Filetype PDF | Posted on 21 Aug 2022 | 3 years ago
Authentication
digital resources
131x Filetype PDF File size 0.82 MB Source: www.boj.or.jp
Bank of Japan Review 2020-E-5
An Overview of Algorithmic Trading in Foreign Exchange Markets
and Its Impacts on Market Liquidity
Financial Markets Department
FUKUMA Noritaka, KADOGAWA Yoichi*
August 2020
In recent years, the foreign exchange market has seen a growing presence of algorithmic trading, that is,
a process of automated transactions based on pre-determined programs. Concurrently, the need to better
understand its characteristics has become more important. In this paper, we construct proxy indicators
of algorithmic trading in the USD/JPY spot market by focusing on its general features - high-speed and
high-frequency transactions. Based on the proxy indicators, algorithmic trading has been on an upward
trend since around 2016 and is more active in European and U.S. time zones than in Japan. Our analysis
shows that algorithmic trading on average helps improve market liquidity in normal times. Its liquidity-
providing function was generally maintained under market stress triggered by the COVID-19 pandemic
from late-February to end-March 2020, though it could have been dampened albeit temporarily in times
of severe stress when the market experienced sudden and sharp price fluctuation.
Introduction In this paper, we outline FX markets’ algorithmic
trading and conduct quantitative analysis on its recent
Looking back at the transition of transaction methods
developments and impacts on market liquidity in the
1
in the foreign exchange (FX) markets , the electronic USD/JPY spot market.
trading, in which buy and sell orders and transactions
are conducted on electronic platforms, has emerged Algorithmic Trading in FX Markets
since the early 1990s in the interbank market where
banks and securities companies (dealers) transact. At Trading Algorithms
the beginning of the 2000s, electronic trading has Algorithmic trading is categorized into two types:
begun to prevail in the dealer-to-customer market “trading algorithms” and “execution algorithms”
3
where dealers trade with customers including
(Chart 1). Trading algorithms are transactions that
institutional investors. At the early stage, human traders automatically implement a series of investment
make final investment decisions in electronic trading. decision-making processes ranging from price and
However, around the mid-2000s, algorithmic trading volume order to timing, in pursuit of profits. Trading
has started to prevail. In algorithmic trading, a series of algorithms mainly comprise “market make,”
transaction processes varying from an investment
decision to execution are conducted automatically [Chart 1] Major types of algorithms
based on pre-determined programs. In recent years, Strategy Contents
Trading Algorithms
algorithmic trading has been on an upward trend
Providing the bid-ask quotes as market maker and generating the
because it enables high-speed and high-frequency Market make profit from the spread between sell price and buy price (bid-ask
spread).
trading, which human traders cannot implement, and
improves trade efficiency. For example, algorithmic Directional Following a trend or momentum of price, or pursing the profit by
digesting news and events in short time.
trading share has went up to approximately 70-80% in
2 Generating profits by the arbitrage in the same financial products,
2019 in the FX spot market transacted on the EBS , one
Arbitrage in the gap between actual price and theoretical price, and in
of the most commonly used electronic broking systems latency (difference in arrival time of information among markets).
in the interbank market. These shifts in transaction Executing pre-determined amounts of trade smoothly in order to
Execution Algorithms mitigate market impacts (e.g., slicing original orders into small
methods have seemed to change the FX rate pricing orders to execute gradually)
mechanism and market functioning. Thus,
Note: The table is made by referring to Advanced Financial
Engineering Center of NTT DATA Financial Solutions (2018),
understanding characteristics of algorithmic trading is
“Unmasking Algorithmic Trading,” (Kinzai, available only in
becoming important. Japanese) and others.
Bank of Japan August 2020
1
“directional,” and “arbitrage,” and market make is said USD/JPY, resulting in a poor execution result that the
to be widely used in the FX spot market. Market make USD was bought against the JPY at higher prices than
automatically offers a bid–ask quote and makes profits expected when the trading decision was made. In order
from the difference between executed buy and sell to mitigate this unwanted price impact (called “market
prices (the bid–ask spread), which automates dealers’ impact”), in general, a dealer slice a customers’ large
traditional market making function (liquidity order into small orders and execute them gradually.
provision). Execution algorithms automate this type of execution
Dealers, particularly large European and U.S. method, and have been widely used in recent years
6
banks, and non-banks including high frequency trading along with trading algorithms.
