Speed bumps have been a popular topic recently.[1] The most recent discussion has centered around asymmetric speed bumps, like the one proposed for CBOE EDGA. Such speed bumps would slow incoming “take” orders by a few milliseconds, effectively giving potential resting counterparties extra time to cancel or modify their existing orders. The benefit to liquidity providers is obvious – they have a greater opportunity to adjust their quote in moving markets and avoid being “adversely selected” by faster traders. The net effect on other traders, however, is less clear.
Obvious Winners and Losers
To better isolate the effects of the speed bump, consider two types of traders: 1) “informed” high frequency trading takers (“HFT takers”) attempting to hit a quote prior to an impending price change, and 2) other traders who are taking liquidity at a point in time simply because they need to trade and not because of some informational advantage (e.g., a VWAP algo sending “take” orders to stay on schedule). For the latter traders, a speed bump lasting a few milliseconds should have little effect on their execution quality. Liquidity providers are only slightly more likely to change their quote than they would have without the delay, since very little typically happens in a few millisecond time frame. And in those cases where the liquidity provider would change their quote during the speed bump delay, presumably the price change would be equally likely to be favorable as unfavorable, resulting in the same price on average.
Contrast that with the HFT taker attempting to exploit a short-term informational advantage, i.e., one lasting only a few milliseconds. Such traders are by definition more likely to send orders just prior to a favorable price move. Without a speed bump, these better-informed HFT takers are able to execute their orders before the slower liquidity provider can receive and respond to that same information. With the speed bump, any informational advantage that dissipates within the period of the speed bump cannot be exploited, reducing the profits of the better informed HFT taker – and the expected loss to liquidity providers.
To summarize, then, in the former case, where a liquidity provider is confronted with uninformed take flow, the expected profits and losses to each side of the trade are unaffected by the speed bump, on average. In the latter case, where HFT takers are attempting to exploit a short-term mispricing, the speed bump results in a “wealth transfer” to liquidity providers from HFTs. So, the most obvious beneficiary are liquidity providers who can update their quotes during the speed bump, and the most obvious loser are HFT takers.
But note that the transfer doesn’t apply to all liquidity providers. Rather, only those liquidity providers able to access and respond to short-term information within the length of the pause will benefit. In that sense, the direct effects of an asymmetric speed bump are felt only by HFTs, with the slightly slower liquidity providers gaining some protection against faster HFT takers.[2] The speed bump does nothing to protect manual limit order traders or other “slow” liquidity providers, including some broker algorithms. For example, while a state-of-the-art broker algo platforms (e.g., one with similar technology and liquidity provision strategies as HFTs) could update their limit prices and reduce their adverse selection costs, the less sophisticated algorithms, such as those designed to “wake up” periodically at some lower frequency (e.g., 500 milliseconds), are still exposed to HFT takers even with a speed bump.
Less obvious winners and losers
While the speed bump directly affects mainly high frequency traders, other traders could benefit indirectly. Specifically, if the market for HF liquidity provision is sufficiently competitive, reductions in losses should lead them to price and size their quotes more aggressively. Such tighter spreads and deeper markets will therefore reduce the costs for buy-side traders whose strategies involve taking liquidity. In effect, then, the speed bump indirectly transfers some of the benefits from HFT takers to “low frequency” buy-side and retail takers via tighter spreads and greater depth.
With that said, the fact that some liquidity providers simply cannot exploit a speed bump puts them at an even greater competitive disadvantage relative to HF liquidity providers who can. But most likely the net effect would be small, as the HF liquidity providers already have a sizable advantage over manual traders and less sophisticated algorithms that the incremental negative impact would likely be quite small, if not negligible.[3]
The Ghost of “Last Look”
One criticism of asymmetric speed bumps is that they are similar in principle to the “last look” functionality available in other asset markets, e.g., FX markets. While both effectively result in liquidity providers avoiding unprofitable trades, they are in fact quite different. In “last look”, the liquidity provider responds only after receiving the order, giving them time to evaluate the market before making a decision. In effect, market makers gain an informational advantage after the fact because of last look. A speed bump, on the other hand, attempts to remove an informational advantage held by faster traders by pausing to allow at least some of the information asymmetry to dissipate. The liquidity provider must still make their decision to move independently of any incoming order flow – and must be fast enough to actually benefit.
