The comparative study of the accuracy of prediction of Support Vector Machine, Bayesian Network and C5 models in prediction underpricing for listed companies at TSE and OTC

Previous research into the short-term performance of the initial public offering reflects the fact that short-term stocks perform better than the market in the short run. Statistical models have been able to make good predictions about the performance of new stocks, but the limiting assumptions of some of these models have been effective! So, other ways to deal with these limitations and improve forecasting performance were introduced. Since initial public offering is an important issue in the capital market, in this study, we investigate different classification models to achieve a model that has high efficiency and accuracy in predicting underpricing of initial public offering (IPO) stocks. To achieve the research goal, systematic elimination sampling method is considered to select 84 companies among all listed companies at Tehran Stock Exchange (TSE) and 54 companies among all listed companies at Over the Counter (OTC) from 2003 to 2017. The results showed that support vector machine (SVM), Bayesian Network and C5 decision tree models are highly accurate in predicting underpricing. The results also showed that the influential variables included assets growth, auditor tenure, auditor specialty in the industry, financing ratio, P/E, CFO ratio, ROA, stock price fluctuate, growth opportunity and audit firm size.