Finding Alphas, 2nd Edition

15Alphas from Automated Search

By Yu Huang and and Varat Intaraprasonk

“Change is the only constant in life,” the Greek philosopher Heraclitus wrote some 2,500 years ago. His words are especially relevant to today's financial markets.

We live in an age of information explosion. With the exponential growth in new sources of data, it is becoming impractical to test all data manually. To tackle this problem, computer algorithms can be used to facilitate the search for alpha signals within the huge data cloud. This computer-aided method, called automated alpha search, can significantly boost the efficiency of the search for signals, producing thousands of alphas in a single day. This comes at a price: not all of the signals found are real alphas. Many of the seemingly great alpha signals discovered by automated searches are noise fitted to the in-sample historical data and have no predictive power. Thus, the focus of any automated alpha search is avoiding overfitting to improve the quality of the output signal. This chapter reviews the process of building an automated alpha search system.

EFFICIENCY AND SCALE

The main focus of an automated search is to find a large number of alpha signals from an exponentially larger number of combinations of inputs and functions. These combinations can also be recursive; combinations of combinations can generate novel alpha signals. Such complexity makes efficiency one of the foremost concerns in an automated search. This chapter presents the ...