investor sentiment captures dimensions of risk and return distinct from those in traditional asset-pricing models. پژوهش‌های اخیر نشان می‌دهد که بخشی از نوسان قیمت سهام متأثر از احساسات سرمایه‌گذار است. از یک سو باوجود شواهد تجربی متناقض دربارۀ اثر مذکور و ازسوی دیگر مطالعات محدودی که به نقش افق زمانی در رابطۀ احساسات و بازدۀ سهام توجه می‌کند، هدف این پژوهش بررسی تأثیر احساسات سرمایه‌گذار بر بازدۀ سهام در سه بازۀ زمانی کوتاه‌مدت روزانه، هفتگی و ماهانه است (ساختار زمانی احساسات). اثر احساسات سرمایه‌گذار بر بازدۀ اضافی سهام در سه تواتر زمانی روزانه، هفتگی و ماهانه با استفاده از رگرسیون داده‌‌های ترکیبی و کنترل اثر بازدۀ اضافی بازار، عوامل ارزش و اندازه آزمون و با یکدیگر مقایسه شده است. شاخص احساسات سرمایه‌گذار با استفاده از چهار سنجۀ غیرمستقیم «نرخ تعدیل‌شدۀ گردش سهام»، «عدم‌تعادل خرید-فروش»، «مبلغ معامله» و «شاخص خط روان‌شناختی» و دو روش «تحلیل مؤلفه‌های اصلی» و «فیلتر کالمن» اندازه‌گیری شده است. یافته‌ها حاکی‌از تأثیر احساسات سرمایه‌گذار بر بازدۀ سهام است؛ اما شدت این تأثیر در گذر زمان از داده‌های روزانه به هفتگی و ماهانه، تقلیل یافته و ساختار زمانی احساسات سرمایه‌گذار تابع نزولی از زمان است. به سخن دقیق‌تر، با گذشت زمان همراه با افشای اطلاعات بیشتر و کاهش عدم‌اطمینان، نقش احساسات در تصمیم‌گیری سرمایه‌گذاران کاهش می‌یابد و عوامل فراگیر ریسک قادر به پوشش محتوای اطلاعاتی احساسات سرمایه‌گذار نیست. نتایج حاصل از مقایسۀ اثر احساسات بر بازدۀ سهام با استفاده از سنجۀ مبتنی بر روش‌های «فیلتر کالمن» با نتایج استفاده از سنجۀ «تحلیل مؤلفه‌های اصلی» همسو است. پژوهش حاضر برای نخستین بار به واکاوی تأثیر متفاوت احساسات سرمایه‌گذار بر بازدۀ سهام در سه بازۀ روزانه، هفتگی و ماهانه می‌پردازد و ساختار زمانی احساسات سرمایه‌گذار را می‌آزماید. در پژوهش‌های پیشین عمدتاً از شاخص ترکیبی احساسات مبتنی بر روش «تحلیل مؤلفه‌های اصلی» استفاده شده است. در این پژوهش برای اولین بار در بورس اوراق بهادار تهران از روش «فیلتر کالمن» برای ساخت شاخص احساسات استفاده شده و ساختار زمانی احساسات براساس دو روش «فیلتر کالمن» و «تحلیل مؤلفه‌های اصلی» آزمون شده است. https://amf.ui.ac.ir/article_29940_364e8c24a043287817aabf5497b220fe.pdf