(HFT) entities, are said to actively use market make Real money (i.e., pension funds and life insurance
4
algorithms (Chart 2). Market make algorithm controls companies) is said to be dominant execution algorithm
the width of spread and volumes finely, offers new users, and dealers including banks are both providers
7
orders, and changes or cancels existing orders and users of the execution algorithm. Additionally,
depending on overall market developments and market details of execution results are recorded electronically,
order changes. Particularly, non-banks repeat these enabling users to analyze the results and enhance
8
transaction behaviors at high speed and frequency. stakeholder accountability of the execution.
These non-banks have rapidly grown in FX markets,
making them comparable with large European and U.S.
Quantitative Analysis of Algorithmic
5
banks. Meanwhile, price-takers, such as hedge funds, Trading
tend to use other types of trading algorithms, including
In this section, we introduce proxy indicators of
directional. Directional can also be conducted at a high
speed and frequency because taking buy and sell quotes algorithmic trading in the USD/JPY spot markets, and
quickly in response to news contents and market analyze impacts of algorithmic trading on market
developments are essential to generate profits. liquidity in normal times and market stress times.
[Chart 2] Structure of FX markets Proxy indicators of algorithmic trading
Dealer-to-Customer Market
Data in the FX market that can identify individual
Customers traders and analyze their trading behaviors in detail are
(Hedge funds, Real money investors, Non-financial corporates, 9
FX retail aggregators, HFTs） limited. This situation reflects the FX market’s
Interbank Market unique characteristics, that is, no specific regulatory
authorities exist, and various participants trade over-
Banks and Banks and
security companies security companies the-counter (OTC) at various venues all over the world.
(Dealers) (Dealers)
PBservice Inter-Dealer PBservice Hence, specifying “algorithmic traders” and analyzing
HFTsand others Platform (IDP) HFTs and others
（Non-bank （Non-bank their transactions in detail is difficult. Under such
market makers） market makers）
circumstances, certain previous studies focus on
Hedge Funds
algorithmic trading’s general features, that is, high-
Multi-Dealer
Platform (MDP)
Single-Dealer Single-Dealer speed and high-frequency transactions relative to
Dealer-to-Customer Market Platform (SDP) Platform (SDP)
human traders. They measure individual contracts’
Customers
(Hedge funds, Real money investors, Non-financial corporates, transaction speed and regard it as an algorithmic
FX retail aggregators, HFTs） 10
trading if transacted faster than a certain threshold.
Note: The solid and dotted lines show electronic trading and voice
This paper refers to these studies, and tick data of
trading, respectively. IDP, SDP, and MDP are electronic trading 11
platforms. PB refers to the prime-brokerage service (refer to EBS , which is a kind of granular transaction data, are
Note 4).
used to construct the following two proxy indicators of
12
algorithmic trading.
Execution Algorithms First, we construct an indicator referred to as “fast-
While trading algorithms automate a series of decision- paced orders,” which captures a market maker behavior
making process of transactions, execution algorithms that cancels a quote below 100 milliseconds (0.1
13
aim to automatically and smoothly execute a pre- second) after it was newly provided. This indicator is
determined amount of buy and sell contract. For assumed to mainly capture market make algorithm
example, when a dealer seeks to execute a customer’s developments, which typically provide new quotes and
large amount of USD buying/JPY selling order, the cancel them at a high speed and frequency. From the
order execution itself puts upward price pressure on the liquidity provider or consumer’s perspective, this
2 Bank of Japan August 2020
indicator focuses on liquidity providers’ (i.e., market [Chart 3] Algorithm trading indicators
makers) behavior. (time series)
Second, we calculate another indicator called “fast
300 (%) (%) 60
executions,” which captures an investor behavior that
Fast-paced Orders
takes a quote below 100 milliseconds (0.1 second) after 250
Fast Executions (right scale) 55
it was newly provided by market makers. This indicator
200
is assumed to capture liquidity consumers’ (i.e., price-
takers) behaviors, who use trading algorithms 150 50
including “directional.” However, this indicator would 100
14 45
also by and large capture cover deals accompanied
50
with market making activities, implying that this is a
comprehensive indicator that involves activities of 0 40
15 10 11 12 13 14 15 16 17 18 19 20 CY
liquidity providers (market makers) as well.
