The TRADE 63 - Q1 2020 | Page 20

[ I N - D E P T H | T C A ] by considering certain factors such as the percent- age of the average daily trading volume, spread size, expected impact, and volatility when orders are routed to them. Comparing like-for-like performance is another important part of the process, therefore all data points are separated, and similar brokers and flow are grouped together. When comparing similar brokers, the analysis goes beyond which broker out- or underperformed the benchmark the most. All contextual metrics are measured, including spread capture, fill size, momentum, impact, and reversion, for example, in order to paint a fuller picture of broker performance. A weighted ranking score system is then pro- duced to reveal the broker winners and losers in the analysis, before trends or potential outliers that are driving the ranking are identified. For the losers, Hermes consults with the brokers to verify the results of the analysis and work towards preventing any continued poor performance. If this low scoring occurs over a small sample size, the broker is placed on a watchlist, which is closely monitored until the results are either proved or discredited. A desk re- view is also undertaken in circumstanc- es where there are difficult market conditions or flow. The process for high-touch and broker analysis is similar to the analysis carried out for algorithms, but Hermes com- 20 // TheTRADE // Spring 2020 pares algorithms side-by-side for actionable insight not only for selection, but for informing any changes made to the algorithm itself. If the algorithm is tweaked in partnership with the broker,