International Review of Finance, 2017

DOI: 10.1111/irfi.12159

Front-Running and Market Quality:

An Evolutionary Perspective on
High Frequency Trading*
THORSTEN HENS†,‡, TERJE LENSBERG‡ AND KLAUS REINER SCHENK-HOPPɇ,§
†
Department of Banking and Finance, University of Zurich, Zurich, Switzerland
‡
Department of Finance, Norges Handelshoyskole, Bergen, Norway and
§
Department of Economics, University of Manchester, Manchester, UK

ABSTRACT

We study front-running by high-frequency traders (HFTs) in a limit order

model with continuous trading. The model describes an evolutionary equi-
librium of low-frequency traders who compete in portfolio management ser-
vices by offering investment styles. The introduction of front-runners inflicts
heavy losses on speculators, while leaving passive investors relatively
unscathed. This encourages investment in the market portfolio and markedly
reduces overall turnover. Speculative trading persists despite its lower profit-
ability. By most measures, market quality is not affected to any significant
extent by front-running HFTs.

JEL Codes: D53; D47; C63; C73

Accepted: 8 September 2017

I. INTRODUCTION

Financial markets are complex ecologies of interdependent market participants.

Investors differ with respect to styles that focus on value, news, momentum,
sectors, factors, and so on. Moreover, markets rely on market makers to provide
liquidity; price discovery by informed traders; speculators to correct temporary
mispricings, as well as gamblers who are willing to lose enough money to pay
for these services. In addition, there are predatory traders who make money by
front-running or bluffing other traders. This diversity poses a challenge to
financial research because (i) regulations and innovations can affect different
types of market participants in widely different ways, and (ii) the interdepen-
dence between them renders partial analyses ineffective.

* We thank LGT Alpha Generix for having initiated this research paper during their workshop “Rule-
Based Cash Equity Strategies,” April 27, 2016. The work was conducted during a research stay at iFM in
Vitznau, Switzerland. Financial support from P&K-foundation and from the Swiss National Science
Foundation grant 149856 “Behavioural Financial Markets” (2014–2016) is gratefully acknowledged.

© 2017 International Review of Finance Ltd. 2017

 International Review of Finance

We believe that evolutionary approaches are well suited to study this type of
interconnected systems, and in this paper, we use one such approach to ask
how the advent of high-frequency trading (HFT) can be expected to change the
ecology of equity trading strategies. This is an interesting question because so
far, despite many strong opinions raised in the public (cf. Lewis 2014), very lit-
tle is known about the impact of HFT on the whole ecology of the market.
The excellent recent surveys of the HFT literature by Menkveld (2016) and
Biais and Foucault (2014) stress several channels through which HFTs can profit
from the relative slowness of other market participants.
HFTs engaged in market-making are able to trade less with informed traders
since higher speed reduces the risk of having limit orders being picked off. As
market makers lose to informed traders, being faster increases HFT market
makers’ profits as well as enables them to offer better prices and provide more
liquidity to uniformed traders.
HFTs focussing on intermarket trading (at present using microwave commu-
nication infrastructure) reap the benefits from statistical arbitrage. Being the
first to trade on information revealed in one venue (e.g., a stock market) at
another venue (e.g., a derivative market), HFTs contribute to consistent pricing
of financial assets across fragmented markets. For an illustration see Laughlin
et al. (2014) who analyze latency in information transmission between E-Mini
S&P 500 futures traded in Chicago and SPY traded in New York.
HFTs can prey on large orders by front running. When front-running large
market orders, an HFT takes liquidty and offers it at a worse price. This front-
running strategy works by anticipating market orders (or marketable limit orders
with price impact) from an investor, buying up liquidity using market orders
and posting it at a less favorable price using limit orders. The original order is
then executed at a worse price than anticipated by the investor. Hirschey (2013)
finds that HFTs are able to anticipate order flow and can trade profitably (and
predatorily) on this information This issue is of major concern to institutional
investors, as for example Norges Bank Investment Management (2013), see also
Tong (2015). In this paper, we aim to quantify the extent to which front-runners
impact market quality and their impact on the market ecology.
Several aspects of the equity market plays into the hands of HFTs seeking to
profit from front-running, see, for example, Adrian (2016) and Biais and Fou-
cault (2014).
Exchanges sell co-location—space in the data center for a server (running
automated trading software) close to the trading venue’s matching engine. Co-
location helps HFTs to submit orders fast but, more importantly, gives faster
access to information on quotes and trades that originates from the matching
engine. This creates an information asymmetry which benefits co-located HFTs.
If co-located HFTs were predominantly market-makers then, as argued above,
co-location reduces the losses to informed traders which benefits uninformed
market participants. These are typically small retail investors but, as observed by
van Kervel and Menkveld (2016), also large institutional investors because HFTs
“lean against the wind” up to a certain volume. Balancing the interests of low

