This study aims to verify whether using artificial neural networks (ANNs) to establish classification probabilities generates portfolios with higher excess returns than using ANNs in their traditional role of predicting portfolio returns. Our sample includes all companies listed on the Toronto Stock Exchange from 1994 to 2014 with a monthly average of 16,324 company-month observations. Results indicate that portfolios based on the classification probabilities yield mean returns ranging from 7.81 to 14.40% annually over a 16-year period and that portfolios based on both predicted returns and classification probabilities generate returns that are superior to the market index. In addition, there is evidence that ranking securities based on their probability of beating the market has some benefit.

中文翻译：

---

使用人工神经网络和分类概率创建投资组合：一项加拿大研究

本研究旨在验证使用人工神经网络 (ANN) 来建立分类概率是否比使用人工神经网络在预测投资组合收益的传统角色中产生的投资组合具有更高的超额收益。我们的样本包括从 1994 年到 2014 年在多伦多证券交易所上市的所有公司，每月平均有 16,324 个公司月观察。结果表明，基于分类概率的投资组合在 16 年期间每年产生 7.81% 至 14.40% 的平均回报，并且基于预测回报和分类概率的投资组合产生的回报优于市场指数。此外，有证据表明，根据其跑赢市场的可能性对证券进行排名有一些好处。