What the broker's execution speed number leaves out
"Average execution: 28 milliseconds." Sounds great. Here is what is missing:
The tail. A broker filling 95 percent of orders in 10 ms and the rest in 600 ms still posts a pleasant average. The slow tail is exactly where trouble hides, since slow fills tend to coincide with fast markets.
The starting point. The clock usually starts when the broker receives the order and stops when it passes it on, not when the trader sees the confirmation. The trip from the client machine and back goes uncounted.
The conditions. Figures are often taken from quiet periods. News and rollover rarely make it into the average.
The sample. Some brokers measure a subset of orders, such as market orders under a certain size.
A better view comes from personal logs. Timestamp the click and the confirmation, and look at the 95th and 99th percentile times rather than the mean.
A good question to put to a broker is how they would explain the gap between average and worst case. A vague answer suggests the numbers are marketing. A specific one, with percentiles and sample sizes, suggests someone has actually looked.
Another useful check is the time of day. Execution speed during a quiet Asian hour says little about speed during the first minute of the US session. If the broker's figures do not say when they were measured, assume the quiet hours.
Personal logs beat any published figure. A one month record of click time, confirmation time, and the quote at each moment answers the question for the specific account, pairs, and hours that matter, and costs nothing but a few columns.
There is also the question of what counts as an executed order. If rejected orders and requotes are excluded from the statistics, the figure describes only the successes, which is the flattering half of the story.
What does the 99th percentile fill time look like on your account compared with the published figure?
What the broker's execution speed number leaves out
What the broker's execution speed number leaves out
It’s Fairman 