2 © 2017 International Review of Finance Ltd. 2017

 Front-Running and Market Quality

frequency trader (LFT) and HFT market participants, places market venues in a
delicate position (see, e.g., Harris 2003, Part II): Markets exist because utility-
motivated traders want to exchange assets. Attracting and serving the needs of
LFTs is therefore crucial to the viability of a venue. Since profit-motivated traders
can facilitate this exchange, venues should attract HFTs that enhance liquidity.
By some estimates the majority of HFTs are market making, for example, Hag-
strömer and Nordén (2013), and Brogaard et al. (2015) find that co-location
improves liquidity. Tong (2015) claims that HFTs provide expensive liquidity.
Fragmentation of equity markets also benefits HFTs. Before 2009 U.S.
exchanges would “flash” orders before forwarding these to the market if unfilled,
thus making these orders potential prey for front-running, for example Securities
and Exchange Commission (2009, p. 39/40). Since flashing is for about half a
second, only HFTs could respond. An HFT could observe the flashed order and,
rather than filling it at the NBBO, take liquidity at the venue with the best price
and repost the volume at a worse price. The flashed order will end up being filled
at a less favorable price than the one available at the time of flashing.
The decentralized structure of the U.S. equity market allows to employ simi-
lar techniques even after the demise of “flashing.” Since orders need to be filled
at the NBBO (or at better prices), a large enough order will “slosh” around the
market, being filled—in the extreme case, lot by lot and revolve across different
venues. Observing the arrival of an order at one venue, an HFT would then
employ latency arbitrage by buying liquidity at other venues and selling it at a
higher price. In response to latency arbitrage, smart order routers (automated
order submission systems) break up large institutional orders into smaller
“child” orders, see for example, Norges Bank Investment Management (2015).
Prior to 2005 some U.S. market venues allowed quoting prices that give a
price improvement of at least 1 of 100 of a cent. Since the minimum tick size is
1 cent, this effectively allows traders to jump the price priority queue. Hence
“subpenny trading.” SEC Rule 612 banned placement of displayed subpenny
quotes. The rule however does not preclude subpenny trading: traders can for
instance place hidden subpenny quotes, for example, in dark pools, or offer
“in-house” price improvement to incoming client orders (Dick 2010).
Other aspects of market venues can also benefit HTFs to the disadvantage of
LFTs. Venues compete in fee structures, see, for example, Harris (2015). Some
markets offer taker–maker pricing, a fee structure that inverts the classical
maker–taker pricing where limit orders get a rebate and market orders pay a fee.
The idea behind the maker–taker pricing structure is to encourage liquidity pro-
vision. By paying a (small) additional amount of money when a limit order is
hit, the liquidity provider (the “maker”) is compensated by more than the bid-
ask spread when the provided liquidity is taken. The counterparty sending a
market order (the “taker”) pays the rebate, plus possibly a small increment to the
market venue.
Taker–maker pricing rewards submission of market orders. Although at first
glance this might seem perverse, one has to look at the market from a market
maker’s perspective. While the bid-ask spread is the main profit generator, flow

© 2017 International Review of Finance Ltd. 2017 3

 International Review of Finance

Table 1 An order book

Buyer depth Bid Ask Seller depth
46 5.98 6.00 25
80 5.96 6.01 115
120 5.95 6.03 70

(volume) is needed as well. Taker–maker pricing encourages market orders, that

is, generates flow. But taker–maker pricing also opens opportunities for HFTs
through subpenny trading (Buti et al. 2015). Taker–maker pricing also enables
subpenny trading because the net quote, that is, the effective quote after
rebates/fees, differs from the displayed quote. One can then offer more aggres-
sive limit orders by placing them at taker–maker rather than maker–taker
venues (Harris 2015, p. 70/71).
We next describe a simple strategy to illustrate the profitability of predatory
front-running. Consider the order book for a stock in Table 1.
Suppose a front-runner sees an incoming market buy order of size 100.
Before it reaches the order book, the front-runner places a market buy order for
100. The order executes and moves the ask from 6.00 to 6.01. The front-runner
next submits a limit sell order for 100 at 6.0099, which executes against the
incoming (front run) market order. Total profit for the front runner (and cost
for the incoming market order) is 6.0099 × 100 − 6 × 25 − 6.01 × 75 = 0.24. It
is positive because the front-runner benefits from walking up the sell order
book before selling to the incoming market order at the new and higher ask.1
The limit sell order for 100 at 6.0099 does not obey the usual tick size. As
explained above this order could be a hidden subpenny order or an in-house
cross with an NBBO-improving price (6.0099 is better than the current ask
of 6.01).
It is important to note that only market orders with market impact can be
profitably front run. Therefore, since larger orders have more market impact,
aggressive LFTs will pay more than passive ones if there are front runners in the
market. Aggressive LFTs include news traders and other short-term speculators.
It is commonly assumed that the activity of such traders improve price discov-
ery and other aspects of market quality. In this paper, we ask whether such
traders will reduce their activity level in the presence of HFT front-runners, and
if so, if market quality suffers as a result.
Our model describes financial markets from an evolutionary perspective to
capture the complex ecology of many interdependent market participants.
Building on Lensberg et al. (2015), we consider LFT investment funds and HFT
front-runners interacting in a limit order market. The model presents a wide

1 The front-runner could make more money by trading more aggressively: First, place a larger
market buy order (140 instead of 100) in order to push up the ask to 6.03. Second, use a mar-
ket sell order to dispose of 40 at the best bid of 5.98. Finally, use a limit sell order for 100 at
6.0299 to sell 100 to the incoming market order. This yields a net profit of 1.04, against 0.24
for the simple strategy. In his paper, we only consider simple front-running strategies.

