How high frequency trading strategies work in reality is a rare discussion. Ran into a fresh website explaining in detail how strategies work in identifying low alpha orders in the market. How to make money with a strategy. Distinguishing humans from robots etc. Of course real valuable secrets from the HFT kitchen won’t be shared, and it’s a simplified post. Not always an easy read, but good to get an idea. If it’s your thing, there are a few more posts to find : mechanicalmarkets.
Following is a guest post from Mechanical Markets. The author has been a cook in the HFT kitchen, so knows what he’s talking about.
Protecting Client Interests: Anonymity in US Equities
Anonymity is the top concern for many market participants and the entire order handling chain has a responsibility to protect traders’ identities. Still, there are plenty of exceptions to anonymous trading in US equities. When is it OK for brokers and exchanges to reveal potentially client-identifying information? In this post I will discuss a few situations where trading is not fully anonymous, briefly outline an HFT strategy that relies on one of these situations, as well as ask some related ethical questions.
Many exchanges allow Broker-Dealers to optionally identify their orders to the entire market (e.g. Nasdaq). In the previous posts, we examined order characteristics in an effort to identify desirable trading partners. In keeping with that spirit, the first thing we’ll explore is the performance of Nasdaq trades identifiable by counterparty MPID (Market Participant ID).
Here is a chart of trades’ performance grouped by MPID relative to the market price. The MPID with by far the most trades is UBSS (UBS Securities).
The orders marked UBSS have dramatically poorer performance over the long term than those marked anonymous. This tendency is robust across stocks with different liquidity profiles (e.g. price, spread, volume). We’ve discussed in earlier posts why the supposedly efficient market has not eliminated this type of disparity.
A gut reaction might be: “Wow, UBS is handling their customers’ orders poorly if they lose that much money when they trade.” I don’t know on whose behalf these orders are generally sent, but it’s a reasonable guess that they’re retail-type clients. Before you say: “typical bank, screwing over the least sophisticated customers,” I’d argue that UBS might be doing exactly the right thing as far as their clients are concerned. If these are orders from less sophisticated individuals, they might well be limit orders with the price chosen by the client instead of the broker; in which case all the client wants is to maximize their probability of being executed at the price submitted, without regard for what happens in the minutes afterward.
By posting them to Nasdaq with the MPID field populated, UBS is essentially putting a giant “hit me” sign on these orders, which should get them a better fill rate, even if the rest of the market is making money off them. Counterintuitive as it may seem, what would be more concerning would be if these orders, instead of performing poorly, actually performed unusually well relative to the market. That would imply that their broker was leaking information to the market by flagging these orders with their MPID, presumably not in the customer’s interest.
Ah, you ask, but this refers to the order’s alpha only when it trades, what if it actually has a poor fill rate and as soon as it’s posted to Nasdaq the price starts to move away from it? Just to be sure this isn’t what’s happening, here is a chart of the market price around the time of submission of said orders, when they are submitted within 10 cents of the NBBO. [1]
UBS hasn’t exactly earned itself a sterling reputation lately (Libor, equities, FX, taxes), but on this one I’m willing to give them the benefit of the doubt.
Unlike Nasdaq, DirectEdge does offer a financial incentive for firms to display their MPID to the market (see “Edge Attribution Incentive Program” here). Unfortunately, DirectEdge charges extra for a subscription to its MPID attributed order data, so I don’t have it available to repeat the analysis for EdgeX. But I would be very curious to see if orders associated with any particular MPIDs have unusual properties, particularly if they perform well versus the market. If so, that would suggest (and of course this is speculation) the possibility that a broker is willing to leak market-impacting client data in exchange for the revenue offered by DirectEdge. Again, I’m not saying this is happening on Edge, but this sort of incentive program is fraught with danger and I can’t help but wonder why it’s even offered. [2]
A Trading Strategy
I won’t bore you with too many details, since we’ve already seen strategies that capitalize on analogous low-alpha orders. The basic idea here is that when we see a UBSS order trade, we send orders trying to trade with any quantity remaining. The below simulation simply does that, sending 1000 share orders on stocks above 30$/shr to mop up the quantity available. Note that the profit margins are about the same if we send 100 or 1000 share orders, which does hint that these (typically large) UBSS orders come from retail clients.
