Alpha Factor Discovery¶
An alpha factor is an expression that returns a vector of real numbers when applied to the cross-section of your universe of stocks. In other words, an alpha factor generates one value per asset per date. In expectation, alpha factor values are predictive of the relative magnitude of future returns. An alpha factor could be built from a rank, or it could be a vector of dimensionless numbers.
Productive alpha research is an iterative process that can be broadly broken into three steps:
- Hypothesize. Create a hypothesis about investor behavior, market structure, information asymmetry or any other potential cause of market inefficiency: for example, earnings announcement timing or political campaign contributions.
- Analyze. Compute the values for your alpha factor over your trading universe, and test whether they're predictive of future returns.
- Revise. Think of ways to improve your factor -- you can try narrowing your universe, adding more data, and more. Return to step #1 and begin another iteration with your potential improvements.
On Quantopian, there are two useful tools for researching alpha factors. First, Pipeline allows you to define and compute alpha factors over your entire trading universe. Once you've computed your alpha factors, you can use Alphalens to analyze them. Alphalens allows you to measure the effectiveness of an alpha factor by comparing factor values to forward-looking returns; it also provides a suite of metrics and visualizations so you can easily analyze the results.
Importantly, Alphalens conducts an analysis without consideration for real world trading limitations like transaction costs, short availability, etc. Instead, Alphalens allows you to quickly iterate on your alpha research, making it easier to distinguish between ideas that you should take to the next step in your research, and those that you can discard right away.
Alpha factor discovery should occur in the Research environment.