What Does Flow Know?
Vol #191: August 22nd, 2024
“What happens if the guy next to you steps on your foot?”
“What if he does it again? What if he jams your toes right before the broker walks in?”
This line of questioning was getting annoying. I got the point, but the trader was still testing if I had enough backbone to handle open outcry trading.
Interviewing is brutal. While it’s difficult to fill any job, trading feels particularly like a craps shoot. There aren’t just technical boxes to check, or personalities to fit in at company happy hour; a trader has to have a certain chemical imbalance that keeps them perpetually thirsty for good risk.
I asked ChatGPT to cook up a practice interview question, and the first result is pretty lame. I’m not sure this LLM knows the difference between a broker and an LMM. Lead Market Makers still beat Large Language Models. (DALL-E should also know that 4 is not a prime number.)
Question: Imagine you are an options market maker for a highly liquid stock. You have an order to buy a significant quantity of out-of-the-money (OTM) puts, which has the potential to impact the implied volatility of that option. How would you approach managing the risk associated with this order? Specifically, discuss how you would hedge the position, the potential impact on your bid-ask spread, and any adjustments you might make to your overall trading strategy in response to this order.
Like most bad interview questions, the spirit is there. A glimpse at how the candidate understands liquidity and orderflow could be useful. But this is a misguided academic exercise. The real test is how good arithmetic and logic function when your brain is pumping on cortisol.
One way to simulate this is a mock trading scenario. Observing how candidates play poker is another high value signal. The internet is abound with sample questions for aspiring candidates, and the below is from StackExchange - where hopefuls went before ChatGPT.
I was told to make a market on the number of prime numbers between 1-100. I was confident the number was around 20-30. (It's 25). So I made my market at 20-25. (Interviewer only allowed a spread of 5). They proceeded to buy my sell offer at 25. Then I moved from 30-35. They bought it again. I moved from 35-40. They kept buying until it was 65-70, then sold at 65.
They then stated they have profited off me and I have in fact lost money as they bought for 25,35,etc.
In this example, there is an objectively true value. That’s a key piece, and something we don’t get the benefit of in trading. If you know that fair value is 25 and the answer must settle at fair value, most right answers have you selling as much as you can above that price.
The best answer has you asking some questions around risk and settlement. Is there some quirk where a wide mark causes my short to blow out before expiration? Are we sure we agree on the definition of a prime number?
When trading implied volatilities or stock valuations, there is no truth other than the price where buyers and sellers meet. For all the discounted cash flows and SABR models, the only price that matters is what someone is willing to pay. Other than the purest of arbitrages, price can go as far “out of whack” as order flow demands.
Mock traders love to drive prices wild, and this was no exception. While the first sale at 25 might be low, the candidate adjusts quickly and sells at 30 and 35 - then all the way to 65.
When a market maker gives a quote, it has only two components. The price and the size. All their opinions about volatility, skew, and risk management must be condensed and expressed through these knobs.
As order flow comes into the market, the trader here walks their bid/ask spread up incrementally. If someone keeps buying from you and you’re only raising offers a nickel at a time, you probably won’t make it to settlement.
Liquidity providers want two sided trading. They charge a vigorish around theoretical price to manage inventory while the market mechanism finds truth. Their opinion on fair value can only be as firm as the rest of the market says it is.
Raise your price in the face of buyers. Raise it even more if you can’t find sellers. While bid-ask was constrained here, in practice you’d widen out as orderflow became imbalanced. First by a nickel, then a dime, and then a quarter.
Sizing is about being nimble to risk and capturing opportunity. A buyer will raise your theoretical price, and the more that deviates from other signals (in this case, mathematical facts) the bigger you should trade on the offer.
Selling one contract at each level from 25 to 65, the average price is 45. Increase the size by one at each turn, and the average price goes up to 51.66. Double the size every turn and it’s 60.08. No risk manager is letting you sell 256 at the top after you sold that many all the way up, but you almost keep up with the market.
Most aspiring traders grasp this idea of sizing up in the face of indiscriminate buyers. The catch comes when you finally do find that two sided flow. You can have a bid over the theoretical price, so long as it’s appropriately sized.
When you’re short 256 prime number contracts be prepared to drop fast. If a 65 bid gets hit, it better be tiny. Unless the buy orders crumble, average buy price will exceed average sell price.
As dealers are transforming their pricing models into quotes around theo, they must also be asking second order questions about the texture of the flow, and their current risk profile.
Several of the responses to this StackExchange post suggested taking into account the idea of informed vs. noise traders. This is elaborated in the Albert Kyle paper from 1985 - Continuous Auction and Insider Trading.
The publication date is important - long before a series of market reforms distinctly curtailed bad actors. (See: Fake Plates). How to model the information content of order flow is much more nuanced in today’s markets.
Customer flow has value because it’s uninformed in the aggregate, even if certain players are not. Market makers are naive to any fundamental valuation of a company, but are highly sophisticated in providing a service to informed customers who have a distinct opinion.
Who is or isn’t informed flow can vary. Dealers are the most sophisticated options players in the options market. When dealers trade with dealers, at least one side starts asking a lot of questions. If it’s an accident they both are.
A price setter who becomes a price taker is signaling they believe the market is mispricing that option. It could be taking risk off, or spreading it around because of a different piece of risk. In ultra liquid and tight markets, this will happen much more often, and “market maker to market maker” flow isn’t as scary. Dealers aren’t just trading options with customers, they’re exchanging liquidity profiles with the market.
In less liquid names - like a 1:1 interview - you have to take what the customer gives you. The only truth is the mano y mano buyer. When there’s little offsetting volume that creates noise and gently vacillating inputs, every order is significant. You have to trade scared, and let the market tell you there are 59 prime numbers under 100.
When the contract settles, you win 35 points if you’ve tactically sold up and the game is as fair as you predicted. But in the edge case you’re missing something, standing firm at 30 because you bet too big on this being dumb paper is a costly mistake. Instead of a flesh wound where a sale at 60.08 settles a tick up (the counterparty did sell 65).
When the bell rings, a good liquidity provider doesn’t have a firm opinion on value. The real answer here should be questioning why we’re making markets on facts. A better example would be to guess the dollar value of a third party’s wallet.