Scholarlink

<p>We examine whether industry classification influences IPO pricing outcomes, a question that remains underexplored in the context of emerging markets. To fill this gap in the literature, we apply a suite of supervised and unsupervised machine learning techniques to analyze how sectoral characteristics interact with market and firm-level variables in shaping IPO pricing anomalies. Our dependent variable classifies IPOs as underpriced (first-day returns &gt; 0%) or overpriced (first-day returns ≤ 0%), enabling binary classification of pricing outcomes. Addressing challenges such as high dimensionality and class imbalance through SMOTE, we employ algorithms including XGBoost, CatBoost, LightGBM, Random Forest, and an ensemble Voting Classifier on a dataset of Indian IPOs issued between 2011 and 2022. Model performance is evaluated using accuracy, precision, recall, and F1-score metrics. XGBoost and CatBoost emerge as the most accurate models (both achieving 78% accuracy), with CatBoost demonstrating superior recall (0.90) for identifying overpriced IPOs and XGBoost excelling at detecting underpriced IPOs (recall: 0.79). K-Means clustering (k = 2), supported by Principal Component Analysis, reveals distinct clusters driven more by investor demand and macroeconomic volatility than clear-cut industry boundaries. Feature importance analysis shows that issue-specific variables, particularly total subscription rate (20.61% importance), dominate pricing predictions (55.49% combined contribution), followed by macroeconomic factors (31.15%) and financial metrics (13.36%). The findings contribute methodological novelty to the IPO valuation literature by demonstrating that demand-driven indicators outperform traditional sector classifications in explaining IPO mispricing. Practically, the results offer valuable insights for investors to prioritize subscription data over sector labels, issuers to optimize timing and demand generation, and policymakers to mandate real-time subscription disclosure for enhanced market transparency.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Sector signals: machine learning insights into IPO pricing dynamics

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Amit Kumar Singh,
  • Monomita Nandy,
  • Suman Lodh,
  • Shivani Kalra

摘要

We examine whether industry classification influences IPO pricing outcomes, a question that remains underexplored in the context of emerging markets. To fill this gap in the literature, we apply a suite of supervised and unsupervised machine learning techniques to analyze how sectoral characteristics interact with market and firm-level variables in shaping IPO pricing anomalies. Our dependent variable classifies IPOs as underpriced (first-day returns > 0%) or overpriced (first-day returns ≤ 0%), enabling binary classification of pricing outcomes. Addressing challenges such as high dimensionality and class imbalance through SMOTE, we employ algorithms including XGBoost, CatBoost, LightGBM, Random Forest, and an ensemble Voting Classifier on a dataset of Indian IPOs issued between 2011 and 2022. Model performance is evaluated using accuracy, precision, recall, and F1-score metrics. XGBoost and CatBoost emerge as the most accurate models (both achieving 78% accuracy), with CatBoost demonstrating superior recall (0.90) for identifying overpriced IPOs and XGBoost excelling at detecting underpriced IPOs (recall: 0.79). K-Means clustering (k = 2), supported by Principal Component Analysis, reveals distinct clusters driven more by investor demand and macroeconomic volatility than clear-cut industry boundaries. Feature importance analysis shows that issue-specific variables, particularly total subscription rate (20.61% importance), dominate pricing predictions (55.49% combined contribution), followed by macroeconomic factors (31.15%) and financial metrics (13.36%). The findings contribute methodological novelty to the IPO valuation literature by demonstrating that demand-driven indicators outperform traditional sector classifications in explaining IPO mispricing. Practically, the results offer valuable insights for investors to prioritize subscription data over sector labels, issuers to optimize timing and demand generation, and policymakers to mandate real-time subscription disclosure for enhanced market transparency.