Physica A 387 (2008) 5182–5188

Contents lists available at ScienceDirect

Physica A
journal homepage: www.elsevier.com/locate/physa

Empirical shape function of limit-order books in the Chinese

stock market
Gao-Feng Gu a,b , Wei Chen c , Wei-Xing Zhou a,b,d,e,∗
a
School of Business, East China University of Science and Technology, Shanghai 200237, PR China
b
School of Science, East China University of Science and Technology, Shanghai 200237, PR China
c
Shenzhen Stock Exchange, 5045 Shennan East Road, Shenzhen 518010, PR China
d
Research Center for Econophysics, East China University of Science and Technology, Shanghai 200237, PR China
e
Research Center of Systems Engineering, East China University of Science and Technology, Shanghai 200237, PR China

article info a b s t r a c t

Article history: We have analyzed the statistical probabilities of limit-order book (LOB) shape through
Received 24 January 2008 building the book using the ultra-high-frequency data from 23 liquid stocks traded on the
Received in revised form 30 April 2008 Shenzhen Stock Exchange in 2003. We find that the averaged LOB shape has a maximum
Available online 10 May 2008
away from the same best price for both buy and sell sides of the LOB. The LOB shape
function has nice exponential form in the right tail. The buy side of the LOB is found to
PACS:
be abnormally thicker for the price levels close to the same best although there are much
89.65.Gh
02.50.-r
more sell orders on the book. We also find that the LOB shape functions for both buy and
89.90.+n sell sides have periodic peaks with a period of five. The 1-min averaged volumes at fixed
tick level follow log-normal distributions except for the left tails which display power-
Keywords: law behaviors, exhibit abnormal intraday patterns with increasing trend, and possess long
Econophysics memory that cannot be explained by the intraday patterns. Academic implications of our
Stock markets
empirical results are also briefly discussed.
Continuous double action
Limit-order book shape
© 2008 Elsevier B.V. All rights reserved.
Microstructure theory

1. Introduction

In an order-driven market, limit-order book (LOB) is a queue of orders waiting to be executed and it is the base of
continuous double auction mechanism. Orders in the book are sorted according to price-time priority. The construction of
LOB is a dynamic process. Effective limit orders whose prices do not penetrate the opposite best price are stored in the book,
while an effective market order with the price penetrating the opposite best immediately causes a transaction and removes
the corresponding orders in the opposite book. In addition, cancelations can also remove the orders in the LOB.
Price levels in the limit-order book are discrete. The difference between two adjacent price levels is the tick size u. It is
0.01 RMB for all stocks in the Chinese market. The price level ∆ at any given time t can be defined as follows

(pb − p)/u + 1

for buy orders
∆= (1)
(p − pa )/u + 1 for sell orders,
where p is an allowed price in the LOB and pb and pa are the best bid and best ask, respectively. According to the definition,
∆ = 1 stands for the position at the best bid (ask) in the buy (sell) LOB. Denote Vb (∆, t ) (respectively Vs (∆, t )) as the volume

∗ Corresponding address: School of Business, East China University of Science and Technology, 130 Meilong Road, P.O. Box 114, Shanghai 200237,
PR China. Tel.: +86 21 64253634; fax: +86 21 64253152.
E-mail address: wxzhou@ecust.edu.cn (W.-X. Zhou).

0378-4371/$ – see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.physa.2008.05.008
 G.-F. Gu et al. / Physica A 387 (2008) 5182–5188 5183