The downside of asymmetric speed bumps: Quotes are not protected
Under Reg NMS, delays like those imposed by EDGA would result in that venue’s quote being “unprotected”.[4] Consequently, the venue’s quotes are subject to trade-throughs and lower fills rates, as traders can effectively ignore unprotected quotes when routing. Practically speaking, though, the loss of protected status probably would not adversely affect non-protected quotes much. Providers of smart routers, for example, would be hard-pressed to pass up better-priced liquidity if the unprotected market were alone at the inside. And even if other markets were matching the unprotected quote, cost-sensitive takers may still prefer routing to the non-protected venue if it were cost-effective, so long as the execution price is expected to be the same on average. In the case of EDGA, for example, the inverted cost structure may place it high on routing tables despite its non-protected status. The main time that a trader would skip the speed bump venue is when they have a short-term informational advantage that the speed bump would thwart – which is exactly the point of the speed bump.
It should be noted that speed bumps are not completely irrelevant to non-HFT takers. For example, consider a non-HFT trader who wishes to sweep the market using a smart router or some other aggressive algorithm. Even though the trader is not trying to exploit a short-lived information advantage, the trader must alter their behavior to account for the speed bump to avoid a worse overall execution. To see why, suppose the trader ignored the speed bump and routed to all venues simultaneously. Liquidity providers posting on the speed bump venue may learn of the sweep by observing fills on other venues during the speed bump’s delay period and move their price to avoid being swept. To avoid the non-fill on the speed bump venue, the router needs to take measures to limit this risk. For example, the router could first route to the speed bump market but then delay the orders routed to the other markets to such an extent that they arrive at the non-speed bump venues around the time their initial order clears the speed bump. By doing this, the liquidity providers learn of the other fills only after they have filled the initial order.
Conclusion
Given these arguments, one could imagine the “less fast” high frequency market makers and non-HFT takers, e.g. buy-side traders, would be in favor of asymmetric speed bumps since the net effect is more aggressive pricing and sizing by HF market makers. Furthermore, buy-side traders could potentially benefit directly from a speed bump by using brokers who are able to actually take advantage of the speed bump, e.g., those that have invested in colocation, direct feeds, event-based processing, etc.
Of course, those that invested in technology in order to exploit these extremely short-lived informational advantages would be losers in a speed bump world, as the return on their investments in “speed” fall. This of course begs the question of whether this is fair? Instead of answering that here, I’ll leave that question for such HFTs to ponder, perhaps as they gaze out the windows of their Uber and watch the empty medallioned taxis roll by.
The author is the Founder and President of The Bacidore Group, LLC. For more information on how the Bacidore Group can help improve trading performance as well as measure that performance, please feel free to contact us at info@bacidore.com or via our webpage www.bacidore.com.
For an overview of our other services, please click HERE.
And please check out our other blog posts available HERE.
Copyright 2019, The Bacidore Group, LLC. All Rights Reserved.
FOOTNOTES
[2] Of course, this also puts the slower high-speed liquidity providers on more equal footing with their faster counterparts by reducing their advantage by a few milliseconds.
[3] The ability for high frequency market makers to avoid losses could, in theory, increase the losses to other, slower liquidity providers. For example, suppose HFT takers were more likely to hit the HF market markets first in the absence of a speed bump, e.g., because HF market makers’ speed consistently puts them at the front of the queue. But under a speed bump, these front-of-queue liquidity providers can avoid losses by updating their quotes during the speed bump delay, leaving the remaining traders to absorb losses they otherwise wouldn’t have. However, when HFTs as a group are attempting to exploit a mispricing, they would likely sweep all of the liquidity with or without a speed bump. Therefore, the incremental effect on slower liquidity providers is probably quite small or non-existent.
[4] The speed bump applied by the IEX, by contrast, is a fraction of a millisecond and was deemed a “de minimis” delay. Consequently, IEX quotes are deemed protected despite the speed bump.