How To Fix The Broken ETF Market

Even after two months, my email box is still regularly inundated with questions about market structure, and specifically, “What happened on Aug. 24?” I’ve also spent a far greater amount of time talking to ETF issuers, exchanges, regulators and market makers about these issues than I honestly ever thought I would in one lifetime.

My conclusion—which is, to be clear, just my opinion—is that the anomalous trading we’ve seen occasionally in ETFs isn’t an ETF issue, per se, it’s simply the cracks that show the deeper faults within the U.S. market structure.

Luckily, I think we already have the fundamental tools to solve these problems, and I think there’s a lot of will in the industry to solve them.

But first, a word about the Dutch …

The Dutch Auction

The term “Dutch auction” gets tossed around a lot in financial circles, but I think there are often significant misperceptions about what it means, and how it works. That’s partially because we use the phrase to mean a lot of different things.

Originally, the term meant a reverse auction—where a single item would go up for bid at a high price, and the auctioneer would lower their price until some raised their paddle, and the auction would be over. It’s a format still used today in the collectibles and occasionally antiques markets.

Economists have written reams on whether it’s a better method (for the seller) than a traditional sealed bid auction (where everyone bids in secret, at once and the highest bidder wins).

But in financial markets, we use the term to refer to almost any configuration of rules that results in crossing a maximum volume of securities at a single price. The Treasury Department uses a system like this to sell bonds. WR Hambrecht uses a similar system to price IPOs, which it did for Google’s initial public offering.

And the New York Stock Exchange uses it every day to open and close stocks.

NYSE’s Version Of A Dutch Auction

At the core, the NYSE version of a Dutch auction simply collects all of the market-on-open and limit orders together into a big book and determines the price at which the maximum number of shares will trade hands. How do they do that? It’s pretty simple actually. Imagine there are 10,000 shares to sell at $10, and nobody has entered in a buy order for $10 or more. There’s no natural crossing price, so no auction occurs and any market orders are priced at the midpoint of the current national best bid and offer (NBBO) (typically, the best price at which one can buy or sell a small number of shares right before the open).

But if there are 10,000 shares to sell at $10, and someone has a buy order in for 5,000 shares at $11, well, then 5,000 shares could change hands anywhere between $10 and $11. The price of the cross will be set closest to the previous trade (so it could be $11 if the last trade was $12 or $10 if the last trade was $9).

On top of this, any market orders will cross at that price, as long as there are sufficient buyers and sellers. Market orders get resolved in the order in which they were received, and anything not traded is called “the imbalance,” and not executed. (In practical terms, that doesn’t happen frequently, as market makers will often place an order explicitly to offset these imbalances.)

It doesn’t actually make the math harder when you have thousands of limit and market orders in the mix—you just find the price where the most shares can trade, and you’re done.

The Good

The beauty of the Dutch auction process is that it gives you fantastic price discovery. If you’re interested in the “fair” price of Citigroup on the open, what better definition could you have than the price on which the largest number of people agree?

Anyone asking too much or offering too little just won’t get a trade done. And surely thousands of shares crossing to set the “opening” price makes more sense than a race to see who can get the first 100-share trade done.

The other great thing about a Dutch auction is that it’s information-rich while actually being somewhat anonymous. While it’s great to know that the closing auction for Ford yesterday crossed 2 million shares at $15.36, it’s even better to know whether the orders that didn’t cross because the limits were set outside the final price, or because there were too many market orders on one side—unfilled buys or unfilled sells.

That order-imbalance information can be extremely useful to traders and market makers who are deciding where to put their own orders in. At the same time, your personal order to buy 1 million shares at $16 or better doesn’t hang out on the book all day since the auction book is separate from the regular Level 2 screen we all stare at.

Finally, there’s a bit of inherent fairness in the process. People who place market orders into the auction know they’ll either get the bid/ask midpoint or they’ll get a price where the majority of active traders actually believe it should be trading. (Of course, constant readers will know that ETF investors should generally be avoiding market orders and trading at the open or close).

The Bad

The problem is that the process is slow, at least by modern market standards. For an auction to happen, the order book needs to be collected over some period of time, and then information about what the book suggests needs to be disseminated to the street.

Every exchange has slightly different variations for different kinds of auctions, but the NYSE closing auction is typical: Starting at 3 p.m., the exchange starts running simulated auctions based on the orders they have, and they publish what the clearing price would be, the volume at that price and where the imbalance would be. They keep updating that in real time right up until the auction actually runs at 4 p.m.

A Modest Proposal

Currently, auctions are used most often for setting opening and closing prices, two periods of time where there can be doubt about marking a “fair” price and crossing large volumes across many trades at a single common price. Since both of those are concerns coming out of a “limit up/limit down” (LULD) halt (the plague of Aug. 24, 2015), why doesn’t the market do the same here?

The reason is because the current rules leave this totally to the exchanges.

