Skip to main content

Goldman's H-Through-L Option Trading Computer Had A Rough Day Yesterday

  • Author:
  • Updated:

Yesterday the computer at Goldman Sachs responsible for trading options whose symbols start with the letters H through L traded a bunch of options at the wrong prices and put Goldman out by a hundred million dollars or so. Today various exchanges are sitting down and pondering whether to give Goldman that money back. This strikes some people as unfair because, y'know, hahaha Goldman you screwed up, but also because someone was on the other side of those trades, made a profit, hedged it out, and will now be sad and possibly screwed when it is unwound.1

Which all seems pretty justified, and the image of a bunch of exchange operators getting together and being all "better cancel these trades, it's Goldman, don't want to make them mad" is in fact disturbing. However! That is not apparently what is happening. From Bloomberg:

The [NYSE Amex Options] exchange said decisions would be made using its rules for obvious and catastrophic errors, which are based on how far the price of a contract deviates from previous trades. If they don’t qualify as an obvious error, U.S. exchange may be reluctant to cancel them, according to Neal Wolkoff, former chief executive officer of the American Stock Exchange.

“Since everyone is under scrutiny and you want to be even-handed, the exchanges tend to be pretty wary about going outside the standard benchmarks for what’s considered erroneous trades,” he said in a telephone interview today.

Now Neal Wolkoff could be wrong, of course. We've seen a number of recent SECactions pretty much to the effect of "sheesh, exchanges, stop favoring big customers by bending the rules for them." I'm with Wolkoff that that sort of scrutiny - and the whole headlines-about-Goldman thing - means that the exchanges here will probably follow their rules to the letter, and do Goldman no extra favors, but you could of course read the facts the other way and say "boy, exchanges really really like doing extra favors for big customers," and I could not fault you for that.

What are the rules? Well, you can read them, here is NYSE Amex Options's rule 975NY.2 Basically if the price of a trade is away from the previous national best bid (for an erroneous sell) or offer (for an erroneous buy) by more than some fixed amount - 25 cents for sub-$2 options, $1 for above-$20 options, in-between amounts for in-between amounts - then the trade is adjusted to be that best bid/offer minus/plus a stupidity penalty of 15 cents (for sub-$3 trades) or 30 cents for (above-$3 trades), unless both parties agree to unwind it instead. (If either side is not a market maker, the default is to unwind the trade.) The rules are perhaps less mechanical than you'd like in a perfect world, but they're pretty mechanical; they rely on big jumps in price, not on Goldman saying after the fact that what it did wasn't what it meant to do.

Here is Bloomberg again:

About 240 September $103 put contracts for the iShares Russell 2000 Exchange-Traded Fund traded at $1 at 9:32 a.m. New York time today, down from as much as $3.32 two minutes earlier, data compiled by Bloomberg show. The next trade was executed at $3.27 at 9:33 a.m.

Sounds like an obvious error! Sounds like Goldman should get a do-over at, I dunno, $3.02 or $2.97 or something like that.3

So how do you feel about that? One weird thing about computerized fat fingers is that really they seem like they should be solvable by redundancy. Like: Goldman had a computer, and the computer's job was to bid and offer for options, and it offered options at ludicrous prices. Bad computer! Bad Goldman!

But other people had computers, and those computers' job was to buy and sell options, and those computers lifted Goldman's computer's offers at ludicrous prices.4 Prices that were so ludicrous, in many cases, that the exchange rules explicitly and clearly provided that the trades would be unwound or repriced. Now: should we be mad at those people and their computers? Meh, not really; it's psychologically difficult for me to fault people (or computers) for just picking up free money, even if there's a pretty good chance they'll have to give it back.5 But on the other hand we shouldn't be too solicitous of their well-being in unwinding these trades. If you program your computer to buy options from Goldman at prices that will clearly be unwound, then that's probably just good business. If you program your computer to buy options from Goldman at prices that will clearly be unwound and then spend all your profits,6 then I can't help you. Just don't do that!

Other redundancies would also help. Like: my understanding of these options exchanges is that they are, y'know, basically computerized order matching systems. Maybe just don't match orders that will be broken up as clearly erroneous?7 A theme of the rise of computerized trading, and of the SEC's recent actions against the CBOE and Nasdaq over their rule-bending behavior, is that exchanges are being pushed to become more mechanical: less a club of gentlemen making ad hoc decisions about good market practices, and more a series of mechanical rules that can be applied impartially and predictably to every trade. Once you've got erroneous-trade rules that are so mechanically specified that a computer could apply and predict them, then having a computer prevent bad trades in advance seems like it might work better than having a committee unwind them the next day.

Goldman faces losses on erroneous trades [FT]
Goldman Sachs Said to Send Stock-Option Orders by Mistake [Bloomberg]
Goldman Options Error Shows Peril Persists One Year After Knight [Bloomberg]
Canceling Erroneous Trades: You're Asking The Wrong Questions [Kid Dynamite]

1.So, Kid Dynamite:

This is friggin’ ridiculous. The right question to ask is not “which contracts should be canceled?” The right question to ask is “Why do you think that market professionals who screw up should be let off the hook?” or perhaps “WHY should any contracts be canceled at all?”

