Newly Developed Flexible Grid Trading Model Combined ANN and SSO algorithm

Wei-Chang Yeh 1,*, Yu-Hsin Hsieh 1, Chia-Ling Huang 2

1
Department of Industrial Engineering and Engineering Management, National Tsing Hua University,
Hsinchu 300044, Taiwan; W.C. Yeh: yeh@ieee.org; Y.H. Hsieh: kypss93003@gmail.com
2
Department of International Logistics and Transportation Management, Kainan University, Taoyuan
33857, Taiwan; clhuang@mail.knu.edu.tw
* Correspondence: yeh@ieee.org

Abstract: In modern society, the trading methods and strategies used in financial market have gradually
changed from traditional on-site trading to electronic remote trading, and even online automatic trading
performed by a pre-programmed computer programs because the continuous development of network and
computer computing technology. The quantitative trading, which the main purpose is to automatically
formulate people’s investment decisions into a fixed and quantifiable operation logic that eliminates all
emotional interference and the influence of subjective thoughts and applies this logic to financial market
activities in order to obtain excess profits above average returns, has led a lot of attentions in financial
market. The development of self-adjustment programming algorithms for automatically trading in
financial market has transformed a top priority for academic research and financial practice. Thus, a new
flexible grid trading model combined with the Simplified Swarm Optimization (SSO) algorithm for
optimizing parameters for various market situations as input values and the fully connected neural network
(FNN) and Long Short-Term Memory (LSTM) model for training a quantitative trading model to
automatically calculate and adjust the optimal trading parameters for trading after inputting the existing
market situation is developed and studied in this work. The proposed model provides a self-adjust model
to reduce investors’ effort in the trading market, obtains outperformed investment return rate and model
robustness, and can properly control the balance between risk and return.

Keywords: Flexible grid trading model, simplified swarm optimization (SSO), artificial intelligence (AI),
Fully connected neural network (FNN), Long Short-Term Memory (LSTM) model.

1. Introduction
The formulation of financial market trading methods and strategies has changed with the continuous
development of network and computer computing technology from the late 20th century to the present.
More and more financial institutions and major market trading are gradually changing from traditional on-
site trading to electronic remote trading, and even automated trading can be performed in a pre-
programmed computer programs (algorithmic trading). According to statistics, the US stock market with
the most developed financial development currently has as many as 60 to 70 percent of its trading

1
automated by programmed programs. In this wave of application of computer computing technology and
financial markets, the combination of quantitative trading and computer algorithm is particularly
prosperous [1].
The main function of the financial market is to provide the current market price of traded commodities
so that market participants can benefit from it. Quantitative trading emerged in the stock market in the late
20th century. In recent years, it has been widely applied in automated trading systems in the stock, currency,
and futures markets. In order to receive excess returns, statistics and mathematical models are used to
obtain the high probability of the market happening in the future and formulate a corresponding set of
modeled trading logic for market trading by observing a large amount of historical data in the past.
The biggest feature of quantitative trading is the use of a fixed set of logic for trading, which is most
often used in stock and futures trading market, in order to obtain a stable and continuous excess return
above the average return. In addition, the aspects considered by quantitative trading are becoming more
and more diverse including market composition, developmental evaluation of investment targets, market
sentiment, etc., which can be used for further analysis and reference. With the development of the
information age, more and more relevant information has been completely collected and retained, which
can provide quantitative trading models for reference and analysis, make up for the inability of the human
brain to quickly digest huge amounts of data and can make decisions more objectively and not be affected
by market sentiment.
Nowadays, with the changes of the times, international trade and the internet have promoted the
active free trade market, which reduces trading costs and expands the scale and scope of trading, and the
relationship between the financial industry and technology is also deepening as well as becoming more
mature and popular [2, 3]. How to use the increasingly developed computer computing as an auxiliary tool
for financial market trading and even becoming the key to the success of traders’ profits has become the
focus of research in the financial community in recent years.
In the early e-commerce applications, more attention was paid to how to transform the physical
trading form into an electronic trading form including inputting and storing trading information by the
form of electronic files, which is convenient for future search and analysis. Such simple trading is changed
from manual services to electronic services, which greatly reduces labor costs.
After the preliminary results of electronic finance in recent years, financial industry players hope that
in addition to using computers to complete simple trading procedures, they can further use mature
computer programs to help people complete more complex trading decisions. Especially in recent years,
artificial intelligence has become increasingly prosperous, and there have been many breakthrough
developments [4-7]. How to use computer programs to imitate human thinking logic, assist trading, and
even think faster, wider and deeper than humans has become the direction of development and research in
the financial field in recent years.
Trading-related programming algorithms have emerged since the late 1980s. In the early days, more
human or statistical tools were used for analysis and decision-making, and then computer programs were
used for trading. But today, in the past ten to twenty years, more and more quantitative trading has modeled