4 © 2017 International Review of Finance Ltd. 2017

 Front-Running and Market Quality

range of potentially important mechanisms whose equilibrium effects can be

measured and compared in a coherent manner.
In the model traders’ risk preferences and specific roles (news trader, value
investor, liquidity provider, etc.) are endogenous. There are no a priori assump-
tions on funds’ equity trading strategies. Market equilibrium and rational
behavior of funds are the outcome of a competitive evolutionary process, as
suggested by Alchian (1950). In equilibrium, funds act “as if” they were rational
and the market is efficient. By pursuing this approach, we can encompass the
intertwined changes in market quality and LFTs’ behavior in response to
changes to the market environment due to front-running.
Within this framework the equilibrium effects of front-running HFTs can be
quantified by comparing two scenarios: a market without resp. with front-run-
ning. The model contains a large number of fund managers who trade the stock
and bonds issued by an aggregate firm. There is substantial endogenous hetero-
geneity among individual funds, and specialization on trading styles such as
news trading, value trading, and market making. We ask to what extent front-
running impacts the portfolio holdings, order flow, liquidity, price discovery,
short-term volatility, and long-term price dynamics.
Evolutionary approaches using genetic algorithms to study trading in finan-
cial markets form a growing literature. Recent notable examples are Allen and
Karjalainen (1999), Arifovic and Gençay (2000), Gradojevic and Gençay (2013),
Chiarella et al. (2015), Arifovic et al. (2016).
Our paper is closely related to Arifovic et al. (2016) who also analyze an evo-
lutionary model of order-driven markets with HFTs and LFTs. Their main
emphasis is on the role of information about fundamental values in the pres-
ence of different speeds of trading. HFTs can have both an informational and a
speed advantage. In our paper, HFTs do not have any superior information
about fundamental values. They only possess the ability to observe the order
flow which is used for front-running. In this sense these two papers explore
complementary issues.

II. MODEL

We give a short explanation of the model and refer the reader to Lensberg
et al. (2015) for further details.
Earnings per share of the aggregate firm are determined by a geometric
Ornstein–Uhlenbeck stochastic process with a time-varying mean μ st , where st
is the (unobservable) state of the economy at time t. The economy is either in
expansion (st = 1) or contraction (st = 0). Bonds are risk-free consols, and divi-
dends equal total earnings minus total interest payments.
Shares are traded against bonds in a continuous market with a limit order
book. Fund managers arrive at the market at random times and can submit mar-
ketable limit orders. An order crossing the spread is a market order and is imme-
diately executed. Partial execution is possible with the unfilled quantity of the

© 2017 International Review of Finance Ltd. 2017 5

 International Review of Finance

order being added to the book. An order that is behind the market, that is, does
not cross the spread, is added to the book where it stays until it is executed or a
new order is submitted by the trader. The usual price-time priority in order exe-
cution applies.
Margin trading is allowed subject to a 50% initial and 33% maintenance
margin requirement. This means that funds can sell short as well as leverage
long but only to the extent that they meet these margin requirements. Margin
violations lead to margin calls and the automatic submission of market orders
to restore the margin requirement. Bankrupt funds are removed from the mar-
ket and their holdings are distributed among all non-bankrupt funds in propor-
tion to their managed wealth.
Each fund has a quantitative trading strategy, represented as a computer pro-
gram that maps input data to limit orders. Input data consist of market data
(bid, ask and quantities at these prices, change of mid-price in the last 24 h),
fundamental information [risk-neutral price (RNP) of the stock], news (change
in RNP in the last 24 h), current portfolio holdings, and the state of the margin
account. Limit orders consist of a price and a quantity.
The model is calibrated to the current U.S. market by either drawing on
empirical observations or choose parameter values that give results consistent
with historical averages and stylized facts of the business cycle, aggregate earn-
ings and earnings surprises [details are provided in Lensberg et al. 2015].
The market is populated with 20,000 funds. This number is motivated by U.-
S. institutional investment industry research that puts the number of mutual
funds at around 8000 [Investment Companies Institute (ICI)‘s 2017 Investment
Company Fact Book] and that of traditional hedge funds at 7500–8000
(eVestment).
Funds compete for survival in a market for portfolio management services,
where high performing funds attract clients from low performing ones. The
evolution of strategies in the market is modeled in terms of a genetic program-
ming algorithm with tournament selection. At the end of every trading day
(250 per year), there are four tournaments. In each tournament, eight funds are
randomly selected and ranked by wealth under management. The computer
programs of the worst two are replaced by copies of the computer programs of
the two best ones. Then, with certain probabilities, the two copied programs
are subject to random crossover and mutation. This yields a 10% p.a. exit/entry
rate which is in line with the attrition rate of U.S. equity fund managers (Busse
et al. 2010) and hedge funds (Preqin, Hedge Funds in 2016). Hedge fund launch
sizes are estimated using data from AIMA/GPP (Emerging Managers Survey
2017) and eVestment (Alternatives Research, Impact of Age and Size on Hedge
Fund Performance, July 2015). Computing the mean assets under management
of new funds relative to the industry average, we obtain estimates of 9% and
32% respectively. Based on these observations we endow new funds with 20%
of the average portfolio.
While individual funds can only increase their managed wealth through
superior portfolio returns, the evolutionary process selects for just those

