随着技术进步和计算机科学的发展，量化投资作为一种创新的投资策略，逐渐赢得了投资者的青睐，并在我国证券市场上流行起来。不管是公募基金还是私募基金，能够实现较高回报且风险较低的投资方法，都会持续受到投资界的欢迎。

本研究聚焦于算力租赁领域的80只股票，采用机器学习与深度神经网络算法构建选股模型，从而得到一个多因子量化选股策略，以获得显著且持续的投资回报。在量化投资策略领域中，多因子选股占据了重要地位，其核心是利用对于股票市场的投资回报有重大影响的因子，并基于这些因子构建投资组合，旨在获得超过市场平均水平的回报。通过相关算法，本研究对因子数据进行筛选和分析，以构建有效的选股模型。

具体来看，本研究首先通过分析股票因子的重要性和因子间的相关性，对于初始因子池进行筛选，并据此构建最终的因子池。接下来，这些关键因子被用于构造多因子选股模型。构建选股模型的过程包括两个关键环节。第一，根据机器学习模型的预测效果，确定一个多因子机器选股模型，并使用该模型对股票池进行筛选；第二，引入深度神经网络模型，通过对大量历史数据的学习和训练，目的是预见股票价格的变动趋势。之后采取每月一次的股票轮换投资策略，挑选出表现突出的股票。最后，对所选股票执行交易，并对该交易策略进行回测，通过计算评价指标评估策略的效果。

根据上述研究框架，本研究使用2018年至2022年的数据进行因子筛选以及选股模型的构建，使用2022年至2023年的数据进行股票交易策略的回测，目的是检验交易策略的有效性。最终发现，本研究构建的多因子选股模型显著获得了超额回报，其中CNN+ LSTM选股模型的表现最为突出，实现了33.1%的收益率。

With the advancement of technology and the development of computer science, quantitative investment, as an innovative investment strategy, has gradually won the favor of investors and become popular in China's securities market. Regardless of whether it is a public or private fund, investment methods that can realize higher returns with lower risks continue to be popular among the investment community.

This study focuses on 80 stocks in the arithmetic leasing sector and uses machine learning and deep neural network algorithms to construct a stock selection model, resulting in a multi-factor quantitative stock selection strategy that achieves significant and consistent investment returns. In the field of quantitative investment strategies, multi-factor stock picking occupies an important position, which centers on the use of factors that have a significant impact on the investment returns of the stock market and the construction of portfolios based on these factors with the aim of obtaining returns that exceed the market average. Through relevant algorithms, this study screens and analyzes factor data in order to construct an effective stock selection model.

Specifically, this study first screens the initial factor pool by analyzing the importance of stock factors and correlations between factors, and constructs the final factor pool accordingly. Next, these key factors are used to construct a multi-factor stock selection model. The process of constructing a stock selection model consists of two key components. First, a multifactor machine stock picking model is identified based on the predictive effectiveness of the machine learning model and the stock pool is screened using this model. Second, a deep neural network model is introduced to learn and train on a large amount of historical data with the aim of anticipating the trend of stock price movements. This is followed by a monthly stock rotation investment strategy in order to select the top performing stocks. Finally, trades are executed on the selected stocks and the trading strategy is backtested to assess the effectiveness of the strategy by calculating evaluation metrics.

Based on the above research framework, this study uses data from 2018 to 2022 for factor screening as well as stock selection model construction, and uses data from 2022 to 2023 for backtesting the stock trading strategy with the aim of testing the effectiveness of the trading strategy. Ultimately, it is found that the multifactor stock picking model constructed in this study significantly earns excess returns, with the CNN+ LSTM stock picking model performing most prominently, realizing a return of 33.1%.