Note: These indicators are calculated in the USD/JPY spot market.
Based on the concept above, we construct time-
The time-series data are shown as yearly average of indicators
within 10 min after release of the U.S. employment report each
series data of these two indicators, focusing on
month. The data in 2020 comprise the average from January to
developments within 10 minutes after release of the March. The value of fast-paced orders and fast executions are
divided by total trading volumes to control the impact of the
U.S. employment report (from 8:30 A.M. to 8:39 A.M.
increase in total trading volumes. The values of fast-paced
orders capture behaviors of cancelling quotes; therefore, the
Eastern Standard Time). As a result, the two indicators values can exceed 100%.
have been are on an upward trend since around 2016, Source: EBS
implying that algorithmic trading has prevailed in the
16
USD/JPY spot market as well (Chart 3). In addition, [Chart 4] Algorithm trading indicators
a calculation of the hourly average of these indicators (hourly-average)
from November 2019 to January 2020 shows that
indicator levels are higher in European and U.S. time
(%) (%)
zone than in Japan (Chart 4). In general, the use of 400 45
algorithmic trading is said be limited by Japanese non- 350
44
financial corporations, although total USD/JPY trading 300
volumes in Japanese time zone are large due to their 250
43
transactions. Conversely, large European and U.S. 200
banks as well as non-banks, which utilize algorithmic 150 42
trading actively, are said to have strong presence in
100 41
European and U.S. time zones. The two indicators we
50
calculated are consistent with these general
0 40
characteristics of algorithmic trading in the USD/JPY Japan time European and Japan time European and
spot market. zone U.S. time zone zone U.S. time zone
Note: The data comprise the hourly average of the algorithm trading
indicators in the USD/JPY spot market from November 2019 to
Impacts on market liquidity in normal times January 2020, excluding Christmas holidays, year-end, and
new year holidays. “Japan time zone” and “European and U.S.
time zone” show the hourly average from 7 a.m. to 3 p.m. and
While the widespread use of algorithmic trading is from 3 p.m. to 7 a.m. the next day in Tokyo time, respectively.
assumed to have various impacts ranging from FX Source: EBS
rate’s pricing mechanism to overall market functioning,
Data within 10 minutes after the release of the U.S.
most previous studies focus on the impacts on market
employment report from January 2014 to March 2020
liquidity. Empirical study results often highlight that
(each variable is a mean value of tick data recorded
the increased presence of algorithmic trading positively
within the 10 minutes) are used to estimate the above
contributes to market liquidity improvement, at least in
normal times.17 regression analysis. We adopt effective spread for the
dependent variable () as a liquidity indicator,
The regression analysis below is conducted based
that is, the spread between traded price and mid-quote
on previous literature methods, using algorithmic
price (best bid and best ask average price) at the same
trading’s two proxy indicators to verify whether the 19
time. The following independent variables are used:
above observations are consistent with the USD/JPY
18 logarithmic form of either of the two algorithmic
spot market.
trading proxy indicators ( ) and degree of surprise
| | = + + |_ℎ |
in the U.S. employment report (_ℎ ; calculated
+ +
3 Bank of Japan August 2020
by dividing the gap between the actual number of Impacts on market liquidity: in market stress
nonfarm payrolls and its market expectation by the times
standard deviation of the gap during the estimation
This section examines whether the results above are
period). Control variables ( ) comprise the consistent even in times of market stress when
lagged value of and logarithmic form of the volatility is high. Previous literature finds that
20
amount of best quotes (so-called “depth,” another
algorithmic trading can deteriorate market liquidity by
liquidity indicator).