6 © 2017 International Review of Finance Ltd. 2017

 Front-Running and Market Quality

strategies which yield superior performance. This leads to an increase in the

number of funds that follow similar strategies. In equilibrium, surviving strate-
gies tend to cluster around a small number of distinct investment styles. The
ecology of the market is given by these styles and the total assets under man-
agement of each style.

A. Dataset
Two scenarios are considered: a benchmark scenario (BAS) with only LFTs, and
a second scenario which also includes a high frequency trader (HFT). The
behavior of the HFT is exogenous to the model: It uses the simple front-
running strategy discussed in connection with Table 1 against all incoming
market orders from LFTs. This assumption on front-running is more convenient
(and simpler) than using more sophisticated (and more profitable) predatory
strategies but having HFTs anticipating only part of the order flow (and thus
miss out on a proportion of profitable trades). To implement the simple front-
running strategy, an HFT only requires knowledge about the size of the arriving
order, the state of the book and the ability to place an order in front of the
incoming one; features that are inherently provided by co-locating and sub-
scribing to fast intermarket connections.
Comparing the market dynamics and LFT behavior between these two sce-
narios, we can assess the impact of front-running HFTs on market quality,
informational efficiency and investor behavior. For each scenario, we do
200 independent runs with the model. Each pair of runs uses identical seeds for
the random number generator. This means, in particular, that sample paths for
the earnings process are identical across scenarios for the same run. In this
manner, we obtain two datasets, each consisting of 200 independent observa-
tions which allow for pairwise comparisons between scenarios at any time
scale.
The BAS scenario is identical to the base case of Lensberg et al. (2015). For
each one of 200 runs, the model is solved by running it for 15 million trading
days. At that point, the pricing process passes some standard tests of rationality
and dynamic stability, that is, the model is judged to have converged to a
“steady state.” The HFT scenario is identical to the BAS scenario during the first
10 million trading days. At that point, we introduce the HFT who continues to
follow its front-running strategy during the remaining 5 million trading days.
After 15 million trading days, we run the models for an additional 10,000
trading days to collect data. The data include measures of trading behavior,
market quality, and front-running costs. The latter are costs to LFTs of doing
business with front-runners. We distinguish between the impact and equilibrium
effects on front-running costs. The impact effect would be the cost paid by LFTs
if they did not change their behavior in response to the introduction of the
HFT, and the equilibrium effect is the cost they pay after having adapted their
behavior to the new situation. The impact and equilibrium effects are measured
with data from the BAS and HFT scenarios, respectively.

© 2017 International Review of Finance Ltd. 2017 7

 International Review of Finance

III. RESULTS

Results on the impact and equilibrium effects of HFT are organized in two parts.
The first part analyzes aspects of market quality: Price discovery; Short-term price
fluctuations; Long swings and Liquidity (Table 2). The second part is concerned
with investor behavior (Table 4). We classify the traders by their portfolio posi-
tion (Short, Only bonds, Mostly bonds, Market portfolio, Mostly stock, Only stock,
and Leveraged long) and analyze the effect of the BAS and the HFT scenarios on
these types: relative frequency, wealth under management, trading activity,
limit versus market order usage, and loss to front-runners.

A. Market quality
i. Price efficiency
Speculators contribute to price discovery by trading mispriced assets. They
make money by trading aggressively on short-term informational advantages.
Speculators are therefore particularly vulnerable to front-running. This leads to
the hypothesis that front-running HFTs will reduce price efficiency. To test the
hypothesis, we regress daily log stock returns on daily log innovations to RNP
(the risk-neutral price of the stock), and use the corresponding R2s as our mea-
sure of price discovery. We also test for asymmetric price responses to good and
bad news by computing upside and downside R2s for days with an increasing
(decreasing) RNP.
Panel (1) of Table 2 contains the results. We find no significant differences
between the two scenarios. Mean stock prices are the same, price discovery (R2)
and price response to good/bad news (R2 down-up) are unchanged. There is no
evidence that front-runners harm price efficiency.

ii. Short- and long-term price fluctuations

We next ask whether front-runners destabilize the market by increasing price
fluctuations. Front-runners’ trading could increase excess volatility in the short
run by triggering cascades of buying or selling. They could also make long-term
swings in the market more pronounced if their presence leads to a flight to
safety by LFTs in bad times.
Panel (2) of Table 2 contains results on short-term price fluctuations. We
report volatility, skewness and kurtosis for daily returns. There are no signifi-
cant differences between the BAS and the HFT scenario.
Long swings in prices across business cycles are measured as the mean per-
centage decline in the stock price from a peak in an expansion to the trough in
the subsequent recession. This is the “Peak-to-trough” variable in Panel (3) of
Table 2. The variables “High” and “Low” are the means of maximal and mini-
mal stock prices across 200 40-year periods, and “Range” is the difference
between “High” and “Low.” We find no significant differences in any of these
results.

