دانشگاه اصفهان نشریه پژوهش های حسابداری مالی 2322-3405 18 1 2026 05 22 Testing Capital Asset Pricing Models with Single Stocks: An Instrumental Variable Approach آزمون مدل‌های قیمت‌گذاری دارایی‌های سرمایه‌ای با تک‌سهام؛ رویکرد متغیرهای ابزاری 45 78 30650 10.22108/far.2026.148364.2210 FA مهدی خرم آبادی استادیار، گروه حسابداری، دانشگاه پیام نور، تهران، ایران سیده زهرا میراشرفی استادیار، گروه حسابداری، دانشگاه پیام نور، تهران، ایران منا میررضائی کارشناسی‌ارشد حسابداری، دانشگاه پیام نور، تهران، ایران Journal Article 2026 02 14 The primary objective of this study was to evaluate the explanatory power of risk-adjusted excess returns in single-stock capital asset pricing models (CAPM) using an instrumental variable approach, compared to portfolio-based factor models, with the aim of identifying the superior pricing model. The models assessed include single-factor CAPM, the three-factor and five-factor Fama-French models, the Carhart four-factor model, the Hou, Xue, and Zhang q-factor model, and the modified q-factor model by Hou et al. These models were examined under three specifications: constant beta, time-varying beta, and single-stock characteristics. Data from 196 companies listed on the Tehran Stock Exchange from 2015 to 2024 were analyzed using Ordinary Least Squares (OLS) and Generalized Method of Moments (GMM). The findings indicate significant differences in the explanatory power of risk-adjusted excess returns between multifactor models and the single-factor model, as well as between models with time-varying betas and those with constant betas. Among the six models with constant betas and no single-stock characteristics, the Fama-French five-factor model demonstrated the highest explanatory power. In contrast, the time-varying beta specification of the Hou, Xue, and Zhang q-factor model exhibited the greatest ability to explain risk-adjusted excess returns. Moreover, factor models incorporating single-stock characteristics showed significantly higher explanatory power than those without such characteristics. Additionally, models estimated via OLS provided more accurate predictions of risk-adjusted excess returns compared to those estimated using GMM. The sensitivity analysis using the GMM approach indicates that while some complex models face overfitting risks within the instrumental variables framework, the augmented q-factor model with time-varying beta and single-stock characteristics remains the superior model and demonstrates high robustness across both OLS and GMM estimation methods, thereby confirming the main findings of the study. <strong>Introduction</strong> This study examines the explanatory power of six prominent asset pricing models in the Tehran Stock Exchange, employing a novel single-stock instrumental variable (IV) approach to address inherent empirical limitations. Traditional linear factor models including the CAPM, Fama-French three-, four-, and five-factor models, the q-factor model, and its enhanced version (q5) are known to face significant challenges such as time-varying systematic risk and errors-in-variables (EIV) bias when estimated betas are used. These issues can distort risk-return relationships and weaken statistical inferences. To overcome these shortcomings, an IV methodology at the individual stock level is applied and contrasted with conventional portfolio-based factor models. The TSE provides a compelling setting for this investigation due to its high volatility, sensitivity to macroeconomic shocks, and evolving market structure. The primary objective is to identify the most robust model for explaining risk-adjusted excess returns in this emerging market context. The findings are expected to contribute to more accurate risk estimation, improved investment decision-making, and enhanced market efficiency. This research underscores the necessity of adapting advanced econometric techniques to address measurement errors and multicollinearity in asset pricing applications within dynamic and complex market environments like Iran's.   <strong>Methodology</strong> This applied, descriptive, and quantitative study employs a comparative approach to test the explanatory power of asset pricing models. Data are gathered from financial statements, the Codal website, and the Rahavard Novin software. The analysis is performed using Ordinary Least Squares (OLS) and Instrumental Variable (IV) regression methods in Excel and EViews. The sample consists of 196 companies listed on the Tehran Stock Exchange, selected through systematic screening based on continuous listing from 2015 to 2024, data availability, exclusion of financial intermediaries, no trading suspension exceeding three months in any year, and consistent fiscal year-end (March 20).   <strong>Findings</strong> The Fama-French five-factor model (2015) with constant beta and no firm-specific characteristics demonstrates the highest explanatory power and lowest forecast error among its category, confirming the first hypothesis. Among models with time-varying beta, the q-factor model (Hou et al., 2015) shows superior performance. Comparative tests confirm that time-varying beta models have significantly higher predictive accuracy than constant-beta models, supporting the second hypothesis. The inclusion of firm-specific characteristics is found to significantly enhance the predictive accuracy of both constant-beta and time-varying beta models. The strongest models within these enhanced categories are identified as the Fama-French five-factor model and the augmented q-factor model (Hou et al., 2021), respectively, confirming the third hypothesis. Regarding estimation methodology, models using Ordinary Least Squares (OLS) provide greater predictive accuracy than those using the Generalized Method of Moments (GMM), confirming the fourth hypothesis. However, sensitivity analysis using GMM (Table 12) reveals that the augmented q-factor model exhibits only borderline validity in the GMM framework (p=0.066), highlighting the importance of cautious interpretation and the need for stronger instrumental variables in future research. A final comprehensive comparison establishes the augmented q-factor model (Hou et al., 2021) with time-varying beta and firm-specific characteristics, estimated via OLS, as the overall superior model for explaining and predicting excess returns in the Tehran Stock Exchange.   <strong>Conclusion and Implications</strong> This study demonstrates the superiority of multi-factor models over the single-factor CAPM for the Tehran Stock Exchange. Models with time-varying beta significantly outperform constant-beta specifications, and the inclusion of firm-specific characteristics further enhances explanatory power. The augmented q-factor model (Hou et al., 2021) with these dynamic and firm-level features, estimated via OLS, is identified as the most robust model. However, its borderline validity in the GMM framework highlights the need for stronger instrumental variables in future research. Therefore, investors and analysts are advised to adopt such dynamic, multi-factor frameworks that integrate both systematic and idiosyncratic risks for more accurate pricing and risk management. Future research should explore alternative instrumental variables, apply machine learning techniques, and incorporate investor sentiment factors to further refine asset pricing models in emerging market.   هدف اصلی پژوهش حاضر آزمون قدرت تبیین بازده مازاد بر ریسک مدل‌های قیمت‌گذاری دارایی‌های سرمایه‌ای با تک‌سهام با استفاده از رویکرد متغیر ابزاری در مقایسه با مدل‌های عاملی مبتنی بر پرتفوی‌های مبنا به منظور انتخاب مدل قیمت‌گذاری برتر است. مدل‌های تک‌عاملی قیمت‌گذاری دارایی‌های سرمایه‌ای، سه‌عاملی فاما و فرنچ، چهارعاملی کارهارت، پنج‌عاملی فاما و فرنچ، q عاملی هو و همکاران (2015)، q عاملی تعدیل‌شدۀ هوو و همکاران (2021)، با بتای ثابت، بتای متغیر با زمان و دارای ویژگی‌های تک‌سهام ارزیابی شدند. داده‌های 196 شرکت پذیرفته‌شده در بورس اوراق بهادار تهران طی سال‌های 1394 تا پایان 1403 با روش حداقل مربعات معمولی و روش گشتاورهای تعمیم­یافته تجزیه‌وتحلیل شد. نتایج پژوهش حاکی از آن است که قدرت تبیین بازده مازاد بر ریسک بین مدل‌های چندعاملی و مدل تک‌عاملی، همچنین بین مدل‌های عاملی با بتای متغیر با زمان و مدل‌های عاملی با بتای ثابت، تفاوتی معنادار وجود دارد. مدل پنج‌عاملی فاما و فرنچ، در بین شش مدل، دارای بتای ثابت و بدون ویژگی‌های تک‌سهام، و مدل q عاملی هو و همکاران (2015) با بتای متغیر با زمان، بیشترین توانایی را در تبیین بازده مازاد بر ریسک سهام شرکت‌ها دارند. قدرت توضیحی بازده مازاد بر ریسک در مدل‌های عاملی دارای ویژگی‌های تک‌سهام و مدل‌های عاملی بدون ویژگی‌های تک‌سهام تفاوتی معنادار دارد و قدرت تببین مدل‌های عاملی دارای ویژگی‌های تک‌سهام بیشتر است. همچنین، مدل‌های برآوردشده به روش OLS دقت بیشتری در پیش‌بینی بازده مازاد بر ریسک سهام نسبت به روش گشتاورهای تعمیم‌یافته دارند. نتایج آزمون تحلیل حساسیت با روش GMM نیز نشان می‌دهد اگرچه در چارچوب متغیرهای ابزاری، برخی از مدل‌های پیچیده با ریسک بیش‌برازش مواجه هستند، مدل q عاملی تعدیل‌شده با بتای متغیر با زمان و ویژگی‌های تک‌سهام، به ‌عنوان مدل برتر پژوهش، از پایداری زیادی در هر دو روش OLS و GMM برخوردار است. <strong> </strong>   https://far.ui.ac.ir/article_30650_602f5eef80ebbbbb8a5d9ce915108010.pdf