2
the decision-making process and designed it into the trading algorithm with the trading logic resulting to
make the algorithm more intelligent and automatic. And it has become a highly practical and popular
research, which has attracted the attention and practice of many governments and financial institutions [8].
Today, there are many rich and diverse research results in the field of trading algorithm research, such
as the use of mean regression to adjust the allocation of stock investment weights [9], the use of long short-
term memory (LSTM) combined with grid trading method (GTM) to predict market trends for currency
trading [10], using box theory combined with support vector machine (SVM) to assist stock trading
decisions [11], using Ichimoku Kinkohyo to analyze foreign exchange trading [12], and the use of market
trends instead of fixed time series to make trading decision for the trading strategy of the execution unit
[13]. It is not difficult to see that the use of computer programs to assist trading and even directly making
trading decisions and executing trading have become an unstoppable trend in financial development.
The most basic definition of grid trading method is an average pending order within a specific price
range to carry out a trading strategy of buying low and selling high [14]. The principle of grid trading is
simple but it can effectively earn spread profits when prices fluctuate. In recent years, this strategy has
been often applied to arbitrage in many financial markets [15-16].
Simplified Swarm Optimization (SSO) was proposed by Yeh [17] in 2009 to improve the problem of
Particle Swarm Optimization (PSO) in solving discrete problems with the core concept of simplicity. SSO
simplifies the algorithm and improves the efficiency of the solution, and has been widely used in solving
problems in many different fields [18-29]. Finding the best parameter combination is one of the SSO
applications. Multiple parameters can be considered and adjusted at the same time to find the best
parameter combination, which has been used to adjust the hyperparameter combination of convolutional
neural network [30], optimize the parameters of solar models [31-33], and adjust the parameters in
artificial neural network (ANN) [34].
Recently, the application of deep learning (DL) and ANN in financial activities has become more and
more diverse, which the most widely used is the prediction of the stock market, exchange rate, and
financial index, etc. [35]. At present, there are still many studies trying to use various neural network
models to assist people in financial forecasting and decision-making [36-38] for making more rational and
accurate judgments.
To sum up, this study proposed a new grid trading model, adopts SSO algorithm to find parameters
suitable for various market situations as input values and labels, trains the fully connected neural network
and LSTM model, and finally, a quantitative trading model can automatically calculate and adjust the
optimal trading parameters for trading after inputting the existing market situation.
Following the rise of e-commerce and quantitative trading mentioned above, the relationship between
people’s economic activities and technology has become more and more inseparable, and many trading
relies on programs to complete. In future trends, programs will not only help people record trading and
complete transfers but even help people make trading decisions automatically including determining the
timing and price of trading. Against the foregoing background, the purpose of this study is to:
1. Provide a new set of grid trading algorithms to improve the shortcomings of premature entry and exit

3
 of existing grid trading models in the market.
2. This trading algorithm can adapt to change in the external environment as time and market conditions
change, and self-adjust the model to reduce investors’ effort in the trading market.
3. Through a set of trading models with logical rules, the irrational decisions brought about by investors’
subjective trading decisions are reduced.
4. Balance the relationship between risk and profit, and get a great reward under a certain reasonable risk.

2. Overview of grid trading, SSO, and DL

2.1. Grid trading

The most basic definition of grid trading method is an average pending order within a specific price
range to carry out a trading strategy of buying low and selling high [14]. There are two key factors in grid
trading that determine the effectiveness of this strategy. The first is price volatility including both ups and
downs. If the price goes up all the way, those who hold the spot make more profit than grid trading. If the
price goes down all the way, spot traders basically cannot have many profit opportunities but short futures
buyers are more able to make profits. If the market goes up and down, the grid trading strategy has the
greatest profitability. The more frequent the fluctuations, the greater the profit rate increases. In most
markets, prices are highly fluctuating and often have a price mean reversion [39-41] so that grid trading
strategies are now being used by more and more market players.
The other key determines the profit size of grid trading is its parameter setting including grid average,
upper bound of the grid, lower bound of the grid, number of grids, initial price, stop loss point, stop profit
point, etc. The setting of the above parameters should be based on product price fluctuations, trading costs,
risk tolerance, and the amount of principal as the setting considerations, which directly affect the final
performance of grid trading.
Grid trading is mainly divided into two types: equal-distance grid and equal-ratio grid. At present,
the grid that many people use in practice is equal-distance grid.
The setting of equal-distance grid is based on the initial price P0, and the setting of upper bound of
the grid Gul, the lower bound of the grid Gll, the total number of grids n, and the calculation of the grid
spacing Gs by Eq. (1).