8 © 2017 International Review of Finance Ltd. 2017

 Front-Running and Market Quality

Table 2 Market quality indicators

BAS HFT p-Value
(1) Price discovery
Stock price 15.37 15.43 (0.486)
R2 (dp = a + b
dRNP) 0.874 0.874 (0.973)
R2 down-up 0.037 0.030 (0.270)
(2) Short-term price fluctuations
Volatility (annualized) 17.36 17.04 (0.369)
Skewness 0.08 0.22 (0.434)
Excess kurtosis 18.34 35.49 (0.149)
(3) Long swings
Peak-to-trough (%) 43.14 42.65 (0.209)
High 21.49 21.32 (0.628)
Low 8.16 8.23 (0.514)
Range 13.33 13.09 (0.551)
(4) Liquidity
Bid-ask spread (bp) 10.18 10.19 (0.948)
Market impact (bp) 1.07 1.15 (0.087)
Roundtrip cost (bp, ex. front-running 12.32 12.50 (0.484)
cost)
Turnover per day (% of outstanding 2.46 1.90 (0.000)
shares)
Days between trades 5.23 5.48 (0.000)
Order size (number of shares) 6067 4882 (0.000)
Order book depth (market impact of 50 K 3.11 3.70 (0.000)
shares)
Notes: (1) Price discovery. For each scenario and run, we compute the R2 obtained by regressing
daily log returns on daily log innovations in the RNP of the stock. Upside and downside R2 are
calculated in the same way, except for restricting the data to days with a non-decreasing and
decreasing RNP, respectively. Downside-minus-upside R2 is the difference between the downside
and the upside R2. (2) Short-term price fluctuations. Annualized volatility, skewness and excess kur-
tosis are computed by run from daily log-returns. (3) Long swings. For each scenario, Peak-to-
trough is the mean percentage decline in the stock price from the peak in an expansion to the
trough in the subsequent recession. An expansion (recession) is defined as an interval of trading
days T = {t1, …, tk} such that the state variable st is 1 (0) on all days in T and 0 (1) on days t1–1
and tk + 1. There are 687 of these events across the 200 independent runs of the model. High
(Low) is the mean across 200 runs of the maximum (minimum) closing stock price across all
10,000 trading days of that run. Range is the difference between High and Low. (4) Liquidity. The
bid-ask spread is the difference between the closing ask and bid. Market impact is the difference
between the current bid (ask) and the average execution price of a market sell (buy) order. Round-
trip cost is the bid-ask spread plus two times the market impact. Turnover per day is trade volume
divided by the number of shares outstanding. Days between trades is the average number of days
between two consecutive trades by the same investor. Order size on a given day is trade volume
divided by the number of trades. Except for Panel (3) Long swings, the datasets consist of
200 run means of daily observations of each variable. p-Values in parentheses are computed from
paired t-tests of differences in means between the HFT scenario and the base case

iii. Liquidity
We measure liquidity in several dimensions: quoted bid-ask spread, price
impact (slippage) of market orders, roundtrip cost, order book depth, daily LFT
turnover, and average LFT order size. For the HFT scenario, all measures include

© 2017 International Review of Finance Ltd. 2017 9

 International Review of Finance

the equilibrium effects on the order book by HFTs, but their immediate effects
are excluded. In particular, the slippage of a particular market order is com-
puted under the assumption that the order will not be subject to front-running.
The immediate impact of front-running HFTs on transaction costs are consid-
ered separately in the next subsection.
Panel (4) of Table 2 collects the results on liquidity. It shows that front-
running by HFTs has a significant negative effect on trading activity and market
depth: Trade volume and order book depth are both reduced, and LFTs trade
less frequently in smaller order size. However, the table also shows that the
market digests these negative volume effects without affecting equilibrium
transaction costs, as measured by the bid-ask spread and market impact.
Summing up, we find that front-running reduces trading activity without
affecting other aspects of market quality. The latter may seem to be at odds
with the empirical literature, which reports significant effects of HFT activity on
price discovery, short-term volatility and transaction costs (Menkveld 2016).
Two differences between our approach and the existing literature could account
for this apparent anomaly: First, we isolate the effect of front-running from
other aspects of HFT behavior, such as market marking, arbitrage, and momen-
tum strategies. Front-running in our model affects about 20% of the trading
volume, which is only a part of the total activity of HFT reported in the empiri-
cal literature (Brogaard 2010). Second, our data are obtained from controlled
experiments where the effects of HFTs are measured across a large number of
very long time periods.
We provide an example to illustrate these points. Many authors have found
evidence that HFT activity increases intraday volatility. To obtain a standard
test of this hypothesis we use data from the HFT scenario only. We represent
HFT activity by HFT trade volume and measure intraday volatility as the root
mean of volume-weighted relative variances of daily transaction prices. Table 3
contains the results, and Model (1) seems to indicate that HFT activity increases
intraday volatility.

