Algorithmic trading strategies are driven by signals that indicate when to buy or sell assets to generate positive returns relative to a benchmark. The portion of an asset's return that is not explained by exposure to the benchmark is called alpha, and hence these signals are also called alpha factors.

Alpha factors aim to predict the price movements of assets in the investment universe based on the available market, fundamental, or alternative data. A factor may combine one or several input variables, but assumes a single value for each asset every time the strategy evaluates the factor. Trade decisions typically rely on relative values across assets. Trading strategies are often based on signals emitted by multiple factors, and we will see that machine learning (ML) models are particularly well suited to integrate the various signals efficiently to make more accurate predictions.

The design, evaluation, and combination of alpha factors are critical steps during the research phase of the algorithmic trading strategy workflow, as shown in the following diagram. We will focus on the research phase in this Chapter 4, Strategy Evaluation, and the execution phase in the next chapter. The remainder of this book will then focus on the use of ML to discover and combine alpha factors. Take a look at the following figure:

This chapter will use a simple mean-reversal factor to introduce the algorithmic trading simulator zipline that is written in Python and facilitates the testing of alpha factors for a given investment universe. We will also use zipline when we backtest trading strategies in a portfolio context in the next chapter. Next, we will discuss key metrics to evaluate the predictive performance of alpha factors, including the information coefficient and the information ratio, which leads to the fundamental law of active management.

In particular, this chapter will address the following topics:

• How to characterize, justify and measure key types of alpha factors
• How to create alpha factors using financial feature engineering
• How to use zipline offline to test individual alpha factors
• How to use zipline on Quantopian to combine alpha factors and identify more sophisticated signals
• How the information coefficient (IC) measures an alpha factor's predictive performance
• How to use alphalens to evaluate predictive performance and turnover