The strategy makes about 100 mils after fees on 4000 trades per day with an average trade size of 250 shares, netting 10k/day (assuming minimal exit costs). Another 5k/day can be had by sending identical orders to EdgeX. And there’s potential for more; the volume can be scaled much higher by submitting the orders at a more aggressive price. This strategy is somewhat latency-sensitive and there’s a noticeable uptick in volume 50us after a UBSS order executes, suggesting that someone is wise to the MPID feature.
The previous posts focused on order properties that could signal the identity of their senders. Orders displaying fewer HFT-like characteristics make for desirable counterparties. Now we see that knowing the broker’s identity also can tell us a great deal about the client. Imagine how much we might make if we knew the broker for every order we encountered!
Ethical Questions
These examples are not the only exceptions to anonymity of the markets. Certainly some dark pools openly advertise that they filter out toxic counterparties or allow users to choose which sub-population of the pool they’d like to interact with (and we’ve seen Barclays get into trouble over their claims of this kind with respect to their dark pool). It’s my understanding that internalizers can see some identifying characteristics of incoming orders. The general idea behind that identification is that it can result in clients receiving a better price on their executions. A market order sent via Charles Schwab may look a lot like a noise trader to Getco and could get executed near the midpoint; another order sent from Goldman immediately afterwards may scare them and won’t get executed at all.
This does sound like good client service, but there are a few issues that deserve consideration. Instead of passing on their market-making revenue to the Schwab client through NBBO improvement, what if some of it were diverted to Schwab itself via payments for order flow? In that case Schwab might have an incentive to share information with Getco that might not always be in the client’s interest. Could Schwab reveal anything that might indicate to the internalizer that the order was sent not from your grandparents’ IRA, but by an executive exercising stock options? Remember from our examples that even mundane information like a timestamp or size can clue us in to who might have sent an order, so Schwab could even unwittingly transmit sensitive data. I have no reason to believe Schwab does so, but it’s a thought that could make investors uncomfortable. [3]
This hypothetical brings us to the definition of front-running. Is it front-running to sell client information to outsiders? If selling the information helps the client, then it seems like the answer should be no. But what if it doesn’t? Let’s say hypothetically the following happens:
- Broker has clients that tend to continue trading on the same side after their first order
- Broker forwards such an order from Trader A to DirectEdge.
- Broker tells DirectEdge to expose it’s MPID to the market. DirectEdge pays broker.
- Trader B pays DirectEdge for the MPID data.
- Trader B sees the newly added order and, anticipating the price to move away from it, sends other orders that degrade the fill quality of Trader A’s current or future orders.
It doesn’t sound so different from a broker calling up a friend upon receiving a client’s instructions, the friend using that info to make money, and then handing the broker a briefcase of cash. I’m not a lawyer, but if something akin to the above occurs it would reflect very poorly on any involved parties who understood the consequences. I’ll stress one more time that this scenario is hypothetical; for all I know DirectEdge doesn’t even have subscribers to their attributed data feed.
One final point about the broader consequences of reduced anonymity in markets: even if clients’ interests are looked after meticulously, de-anonymization of traders can result in a two-tiered market with heavy fragmentation. Because noise traders can receive better prices with their identities partly revealed, orders remaining fully anonymous will tend to come from large institutional traders. You could argue that this is exactly what’s happened to today’s markets; retail flow goes through internalizers and institutions are left to fight each other elsewhere, with HFT connecting the two.
[1] “NBBO” here means an approximation of the real NBBO, essentially an estimate based on a subset of exchanges. Note that orders with the UBSS MPID do exhibit significant toxicity if they are submitted at prices more aggressive than the NBBO. Perhaps this is a sign of information leakage coming from the MPID, but given the large size of these orders (~10 times the size of a typical Nasdaq order that improves the NBBO) and their very high fill rate near 80%, I don’t find that thought particularly convincing. The posted plot includes these orders in its averages.
Source : mechanicalmarkets.wordpress.com