Abstract:

Although the risk of a company delisting is a major concern for IPO investors who prefer a medium to long investment horizon, about one-third of the newly listed companies fail within the first 5 years after the IPO date. Under these circumstances, the main focus of the papers studying the likelihood of IPO survival has become the identification of characteristics that can ex-ante signal failure-prone companies. Various aspects characterizing the company or the timing of listing have been identified as reliable predictors of IPO failure. Negative relationships have been demonstrated between the size of the offering, the age of the company and the reputation of the investment bank that brokered the IPO and the likelihood that the company will delist. The database used in this study is composed of a set of 15 variables characterizing 357 software IPOs listed on the main US stock exchanges, namely the New York Stock Exchange (NYSE) and NASDAQ, over the period from January 1997 to December 2020. Two non-parametric models, namely Random Forest (RF) and Gradient Boosting (GB), have been used in this paper due both to their high degree of accuracy and their ability to handle databases characterized by a limited number of observations and a significant number of explanatory variables. Their predictive power was tested on out-of-sample data using the AUC measure. In terms of explanatory factors, the model identifies as the main cause of failure the size of the companies at the time of listing. Similar results have been obtained in the literature. Given that another important factor is the number of IPOs listed in the previous period before the listing, as a proxy for investor enthusiasm, we can assume that there were also periods of opportunity when speculative companies which were not well prepared for such a step, have tried to take advantage of overly enthusiastic investors.