Testing in Prod
Vol #194: September 12th, 2024
One dollar is the usual amount to bet.
It’s a sporting interest, just a nick of skin in the game to keep you interested and buy the winner one quarter of a barista poured coffee. A nice round greenback says there’s interest here - but the price is not the point.
There are only a few physical things you can buy for an actual dollar - disposable pens, dental floss, a few pieces of penny candy. A decade ago my wife caught Keanu Reeves buying ice cube trays at a dollar store in the East Village. Inflation is a bear.
Plenty of options get quoted for a dollar. Yesterday’s closing prices had 13,321 different calls and puts that marked within a nickel of that hallowed price point. Being a believer in efficient markets, I’m inclined to think those are generally good prices.
But I certainly wouldn’t want to own ALL of them for a dollar. Or sell them all at a dollar. And even more so, if I saw someone that was trading that way - take a step back and reassess.
A spinning trade list on the open is a market maker's worst nightmare. During the trading day the market ebbs and flows more and less smoothly. There’s kinetic friction as orderflow adjusts pricing inputs, but it’s nothing like the static start in the cold, 16 hours since the last options trade.
Something is priced wrong. You didn’t adjust earnings vol, or the new LEAP contract is off by a click and costs you bigly. One little point of vol makes a significant difference a whole year out. Perhaps it’s the Allegheny International Group who got a hot tip that the ETF was closing and knows vol is going to zero.
Famous last words - but this time was different. There was no pattern to the symbols, and there were both buys and sells. All for a $1? As the full textured drop copies trickled in eonic seconds later, it was further head scratching. Goldman Sachs was the counterparty?
The kind of opportunities you read about in blogs like this only last for a few seconds. Decisiveness in those fleeting moments is the difference between predator and prey. There are more dimensions of uncertainty than any model could possibly account for. But the truly great traders - the guys with more commas than name recognition - intuitively know which dollars are good and bad.
On August 20, 2013 - Goldman Sachs handed over $38M to the equity options markets. Two years later they were penalized an additional $7M by the SEC for causing this market disruption. All because they deployed a faulty piece of code, and compounded their mistake with weak operational controls.
When 16,000 bad orders hit the market, it takes some time to parse out what happened. There are plenty of rules about what constitutes an error trade, but they’re designed for one off scenarios. The occasional fat finger, not a rogue gatling gun.
While there was one perpetrator, there were many “victims” and those counterparties were operating blindly with a pile of risk on their books. What was going to stick? Stocks are moving, and there isn’t exactly a help desk that can provide answers in the time required to mute your risk.
Dealer hedging further complicates the error PnL calculation. Getting hit on a trade that’s obviously bad, and you might want to undo the hedge - your options leg is obviously getting busted. But what about everything at the margin - is a customer paying $1 for a $0.50 option going to stand? Is this SPY or DJT? Where was the line getting drawn?
The Knightmare on Wall Street was barely a year in the rearview mirror. The storied high frequency firm had just deployed software for a new retail price improvement feature on the NYSE. Equities went haywire as Knight Capital slammed $3.15 billion of bids and lifted $3.5 billion of offers on the open. That spread plus the movement cost them $460M - $100M more than their cash balance securing the positions.
Options market makers were tucked back in the corner, but we could hear the roil of cratering giant on the Big Board’s floor. It took 45 minutes to find the kill switch. Even worse they had missed 97 alert emails that morning starting at 8am.
DevOps salaries doubled. Now it was the options guys turn to double them again.
By noon we were cobbling together a list of all our trades to request reimbursement from the exchange. There were generous interpretations on both sides. Obviously you only put the losers down and ignore hedging. But for all the line drawing that tested the limits of obvious error rule precedence, it did reduce the ultimate bill for the Giant Squid. Estimates at the time put this potential error cost even bigger than Knight’s.
I’m proud to say we took well more than our market share worth of those errors. Even internally, as with most things in finance, there was a power law distribution. Our top trader knew exactly he could take and keep. It wasn’t the nickels that were ninety five cents good, those got busted. The play was in $.80 trades in liquid products you could do in size and call it fair. When shit hits the fan, play the exchange and the rule book, not the price.
Participants were learning just how severely the hot path burns. As volumes scaled and automation increased, the recursive externalities shown by the Flash Crash and the risks of code deployment were stark. Internal controls where a lackey blindly pushes a “release” button or is numbed to spam e-mail, up just wouldn’t cut it anymore. Venues needed to bolster their own risk management as the arbiters of transactions.
After a long slog of beating the bugs out of code, it’s exciting to fire the laser. But there’s almost certainly a catch. It could be the rounding error from Superman III / Office Space. A flag that erroneously flips contingent orders to live. Murphy’s law of DevOps would be something like - if it can go wrong, it will go wrong at scale.
With the various APIs and programmable trading opportunities for retail traders these days, there is plenty of room for good automation. I love having a bot that automatically bids out nickels to close short condor legs (independently of course). And while fortunately buy side customers have several prophylactic layers between them and direct market access, we also don't have accounts that can weather eight figure losses.
Judo your advantage here. Unlike Knight Capital or Goldman Sachs - you have no obligation to trade at all, or in any given size. Of course you test in the sandbox, over and over again. Absolutely double check the decimal points. But just do a one lot. Flip a single share.
Testing has a budget in both time and dollars, so you have to be prepared to lose a little bit. But there’s no obligation to jump in feet first. This goes for non technical implementations also. It doesn’t quite have the neck snapping feel of a production error at the opening bell, but rolling into a new strategy should be a zero entry gradient.
Whether it’s converting lump sum cash into equity, or starting to balance out your share value with hedged equity, a pinky toe in the water goes a long way. An extra million dollars in your account is a problem we’d all like to have. Cold logic says Even God Couldn’t Beat Dollar Cost Averaging so buy at the open and you have 2,174 additional shares of QQQ. Yesterday’s ride took that day's trade down $17k to close up $19k. Are you really ready for $36k more daily swing in your account? That 4% interest rate on cash was so cozy.
Overlays have the inverse problem - after having watched equities appreciate year after year, how does it feel to be a little bit covered? The emotional experience of the trade takes some time getting used to. FOMO is real.
There’s no need to flip the switch and pipe your new code through the hot path all at once. Rebalancing can be a process, not a point in time. A big reason why we tranch across multiple expirations is to smoothen the PnL streams.
Eliminating unforced errors keeps firms in business, and investors in the black. Trade, experiment, and dabble in new strategies - but be careful flipping the switch. The most robust way to implement a theoretical success is slowly and surely. Let the tech, results, and experience prove themselves dollar by dollar.