Yuval Be careful with passive NBBO pegs, this can cause you to get toxic fills
Here are some details of: Fidelity’s price improvement statistics.
This includes SP500 vs non-SP500 orders and breaks down price-improvement by order sizes which people have expressed an interest above.
This is produced by the Financial Information Forum (FIF). Sadly, Fidelity and 2 Sigma Securities are the only organizations providing information to FIF at this time (there used to be more).
I’ve skimmed the study, and it’s interesting for sure, but there are several reasons why I would take it with a grain of salt and not rush to move all your trading accounts over to TD Ameritrade tomorrow.
-
As Matt Levine noted in his column, the trade sizes of $100 are not just small, but comically small. If one were only ever managing position sizes of $100 with market orders or marketable limit orders, then this study may be pretty representative for them. But even in the Fidelity price improvement report linked above, you can see the price improvement metrics drop as order size increases.
-
Their program is designed to randomly send buy orders for and then liquidate 30 minutes later. This randomized order flow is the definition of uninformed order flow that PFOF providers love and will happily eat up. The more informed, alpha-driven flow would likely not get the same beneficial fills from PFOF participants.
-
They are not doing a fully apples-to-apples comparison across the same date range. Transaction costs can vary greatly over time, so this is a bit of a sin. It looks like they ameliorate this a bit by providing some pairwise broker comparisons across the intersection of their common dates, but as a whole the data consists of different date segments. For example, their IBKR Pro and Lite data is completely disjoint, as it looks like they had a single IBKR account that they just flipped from Pro to Lite in April, so the IBKR data isn’t comparable at all.
-
It looks like they are estimating price improvement metrics and on vs off exchange execution by trying to locate their orders in the TAQ data stream. This seems to be an error prone and non-scalable approach. For example, if they were sending more realistic round lot orders, it would be impossible to pick out their orders reliably from the stream. Likewise, calculating price improvement using the quote that was in effect just before the trade in the TAQ data would skew your results depending on broker latency. The more standard way would be to benchmark the order to the quote at the time of submission or routing decision.
In fact, that’s what I would recommend each person do, if possible – measure their total transaction costs for their own order flow. If you use IBKR, they provide fairly decent transaction cost analysis (TCA) reports for you. Moreover, if you send algorithmic orders, then the cost of the entire order relative to the quote at the time your order was accepted (its arrival price) is shown in the report. It’s not as clean if you are slicing and sending your own limit orders, as the order linkage may not be clear, and the market impact of your manual order working may not be measurable to them – you’d likely have to measure this yourself.
This is another reason I’d encourage P123 to expand their support for the full range of algo orders on IBKR. Arguably the biggest industry standard, the Arrival Price order, is not yet supported. And perhaps more importantly, IBKR supports algo orders for other brokers, notably CSFB, Fox River, and Jefferies, who are well-established in the execution space. Having this wider order support, and the ability to randomize order types would be a huge step up.
Almost no, and I mean literally zero, retail market orders have legal informational edge over the market makers. The SEC legalized the implicit pricing of how dumb retail orders are leading to PFOF. In other words, whoever (Jump, Citadel, Jane St. etc.), can extract the most value from how mispriced your order is gets to fill your order. Why else do they pay the brokerage house for the right to fill your order? As soon as the market maker deems your order to not be mispriced (like, when they think you may win on the trade, the let it go to the exchange for a worse fill). So you should think of it as losing less for only market orders.
In regards to IB, Petterfy rightly believes that minimizing trading costs must lead to routing everything to a central exchange for auction. Of course, we don’t have a central exchange and PFOF routes less to exchanges, which widens the quotes, which artificial leads to better price improvement. Why does this toxic cycle exist? Because that maximizes market maker profits in the age of digitization and penny wide quotes.
TL:DR Don’t place market orders. But if you do, place them at brokerage houses that sell your orders for internalization (i.e never go to an exchange).
Korr,
Thank you. Can you expand on this for “marketable limit orders” and also on VWAP orders (and percent-of-volume orders) if this has been an interest of yours? I think a VWAP order is essentially multiple “marketable limit orders” made throughout the day.
I think a marketable limit order MAY get price improvement through internalization at some brokers but also there could be a rebate for being a “market maker” for some of the limit orders (the broker getting some of the rebate) and this could be a factor in where the order is directed. Any data on this in the literature or on the broker’s web sites is limited. E.g., the original paper does not directly address this.
Thank you in advance if you have anything to add with regard to “marketable limit orders” or algorithms like VWAP that may involve a marketable limit order.
Best,
Jim
Hi Jrinne,
I have no experience with marketable limit orders so I can’t speak to personal experience. The reason I don’t is because every time one crosses the spread, one overpays – paying a convenience (aka liquidity) fee to the market makers.
I have experience and some knowledge with VWAP orders. This is a highly competitive market as VWAP ex ante is theoretical - that is the benchmark by which the order is measured. So anyone who has designed a VWAP algo is continually improving the mechanics of the algo to remain competitive. I believe, but cannot be certain – all these mechanics are held private – all utilize a variety of order types. The idea is that you want to trade when volume is heaviest as you trade against customers, not dealers, so overall costs are minimized. Since book depths are notoriously misleading , one must minimize order size all else equal, resulting in having multiple round lot orders.
What’s the best way to get filled and minimize both moving the spread and crossing it? The consensus has been and continues to be VWAP. This is not to say other order types, including algos but excluding market orders, are worse. Just depends on what you are you trying to achieve.
Hope that is informative for you.
Korr,
Thank you. I get what you are saying about sending a limit order to an exchange (and internalization). I appreciate your further insights.
Jim
Thomas Peterffy–the CEO of IB–just sent out an email regarding trading costs. Among other things, he explained that when a marketable order trades with a limit order on IBs dark pool, the price will favor the limit order. That’s by design.
In other words, the study that I quoted to start this thread is not applicable to limit orders (or any other non-marketable order) at all.