以機器學習方法進行科技新創價值預測

Abstract

創業投資(Venture Capital)對新創公司早期階段的生存具有重要的影響，近年來持續增長的投資額和交易量顯示了創業投資對新創公司的重要性，但是在選擇具有潛力的新創公司時，創業投資家面臨巨大的風險，包括新創公司的高失敗率和與新創公司之間的資訊不對稱。因此，開發一項能夠有效幫助創業投資家識別新創公司潛在價值的技術至關重要。 在本研究中，我們開發了一項預測模型，該模型採用機器學習算法，通過採用更全面的變項來預測科技新創公司的價值。我們將預測變數分為三個種類，包含：基本變數、專利相關變數、上一輪相關變數。實證結果顯示，添加專利相關變項或上一輪相關變項都可以提高新創公司價值的預測準確性；此外，結合了所有變項的表現最佳。此技術有利於創業投資家評估科技新創公司的價值(特別是在其不熟悉的產業中)；與此同時，它也能夠幫助新創公司制定募資談判策略、以及決定最適宜的募資時機。

Venture Capital is an important financing strategy for helping startups to survive in their early stage. Recently, the growth of VC investments and deals has been showing the increasing importance of startups. However, when selecting potential startups for investments, VCs are exposed to great risks, including the high failure rate of startups and the information asymmetry between VCs and startups. Thus, it is essential to develop a prediction technique that can effectively identify the value of a technology startup. In this research, we follow the machine learning approach to develop a predictive model that encompasses a more comprehensive set of predictors (independent variables) for predicting the value of technology startups. We categorize these variables into three categories, including basic variables, patent-related variables, and last-round related variables. As expected, our evaluation results show that adding patent-related variables or last-round related variables improves the accuracy of startup value prediction. In addition, combining all the features performs the best. Our proposed prediction technique is expected to benefit VCs in evaluating the value of technology startups, especially in unfamiliar industries. At the same time, it is expected to help startups to develop the fundraising negotiation strategies and identify the most appropriate timing for fundraising.