Table 3 Effect of HFT trade volume on intraday volatility, defined as the root mean
across all trading days of volume-weighted relative variances of daily transaction
prices
(1) (2) (3)
HFT trade volume 0.322 (0.015) 0.222 (0.077) 0.193 (0.129)
Market distress 0.390 (0.000) 0.365 (0.000)
LFT aggressiveness 0.220 (0.000)
Observations 200 200 200
Adjusted R2 0.099 0.238 0.282
Notes: Market distress is represented by the state variable (1 − st), which is 1 if the economy is in a
downturn. LFT aggressiveness is the fraction of LFTs who hold short or leveraged long positions
(margin traders). Such traders are conspicuous for trading frequently in large size, compare
Section III.B. OLS on standardized variables with robust standard errors. p-Values in parenthesis

10 © 2017 International Review of Finance Ltd. 2017

 Front-Running and Market Quality

But this effect is driven by exogenous differences between business cycles

across runs, with corresponding differences in market distress, volatility, and
aggressiveness of LFTs. Since the front-running HFTs prey on aggressive LFTs,
we observe a spurious positive relationship between volatility and HFT activity.
Table 3 supports these claims by showing that nothing significant is left to be
explained by HFT activity if one controls for Market distress and LFT aggressive-
ness. This agrees with the results on daily volatility in Table 2, where the con-
founding factor has been eliminated by comparing data across 200 identical
pairs of business cycle realizations. We can conclude that if HFTs increase
short-term volatility, it must be due to other HFT activities than front-running.

B. Investor behavior
We now turn to the effect of HFT on the behavior of investors. In particular we
want to understand why and how trading activity is reduced, as observed in
Section III.A.
Traders are classified by portfolio position which are categorized into seven
types: Short, Only bonds, Mostly bonds, Market portfolio, Mostly stock, Only stock,
and Leveraged long. For instance, Short comprises investors who sold the stock
short and have a negative position in the stock. Leveraged long are all investors
who borrow to buy the stock. Lensberg et al. (2015) found marked differences
in the behavior across these types. We therefore ask here whether and how they
react differently to HFT.
Table 4 contains results on the effect of HFTs on the investor population, the
behavior of different types and their trading costs. The table contains four
panels of variables: (1) three frequency distributions of trader types; (2) order
usage for each trader type (limit, market and no order), (3) two measures of
trading activity, and (4) two measures of front-running (FR) costs.
The top panel (1) of Table 4 shows that close to half of total wealth under
management is held by funds who invest in the market portfolio. Only about
15% of the wealth is held by traders who are long in one asset only (Short,
Only bonds, Only stock and Leveraged long), but these traders represent almost
half the population and they submit 50% of all orders.
These findings reveal a market populated by investors who range from small
active funds with extreme positions (Speculators) to large passive funds who
hold the market portfolio (Index funds). The effect of introducing front-running
HFTs is a small, but significant increase in the number of index funds (as well
as a matching increase in the proportion of submitted orders), and a corre-
sponding reduction of traders with speculative positions.
Panel (2) of Table 4 contains information about order usage. Overall, the
traders place 2–3 times more limit than market orders, and index funds are least
likely to use market orders. Speculative traders respond to front-running by a
slight, but significant reduction in their use of market orders. Index funds do
not change the composition of their order flow.

© 2017 International Review of Finance Ltd. 2017 11

 Table 4 Summary statistics by traders’ portfolio positions

12
Trader portfolio position
Only Mostly Market Mostly Only Leveraged All
Variable Scenario Short bonds bonds portfolio stock stock long positions
(1) Frequency distributions (%)
Managed wealth BAS 3.9 2.5 13.7 45.5 24.1 5.4 5.0 100
HFT 3.9 2.1 13.8 46.7 23.7 5.1 4.7 100
Traders BAS 18.8 5.2 9.9 28.9 13.0 6.6 17.7 100
HFT 18.6 4.9 10.1 29.9 12.6 6.1 17.7 100
Submitted orders BAS 21.4 4.3 10.5 27.1 12.6 5.2 18.8 100
HFT 20.9 4.1 10.7 28.2 12.4 4.9 18.9 100
(2) Order usage (%)
Limit orders BAS 62.9 42.1 50.9 51.2 52.0 42.1 52.8 52.6
HFT 62.5 42.9 51.4 51.5 52.7 42.0 53.9 53.1
Market orders BAS 21.1 21.6 25.4 18.1 20.2 20.9 25.5 21.1
HFT 20.3 19.9 24.6 17.7 19.3 20.7 24.5 20.2
No order BAS 16.1 36.4 23.7 30.7 27.8 37.0 21.6 26.4
HFT 17.3 37.2 24.0 30.8 28.1 37.3 21.6 26.7
(3) Trading activity
Market order size (value/1000) BAS 102 183 110 60 69 119 100 89
HFT 88 154 85 46 58 103 83 72
Daily turnover (% of wealth) BAS 8.37 7.64 1.60 0.48 0.56 3.01 7.84 1.40
HFT 6.92 6.31 1.23 0.38 0.45 2.76 6.55 1.09
International Review of Finance

