Many complex systems observed in nature and society can be described in terms of network. A salient feature of networks is the presence of community patterns. Network-based models have already been applied in the analysis of data from very diverse areas, from epidemics modeling to periodicity detection in meteorological data. In this paper, inspired by the formation of community structures, such as the metabolic networks and the anatomical and functional connectome observed in biological neural networks, we present a model which makes use of connector hubs to detect price trend reversals and to automatize decision-making processes in stock market trading operations for selecting a good investment strategy and improve the returns. It starts by mapping the historical stock price time series as a network, where each node represents a price variation range and the edges are generated according to the time sequential order in which these ranges occur. Afterwards, communities of the constructed network so far are detected, which represent the up and down trends of the stock prices. The model has two phases: (1) Trend detection phase, where the price trend communities are detected and trend labels are generated; and (2) Operating phase. In this phase, the proposed technique predicts trend labels to future stock prices, in such a way that these trends can be used as triggers to perform buying and selling operations of the stock. We evaluate the model by applying it on historical data from 10 of the most traded stocks from both NYSE and the Brazilian Stock Exchange (Bovespa). The obtained results are promising, with the model’s best returns being able to outperform the stock price returns for the same period in 15 out of the 20 cases under consideration.

可以用网络来描述自然界和社会中观察到的许多复杂系统。网络的一个显着特征是社区模式的存在。从流行病模型到气象数据的周期性检测，基于网络的模型已被用于分析非常不同领域的数据。在本文中，受社区结构（例如代谢网络以及在生物神经网络中观察到的解剖学和功能性连接体）形成的启发，我们提出了一个模型，该模型利用连接器枢纽来检测价格趋势反转并自动进行决策股票交易操作中的流程，以选择良好的投资策略并提高收益。首先将历史股价时间序列映射为网络，其中每个节点代表一个价格变化范围，并且根据这些范围出现的时间顺序生成边。之后，将检测到到目前为止已构建网络的社区，这些社区代表向上和向下的股票价格走势。该模型分为两个阶段：（1）趋势检测阶段，其中检测价格趋势社区并生成趋势标签；（2）运营阶段。在此阶段，所提出的技术以将来的股票价格预测趋势标签，从而可以将这些趋势用作执行股票买卖操作的触发器。我们通过将其应用于来自纽约证券交易所和巴西证券交易所（Bovespa）的10只交易量最大的股票的历史数据来评估该模型。获得的结果令人鼓舞，在所考虑的20个案例中，有15个模型的最佳收益能够超过同期的股票收益。