نوع مقاله : مقاله علمی پژوهشی
نویسندگان
1 گروه حسابداری، واحد بروجر ، دانشگاه آزاد اسلامی، بروجرد ، ایران .
2 گروه حسابداری ، دانشکده اقتصاد و حسابداری، دانشگاه رازی، کرمانشاه ، ایران
3 گروه مهندسی کامپیوتر، واحد بروجرد، دانشگاه آزاد اسلامی، بروجرد، ایران
4 گروه حسابداری ، واحد بروجرد، دانشگاه آزاد اسلامی ، بروجرد ، ایران
چکیده
کلیدواژه‌ها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Objective: This research aims to develop a novel model based on the Gradient Boosting algorithm and Transformer architecture for accurately predicting corporate valuation ratios at the time of Initial Public Offering (IPO) in the Iranian market. Contrary to traditional approaches focusing on first-day returns, this study evaluates the relative pricing of companies through three key ratios: Price-to-Book Value (P/B), Enterprise Value-to-Total Assets (EV/TA), and Enterprise Value-to-Sales (EV/S). Given the complexities of the Iranian market, including severe economic volatility, chronic inflation, economic sanctions, and regulatory restrictions, this research provides an efficient tool for investors, underwriters, and regulatory bodies to make more informed decisions.
Methodology: This quantitative-applied study utilized data from 163 companies (42 listed on the Tehran Stock Exchange and 121 on the over-the-counter market - Farabourse) spanning the period 1392 to 1402 (approximately 2013-2023). This period encompasses diverse economic conditions, including severe sanctions, currency fluctuations, recession, and economic booms. Independent variables included firm-specific financial features (size, debt ratio, profitability, return on assets, profit margin, cash flow ratio, financial leverage, and weighted average cost of capital) and macroeconomic variables (inflation rate, main stock market index, sanctions index, exchange rate, and interest rate). The proposed architecture is a hybrid of Gradient Boosting Decision Trees, Long Short-Term Memory (LSTM) networks, and self-attention mechanisms, optimized through an Automated Machine Learning (AutoML) architecture search. Optimal parameters included a learning rate of 0.001, 150 decision trees, 4-time steps, and 3 self-attention layers. The model was compared with traditional methods (Linear Regression) and advanced methods (standard Gradient Boosting algorithms, Random Forest, and Recurrent Neural Networks) and validated using ten-fold cross-validation and bootstrapping.
Findings: The results demonstrated the decisive superiority of the proposed model in predicting all three valuation ratios. For the Price-to-Book Value ratio, the developed model achieved an R2 of 0.853 and a Root Mean Square Error (RMSE) of 0.497, representing a 60% improvement in prediction accuracy compared to Linear Regression (R2 of 0.534). In predicting the Enterprise Value-to-Total Assets ratio, an R2 of 0.865 and a 69% reduction in Mean Squared Error (MSE) (from 0.152 to 0.047) were obtained. For the Enterprise Value-to-Sales ratio, an R2 of 0.846 was achieved, showing a 67% improvement in accuracy over traditional methods. Other advanced models also performed acceptably but did not reach the level of the proposed model. Cross-validation tests confirmed the model's high stability with a very low standard deviation (around 0.01), indicating its generalizability and reliability under various conditions. Variable importance analysis revealed that the inflation rate acts as the second most important factor across all models, with an average importance of 0.19, indicating the profound impact of macroeconomic conditions on corporate valuation in the Iranian market. Return on Assets (0.21) was identified as the most important financial performance indicator for the P/B ratio, Profit Margin (0.225) for the EV/S ratio, and Cash Flow to Assets Ratio (0.2) for the EV/TA ratio.
Conclusion: The findings emphasize the importance of employing novel machine learning techniques in predicting IPO valuation ratios. The developed model, by reducing prediction error by 60% to 69%, provides an advanced tool for underwriters, investors, and fund managers to mitigate the risk of undervaluation or overvaluation. The identification of the pivotal role of the inflation rate highlights the high sensitivity of valuations to macroeconomic conditions in the Iranian market. This research contributes in three key areas: introducing a novel hybrid architecture, providing a comprehensive validation framework, and identifying the differential importance of variables for various valuation ratios under unstable economic conditions.
کلیدواژه‌ها [English]
)