(4) Front-running (FR) costs

Unit FR cost (bp of trans. Value) IMP 1.99 1.49 1.24 1.14 1.26 1.25 1.61 1.46
EQ 1.98 1.46 1.06 0.98 1.12 1.16 1.55 1.39
Annual FR cost (bp of wealth) IMP 35.9 25.5 4.5 1.3 1.6 8.1 26.1 4.8
EQ 29.1 21.7 3.1 0.8 1.1 7.1 21.6 3.5
Notes: Let ðα,βÞ≔ðS=S,B=BÞ for a portfolio with S stocks and B bonds, where ðS, ℬÞ is the current number of stocks and bonds outstanding. If
α + β > 0, we define λ = α/(α + β) and classify the portfolio as Short if λ < −0.05; Only bonds if −0.05 ≤ λ < 0.05; Mostly bonds if 0.05 ≤ λ < 0.35;
Market portfolio if 0.35 ≤ λ < 0.65; Mostly stock if 0.65 ≤ λ < 0.95; Only stock if 0.95 ≤ λ < 1.05; and Leveraged long if λ ≥ 1.05. If α + β ≤ 0, the
portfolio is classified as Short if α < 0, and as Leveraged long if β < 0. Investors with portfolios such that α ≤ 0 and β ≤ 0 are bankrupt and
excluded from the classification. For each scenario, run, and variable, we compute a histogram on the bins of this classification. The number

© 2017 International Review of Finance Ltd. 2017

of observations is 200 in each scenario. p-values of t-tests of differences in means between the two scenarios are represented as p < 0.01
and p ≥ 0.01.
 Front-Running and Market Quality

Panel (3) depicts market order size and daily turnover across trader types.
Speculators submit orders almost twice as large as those of index funds, and
their turnover (measured as percentage of wealth under management) is more
than 15 times larger. Front-running by HFTs causes most trader types to reduce
the size of their market orders by 15–20%, which leads to a similar reduction in
their turnover.
Panel (4) shows the costs incurred by trading with front-runners. IMP and
EQ denote the impact and equilibrium costs, computed from scenarios BAS and
HFT respectively. The impact cost is the cost to LFTs if they did not change
their behavior in response to front-running. The equilibrium cost is the actual
cost incurred.
Annual impact costs relative to managed wealth range from 1.3 basis points
for index funds to 35.9 bp for traders with short positions. Equilibrium costs
are lower than impact costs for all trader types, but with substantial variations
in relative difference. By changing their trading behavior, speculative traders
save 20% on their impact cost against 40% for index funds. This reduction in
cost is a combination of many factors: trade frequency, order aggressiveness,
order size relative to liquidity and order timing. Measuring the difference
between impact and equilibrium cost relative to transaction value (unit FR
costs), we find that index funds save 14% over their impact costs, while for
speculative traders there is no significant difference.
Our results show that speculators respond less strongly to front-running than
index funds. The latter markedly reduce their losses to front-runners by chang-
ing trading behavior. Speculators, in contrast, are less successful at reducing
losses due to front-running.
Since many speculators make money by trading large size on short-lived
information advantages (see Lensberg et al. 2015, Table 9 and discussion on
p. 113), they have limited opportunities to search or, in the present model,
wait, for good liquidity. Index funds do not face this tradeoff because they can
trade less with greater patience. In the presence of front-runners this makes a
substantial difference. Although speculators lose up to 29 bp p.a. to front-run-
ners, their number and wealth under management remains nearly unchanged.
This is consistent with our finding that front-runners do not reduce the overall
market quality.
The absence of a major change in the composition of the investor popula-
tion shows that the BAS scenario is quite robust to small increases in unit trans-
action costs. This result can be interpreted from an evolutionary perspective.
Small funds must deliver superior performance in order to attract clients, but in
an efficient market there is no strategy that can be expected to achieve that
goal. In particular, hiding in the crowd of index funds is not a viable alterna-
tive. Superior performance can only be achieved by chance, and this requires
some amount of gambling. Our results on unit FR costs in panel (4) of Table 4
show that small funds are willing to pay a net fee of one basis point for a
chance to beat the market portfolio.

© 2017 International Review of Finance Ltd. 2017 13

 International Review of Finance

IV. CONCLUSION

Front-running by HFTs extracts rents from investors who need to trade large
size quickly. In our model, these investors are speculators who seek to gain from
news events and short-term mispricing. Index funds who tend to hold large
market portfolios are much less affected. They become more patient and trade
more carefully in the presence of HFTs.

