Thanks to everyone who sent feedback on last week's experimental post about financial history. We're doing it again!

Academic finance is a strange domain because researchers are working with some of the same datasets as industry practitioners, and are working on the same basic task of figuring out why asset prices move the way they do. But because it's a description of a live, adversarial process, it has some quirks. For example, in most fields if you ran an experiment and found that X was true, and then someone else ran the same experiment a year later and found that X was false, you'd say that this raised serious questions about the validity of the original study. But when the subject is finance, it's the exact opposite! If someone publishes a paper saying that you can make money buying small-cap stocks, or that the embedded option in convertible bonds tends to be mispriced, etc., then a subsequent study finding the same thing reveals that the original study was probably wrong, because if it were right, the arbitrage would have closed.

Often, what these papers actually find is a strategy with hidden, second order constraints: stocks go up at night and are flat during the day on average, for example, but the transaction cost of trading on this anomaly eats the profits. Heavily-shorted stocks do underperform the rest of the market, but the cost to borrow them reflects this. Sometimes, the excess return is an unacknowledged risk premium: in the 80s, there was some applied research in the direction of creating cheap synthetic put options by dynamically hedging (the basic intuition is that you can mimic the payoff of owning put options on an index by shorting index futures as it declines). This was true in a stable market, but in the crash of 1987, the overall market dropped 20.5%, but futures were at one point about 18% lower than the index itself.¹

And sometimes, the reason an anomaly sticks around is a different kind of transaction costs: the kind that are imposed by participants in a price-fixing scheme. There's a 1994 paper from the Journal of Financial Economics noting an odd feature of Nasdaq stocks around that time: market-makers wouldn't quote odd-eighths. For context, since the early days of US trading, stocks have been quoted in eighths, probably reflecting the fact that Spanish dollars, the main currency in early American commerce, were worth eight Spanish reales. Exchanges would later voluntarily switch to quoting sixteenths, before regulators required them to decimalize.² And, peculiarly, Nasdaq market-makers in the early 90s were much more prone to quoting quarters and halves than quoting eighths.

And there was a pretty simple explanation for this, elaborated on in a subsequent publication, this time by the SEC: they were colluding to widen spreads so they'd earn more on trades, and the easiest wide-scale way to do this was to just refuse to quote odd eighths.

In a normal market, this would be hard to pull off. You and I could agree that we'll only quote some stock in ten-cent increments, but if the natural spread is eight cents then a competing market-maker will wedge itself between our quotes and get all the volume for itself. But at the time, the main venue for quoting trades identified the market-maker in question, so it was easy to see who was participating and who wasn't. What market-makers basically did was to spread vague rumors about unprofessional behavior on the part of the non-participants, and also to refuse to spread market-moving rumors to them.

Manual trade execution is a job where information gets transmitted in two forms: by prices, and by people sharing interesting tidbits with one another over the phone. For a market-maker, these rumors are generally about who's selling and how much (if you're a buyer, you care whether the big seller who's pushed the price down this morning is a random fat-finger trader or a big fund exiting a position. You aren't supposed to know the details, but "market color" is a fuzzy concept. In the voice trading era, it was more common for this to leak out directly, and to go to favored counterparties. Today, the same kind of information still exists, but it's in aggregated, anonymized from—prime brokers produce reports on positioning, sentiment, and general market behavior, which puts individual counterparties on a more level playing field.) It's pretty hard to be in the business of passively taking on market risk when your competitors have better access to information than you do; that basically guarantees a winner's curse.

So once this got going, it was a surprisingly durable arrangement. It's hard to tell how it got started, but one plausible mechanism is that market-makers in less liquid stocks always quoted them in wider, round-number increments (you can still find plenty of microcaps that trade in the low-dollars range and have a spread of ten or twenty cents). And when liquidity grew to the point that it was optimal to quote in eighths instead, all that had to happen was for the first market-maker to do it to get some irate complaints from counterparties who were annoyed that they'd have to accept slimmer margins. That can slowly harden into an industry convention, and eventually a taboo: if you know that bidding 10 3/8 instead of 10 1/4 is unprofessional, and that it's amateurish to even ask why, you can walk backwards into participating in a cartel.

This was already eroding by the time the SEC got involved. As the SEC report notes, the Instinet electronic trading system was more anonymous, and offered tighter spreads than the Nasdaq itself. Volume was already moving there. But industry self-regulatory bodies love preserving high margins, especially if that protects them from internal competition and transfers wealth from customers to them. So it was quite helpful for academics and then regulators to give them a kick in the pants.

The irony of all of this is that market-making is a much bigger, more lucrative business now than it was then. When trading is absurdly cheap, people trade absurdly often, and cheaper trades encourage more uninformed trading. In the 90s, if you were buying, say, a specific oil stock, and then shorting other oil stocks to hedge, you had to be at least a little careful about how you hedged because you were paying high transaction costs. Now, it's trivial to use an ETF to net out that industry exposure, and the market-makers who keep that ETF priced in line with its holdings are going to make lots of minimal-information-content trades in a bunch of oil stocks, paying tiny execution costs when they do so. Trading happens to be a domain where almost no one has good intuitions about the elasticity of demand, in part because liquidity in one market is complementary to liquidity in another, and the people who take advantage of this do so in unpredictable ways. Nasdaq market-makers made some temporary excess profits for a while, got in trouble for how dishonestly they did it, and ended up structuring their business around a version of it that had a fraction of the potential of its later form.

This one cuts across a few different Diff themes:

• On Monday, we looked at why cartels tend to be unstable.

• We’ve considered when companies own a standard, and when industries collaborate to set one.

• Bonds are going through the same voice-to-electronic transition equities went through ($).

• Hard-to-scale trades are often the root of bigger trading companies.

• What happens when the minimum tick size drops ($)?

Join the discussion!

1  "About" is accurate here, because one feature of the crash was that many stocks hadn't opened, and some quotes were stale. Futures kept trading, so they were the liquid and certain proxy for an illiquid and uncertain underlying. Still, they got way too low, and demonstrated that while you could synthesize cheap close-to-the-money options, you would have done better buying further out-of-the-money options instead.

2  Ironically, as trading got digitized we moved away from units that are convenient for computers. If anything, we should have switched to quoting in eight- and then sixteen-bit floating-point numbers.