Comparing a Decade of Order Times and Cancel Rates

The SEC MIDAS dashboard started nearly a decade ago in the midst of increasing electronic trading and message traffic in the market.

Today we use MIDAS data to look at how order cancel times today compare to markets from almost 10 years ago. 

Order cancel rates still seem largely “clock” driven

One thing that surprised many in 2013 when the data was first published was the spikes at very round time intervals for order lifetimes. 

Today’s update shows, despite the advancement of technology, that phenomenon has not changed much. In fact, the new data shows that the spikes in cancels occur at around the same round time intervals as back in 2013 (Chart 1).

Note: The zigzag shape we see at each factor of 10 is not because of a persistent pattern of cancels over time but rather a mathematical result of how the data is collected. As time gets bigger, the data buckets become less granular. In short, the 101-millisecond (ms) bucket represents a timespan that is 10-times as long as the 99.9ms bucket, so it makes sense that the wider data point has around 10-times more cancels.

Chart 1: Cancels cluster at very round lifetimes, which are consistent across the whole 10-year period

Academic research showed electronic trading helped market quality

To keep things in perspective, there are 23,400 seconds each trading day and over 75 million trades across over 12,000 symbols. That averages to over 3,000 trades each second. Clearly, the addition of computers to trading means it’s possible to price and trade a lot of liquidity very fast. 

The initial burst in market activity that the initial wave of electronic trading created was met with skepticism about how bad computerized quoting would be for the markets. 

That, in turn, attracted a wave of academic research into “HFT” traders. Despite the skepticism, on average, they found a positive impact on markets from electronic trading. Many papers found that spreads tightened, liquidity increased, and assets were mispriced less and corrected faster. In general, electronic traders were mostly good for market quality, with positive spillover effects on asset pricing and transaction costs.  

Some studies found that HFTs provided liquidity when it was scarce, and others show that almost all traders these days are electronic, so the impact of latency depends on your trading strategy. Researchers also found that all the “latency arbitrage” people said existed turned out to add to just $14 million in profits over almost a year once you accounted for the transmission times (the speed of light) of actually trying to trade on mispriced stocks across markets in different locations. In other words, the incidents of mispricing were so fleeting that they were mostly corrected (or canceled) before an arbitrageur could trade. 

Markets have gotten faster

Despite what the data in Chart 1 suggests, where clock-based order durations seem fairly consistent over the past 10 years, markets have undoubtedly gotten faster over the nearly 10 years in question. 

For example, the average time it takes the UTP SIP to compile the NBBO has decreased 99% (from 1.15ms in 2013 to around 14 microseconds (µs) today), while at the same time, its capacity to process transactions has almost quadrupled (from 5.5 billion per day in 2013 to around 20.4 billion per day today).[HL1] [HL2] 

Looking at the MIDAS data, we also see a change, with more orders having a shorter lifetime. 

• The median lifetime of a canceled order has shrunk from 847ms in 2013 to 99.7ms in 2022. 
• Even the longer-life orders have shrunk from 120 seconds to 20 seconds in 2022 (in Chart 2, the dot at the top of the chart represents the 95th percentile of cancel duration).

You can read more about the difference between microseconds (µs) and milliseconds (ms) here.

Chart 2: More orders have a shorter life now than in 2013

Does fragmentation contribute to message traffic?

Midas also lets us look at cancel-to-trade ratios over time.

High cancel-to-trade ratios, alone, are not necessarily bad. Data shows they go up when:

• Prices are more efficient as limits need to be modified more frequently to reflect more refined valuations.
• No trades happen, so the divisor is small, making the ratio larger (see Chart 3).

Market fragmentation will also increase cancel-to-trade ratios, as there are more markets to price correctly – and fewer trades on each exchange. We can see that by looking at the cancel-to-trade ratios by exchange, which shows exchanges that trade less generally have higher cancel-to-trade ratios.

Clearly, the addition of new venues comes with costs as market makers need to keep more quotes aligned across multiple regions and connect to more data feeds. Both of which add to hardware costs in the industry.

Chart 3: Cancel-to-trade ratios (Q3 2022) are higher for exchanges with fewer trades

Strangely, IEX, which is the only exchange allowed to cancel (or fade) protected quotes before they can be traded based on changes in away markets, is missing from the SEC’s MIDAS website. Even though newer exchanges MEMX and MIAX have been added.

Markets are fast, but that’s (mostly) OK

The short story here is that the MIDAS data shows just how quickly and accurately electronic market makers update prices and quotes. We can also see that although markets seem to adjust orders more quickly now than 10 years ago, some clock-based behaviors seem to have changed very little. 

Other Topics

Data and Analytics US Markets Markets

Phil Mackintosh

Nasdaq

Phil Mackintosh is Chief Economist and a Senior Vice President at Nasdaq. His team is responsible for a variety of projects and initiatives in the U.S. and Europe to improve market structure, encourage capital formation and enhance trading efficiency. 

Read Phil's Bio