When limit-down occurs, technically it’s the listing exchange that notices a stock trading outside its band and “calls the halt.” All exchanges (and derivative exchanges too, like the options market) must cease trading in that name. Once five minutes have passed, the exchange then has 10 minutes to get the security open for trading again, using essentially whatever rules they happen to have in place.

If it takes them more than 10 minutes, all bets are off and trading resumes simultaneously on all exchanges.

And here’s where the real issues emerge. Right now, the NYSE rules around things like reopening a stock leave significant room for interpretation. Because of its legacy as a physical exchange, the actual rule for reopening (123D) is positively quaint, including notes like:

“Brokers must recognize that orders or cancellations merely dropped on the counter can be lost or misplaced, and should hand the order directly to the DMM [Designated Market Maker] or his or her assistant and orally state the terms.”

Obviously, it’s been a long time since people crowded around a physical post waiting for the DMM to sort through the paper being thrown at him to reconstruct a viable opening price. Instead, the process—again, with seemingly a large amount of discretion on behalf of the DMM—is to simply reopen the book, cross as many shares as possible, and resume trading.

More modern exchanges, birthed in the electronic trading era, actually use a formal Dutch auction process explicitly to reopen a halted security. BATS runs a simple, clean process that won’t reopen a security until all the market orders in the system can be executed at a common price, regardless of pricing bounds. The end result is a process where a true, optimal clearing price can be established. If that price is lower, or higher, so be it.

I believe this process, or something like it, should be the universal, mandated standard for coming out of a LULD trading pause for anything, including ETFs.

The Bugaboo: Fragmentation

But even if the SEC mandated a set Dutch auction reopening process that every listing exchange would have to follow, it wouldn’t necessarily matter, because market fragmentation complicates things.

Under the current rules, if the BATS process took more than 10 minutes, well, the chaos button would be hit and it would be a free-for-all again. And since there’s no guarantee that I, as a retail investor, or even as an institution, can explicitly route my market or limit order to BATS in the middle of the five-minute trading halt, the reopening auction will always just be a fraction of the true market demand—and information—at that time. Even in the absence of a Dutch auction process, this creates huge problems.

Consider what happened on Aug. 24 when my poster-child ETF for trade forensics, the Guggenheim Equal Weight S&P 500 ETF (RSP | A-83) was coming out of halts:

• At 09:42:13.840, trading was halted after the NBBO “rested” at or below the limit-down band. The last trades were at $53.24.
• After 5 minutes and 45 seconds, the DMM on the NYSE reopened RSP with a flurry of small trades crossing at the official reopening price of $52.71. It takes the NYSE 9 milliseconds to print 326 individual small trades, for a total share count of 71,474 shares.
• However, during the 9 milliseconds it took for the NYSE to even print those reopening trades, BATS crossed 1,047 shares at $50. Direct Edge processed 13,021 shares at $53.17.

What Was RSP’s Fair Price?

So what was the true “fair” price of RSP after the halt? Was it $52.71? Who knows? But the answer actually matters, since the bands for halting the stock (which would happen again in a few minutes) are reset based on this official “reopen” price.

In my opinion, the whole point of a trading halt is to give the market something that is in incredibly short supply in the modern trading environment: time. So when a limited up/limit down event happens, I suggest we try this:

• During a LULD halt, all order flow routes to the listing exchange.
• Exchanges use a common agreed-upon Dutch auction process.
• Exchanges publish all the order imbalances and proposed clearing prices during the halt.
• The security only reopens when all market orders can clear.
• Nobody gets to trade until the security reopens on the listing exchange.

It’s really two pieces meshed together. Without collapsing down to a primary exchange, it’s difficult to see how any price discovery in a time of extreme volatility isn’t corrupted by issues of latency. Without running a Dutch auction, we’re essentially rewarding speed at 100-share executions over volume in setting the new reference price for the LULD process. Neither seems great to me.

The Penalty Kick?

While I expect at least a dozen emails telling my why my naive idea won’t work (it’s not a Wednesday if someone doesn’t call me an idiot at least once), I’ll be hard to sway. Infrastructurewise, I don’t believe this is a monumental task to implement. The exchanges already have order-routing capabilities, and strong communication among themselves.

Everyone wins in this situation. Investors win because if they’re trading in this chaos, they get a fairer price—with complete recognition that this has nothing to do with the value of the “underlying” securities in an ETF, just “fair” in the sense that it’s based on the maximum possible participation in the price discovery process.

Listing exchanges win because they get another thing to compete on, and increased importance in an ever-fragmented market (secondary exchanges, not so much, I admit). And issuers win because the likelihood of weird, illogical trading shrinks when the whole market slows down and has a good long think about fair prices.

This goes against the trend, for sure, which is toward radical decentralization. But sometimes, maybe just sometimes, it makes sense to have just one player on the field, working things out, instead of scrumming the ball like kids playing soccer. Think of it as the penalty kick after the LULD foul.

At the time of this writing, the author held no positions in the security mentioned. You can reach Dave Nadig at [email protected], or on Twitter @DaveNadig.