stopping the liquidity-providing function in times of
Estimation results are as follows. First, estimation
market stress not well assumed in the algorithm
22
results using fast-paced orders as a proxy indicator of program. For example, in times of market stress
algorithmic trading show a negative and statistically
where JPY appreciates sharply, the following risks
significant coefficient, whereas other coefficients
increase for market makers: inventory risk (i.e., holding
satisfy the expected signs (Chart 5). In other words, the
considerable inventories due to market makers’ biased
more algorithmic trading by liquidity providers (fast- position toward JPY short) and FX risk (i.e., valuation
paced orders) are used, the tighter is the spread (better
losses of inventories caused by
market liquidity). The higher the absolute degree of
additional JPY appreciation). In times of such stress,
surprise in the U.S. employment report, the wider the
market makers are said to (1) keep providing buy and
effective spread in the previous month, and a
sell quotes with wider bid–ask spread and then (2) stop
deterioration in another liquidity indicator (lower depth
providing liquidity when market fluctuation degree
23
level) tends to lead to wider spread (worse market
exceeds a certain maximum threshold. By contrast,
21
liquidity). Next, estimation results using fast
other previous literature claims that algorithmic
executions as the proxy indicator of algorithmic trading
trading’s liquidity-providing function was maintained
24
show a negative but not statistically significant
in times of market stress. These findings show that a
coefficient on the indicator. As discussed in the
firm consensus on algorithmic trading’s functions in
previous section, this proxy indicator contains both
times of market stress has not been reached. Stress
liquidity consumers and providers’ behavior. This may
events do not emerge frequently, and their degree,
be the reason of statistically insignificant impact on
duration, and impact on FX rates including USD/JPY,
market liquidity. In sum, it can be concluded that
diverge across events. Under such circumstances,
algorithmic trading, particularly market make (liquidity
algorithmic program and operation have been gradually
provision), contributes to improving market liquidity in
advanced in response to stress events, making it
the USD/JPY spot market in normal times. This finding difficult to perform an objective evaluation.
is supported by the fact that regression coefficient is
Based on the above understanding, we here try to
interpreted as the average value throughout the
capture algorithmic trading developments from late
estimation period.
[Chart 6] Market environment in the USD/JPY
[Chart 5] Estimation results market (from January to March, 2020)
Dependent variables Effective Spread
Explanatory variables
Fast-paced orders *** (%) (pips) (million USD)
Algorithm -0.075 25 4.2 Bid-ask spread 0
trading (0.018)
Fast executions -0.253 3.8 Depth (right scale, 2
indicators 20 reversed)
(0.172) 3.4
Higher Lower 4
Degree of surprise in the 0.025 * Volatility 3.0 Liquidity
0.040
U.S. employment report (0.019) (0.021) 15 2.6 6
Effective spread in *** ***
0.538 0.730 2.2 8
previous month (lag-term) (0.086) (0.076) 10
Depth *** ** 1.8
-0.232 -0.154 10
(0.059) (0.068) 5 1.4
Constant *** * 12
0.953 1.369 1.0
(0.197) (0.766) 0 0.6 14
Adjusted-R2 0.67 0.61 1/6 1/21 2/5 2/20 3/6 3/21 1/6 1/21 2/5 2/20 3/6 3/21
Sample size 75 75 m/d m/d
Note: ***, **, and * indicate statistical significance at 1%, 5%, and 10% Note: The left panel shows the one-month implied volatility in the
levels, respectively. Standard Error has been provided in USD/JPY market. On the right panel, the bid-ask spread is
parenthesis. shown as daily average of the spread in each minute (from 5
Source: EBS, Bloomberg p.m. to 5 p.m. next day in NY time). Depth indicates daily
averages of the total volumes in best bid and best ask. The
latest data are as of March 31, 2020.
Source: EBS, Bloomberg
4 Bank of Japan August 2020
no reviews yet
Please Login to review.