at level ∆ in the buy (respectively sell) LOB at event time t. Vb (∆, t ) and Vs (∆, t ) can be viewed as the instant LOB shape
functions on the buy and sell sides, respectively.
The LOB shape function is of crucial importance in the research of market microstructure theory of order-driven markets.
A brief discussion is in order. The shape of the LOB affects a trader’s strategy and thus influences order aggressiveness [1].
Also, the LOB shape determines the virtual price impact. The price impact I (ω) of a virtual market order of size ω can be
determined as follows [2–4]
( )
n
X
I (ω) = u × sup n : V (∆, t ) 6 ω . (2)
∆=1
It is found that the virtual price impact is much stronger than the actual impact [4] and large price fluctuations are not
necessarily caused by large orders but rather the liquidity [5,6]. It is rational that a large trader prefers to split his large
order and submit when the opposite LOB is thick such that the price does not change much. In contrast, an impatient small
trader might submit a small order when the opposite LOB is thin for small ∆’s, since usually he does not have ensuing orders.
The optimal trading strategy of a large order also depends on the average LOB shape [7,8], which could be improved if one
considers the instant LOB shape function rather than the average.
When we want to investigate the topics above analytically, the LOB shape function is usually treated as being continuous.
In the derivation of an optimal execution strategy, many unrealistic LOB shape functions have been proposed [7,8]. This
makes the framework less useful in practice and calls for a realistic shape function. Indeed, the empirical LOB shape function
has been investigated in different stock markets. Bouchaud et al. found that the LOB shape of individual liquid stocks on
the Paris Bourse (February 2001) is symmetrical for buys and sells and has a maximum away from the current best bid
or ask (∆ = 1) [9]. They also found that the distribution of order size at the best bid or ask can be fitted by a gamma
distribution [9]. Potters and Bouchaud investigated three stocks traded on the Nasdaq Stock Market and found that all the
LOB shape functions are buy/sell symmetric and only one stock reaches a maximum before relaxation [10]. Similar results
on the shape function are also reported using other market data [2–4,11].
In this paper, we shall study in detail the LOB shape of 23 liquid stocks traded on the Shenzhen Stock Exchange (SZSE)
in China. The rest of the paper is organized as follows. In Section 2, we describe briefly the database we adopt. Section 3
introduces the average shape of buy and sell sides of the LOB. We then discuss in Section 4 the probability distributions and
time dependency of volumes at the first three tick levels. The last section concludes.

2. Data sets

The Chinese stock market is a pure order-driven market where orders are matched resulting in transactions. Our data
contain ultra-high-frequency data1 of 23 liquid stocks listed on the Shenzhen Stock Exchange in 2003 [13]. We find that the
results for different stocks are qualitatively similar. Hence we will present the results only for a very liquid stock. In 2003,
only limit orders were allowed to submit and the market constituted opening call auction, cooling period and continuous
double auction. We focus on the LOB in continuous double auction.
As an example, our presentation is based on the order flow data for a stock named Shenzhen Development Bank Co., LTD
(code 000001),2 whose time stamps are accurate to 0.01 s including details of every event, with the information containing
date, order size, limit price, time, best bid, best ask, transaction volume, and aggressiveness identifier (which identifies
whether a record is a buy order, a sell order, or a cancelation). The database totally records 3,925,832 events, including
1,718,156 buy orders, 1,595,961 sell orders, 598,750 cancelations and 12,965 invalid orders. Using this nice database, we
can rebuild the LOB according to the trading rules [14] and study the statistical probabilities of LOB shape.

3. Averaged shape

In the continuous double auction mechanism, order placement adds volume to the book, while order cancelation or
transaction removes volume from the book. It is clear that these three types of events (order placement, order cancelation
and transaction) can change the shape of the LOB. In what follows we use event time, not clock time. In this way, the event
time t advances by 1 when an event occurs. At every time t, we have an instant LOB shape Vb,s (∆, t ) on each side (buy or
sell). The averaged shape of the buy (sell) LOB can be calculated as follows:
M
1 X
Vb,s (∆) = Vb,s (∆, t ), (3)
M t =1
where M is the number of total events in 2003 for the stock we analyzed.

1 According to Robert F. Engle, ‘‘ultra-high-frequency data is defined to be a full record of transactions and their associated characteristics’’ and is
associated with the ‘‘limiting frequency’’ that is achieved when all transactions are recorded [12]. However, the sampling frequency at the transaction
level is actually not the limiting frequency. Our data set contains the flows of orders and their associated characteristics, which has higher frequency. We
nevertheless term our data as ultra-high-frequency data.
2 Each Chinese stock has a unique numeric symbol (stock code) that is the standard and official ticker symbol and is used in trading softwares. The full
company name corresponding to a code can be retrieved from the web site of the SZSE at http://www.szse.cn.
5184 G.-F. Gu et al. / Physica A 387 (2008) 5182–5188

Fig. 1. (Color online) LOB shape V (∆) as a function of relative distance ∆ for buy and sell limit-order books in log-linear coordinates (a) and linear-log
coordinates (b).

