The TRADE 60 | Page 28

[ T H O U G H T L E A D E R S H I P There are several other features that we can construct to help de- scribe the trading strategy. The key is to have as few features as possi- ble while maximising information capturing, meaning we need the important features, but we want to get rid of the superfluous features. We can use some ML tech- niques to decide on which features are important, so it is better to try as many features as possible, while only keeping those with the best explanatory | L I Q U I D M E T R I X ] power. There are other things we can use these features for. As an example, maybe we want to compare the trading strategies of two different algos, or how an algo’s trading strategy may vary for securities that have different liquidity characteristics. We don’t have to use ML to get value out of features like these, but the features are very useful if and when we decide to implement an ML program. By computing features like Figure 1: The set of four charts on the left show 4 trading strategies. The top left shows a front-loaded strategy where volume is traded quickly at the begin- ning of the order and then the pace of trading slows down as the order progress. The top right strategy shows a back-loaded strategy where the pace of trading accelerates as the order is completed – like trading a ‘Close’ strategy. The bottom right strategy is uniform in time where trading occurs at a constant pace and the volume profile is symmetric in time. The bottom 28 // TheTrade // Summer 2019 these and then curating them, we can build up libraries of analytics that can help us not only apply ML algorithms to find patterns in our data, but also to compare how those analytics vary through time as market structure evolves. It is important to keep in mind that when our order is being executed in the market, the market doesn’t care if the algo had a label, it cares instead about the orders that interact with the order book and potentially become fills. left strategy is symmetric but trades at a variable pace. The charts on the right show the unfilled portion of the order over time. The area under the curve represents the ‘exposure’ of the unfilled shares to opportunity risk. The top left has a low value, the top left has a high value and the bottom row are both intermediate. The bottom right shows a smooth profile while the bottom right is rougher. Our second metric, ‘roughness’ characterises the difference of the trading strategy from a smooth strategy.