AI For Trading:Alpha Factors Summary (74)

We first propose and generate alpha factors,then evaluate them to find those that might show some promise.

Then we perform out-of-sample testing of the alpha factors using historical data that wasn't used to construct the alpha vector.

If that looks promising,then we would conduct paper trading in which we don't use real money,but we follow the factor as if we're making theoretical trays over time on newly arriving live market data for some period.

If that showed promise,then we would put the alpha into production in a real portfolio with real money.The alpha at that stage would be blended with other alphas and the final alpha vector would pass through a portfolio optimizer.

We would likely start by giving that alpha factor a small weight in the combined vector, and given more weight if it performed and improved portfolio performance.

We would monitor the alpha factor over time knowing that at some point, the factor's usefulness will erode because we're trading in a competitive market. Then, we would remove the alpha factor and go back to the beginning to search for new promising alpha factors.