Several weeks ago the SEC implemented the new single stock circuit breakers which temporarily halt a stock if its trade price moves more than 10% over a 5 minute period. Astute observers have already noticed a few cases where stocks have been halted due to erroneous prints - most notably Citigroup (C) on June 29th, and APC on July 6th. C was halted when it was trading with an inside market (NBBO) of $3.79 - $3.80, and an erroneous print of 8820 shares @ $3.3174 triggered a halt. APC was halted when it was trading $39.14 - $39.15, and an erroneous print of 200 shares at $100,000 triggered a halt.
Back when the SEC asked for initial comments on this issue, commenter Peter mentioned in my comments section and in a letter he sent to the SEC that trading curbs would be better served if they were implemented off of changes in NBBO (national best bid offer) rather than trade price. At the time, I was kinda burnt out with the whole thing, and didn't take the time to think through his points, but the recent halts based on erroneous trades highlight the message Peter was trying to get across.
First, let's look at what the circuit breakers are trying to accomplish: the goal is to avoid May 6th type "flash crashes" when liquidity dries up. However, the result of the implementation of the rules has been that erroneous prints are triggering halts. Let me define "erroneous print." Every trade in the equity markets gets reported, or "printed" to the "tape." Ticker, shares, price, and some other additional information. Much of this reporting is automated, but there are still situations where trades are reported "by hand." When I was in the business, we had an "ACT Terminal" on our desk, and if a client came in to buy 100,000 SPY, which we sold to him, someone would walk over to this machine and manually key in the details to the ACT Terminal: SOLD 100,000 SPY @ 110.56, include any additional modifiers regarding early settlement, or perhaps that the trade was an average price trade occurring over a period of time, and then hit SEND. Everyone could then see the trade scroll by on the ticker tape (digitally!).
Of course, if the person typing in the trade details had "fat fingers," he might accidentally type 1000000 SPY, adding a zero, or perhaps 11056, missing the decimal point, and a bad trade would hit the tape, which would have to be corrected.
Nowadays, although I believe ACT Terminals still exist, most of this reporting logic, even when entered manually, is entered from a trader's order management system, which is tied into a tape reporting facility.
Why does all of this matter? Because what happened in the C halt on June 29th is very different from what happened to ACN on May 6th. To review, on May 6th in ACN, sell orders without restriction on price overwhelmed buy orders, and every bid was smacked, all the way down to 1c. On June 29th, in C, this is very much NOT what happened - there were $3.79 bids the entire time - in other words, the order book didn't get "swept" - it was simply a bad price entered in the reporting of a trade. Unfortunately, this bad report still triggered a trading halt. (note: FINRA confirmed that this trade was manually reported by a broker dealer)
Similarly, with APC, you did not see every single offer up to $100,000 lifted. On the contrary - a bad 200 share print at $100,000 triggered the halt. It's important for readers to understand the difference between a "bad print" that results in a hiccup in a stock chart, and an order which results in real trades that decimate a thin order book - which is to say, an order which is to buy more shares than are offered in aggregate, or to sell more shares than are bid for in aggregate - that is what happened on May 6th. It's possible to have a "fat finger" error on a real trade too - that happens as well - although I don't think it's what happened on May 6th.
So, we see that trading halts are being erroneously triggered. Peter shares his concerns in another comment letter he sent to the SEC, in which he analyzed data from the last 18 months and determined that "Analysis shows that the current pilot circuit breaker logic would have triggered incorrectly on 158 securities at least 238 times in the past 18 months." He goes on to argue that "Erroneous halting of certain highly liquid and systemically important ETFs may further destabilize the markets. The analysis shows that SPY would have been erroneously halted twice in the past 18 months, potentially destabilizing the markets."
Another problem with the current circuit breakers is that anyone (with trade reporting capability) can effectively halt any stock any time they want just by throwing up a bad print on the tape! Talk about the ability to manipulate markets... This is definitely not a good thing, and further undermines fears people have about market veracity. So how do we fix it? As Peter suggested initially, one way is to make the halts based on the NBBO, rather than the last trade. So, since C was $3.79 bid the entire time when the erroneous print was reported, the stock would not have been halted, since the NBBO did not move enough to trigger a halt. If the computational power to maintain a constantly rolling evaluation of the 5 minute change in NBBO is too great, I think the same effect could be attained if the reference NBBO base price was updated occasionally, say, every 5 or 10 minutes (instead of continuously), whatever frequency data processing capabilities allow.
Perhaps another option would be to treat manually reported trades differently than trades reported based on automated logic. Although it's certainly possible that bad logic or bad data could result in erroneous prints from automated methods, it seems that the likelihood of such erroneous reporting is lower.
Finally, there has been some discussion of having individual stock limits - like we currently have for futures markets - limit up and limit down. Japan actually does a form of this with their individual stocks. I don't really like this idea because any limits would have to be symmetrical on the downside and the upside. If we decided to set a down limit of 20% for individual stocks, then we'd also have to set an up limit of 20% - this would be problematic on the upside in the case of announced mergers with large premiums, or on the downside in cases of big lawsuits or failed drug approvals for biotech stocks.
In any case, I think it's clear that some relatively simple changes need to be made to the current circuit breaker rules so that erroneous prints do not disrupt trading.