Of all the ways in which financial markets have recently changed, one of the most significant has been a drastic increase in speed. Exchanges process orders faster than ever before, and the so-called ‘high-frequency traders’ have specialised in developing the ability to react as quickly as cutting-edge technology permits. At the time of this writing, the NYSE reports its median latency as 26 microseconds and 65 microseconds at the 99th percentile.1 To illustrate, a fighter jet travelling at Mach 1 would move just 1–2 centimetres during that time!
What are the implications of these faster speeds? Several studies have examined the data for answers to that question. While there are some exceptions, the bulk of the empirical evidence associates faster speeds with smaller bid-ask spreads (e.g. Hendershott et al. 2011).2,3 This is a positive effect of faster speeds. Despite evidence that faster speeds accelerate the speed at which available information is worked into prices, recent empirical research has highlighted that faster speeds might reduce the amount of information that becomes available in the first place, leading to less informative prices in the end (e.g. Weller 2018). This is a negative effect of faster speeds.
Taking those two effects together, faster speeds seem to induce a trade-off – they lead to smaller spreads but may also lead to less informative prices.4 In recent research (Baldauf and Mollner 2019), we capture this trade-off within a theoretical model of high-frequency trading in modern financial markets. We then use the model to evaluate some potential market design responses to high-frequency trading that are currently in debate. But before discussing those responses, let us describe the model and explain why the aforementioned trade-off arises.
The model features a single security that is traded on multiple exchanges, latency (that may be random in length), and three types of traders: (i) ‘liquidity-investors’ who trade for exogenous hedging, saving, or borrowing motives; (ii) ‘information-investors’ who may, through costly research, obtain and subsequently trade on private information; and (iii) high-frequency traders.
In equilibrium, as in other literature (e.g. Budish et al. 2015), high-frequency traders play two roles. One type, the ‘liquidity provider’, facilitates trade by posting quotes at all exchanges. Others, ‘snipers’, wait to trade until order flow reveals a sufficiently strong signal of the value of the security. Similar to other literature (e.g. Easley and O’Hara 1987), information-investors trade larger quantities than liquidity-investors in equilibrium, so that order flow signals the investor’s type.
Because latency is random, an information-investor cannot ensure simultaneous processing of orders sent to several exchanges. Thus, if high-frequency traders are sufficiently fast, they may observe the trade generated by the first such order to be processed by an exchange and react to that signal on the remaining exchanges before the orders of the original information-investor are processed there. In equilibrium, the liquidity provider reacts by sending cancellations for her remaining quotes, which we call ‘passive-side order anticipation’. Snipers react by sending orders to trade against the remaining quotes, which we call ‘aggressive-side order anticipation’. The resulting winner-take-all races may be won by the information-investor, the liquidity provider, or a sniper.
Faster speeds enable high-frequency traders to be more successful at order anticipation, which produces the trade-off introduced above. A negative effect is a reduction in information production. Intuitively, order anticipation reduces the amount of economic rent that information-investors can extract by trading on a piece of information, thereby weakening the incentive to obtain such information. Less fundamental research is then conducted so that market prices become less informative about the fundamental value of the security.
However, a positive effect is an improvement in liquidity as measured by the bid-ask spread. To see why order anticipation achieves this, first note that a spread arises because the liquidity provider must offset the losses created by adverse selection (i.e., the fact that information-investors and snipers trade against her quotes only when they are mispriced). Order anticipation reduces the spread by lessening the adverse selection faced by liquidity providers through two channels: (i) passive-side order anticipation is itself successful avoidance of adverse selection; and (ii) through its effect on research, order anticipation reduces the amount of asymmetric information that is available to create adverse selection.
An inefficiency of high-frequency trading
Two of the things that one would most desire from our markets are small spreads and informative prices. Sometimes these two objectives conflict, so that there may be a trade-off between them. Indeed, high-frequency trading itself illustrates this trade-off – faster speeds lead to smaller spreads but also to less-informative prices. Nevertheless, one would ideally like our markets to approach the frontier of that trade-off.
In our model, the baseline scenario – in which exchanges operate conventional limit order books – is not on the frontier of this trade-off between small spreads and informative prices. The reason is what we referred to above as aggressive-side order anticipation – situations in which a high-frequency trader, reacting to order flow or public news, trades against a quote in the instants before it would have been cancelled.
The intuition is the following. To reach the frontier, it is necessary that all economic rents from informed trading flow to the traders who actually produce that information. However, the effect of aggressive-side order anticipation is that certain high-frequency traders divert some of these rents and profit from information that they did not themselves produce.
Because aggressive-side order anticipation forces outcomes off this frontier, it may be desirable to find ways to eliminate it. The so-called ‘asymmetric speed bump’ can do just that.5,6
Asymmetric speed bumps
An asymmetric speed bump is a (typically short) delay that an exchange applies to the processing of some, but not to all, order types.7 For example, an exchange might delay all orders except cancellations.8 Consequently, a cancellation received slightly after an order of a different type might be processed first, whereas conventional limit order books would process orders strictly in the order received.9
Why would such an asymmetric speed bump eliminate aggressive-side order anticipation? Recall how high-frequency traders react to market signals – the liquidity provider sends cancellations for her remaining quotes (i.e. passive-side order anticipation), and snipers send orders to trade against the remaining quotes (i.e. aggressive-side order anticipation). If delays are applied to the orders sent by snipers but not to the cancellations sent by liquidity providers, then all mispriced quotes are cancelled before snipers can trade against them. Snipers will never be successful, and aggressive-side order anticipation will not occur on path.