Goldman Sachs is not some Mom & Pop retail schmuck. They are professionals. Many would consider them to be amongst the smartest of The Smart when it comes to trading on Wall Street. They should know better. If Goldman Sachs is rushing their code into production without checking it carefully enough, they should be responsible for the outcome. If Goldman Sachs is writing code that sends out bad orders, they should be responsible for the outcome.

Or, from Bloomberg:

Canceling the Goldman trades won’t guarantee nobody will suffer losses, according to Ophir Gottlieb, managing director of options analytics firm Livevol Inc. in San Francisco. Market makers who hedged the options transactions in the stock market may still be at risk, he said in a telephone interview.

“The firm or firms that created the error have absolutely no wealth at risk because they get their options trades canceled,” Gottlieb said. “But the people that facilitate market liquidity get crushed.”

2.A generous sample:

(1) Definition of Obvious Error. For purposes of this rule only, an Obvious Error will be deemed to have occurred when the execution price of a transaction is higher or lower than the Theoretical Price for the series by an amount equal to at least the amount shown below:

Theoretical Price Minimum Amount
Below $2 .25
$2 to $5 .40
Above $5 to $10 .50
Above $10 to $20 .80
Above $20 1.00

(2) Definition of Theoretical Price. For purposes of Rule 975NY(a), the Theoretical Price of an option is:

(A) if the series is traded on at least one other options exchange, the last bid price with respect to an erroneous sell transaction and the last offer price with respect to an erroneous buy transaction, just prior to the trade, that comprise the National Best Bid/Offer ("NBBO") as disseminated by the Options Price Reporting Authority ("OPRA"); or

(B) if there are not quotes for comparison purposes, or if the bid/ask differential of the national best bid and offer for the affected series just prior to the erroneous transaction was at least two times the permitted bid/ask differential pursuant to Rule 925NY(b)(4) as determined by a designated Trading Official.

(3) Adjust or Bust. The Exchange will determine whether there was an Obvious Error as defined above. If it is determined that an Obvious Error has occurred, the Exchange shall take one of the following actions listed below. Upon taking final action, the Exchange shall promptly notify both parties to the trade.

(A) Where each party to the transaction is a Market Maker. The execution price of the transaction will be adjusted by the Exchange to the prices provided in paragraphs (i) and (ii) below unless both parties agree to adjust the transaction to a different price or agree to bust the trade within ten (10) minutes of being notified by the Exchange of the Obvious Error.

(i) Erroneous buy transactions will be adjusted to their Theoretical Price: plus $.15 if the Theoretical Price is under $3 and plus $.30 if the Theoretical Price is at or above $3.

(ii) Erroneous sell transactions will be adjusted to their Theoretical Price: minus $.15 if the Theoretical Price is under $3 and minus $.30 if the Theoretical Price is at or above $3.

(B) Where at least one party to the Obvious Error is not a Market Maker. The trade will be busted by the Exchange unless both parties agree to an adjustment price for the transaction within thirty (30) minutes of being notified by the Exchange of the Obvious Error.

3.Here's Bloomberg's next example:

For the October $91 put contracts on the ETF, 993 options traded at $1 at 9:30 a.m., compared with 81 cents yesterday. The next trade 8 minutes later was for 15 contracts at 77 cents.

On my quick reading of the rules it sounds like Goldman's gotta eat that one though? Of course last trade yesterday / next trade 8 minutes later are not proof of the "Theoretical Price" but it'd be hard for Goldman to argue it was much lower than 77 cents. And at 77 cents they're only off by 23 cents, which seems like not enough to bust the trade.

Now I confess I'm exaggerating the simplicity here. This is probably right:

“To bust a trade in equities it’s relatively straightforward, to bust a trade in options it would take more time,” Howard Tai, a Kansas City, Missouri-based analyst with Aite Group LLC, said in an interview. “You need to look at each one of the factors and then run through a sanity check, and say, ‘Beyond the cash equity price at the time it happened, how did everything else affect it?’”

Still the more mechanical those factors are, the better.

4.Or maybe actual humans made those decisions but I doubt it. Anyway it comes to the same thing; the humans are supposed to know the rules too.

5.Also if you and GS are both market makers then on my read if you do this trade and it's repriced you get the 15/30 cent stupidity penalty that Goldman pays, so it's still a profitable trade and you should do it. Against this there is the whole "these trades destabilize markets" thing but that is not necessarily a great reason for any particular counterparty not to lift Goldman's below-market offer; after all if you don't someone else probably will, so you might as well get the stupidity bonus for yourself.

6.By which I mean loosely that the right algorithm is if, say, the NBBO is $3.30 and Goldman sells you options at $1, your computer should assume you bought at $3 (the penalty price) and hedge from there. Assuming that you bought at $1 and just got free money from a trade that will be unwound tomorrow is dumb.

7.[Update:This guy agrees:]

"Why do they not define what a valid trade is and program their matching engine accordingly?" said Thomas Peterffy, chief executive of Interactive Brokers Group, one of the largest traders of stock-options in the U.S. and a pioneer of electronic options trading. "Brokers are required to have software that rejects invalid orders. Why not exchanges?"

Well? Is there an answer to that question?


Goldman Traders Never Had a Down Day

Goldman Sachs just revealed in an SEC filing that its traders made money on every single trading day last quarter, a record for the firm. Net revenue for trading was $25 million or higher in all of the first quarter’s 63 trading days with 35 of those days bringing in more $100 million, according to the filing.