另类Alpha：基于供应链数据的量化因子挖掘

标星★公众号     爱你们♥

近期原创文章：

♥ 5种机器学习算法在预测股价的应用（代码+数据）

♥ Two Sigma用新闻来预测股价走势，带你吊打Kaggle

♥ 2万字干货：利用深度学习最新前沿预测股价走势

♥ 机器学习在量化金融领域的误用！

♥ 基于RNN和LSTM的股市预测方法

♥ 如何鉴别那些用深度学习预测股价的花哨模型？

♥ 优化强化学习Q-learning算法进行股市

♥ WorldQuant 101 Alpha、国泰君安 191 Alpha

♥ 基于回声状态网络预测股票价格（附代码）

♥ 计量经济学应用投资失败的7个原因

♥ 配对交易千千万，强化学习最NB！（文档+代码）

♥ 关于高盛在Github开源背后的真相！

♥ 新一代量化带货王诞生！Oh My God！

♥ 独家！关于定量/交易求职分享（附真实试题）

♥ Quant们的身份危机！

♥ 拿起Python，防御特朗普的Twitter

♥ AQR最新研究 | 机器能“学习”金融吗？

作者：ChinaScope

在量化交易中，如何获取适当的数据用于开发和测试交易策略，往往是投资者面临的难题。随着技术的发展，获取大数据的成本不断降低，但历史价格等传统数据已完全无法满足投资者需求，可挖掘Alpha已基本消失。因此尝试从另类数据（Alternative Data）中提取交易信号逐步成为业界主流趋势。在海外市场，量化投资领域对另类数据的应用在过去两三年内已实现阶段性发展，另类数据如资讯情绪（Sentiment），产业链及供应链数据等已被广泛纳入量化策略。

随着中国金融市场的进一步开放及交易规则的逐步成熟，越来越多的海外量化投资机构已开始着手将海外市场中的另类数据策略复制到中国市场，而A股不断增量纳入MSCI及FTSE指数的趋势也加速了这一进程。同时本土头部金融机构对于使用另类数据形成有效交易因子并整合入现有量化策略这一趋势也已形成高度共识，另类数据的应用增长趋势正在形成。

作为一家专注于数据智能领域超过十年的公司，数库在另类数据领域拥有深厚的积累。由于数库对外提供的数据流服务均由自研DAS数据自动化生产平台产生，该平台拥有非常严格的质检体系及数据标准化能力，进而保障了数据流的稳定性及连贯性，确保了数据流在量化领域中的可应用性。目前数库生产的新闻情绪（Sentiment），产业链及供应链数据已在海外被大量头部量化机构采纳并登陆了如纳斯达克Quandl等专业另类数据平台，为投资中国市场的机构提供了专业的另类数据服务。本篇文章呈现了数库对于供应链数据在量化投资领域的研究成果，由于针对人群的阅读习惯原因，文章主体以英文呈现。

获取原报告，点击阅读原文

以下为文章正文：

ChinaScope Supply-Chain Data

• Data Source:The supply chain data is constructed on the basis of provisional announcements and periodic reports disclosed by listed companies.

• Coverage Area:Relationships centered on all listed companies on the Shanghai Stock Exchange and the Shenzhen Stock Exchange.

• Related tables in this report: 

    • equity_customer

    • equity_supplier

Table: [equity_customer] - Supply Chain (Customers)

This data table records two things:

1、Major customers as disclosed by the listed company for the period cover by the exchange filing;

2、Customers who have accounts receivables with the listed company, the vintage of which has already surpassed one yea.

Table: [equity_supplier] - Supply Chain (supplier)

This table presents three key elements:

1、Major suppliers as disclosed by the listed company;

2、Suppliers to whom the listed company has accounts payables, the vintage of which has already gone over a year;

3、Suppliers to whom the company has prepayment.

Data Statistics

Data Pre-processing and Statistics

Process the data of equity-customer and equity-supplier by following steps:

The subject means the supplier on the equity-customer table and the customer on the equity-supplier table. 

The object means the customer on the equity-customer table and the supplier on the equity-supplier table.

1. Only reserve the data where the object is a listed company;

2. Delete the data where the object is a listed company on NEEQ ;

3. Delete the data where the object is a listed company on China B-Share ;

*the figure gives the statistics of the number of subject where their object is a listed company(filtered as the up steps)

Data Pre-processing and Statistics

Factor Mining

Factor Mining: Customer-Momentum

Definition:

• Assume that the stock price of a listed company will affected by its downstream companies, which are their customers, to some extent. 

• We defined this factor as customer-momentum:

    • which means the customer’s stock price fluctuation will transfer and affect the supplier’s stock price.

Factor Mining: Customer-Momentum

The Steps:

• All the data is from the equity-customer table:

1. Only keep the data where the customer is listed on China-A, HK, US market;

2. For the failure of the getting the related price data, delete the data where the customer is one of the following company:

    • 'YZC', 'LOXO', 'XL', 'LDK', 'CSUN', ‘ISS'，'8058', '2016', '1819', '3928', '8253', ‘0739’

3. Match the report-date to the report publish date，and set the data-available date as the next calendar day after the financial report publish date;

• At the end of each holding period，get the newest customer list for each subject company, if the customer was released more than two years ago, do not include it on the customer list.

• Calculate the mean return of each subject’s customers for the former holding period，this return is the customer-momentum factor;

• Conduct factor analysis for the customer-momentum factor.

Factor Mining: Customer-Momentum

rate>=4%，IC Series of Different Holding-Period

rate>=4%，long-short performance of different holding-period

Factor Mining: Supplier-Momentum

Definition:

• Assume that the stock price of a listed company will affected by its upstream companies, which are their suppliers, to some extent. 

• We defined this factor as supplier-momentum:

    • which means the supplier’s stock price fluctuation will transfer and affect the customer’s stock price.

The Steps:

• All the data is from the equity-supplier table:

1. Only keep the data where the supplier is listed on China-A, HK, US market;

2. For the failure of the getting the related price data, delete the data where the supplier is one of the following company:

    • 'YZC', 'LOXO', 'XL', 'LDK', 'CSUN', ‘ISS'，'8058', '2016', '1819', '3928', '8253', ‘0739’

3. Match the report-date to the report publish date，and set the data-available date as the next calendar day after the financial report publish date;

• At the end of each holding period，get the newest supplier list for each subject company, if the supplier was released more than two years ago, do not include it on the supplier list.

• Calculate the mean return of each subject’s suppliers for the former holding period，this return is the supplier-momentum factor;

• Conduct factor analysis for the supplier-momentum factor.

rate>=2%，IC Series of Different Holding-Period

rate>=2%，long-short performance of different holding-period

如需以上PDF版本

点击阅读原文

—End—

量化投资与机器学习微信公众号，是业内垂直于Quant、MFE、CST、AI等专业的主流量化自媒体。公众号拥有来自公募、私募、券商、银行、海外等众多圈内10W+关注者。每日发布行业前沿研究成果和最新量化资讯。

你点的每个“在看”，我都认真当成了喜欢