Asymmetric speed bumps improve markets by eliminating aggressive-side order anticipation, which, as we argued above, represents an inefficient form of high-frequency trading. Asymmetric speed bumps do not, however, eliminate passive-side order anticipation. But that may be a feature rather than a bug. Passive-side order anticipation is not inefficient in the same way – indeed, it is largely the reason that spreads have become smaller over the same period in which markets have been reshaped by faster speeds.
In recent years, asymmetric speed bumps have been implemented by a variety of venues, including the Aequitas NEO Exchange (Canadian equities, 2015), the TSX Alpha Exchange (Canadian equities, 2015), the Eurex Exchange (German and French equity options, 2019), and the Moscow Exchange (US dollar/ruble foreign exchange, 2019).
Still other venues have received approval for such asymmetric speed bumps but have not yet implemented them, including ICE Futures US Exchange (gold and silver futures) and the London Metal Exchange (precious metals).
Notably absent from this list are all 13 equities exchanges in the US. But that may be about to change. The Chicago Board Options’ (Cboe) EDGA Exchange has recently proposed to add its own asymmetric speed bump, which is currently pending approval by the Securities and Exchange Commission (SEC).
These changes have not been without controversy.10 For example, some have questioned whether it is ‘fair’ for a delay to discriminate based on order type – to delay certain order types but not others. We would, however, emphasise that these delays do not discriminate based on a trader’s identity.
Others have expressed concern over the fact that the immediate beneficiaries of an asymmetric speed will primarily be high-frequency liquidity providers. But formal, model-based analysis can be useful in determining the full equilibrium effects of such changes and identifying who the ultimate beneficiaries are likely to be. In particular, our model suggests that, although high-frequency liquidity providers will face better conditions because of the speed bump, competition among them will likely lead them to quote tighter and deeper markets, thereby passing some or all of the benefits on to other traders.
In developing new ways to organise trading, we prefer a scalpel to a hatchet. After all, the functioning of financial markets is of paramount importance – both for bringing buyers and sellers together, and for posting up-to-date prices that reflect information – and we should be risk averse about big changes that may prove disruptive. For similar reasons, it might be difficult to persuade exchanges and market participants to accept a big change.
Asymmetric speed bumps change the rules of the game but in a small, surgical way. This might explain why exchanges and regulators have been willing to experiment with them so far. Nevertheless, even tiny speed bumps could have a big impact, allowing us to harness even more of the benefits of well-functioning markets and without disrupting the status quo.
Baldauf, M and J Mollner (2019), “High-frequency trading and market performance”, Journal of Finance, forthcoming.
Budish, E, P Cramton and J Shim (2015), “The high-frequency trading arms race: Frequent batch auctions as a market design response”, Quarterly Journal of Economics 130(4): 1547-1621.
Easley, D and M O’Hara (1987), “Price, trade size, and information in securities markets”, Journal of Financial Economics 19(1): 69-90.
Hendershott, T, C M Jones and A J Menkveld (2011), “Does algorithmic trading improve liquidity?”, Journal of Finance 66 (1): 1-33.
Kirilenko, A A, A S Kyle, M Samadi and T Tuzun (2017), “The flash crash: The impact of high frequency trading on an electronic market”, Journal of Finance 72(3): 967-998.
Menkveld, A J and M A Zoican (2017), “Need for speed? Exchange latency and liquidity”, Review of Financial Studies 30(4): 1188-1228.
Weller, B M (2018), “Does algorithmic trading reduce information acquisition?”, Review of Financial Studies 31(6): 2184-2226.
Zoican, M (2014), “Finance at the speed of light: Is faster trading always better?,” VoxEU.org, 20 September.
 See https://www.nyse.com/pillar (reported statistics are for the binary protocol). A microsecond is a millionth of a second.
 The bid-ask spread is the gap between the price at which one could immediately buy one share and the price at which one could immediately sell one share. The spread is a measure of liquidity, with a smaller spread corresponding to a market that is more liquid and in which transaction costs are smaller.
 One such exception is Menkveld and Zoican (2017). See also Zoican (2014).
 Of course, there are also other implications of faster speeds. For example, some have linked faster speeds with more volatile prices, including the so-called ‘flash crashes’ (e.g., Kirilenko et al. 2017).
 The ‘asymmetric speed bump’ idea corresponds to what we call a ‘non-cancellation delay’ in the paper.
 The so-called ‘frequent batch auctions’ constitute another way of eliminating aggressive-side order anticipation. See Budish et al. (2015) or Section 6.2 of Baldauf and Mollner (2019) for details.
 Note that this differs from applying a delay to all incoming orders, as has been done by venues such as IEX (American equities, 2013) and NYSE American (American equities, 2017). Nevertheless, there are some similarities in terms of the basic economics. See Appendix F.1 of Baldauf and Mollner (2019) for details.
 The basic economics would remain the same if certain other orders (e.g., non-marketable limit orders) were also exempted from the delay. Many of the asymmetric speed bumps that have recently been implemented in practice do exempt other orders beyond cancellations.
 Another question is whether the speed bump ought to be deterministic or stochastic in length. As we discuss in the paper, the answer likely depends on how one prioritises a small spread versus informative prices.
 See, for example, the comments submitted to the SEC concerning Cboe proposed speed bump for the EDGA exchange: https://www.sec.gov/comments/sr-cboeedga-2019-012/srcboeedga2019012.htm.