Gul  Gll
Gs  (1)
n

After the setting of the grid model, the trading is based on this grid to buy and sell financial products
with the rise and fall of prices. For an example of the equal-distance grid, please refer to Figure 1.

4
 Price
Gul

P0+5Gs

P0+4Gs

Sell Sell
P0+3Gs

P0+2Gs Sell Buy Sell Sell Buy

Sell Buy Sell Buy Sell Buy

P0+Gs

Buy Sell Buy Sell Buy

P0

Buy Sell Buy Sell

P0-Gs

Buy Buy
P0-2Gs

P0-3Gs

Time

Figure 1. Schematic diagram of equal-distance grid trading

The grid spacing Gs of equal-ratio grid is a fixed ratio and the equal-ratio grid is calculated at this
ratio. For an example of the equal- ratio grid, please refer to Figure 2.

Price
Gul

P0  Gs5

P0  Gs4

P0  Gs3 Sell Sell

P0  Gs2 Sell Buy Sell Sell Buy

Sell Buy Sell Buy Sell Buy

P0  Gs1

P0  Gs0 Buy Sell Buy Sell Buy

Buy Sell Buy Sell

P0  Gs1

Buy Buy
P0  Gs

P0  Gs

Gll Time

Figure 2. Schematic diagram of equal-ratio grid trading

5
2.2. SSO
Simplified Swarm Optimization (SSO) was proposed by Yeh [17] in 2009 bases on the concepts of
personal best (pbest) and global best (gbest) and adds random numbers resulting that the solution owns
the opportunity to escape from the local optimal solution and improve the diversity of the solution.
The major difference between SSO and other heuristic algorithms is its special update mechanism.
There are three particularly important parameter settings including Cg, Cp, and Cw, where Cg > Cp > Cw.
The update mechanism is shown in Eq. (2). According to the correspondence between the random number

ρ and Cg, Cp, and Cw, the next generation solution xi,jt is determined, which may be gbest, pbest, the

current solution or a random number, respectively.

 g j ,if  ρ  [0,Cg 

t 1  pi,j ,if  ρ  [C g ,C p )
xi,j   t (2)
 xi,j ,if  ρ  [C p ,Cw )
 x,if  ρ  [C ,1)
 w  

where, let xi,jt  xi,t 1 , xi,t 2 ,..., xi,jt be the ith solution with j variables in the tth generation; ρ is a random

number subjects to uniform distribution between [0, 1].

The update method is that when ρ is between [0, Cg), the variable xi,jt+1 is replaced by the global

best solution gj that is the best performing solution among all the solutions at present; when ρ is between
[Cg, Cp), it is replaced by the best solution in the region pi,j that is the optimal solution in the past
generations of the variable; when is between [Cp, Cw), it maintains the solution of the previous generation

xi,jt ; when is between [Cw, 1), it is replaced by x that is a random number generated in the upper and lower
bounds of the variable. The purpose is to reduce the chance of getting trapped in local optimal solutions,
while also increasing the diversity of solutions.

2.3. DL
The ANN, Back-propagating method, and Long Shout-Term Memory (LSTM) adopted in this study
have an overview in this section.

2.3.1. ANN
In recent years, DL has been widely used in medicine, industry, transportation and other fields, which
can help the completion of speech recognition and machine vision. It is a machine learning (ML) algorithm
based on ANN, which learns through data features.

6
 The computing architecture of ANN was first proposed by WS. McCulloch and W. Pitts [42], and
subsequently it has been continuously improved by many outstanding scholars [43, 44], and it has become
a famous ML model in artificial intelligence (AI).
The basic structure of ANN is divided into three parts including input layer, hidden layer and output
layer. Each node of the input layer corresponds to the input predictor variable, each node of the output
layer corresponds to the objective variable, and the hidden layer is sandwiched between input layer and
output layer.
Except for the nodes in the input layer, each node in the ANN is connected to several nodes in front
of it, which is this model is called fully connected neural network (FNN), and each node has a
corresponding weight.