It is known that traders tend to place their orders on the same best price [9,10,15–17]. On the other hand, the orders
placed near the same best have a higher execution probability, and impatient traders are likely to make a cancelation when
these orders are not executed immediately. It is thus not clear what is the LOB shape under these opposite forces. Fig. 1
shows the shapes for both sides of the LOB.
In Fig. 1(a), we in general find that the LOB shape function has a maximum away from the same best (∆ = 1) and is
roughly symmetrical to the maximum, which consists with the result of Bouchaud et al. [9]. The LOB shapes are asymmetric
between buy orders and sell orders. The LOB shape V (∆) increases when ∆ 6 ∆max and decreases afterwards, where
∆max = 4 for buys and ∆max = 11 for sells. We note that only two (000088 and 000539) of the 23 stocks do not have clear
maxima and the values of ∆max vary from stock to stock. In addition, the total volume of sell orders is greater than that of
buy orders, which is especially visible for large ∆. This phenomenon is also observed for other stocks except that two stocks
(000088 and 000089) have comparable buy and sell volumes, which is consistent with the fact that the Chinese stock market
in 2003 was in the middle of a long-lasting bearish antibubble from 2001 to 2005 [18] and more market participators tended
to sell their shares.
There are two more features arise in the empirical LOB shape function. Although there are more sell limit orders in the
book, the buy side of the LOB is still thinner than the sell side for small ∆ in Fig. 1(a). In 2003, only the information on the
first three visible levels (∆ = 1, n1 , and n2 such that the instant LOB shape function V (1) 6= 0, V (n1 ) 6= 0, V (n2 ) 6= 0 and
V (∆) = 0 for other relative distances less than n2 ) were disposed to traders. We find that, 10 stocks have thicker sell books,
10 stocks have thicker buy books, and the other three have comparable book thickness. This observation is very interesting
since the traders faced a very strong illusionary signal that there were more buy orders while the market was bearish.
Another interesting feature is the presence of periodic peaks at ∆ = 5n + 1 for n = 0, 1, 2, . . ., which are observed in all 23
stocks. The periodic peaks are higher for sell orders than buy orders. The underlying mechanism of this universal behavior is
unclear, which might be related to the trading strategy of larger traders or people’s irrational preference of some numbers
like 5, 10 or their multiples [19]. This periodic strip pattern was also observed in the snapshot (but not in the average shape)
of the LOB for stocks traded on the London Stock Exchange [11]. These two features call for further investigation, which is
however beyond the scope of this work.
In Fig. 1(b), we show the shape functions in linear-log coordinates to study the functional form for large ∆. The volumes
on both buy and sell sides decrease exponentially,

Vb,s (∆) ∼ e−βb,s ∆ . (4)

Using least-squares fitting method, we obtain that βb = 0.044 ± 0.0004 for the buy side of the LOB and βs = 0.025 ± 0.0002
for the sell side of the LOB. The decreasing speed for the buy side of the LOB is faster than that for the sell side, which means
that there is a larger proportion of more aggressive orders on the buy side of the LOB than on the sell side. It seems that buyers
pay more attention to the execution probability, while sellers consider the return of their investigation more important. We
notice that most of other stocks have similar exponentially decreasing shapes. In contrast, Bouchaud et al. have found that
the LOB shape tails have power-law behaviors for the three liquid stocks traded on the Paris Bourse [9]. In addition, Vb,s (∆)
abruptly plummets to zero at the tail ends, which is caused by the 10% price fluctuation limitation compared to the close
price on the previous trading day. In the left panel of Table 1, we present the estimated values of βb,s for the 23 stocks. The
orders of magnitude of each βb,s are comparable to one another for the 23 stocks. For each stock, we find that βb > βs . The
means of βb and βs are 0.052 ± 0.023 and 0.031 ± 0.012, respectively.
We have studied the event-time averaged volume placed at each tick levels in the LOB. However, the volume may have
large fluctuations and greatly deviate from the mean. It is necessary to analyze the fluctuations of volumes at each tick levels.
 G.-F. Gu et al. / Physica A 387 (2008) 5182–5188 5185

Table 1
Estimated values of βb,s in Eq. (4) and β∆ in Eq. (9) for all 23 stocks

βb,s in Eq. (4) β∆ in Eq. (9)