Overall, turnover is lower in the presence of front-running HFTs, and there is

a small but significant tendency to invest more in the market portfolio. By most
measures, market quality is not affected to any significant extent by front-
running HFTs. Speculative trading persists despite its lower profitability because
market efficiency forces small funds to gamble in order to survive.

Thorsten Hens
Department of Banking and Finance
University of Zurich
Plattenstrasse 32
Zurich 8032
Switzerland
[email protected]

REFERENCES

Adrian, J. (2016), ‘Informational Inequality: How High Frequency Traders Use Premier
Access to Information to Prey on Institutional Investors’, Duke Law & Technology
Review, 14, 256–79.
Alchian, A. A. (1950), ‘Uncertainty, Evolution, and Economic Theory’, Journal of Politi-
cal Economy, 58, 211–21.
Allen, F., and R. Karjalainen (1999), ‘Using Genetic Algorithms to Find Technical
Trading Rules’, Journal of Financial Economics, 51, 245–71.
Arifovic, J., and R. Gençay (2000), ‘Statistical Properties of Genetic Learning in a
Model of Exchange Rate’, Journal of Economic Dynamics & Control, 24,
981–1005.
Arifovic, J., C. Chiarella, X. He, and L. Wei (2016), ‘High Frequency Trading and
Learning’, Working Paper, Available at SSRN: https://ssrn.com/abstract=2771153
Biais, B., and T. Foucault (2014), ‘HFT and Market Quality’, Bankers, Markets & Inves-
tors, 128, 5–19.
Brogaard, J. (2010), ‘High Frequency Trading and Its Impact on Market Quality’,
Working Paper.
Brogaard, J. et al. (2015), ‘Trading Fast and Slow: Colocation and Liquidity’, Review of
Financial Studies, 28, 3407–43.
Busse, J. A., A. Goyal, and S. Wahal (2010), ‘Performance and Persistence in Institu-
tional Investment Management’, Journal of Finance, 65, 765–90.
Buti, S., Consonni, F., Rindi, B., Wen, Y. and Werner, I. M. (2015). Sub-Penny and
Queue-Jumping. Working paper, Available at SSRN-id2350424.

14 © 2017 International Review of Finance Ltd. 2017

 Front-Running and Market Quality

Chiarella, C., X. He, and L. Wei (2015), ‘Learning, Information Processing and Order
Submission in Limit Order Markets’, Journal of Economic Dynamics and Control, 61,
245–68.
Dick, D. (2010), ‘The Hidden Cost of “Sub-Pennying”,’ CFA Magazine, Jan–Feb, 8–9.
Gradojevic, N., and R. Gençay (2013), ‘Fuzzy Logic, Trading Uncertainty and Techni-
cal Trading’, Journal of Banking & Finance, 37, 578–86.
Hagströmer, B., and L. Nordén (2013), ‘The Diversity of High-Frequency Traders’, Jour-
nal of Financial Markets, 16, 741–70.
Harris, L. E. (2003), Trading and Exchanges: Market Microstructure for Practitioners.
New York: Oxford University Press.
Harris, L. E. (2015), Trading and Electronic Markets: What Investment Professionals Need
to Know. Charlottesville, VA: CFA Institute.
Hirschey, N. (2013), ‘Do High-Frequency Traders Anticipate Buying and Selling Pres-
sure?’ Working Paper.
van Kervel, V. and A. J. Menkveld (2016), ‘High-Frequency Trading around Large Insti-
tutional Orders,’ Working Paper.
Laughlin, G., A. Aguirre, and J. Grundfest (2014), ‘Information Transmission Between
Financial Markets in Chicago and New York’, Financial Review, 49, 283–312.
Lensberg, T., K. R. Schenk-Hoppé, and D. Ladley (2015), ‘Costs and Benefits of Finan-
cial Regulation: Short-Selling Bans and Transaction Taxes’, Journal of Banking &
Finance, 51, 103–18.
Lewis, M. (2014), Flash Boys: A Wall Street Revolt. New York: Norton and Norton
Publishers.
Menkveld, A. J. (2016), ‘The Economics of High-Frequency Trading: Taking Stock’,
Annual Review of Financial Economics, 8, 1–24.
Norges Bank Investment Management (2013), ‘High Frequency Trading—An Asset
Manager’s Perspective,’ NBIM Discussion Note #1–13.
Norges Bank Investment Management (2015), ‘Sourcing Liquidity in Fragmented
Markets,’ NBIM Discussion Note #1–15.
Securities and Exchange Commission (2009), ‘Elimination of Flash Order Exception
from Rule 602 of Regulation NMS,’ RIN 3235-AK40.
Tong, L. (2015), ‘A Blessing or a Curse? The Impact of High Frequency Trading on
Institutional Investors,’ Working Paper.

© 2017 International Review of Finance Ltd. 2017 15