Sniping makes your trade look fat
In an ideal stock market, all people who want to trade would meet a single location to arrange their trades. In the old days, the trading floors of exchanges such as the NYSE were such a place. Over time, however, people got weary of the NYSE and its outsized influence and asked for competition. They got it: trading of entirely fungible assets today is spread over dozens of locations. But — in today’ digital world, computers should be able to integrate the different markets seamlessly, shouldn’t they? That is the idea behind U.S. regulation “National Market System” (NMS), which established rules to generate a virtual national market. The problem is that it is fundamentally impossible to integrate geographically separate marketplaces seamlessly. Why? Because even at the speed of light, information and data take time to travel between the data centers that host the trading systems. For traders, this presents an enormous challenge, and, allegedly, for some even an opportunity.
The influence of the geography of markets has been a hotly debated issue in the trading community and the popular press, especially after Michael Lewis described in “Flash Boys” how orders that “arrived at different times at different exchanges” could be “front-run.” Yet the genuine statistical evidence on the impact of geographic market fragmentation is scarce. More than that, we know very little about how market participants trade across multiple trading venues.
Katya Malinova and I set out to study this issue (click here for the full text), and we are now, after years of work, finally ready to share our findings.
Our first goal was to study trading behavior in a multi-market setting and we were particularly interested in trades that use multiple markets. Thanks to the data that the Canadian regulator IIROC gave us access to, we could string together trades across multiple marketplaces, something which we believe hasn’t been done before. A first striking observation is that there is a fast and robust reaction by other market participants to multi-market trades. Within 50 milliseconds, 45% of trades are followed by more aggressive orders from other traders, meaning that, for instance, buys follows a buy. There is also a flurry of cancellations and order re-submissions across all markets. These multi-market trades move the price twice as much as single-market trades. And it’s not because these trades are likely driven by institution — even multi-market orders retail investors have twice the price impact. Size isn’t the issue either — similar-sized single market orders have half the price impact. In other words, on the tape, multi-market trades look much bigger than they are.
Katya’s and my first instinct was to see if there is something special to multi-market orders. When we looked at the data carefully, however, we noticed something else that was rather striking: about 2/3 of the follow-on trades come from a group of just fourteen traders (of almost 3,000). Circling back to the trades, when we control for any number of features, it turns out that these snipers’ presence is the most significant marker for a larger price movement. If they aren’t there, there’s almost no difference between multi- and single-market trades.
Other authors have called these quick-moving traders snipers, so we adopt that term here, even though we don’t quite like the implied judgment of malfeasance. Indeed, everyone please take a deep breath —our findings are not a smoking gun that shows that these snipers cause an excess price impact (and thereby make trading more expensive, in particular for institutions). The snipers may be good at identifying the trades that matter, that will move the price, and they are just faster at reacting to such trades.
Teasing out the causality is a pretty hard problem. What we’d love to have is two situations: one with and one without snipers. Although we don’t have such a setting, in our data, there was a market organization change that exogenously made it harder for these snipers to react.
What happened specifically was that one of the main markets in Canada moved into the same datacentre and, as we understand it, onto the same system as another main market. The result is that it became much easier for “normal” traders to reach both these markets at the same time. That this is true is visible in the data as many more trades happen on two markets at the same time, up to the millisecond. We also observe that there are fewer follower trades by the snipers; it’s not a huge drop, but it is significant.
Now, how can we identify the snipers’ effect? First, the overall price impact of trades that are not followed by snipers declines (though the effect is statistically weak). That’s unlikely to occur because there are fundamental changes to these trades — most investors had no idea that there had been a change. Instead, we believe that this indicates that the absence of the snipers lowers the price impact. Trades that are followed by snipers exhibit no changes in their price impact. The last observation suggests that when fast traders are present, they continue to create the same effect, and it further supports the view that the event did not affect the informativeness of “end-investor” trades. Taken these findings together, we interpret them as evidence that the snipers cause some of the excessive price impacts.
So there you have it. Snipers cause at least a bit of a larger price impact. It’s not clear how much, and it’s not clear that they aren’t also better at detecting trades that will move prices eventually. But, as they latch on to your activity, they do make your trade look fat.