Stock code βb βs β1 β2 β3
000001 0.044 ± 0.0004 0.025 ± 0.0002 4.19 ± 0.09 2.61 ± 0.03 2.67 ± 0.05
000002 0.046 ± 0.0003 0.039 ± 0.0004 2.84 ± 0.08 2.02 ± 0.10 2.13 ± 0.12
000009 0.100 ± 0.0015 0.066 ± 0.0024 3.55 ± 0.15 2.67 ± 0.08 2.85 ± 0.11
000012 0.047 ± 0.0006 0.030 ± 0.0005 2.38 ± 0.14 2.18 ± 0.16 2.18 ± 0.16
000016 0.068 ± 0.0007 0.038 ± 0.0005 2.38 ± 0.14 2.36 ± 0.14 2.41 ± 0.15
000021 0.030 ± 0.0003 0.025 ± 0.0003 2.46 ± 0.16 2.05 ± 0.10 2.20 ± 0.12
000024 0.048 ± 0.0012 0.035 ± 0.0006 1.56 ± 0.17 1.85 ± 0.21 1.97 ± 0.25
000027 0.058 ± 0.0007 0.036 ± 0.0006 2.26 ± 0.12 2.15 ± 0.11 2.06 ± 0.09
000063 0.027 ± 0.0003 0.016 ± 0.0003 1.87 ± 0.20 1.08 ± 0.10 1.29 ± 0.10
000066 0.045 ± 0.0006 0.030 ± 0.0005 2.57 ± 0.14 2.14 ± 0.14 2.10 ± 0.15
000088 0.027 ± 0.0005 0.022 ± 0.0003 1.06 ± 0.25 1.20 ± 0.21 1.15 ± 0.21
000089 0.080 ± 0.0019 0.030 ± 0.0007 2.11 ± 0.18 1.98 ± 0.13 1.92 ± 0.12
000406 0.051 ± 0.0012 0.030 ± 0.0004 3.21 ± 0.22 2.41 ± 0.10 2.51 ± 0.12
000429 0.098 ± 0.0016 0.052 ± 0.0015 2.23 ± 0.15 2.33 ± 0.16 2.27 ± 0.13
000488 0.039 ± 0.0005 0.030 ± 0.0003 0.98 ± 0.33 1.14 ± 0.30 1.25 ± 0.31
000539 0.052 ± 0.0016 0.025 ± 0.0008 1.09 ± 0.24 1.90 ± 0.37 1.60 ± 0.26
000541 0.036 ± 0.0012 0.017 ± 0.0010 1.38 ± 0.24 2.06 ± 0.26 2.00 ± 0.24
000550 0.028 ± 0.0004 0.024 ± 0.0002 2.12 ± 0.15 1.91 ± 0.09 2.04 ± 0.12
000581 0.044 ± 0.0009 0.020 ± 0.0006 2.11 ± 0.35 1.82 ± 0.21 1.86 ± 0.22
000625 0.024 ± 0.0002 0.018 ± 0.0002 2.04 ± 0.12 1.39 ± 0.08 1.39 ± 0.07
000709 0.101 ± 0.0012 0.048 ± 0.0013 2.58 ± 0.13 2.54 ± 0.10 2.69 ± 0.12
000720 0.036 ± 0.0005 0.034 ± 0.0004 2.95 ± 0.54 2.53 ± 0.21 3.04 ± 0.31
000778 0.056 ± 0.0010 0.020 ± 0.0004 2.62 ± 0.29 2.46 ± 0.24 2.04 ± 0.16
Mean 0.052 ± 0.023 0.031 ± 0.012 2.29 ± 0.80 2.04 ± 0.47 2.07 ± 0.50

Fig. 2. (Color online) Plot of the standard deviations σ (∆) as a function of the relative distance ∆ for uy and sell sides of the LOB.

Here, we study the standard deviation σ as a function of the relative distance ∆, that is,
q
σb,s (∆) = hVb,s (∆)2 i − hVb,s (∆)i2 . (5)

The standard deviations for both sides of the LOB are presented in Fig. 2. We find that the functional form of σ (∆) is very
similar to that of the shape for both buy and sell sides of the LOB. The standard deviation σ (∆) increases with ∆ at the first
few levels and then decreases exponentially. When comparing the buy and sell sides of the LOB, the sell side of the LOB is
found to be thicker with larger fluctuations.