2.3.2. Back-propagating method

In DL, the back-propagating method [45] is an extremely important key to make the model complete
and is often used to train and optimize ANN. Using the back-propagation, the gradient of the loss function
to the weight can be efficiently found, and then the gradient descent method [46] is used to solve each
weight. The “loss” in the loss function refers to the error of the actual value and the predicted value. The
main concept of back-propagation is to return the error resulting that the weight can use the error size to
perform gradient descent method to obtain and update the more suitable weight, further reduce the error
and optimize the weight.

2.3.3. LSTM
LSTM is a model generated to improve the short-term memory of recurrent neural network (RNN).
It is mainly composed of four units including memory cell, input gate, output gate and forget gate.
The input gate controls whether it is input into the memory unit this time, the memory unit is
responsible for storing the calculated value, the forget gate controls to clear the memory, and the output
gate controls whether to output the operation result.

3. Proposed Approach
The approach proposed in this study is presented in sequence. The initial setting and operation
mechanism of grid trading is introduced in Subsection 3.1. Subsection 3.2 presents the concept and setting
method of flexible grid. The use of SSO to obtain the optimal flexible grid parameters in different
situations is shown in Subsection 3.3. As the training basis for DL, a neural network is finally trained that
can automatically output the suitable grid parameters for trading by inputting recent market information
to obtain excess returns from market fluctuations is represented in Subsection 3.4.

3.1. Operation mechanism of grid trading

This section presents in detail how grid trading practices in this study, including initial parameter
setting and calculation, and subsequent operation mechanisms and processes.

3.1.1. Initial parameter setting of grid trading

A total of five basic parameters need to be set, namely the total investment capital F0, the initial price
7
P0, the upper bound of the grid Gul, the lower bound of the grid Gll, and the total number of grids n, before
operating a grid trading model. The total investment capital and initial price are set as control variables in
this study, i.e., when comparing the results, the total investment capital and initial price used in any method
will be set to the same fixed value, and compared on the same standard.
Before the grid trading model runs, the unit price difference Gs, which is the grid spacing, needs to
be calculated using Eq. (1) in Section 2 if the grid is an equal-distance grid. Gs is a certain value in equal-

distance grid. But it is a fixed ratio in equal-ratio grid, for example, the value of the next grid is P0  Gs1 =110

and the next grid is P0  Gs2 =121 when P0=100 and Gs=1.1. And this ration can be calculated from Eq. (3)

below.

Gul  Gll
Gs  n 1 (3)
Gll

In addition, it is also necessary to calculate how much funds should be used to purchase the spot
before grid trading, i.e., the initial purchase of spot S0, so as to sell for profit when the price rises. And
how much funds must be kept on hand to buy sopt when the price falls, here C0 means to start holding
cash. Hence, the total investment funds are divided into two parts that is represented by the following Eq.
(4). The spot and funds held in each subsequent period are represented by Sj and Cj, where j represents the
jth period.
F0 = S0 + C0 (4)

In order to calculate the initial purchase of spot S0 and the initial holding of cash C0, it is necessary
to first calculate the number of upper grids nu and the number of lower grids nl, the sum of which is
equivalent to the total number of grids n such as shown in Eq. (5). The values of nu and nl can be obtained
by Eqs. (6)-(7), respectively. Here, the equal-distance grid is used as an example.

n = nu + nl (5)

nu = (Gul- P0) / Gs (6)

nl = (P0- Gll) / Gs (7)

After calculating the number of upper grids nu and the number of lower grids nl, and then further use
Eqs. (8)-(9) to calculate the initial purchase of spot S0 and the initial holding of cash C0. At this time, the
single-cell trading volume Gv is still an unknown value.

8
 S0 = Gv ∙nu ∙P0 (8)

C0 = Gv ∙[(P0 - Gs) + Gll] / 2 ∙nl (9)

Finally, substitute Eqs. (8)-(9) into Eq. (4) to get the single-cell trading volume Gv, which can be
calculated by Eq. (10).

Gv = F0 / {[(P0 - Gs) + Gll] / 2∙nl + nu ∙P0} (10)

Finally, calculate the price of each grid cell gi, i from 1 to n using Eq. (11) (equal-distance grid) or
Eq. (12) (equal-ratio grid).

gi = Gll + Gs ∙(i - 1) (11)

gi = Gll ∙ Gs( i 1) (12)

3.1.2. Operation mechanism of grid trading

The following Figures 3 (a)-(e) illustrate in detail how the grid trading model updates and adjusts
with the market price after the initial parameters are set, and how to settle at the end.
1. When the grid is initially running, the current price is used as the benchmark, and the above grid price
is placed on a sell order, and the following grid price is placed on a buy order as shown in Fig. 3 (a).

