Finding Alphas, 2nd Edition

4Alpha Design

By Scott Bender and and Yongfeng He

This chapter will lay out the process of designing an alpha, starting with raw data. We will discuss some of the important design decisions you need to make when creating an alpha, as well as how to properly evaluate an alpha. At the end of this chapter, we will highlight some issues that can arise after alphas have been developed and put into production.

DATA INPUTS TO AN ALPHA

Alphas are fueled by data. The edge sought for an alpha may come from identifying high-quality pieces of publicly available data, superior processing of the data – or both. Some typical data sources are:

• Prices and volumes. Technical analysis or regression models may be built based on this data.
• Fundamentals. By automating the analysis of key metrics for each company, you can build alphas that typically have very low turnover.
• Macroeconomic data, such as GDP numbers and employment rates, that have market-wide effects upon their release.
• Text, such as Federal Open Market Committee minutes, company filings, papers, journals, news, or social media.
• Multimedia, notably relevant videos or audio. There are mature techniques to process such data – for example, converting audio into text that can be used to build models.

Sometimes data sources aren't used to generate a directional signal but to attempt to reduce noise in predictions and refine other alpha signals. Examples are:

• Risk factor models. By controlling risk exposure or eliminating exposure to some ...