4. Statistical properties of volumes at individual tick levels

4.1. Probability distribution

We have analyzed the averaged shape function in the previous section. Here we focus on the time averaged volume over
a fixed clock time interval δ t at individual levels
N
1 X
vb,s (∆, t ) = Vb,s (∆, ti ), (6)
N i=1
5186 G.-F. Gu et al. / Physica A 387 (2008) 5182–5188

Fig. 3. (Color online) Probability density functions f (ln v) of 1-min averaged logarithmic volumes at the first three tick levels on the buy side of the LOB in
a linear-linear scale (a) and linear-log scale (b). The curves corresponding to ∆ = 2 and ∆ = 3 in (b) have been vertically translated downward for clarity.
The results are similar on the sell side.

where ti is the time moments of the N events occur in the interval (t − δ t , t ] and N is a function of t and δ t. We use δ t = 1
min to calculate the time-averaged volume at each price level.
Fig. 3 shows the probability density functions (PDFs) for ∆ = 1, 2, and 3. In Fig. 3(a), we find that ln v in general is
normally distributed

(ln v − µ)2
 
1
f (ln v) = √ exp − , (7)
2π σ 2π σ 2
that is, v is log-normally distributed with the PDF being3
p(v) = f (ln v)/v. (8)
With the increase of the relative distance ∆, the mean of ln v , µ, increases, which is line with the result in Fig. 1. We can also
project that µ decreases for large ∆. More generally, we find that the 1-min volumes at other tick levels for different stocks
are basically log-normally distributed. The log-normal distribution was also adopted to model Nasdaq stocks [3].
When v is small, we find that the empirical curves deviate from the log-normal distribution f (ln v). We plot the
probability density functions f (ln v) of ln v in a linear-log scale, which is presented in Fig. 3 (b). It is clear that the small
volumes v deviate from the corresponding log-normal distributions and exhibit power-law behaviors

f (ln v) ∼ v β∆ or p(v) ∼ v β∆ −1 . (9)

Using least-squares fitting, we obtain that β1 = 4.19 ± 0.09 (2.2 < log10 v < 3.5) for ∆ = 1, β2 = 2.61 ± 0.03
(2.1 < log10 v < 4.2) for ∆ = 2, and β3 = 2.67 ± 0.05 (2.1 < log10 v < 4.2) for ∆ = 3. In the right panel of Table 1, we
present the estimated values of β∆ for the 23 stocks. The orders of magnitude of each β∆ are comparable to one another for
the 23 stocks. For each stock, we find roughly that β1 > β2 ≈ β3 . The means of β1 , β2 and β3 are 2.29 ± 0.80, 2.04 ± 0.47
and 2.07 ± 0.50, respectively.
This is also different from the Paris Bourse stocks where the volumes on the best are distributed according to a Gamma
distribution [9]. We have also fitted the data sets using Gamma distributions. The results (not presented) show that Gamma
distributions can well fit the bulk of the data but have larger discrepancies on the two tails than the log-normal model.
Hence, the log-normal model is slightly better than the Gamma model. However, neither models can explain the power-law
tails.

4.2. Intraday patterns

It is well-known that many financial quantities exhibit intraday patterns. For instance, the intraday patterns are well
documented in the trading volume of shares [20]. However, to the best of our knowledge, this phenomenon has not been
investigated for the volumes at individual tick levels. We present such an effort here. In Fig. 4(a), we show the periodograms
of 1-min averaged volumes v at the first three tick levels ∆ = 1, 2, 3. One can observe that there are evident peaks above
the noise level at frequencies of 1, 2, . . . day−1 , which implies a period of one trading day. The highest peak close to zero
frequency corresponds to the inverse of the size of v , which is trivial. Fig. 4(b) illustrates the corresponding intraday patterns
A(t ). We find that A(t ) has an overall increasing trend. This unusual shape of intraday patterns in v seems very different from
those U shapes or L shapes frequently observed in absolute returns, trading volumes, and bid-ask spreads.

3 Denote g (y) and h(x) the PDFs of y and x, respectively. If y is a function of x, we have g (y)dy = h(x)dx. It follows immediately that h(x) = g (y)dy/dx =
g (ln x)/x.
 G.-F. Gu et al. / Physica A 387 (2008) 5182–5188 5187

Fig. 4. (Color online) Detecting intraday patterns in the 1-min averaged volumes. (a) Periodograms of 1-min averaged volumes at the first three tick levels.
The unit of frequency is 1/day. (b) The intraday patterns in the 1-min averaged volumes at the first three tick levels.

