Phantom Fills
Vol #175: April 18th, 2024
Variable rewards lead to addictive behaviors.
Refresh the timeline and see who pops up. Turn over the next card. Send in an option order and get back an unexpected price.
Putting PnL aside, is part of the joy of trading not just the confetti on a fill, but the randomly meted dopamine of seeing pennies shaved and given?
There is so much precision that goes into a trade ex-ante, that the plinko board of execution is maddening. It’s a Heisenberg fuzz where you don’t quite know where the price is actually going to be until you probe the depths.
Many a veteran trader will look you in the eye and swear that someone saw his mouse click before the order hit the market. More than just your phone is listening. Part of the market maker conspiracy lore from retail traders is that they know how to front run your fill.
Traders see the complete lack of determinism and it seems to firmly seal the debate in favor of free will.
It is true, as you peel back the layers of microstructure there are courses and processes that are essentially random. There are marginally shorter and longer sticks to be had. But there are rules. The functions return consistent results. The finely tuned processes that define options execution just happen so fast, it can look like chaos.
Sitting at a blinking terminal, there are dozens of switches refracting your perception of the market, and a sensitive limbic system processing it all. But the computers know that when orders edge each other out by picoseconds, the weather in Peoria matters.
The worst possible experience when sending in an order is not getting a fill. Unless you’re spoofing (which is illegal), orders are meant to be executed. A firm commitment to buy and sell at that price. There’s no “cancel if touched” order type, you just got hit.
Posting a limit order away from the opposite side has a different expectation. You’re trying to be cheeky and get someone to pay up. Maybe you want a little bit of maker credit on top. If that doesn’t fill, well then your price wasn’t good enough for a counterparty either.
Trying to lift the offer only to become the bid is gutting. You agreed to a price, stepped so painfully far across your fair value, only to learn that was no longer possible. Desirous taker becomes forlorn maker.
This happens for several reasons, independent of your karma. At a physical level, there is a trade in a matching engine that gets updated in market data, and processed by a machine, which updates an LCD. Today, all that should take less than a millisecond. Customized GPU chips, light speed pipes, the works.
But the dodo behind the keyboard will take at the very least 13 milliseconds (that’s just for light and movement) and up to 50 or even 600 millisecond in the worst cases to process a new number. Don’t scratch your head for too long. Traders have some super-human powers, but dozens of trades could happen at an exchange in the time it takes the best of us to see one.
At the margin, that really does matter. Perhaps not as often as it feels like it does, but for traders operating on a human time scale there are a lot of variables. Even the clunkiest algo code will smoke you.
Once you’ve realized you didn’t make a trade, or traded less than you thought, it's worth checking the tape to see if someone else did. Getting beat is fair game, but what if that liquidity vanished? That’s when you start checking for hidden cameras.
Accessing the markets directly, you may be faced with the need to take liquidity on multiple exchanges. The difficulty here is timing all of those routes. All those milliseconds between datacenters mean randomness when you’re watching the screens, but a well oiled computer can nimbly dodge adverse inbound orderflow.
One of the results of market fragmentation is that liquidity providers are forced to split their resources amongst many venues. If they’re willing to show a given number of contracts at the prevailing price, how should it be split up? If an order comes into one venue, a dealers gotta deal, and wants to trade as much of it as he can.
This incentivizes showing more size across all venues then you’d be comfortable with, and a race to get out of the way if some of it gets taken. Depending on where their servers are, your servers are, and how all the messaging finds a home, an order to buy 100 shown could only land you 50 from the first venue that got hit. The solution to this is closely monitoring round trip times and introducing lags to synchronize the pounce.
Most retail orders don’t directly access the market, and are executed via auctions. In the vast majority of scenarios this results in improved prices. Contractually and economically so. They may come randomly, but there are billions of pennies showered on retail trades.
What makes those figures less assuaging might be the scenarios where it feels like you could have gotten better. On any given order, it may go to one market maker or another to initiate the price improvement mechanism. They may have a particular circumstance that causes them to pass, and the order just goes to the market.
Risk could be a reason for passing. Too much short stock or the head trader is off the desk sometimes dictate interactions, but the scale of a random electronic order compared to the infrastructure and balance sheets of top market makers is not even of the same magnitude.
If it’s insulting that your order fell between the cracks of a broad wall of price improvement, it's downright confusing when you get a better price in an auction than you had just offered in the market. It’s common to walk up a bid or down an offer, but to bid $1.19 on the screens, then pay up to the $1.20 offer and see a $1.17 fill doesn’t make a lot of sense.
The reason this quirk happens has to do with the complexity of fee schedules for different participants. A broker will post an order where they tend to get the highest rebate, but that conversely looks like a fee to a taker, who could lose the better part of a penny of edge. Further, the wholesaling arrangements only relate to marketable flow. You get no credit for carping the bid/ask on a midmarket order, so save your bullets for elsewhere.
Each player in this system has a cold hard logic. Exchanges have stamped out every bit of irregularity in their market access, whether it’s spooling cables or designing high performance systems with perfect determinism on order sequencing. Unfortunately that doesn’t necessarily make any specific interaction more predictable in this domain.
Fill prices are the first piece of feedback you get on a trade, but it shouldn’t be the most significant. There are plenty of pennies to shave by being tactical about execution, but it’s not an edge for anyone trading less than 100k contracts a day. So don’t sweat it, it shouldn’t be the reason you came to the market.