How Informative are BBO Cancels?

Two weeks ago I looked at the likelihood that an order on the BBO would be canceled as a function of its queue position. We saw that orders with good queue position (in particular with many orders behind them) were less likely to get canceled. This demonstrates that the order owners do consider the orders' queue value in their cancel decisions.

We can take this one step further: Suppose you observe an order with good queue position being canceled. Surely the owner must have a good reason for doing so! Can we use this as a predictive signal?

Fig. 1 below shows how canceled orders on the BBO would have marked out if kept until filled. The x- and y-axes are the number or orders ahead and behind at the time of the cancel, respectively. Apologies for the color scale - red indicates a positive markout vs. the T+10 s mid price.

The total number of orders on the BBO increases towards the upper right. The number of orders behind the canceled order increases towards the top. The best orders (which are near the front of a long queue) are in the top left corner of the plot.

Remember that these orders rarely get canceled. But even if they do, it does not mean that the market is about to move against them. On the contrary, they would have done quite well even if left "unattended" until an eventual fill (note that keeping the order means retaining the option to cancel it at a later point).

For Fig. 1 I weight each cancel equally, i.e., the average markouts ignore the quantities of the canceled orders. I leave it as an exercise to the reader to see if possibly adding the order volume yields a good signal.