Fig. 5. (Color online) Plot of the detrended fluctuation functions F (`) of 1-min averaged volumes (a) and 1-min averaged volumes without intraday
patterns (b) at the first three tick levels on the buy limit-order book. The results corresponding to ∆ = 2 and ∆ = 3 have been vertically translated
downwards for clarity.

4.3. Long memory

Temporal correlations in a time series can be quantitatively assessed by the autocorrelation function C (`), which
describes the average correlation between two points with time lag `. Many processes have the autocorrelation function
decaying exponentially (C (`) ∼ e−`/`0 for ` → ∞), which means that these processes exhibit short memory with
a characteristic timescale `0 . On the other hand, when the autocorrelation function is not integrable, for example, C (`)
decaying as a power-law behavior (C (`) ∼ `−γ ), the process has long memory without any characteristic timescale, which
means that the values in the past have potential predictive power for the future.
Temporal correlations can be equivalently characterized by the Hurst index H, and the relationship between the
autocorrelation exponent γ (assuming C (`) ∼ `−γ ) and the Hurst index H can be expressed by γ = 2 − 2H [21,22].
Detrended fluctuation analysis (DFA) is a popular method to estimate the Hurst index [21,23,24]. We perform DFA on both
the 1-min averaged volumes and 1-min averaged volumes after removing the intraday pattern at the first three tick levels
on the buy side of the LOB. The detrended fluctuation functions F (`) are presented in Fig. 5. Sound power-law relations are
observed in the six curves and the Hurst indexes are H1 = 0.76 ± 0.01 for ∆ = 1, H2 = 0.83 ± 0.01 for ∆ = 2, and
H3 = 0.81 ± 0.01 for ∆ = 3 for the 1-min averaged volumes and H1 = 0.78 ± 0.01 for ∆ = 1, H2 = 0.84 ± 0.01 for
∆ = 2, and H3 = 0.82 ± 0.01 for ∆ = 3 for 1-min averaged volumes without intraday patterns. We find that the intraday
patterns have little impact on the long memory. With the Hurst indexes H significantly larger than 0.5, we argue that the
1-min averaged volumes at the first three tick levels exhibit long memory. Quantitatively similar results are observed for
the sell side of the LOB and for other stocks. This agrees well with the fact that order signs have long memory [25,26].

5. Conclusion

We have investigated the limit-order book shapes of 23 stocks traded on the Shenzhen Stock Exchange in the whole year
2003. For brevity, we presented the results of a very liquid stock (Shenzhen Development Bank Co., LTD, 000001). For most of
the stocks, the averaged shape has a maximum away from the same best and the volumes in the LOBs decrease exponentially.
5188 G.-F. Gu et al. / Physica A 387 (2008) 5182–5188

The LOB shapes are asymmetric between buy and sell orders and the sell side of the LOB shape relaxes much slower. The
probability density functions of 1-min averaged volumes at the first three tick levels are log-normal together with a power-
law behavior for small volumes in the left tails. Using detrended fluctuation analysis, we confirmed that the 1-min averaged
volumes at a fixed tick level on the LOB exhibit long memory, which can not be explained by the intraday patterns. When
compared with the Paris Bourse stocks [9], we find that the LOB shapes are qualitatively similar but quantitatively different.
Several problems arise that need to be addressed: why the buy side of the LOB is abnormally thicker for the price levels
close to the same best and why there are relatively large volume on the tick levels of ∆ = 5n + 1? It is also noteworthy that
our results on the empirical LOB shape functions can be used to develop more realistic optimal trading strategy for large
traders.

Acknowledgments

We thank the referees for their insightful comments and suggestions. This work was partly supported by the National
Natural Science Foundation of China (Grant Nos. 70501011 and 70502007), the Fok Ying Tong Education Foundation (Grant
No. 101086), and the Program for New Century Excellent Talents in University (Grant No. NCET-07-0288).