Sell

Sell

Sell

Current price

Buy

Buy

Buy

Fig. 3. Grid trading operation chart (a)

2. If the price rises until it hits the first grid line, make a sell action, update the spot volume and funds held,
and place a buy order at the original grid position as shown in Fig. 3 (b).

9
 Sell

Sell

Sell

Current price

Buy

Buy

Buy

Fig. 3. Grid trading operation chart (b)

3. If the price falls back to the initial grid line, make a buy action, update the spot volume and funds held,
and place a sell order at the original grid position as shown in Fig. 3 (c).

Sell

Sell

Sell

Current price

Buy

Buy

Buy

Fig. 3. Grid trading operation chart (c)

4. If the price continues to drop to a grid line, make a buy action, update the spot volume and funds held,
and place a sell order at the original grid position as shown in Fig. 3 (d).

10
 Sell

Sell

Sell

Sell

Current price

Buy

Buy

Fig. 3. Grid trading operation chart (d)

5. Continue to trade with the above mechanism. Although the price has returned to the original point of
grid trading, it has successfully arbitraged 7 times that is equivalent to 7 grids of grid spread profits as
shown in Fig. 3 (e).

Sell

Sell Sell Buy Sell

Current price

Buy Sell Buy Sell

Buy Sell Buy

Buy

Fig. 3. Grid trading operation chart (e)

6. When the grid trading model is to be closed, there are two ways to end it. One is to directly keep the
current spot and funds held, and the other is to sell the spot at the current price and convert it into cash.
The former is recommended to be used when the market price is low, and the latter is not recommended.
In this study, the grid is closed and settled in the second method.

For the overall process of grid trading operation, please refer to the grid trading operation flow chart
in following Fig. 4.

11
 Start

Set
F0, P0, Gul, Gll, n

Make an initial pending order

Enter new price

No

Hit any grid line

Yes

Die the price go up

No Yes No

Sell spot, update Sj and Cj , Buy spot, update Sj and Cj ,

and make pending orders and make pending orders

Whether to stop grid trading

Yes

End
Fig. 4. Grid trading operation flow chart.

3.2. Concept and architecture of flexible grid

Based on the equal-distance and equal-ratio grid trading models used in the financial market today,
this study proposes a new and adaptive grid trading model: flexible grid. Through a more flexible grid
structure combined with SSO algorithm and DL, a grid trading model that can adjust parameters according
12
to market changes and adapt to various external conditions is constructed.
Under the framework of the equal-distance grid trading model, it can exert its maximum model
benefits when the market fluctuates. Even if the price returns to the original point, it is still possible to
arbitrage from the volatile market situation based on the adoption of the mean reversion strategy in
quantitative trading [47]. On the other hand, equal-ratio grid trading can obtain better returns in the volatile
rise based on the combination of the two strategies of mean reversion and rend following [48].
The flexible grid proposed in this study captures the advantages of the equal-distance grid and the
equal-ratio grid at the same time. It can outperform the traditional grid trading structure of the past whether
the market is sideways, rising or even falling.
When the flexible grid is initially set, it also needs to set its total investment capital F0, the initial
price P0, the upper bound of the grid Gul, the lower bound of the grid Gll, the number of upper grids nu and
the number of lower grids nl. There are two main differences compared with other models in the initial
setting:
1. The number of upper grids nu and the number of lower grids nl in Eq. (5) are no longer calculated with
Eq. (6) and Eq. (7) but can be set initially.
2. The grid is divided into upper and lower parts with the initial price P0 as the boundary. The upper part
and the lower part can set the number of the grids and have their own grid spacing ratio. The ratio of
the upper grid spacing is Gsu and the lower grid spacing is Gsl. It should be noted here that Gsu must be
a number greater than 0 and less than 1, and Gsl must be a number greater than 1. The feature of this
is that the upper grid spacing becomes smaller and smaller as the price is higher, i.e., the trading
frequency becomes more and more frequent. Similarly, the lower grid spacing becomes smaller and
denser when the price is lower

For the specific calculation method of grid value, please refer to Eqs. (13)-(14) and the over structure
can refer to Figure 5 below.