References

[1] A. Ranaldo, Order aggressiveness in limit order book markets, J. Financ. Mark. 7 (2004) 53–74.
[2] D. Challet, R. Stinchcombe, Analyzing and modeling 1 + 1d markets, Physica A 300 (2001) 285–299.
[3] S. Maslov, M. Mills, Price fluctuations from the order book perspective - Empirical facts and a simple model, Physica A 299 (2001) 234–246.
[4] P. Weber, B. Rosenow, Order book approach to price impact, Quant. Finance 5 (2005) 357–364.
[5] J.D. Farmer, L. Gillemot, F. Lillo, S. Mike, A. Sen, What really causes large price changes? Quant. Finance 4 (2004) 383–397.
[6] P. Weber, B. Rosenow, Large stock price changes: Volume or liquidity, Quant. Finance 6 (2006) 7–14.
[7] A. Obizhaeva, J. Wang, Optimal trading strategy and supply/demand dynamics, J. Financ. Mark. XX 2008 (in press).
[8] A. Alfonsi, A. Schied, A. Schulz, Optimal execution strategies in limit order books with general shape functions, arXiv:0708.1756v2 (2007).
[9] J.-P. Bouchaud, M. Mézard, M. Potters, Statistical properties of stock order books: Empirical results and models, Quant. Finance 2 (2002) 251–256.
[10] M. Potters, J.-P. Bouchaud, More statistical properties of order books and price impact, Physica A 324 (2003) 133–140.
[11] Z. Eisler, J. Kertész, F. Lillo, The limit order book on different time scales, in: J. Kertész, S. Bornholdt, R.N. Mantegna (Eds.), in: Proc. SPIE, vol. 6601, AIP,
Florence, Italy, 2007, p. 66010G.
[12] R.F. Engle, The econometrics of ultra-high-frequency data, Econometrica 68 (2000) 1–20.
[13] G.-F. Gu, W. Chen, W.-X. Zhou, Empirical distributions of Chinese stock returns at different microscopic timescales, Physica A 387 (2008) 495–502.
[14] G.-F. Gu, W. Chen, W.-X. Zhou, Quantifying bid-ask spreads in the Chinese stock market using limit-order book data: Intraday pattern, probability
distribution, long memory, and multifractal nature, Eur. Phys. J. B 57 (2007) 81–87.
[15] J.-I. Maskawa, Correlation of coming limit price with order book in stock markets, Physica A 383 (2007) 90–95.
[16] S. Mike, J.D. Farmer, An empirical behavioral model of liquidity and volatility, J. Econom. Dynam. Control 32 (2008) 200–234.
[17] G.-F. Gu, W. Chen, W.-X. Zhou, Empirical regularities of order placement in the Chinese stock market, Physica A 387 (2008) 3173–3182.
[18] W.-X. Zhou, D. Sornette, Antibubble and prediction of China’s stock market and real-estate, Physica A 337 (2004) 243–268.
[19] S.N. Dorogovtsev, J.F.F. Mendes, J.G. Oliveira, Frequency of occurrence of numbers in the World Wide Web, Physica A 360 (2006) 548–556.
[20] P. Gopikrishnan, V. Plerou, X. Gabaix, H.E. Stanley, Statistical properties of share volume traded in financial markets, Phys. Rev. E 62 (2000)
R4493–R4496.
[21] J.W. Kantelhardt, E. Koscielny-Bunde, H.H.A. Rego, S. Havlin, A. Bunde, Detecting long-range correlations with detrended fluctuation analysis, Physica
A 295 (2001) 441–454.
[22] D. Maraun, H.W. Rust, J. Timmer, Tempting long-memory - on the interpretation of DFA results, Nonlinear Proces. Geophys. 11 (2004) 495–503.
[23] C.-K. Peng, S.V. Buldyrev, S. Havlin, M. Simons, H.E. Stanley, A.L. Goldberger, Mosaic organization of DNA nucleotides, Phys. Rev. E 49 (1994) 1685–1689.
[24] K. Hu, P.C. Ivanov, Z. Chen, P. Carpena, H.E. Stanley, Effect of trends on detrended fluctuation analysis, Phys. Rev. E 64 (2001) 011114.
[25] J.-P. Bouchaud, Y. Gefen, M. Potters, M. Wyart, Fluctuations and response in financial markets: The subtle nature of ‘random’ price changes, Quant.
Finance 4 (2004) 176–190.
[26] F. Lillo, J.D. Farmer, The long memory of the efficient market, Stud. Nonlinear. Dyn. Econom. 8 (3) (2004) 1–33.