1
Gsu  (13)
Gul  P0
nu 1
P0

P0  Gll
Gsl  nl 1 (14)
Gll

13
 Price
Gul
⁝ Sell Sell
Sell Buy Sell Buy
P0  G 3
s

Sell Buy Sell Sell Buy

P0  Gs2

Sell Buy Sell Buy Sell Buy

P0  Gs1

P0 Buy Sell Buy Sell Buy

Buy Sell Buy Sell Buy

P0  Gs1

Buy Sell Buy Sell

P0  Gs

Buy Sell Buy

P0  Gs
Buy
P0  Gs
⁝
Gll
Time

Figure 5. Schematic diagram of flexible grid trading.

The core concept in the design process of flexible grid is to perform more frequent sell actions when
the price is high, and buy the cheaper spot more frequently when the price is lower. The flexible grid has
a similar structure to the equal-distance grid in the middle section of the model, while the lower half of
the model is more similar to the equal-ratio grid, and the upper half of the model is different from the
current common equal-distance grid and equal-ratio grid trading model with higher mesh density at higher
prices. The main purpose of this model is to capture the advantages of equal-distance grid and equal-ratio
grid, and can have better performance than the current existing model whether the market trend is volatile,
continuous rise or fall.
However, in order to truly take advantage of the flexible grid model architecture, optimal parameter
setting and self-adjustment in line with the market situation are crucial determinants for the success or
failure of the model. Therefore, this study uses a SSO algorithm to determine the appropriate parameters
of the flexible grid in different situations that is described in Section 3.3, and input the calculated parameter
combination and the corresponding market situation into the artificial neural network (ANN) for model
training that is presented in Section 3.4. Finally, this study produces a trained DL model, which can
automatically adjust to the grid parameters most suitable for the current market situation by simply
inputting the current market situation.

3.3. SSO for optimal parameters

In this subsection, the SSO algorithm is adopted to adjust the optimal solution of the flexible grid
14
under different market conditions based on the flexible grid constructed in the previous subsection. The
market situation and the optimal parameters are input into the ANN as the training set for model training
in subsection 3.4.

3.3.1. Objective and Constraint

The main objective pursued by this study is to maximize the investment return. Therefore, the market
investment commodity price is set by the SSO algorithm after the j period of update. And the objective
model please refer to following Eq. (15).

Max Sj × Pj × Cj (15)

where Sj, Pj, and Cj represent the spot quantity, the commodity market price, and the funds held in the last
period after the initial purchase of spot S0 and initial holding of funds C0 are inputted through the price of
a total j-periods.
In addition, in the flexible grid model, it is necessary to ensure that the profit of each trading is greater
than the trading cost h% (usually the trading fee rate) so that the following constraint Eq. (16) is set, where
i is from 0 to n.

s.t. gi+1 - gi > h% × gi+1 (16)

The calculation of gi can refer to Eqs. (11)- (12). After substituting it, it can be found that this
constraint is actually a constraint on the grid parameters Gul, Gll, nu, and nl.
On the other hand, additional constraints are placed on the grid upper and lower bounds Gul, Gll, nu,
and nl for experiment 1 and experiment 2 as shown in Eqs. (17)- (19), where x is from 0 to j.

s.t. (experiment 1) P0 × 105% < Gul < P0 × 130% (17)

P0 × 70% < Gll < P0 × 95% (18)

s.t. (experiment 2) P0 × 105% < Gul < P0 × 150% (17)

P0 × 50% < Gll < P0 × 95% (18)

10 < nl < [P0 × 100/(maximum Px × 1.3)] – 10 (19)

The above four constraints are to keep the error space and avoid the overfitting of subsequent ANN
training, which cause the price to easily exceed the upper and lower limits of the grid resulting to lose
arbitrage opportunities, and control the parameters within a reasonable range.

15
3.3.2. Solution encoding
According to subsection 3.3.1, the construction of a grid trading model must be set by the total
investment F0, initial price P0, grid upper bound Gul, grid lower bound Gll and total number of grids n. In
this study, F0 and P0 are control variables, hence, the parameters to be solved by SSO are Gul, Gll and n.
In flexible grid, the total number of grids n can be divided into the number of upper grids nu and the
number of lower grids nl.
And the solution range can refer to Table 1 below.
Table 1. Upper and lower bounds of grid trading parameters
Scope of use Parameters Variable upper bound Variable lower bound
Experiment 1 grid upper bound Gul P0 × 130% P0 × 105%
grid lower bound Gll P0 × 95% P0 × 70%
Experiment 2 grid upper bound Gul, P0 × 150% P0 × 105%
grid lower bound Gll P0 × 95% P0 × 50%
In common use number of upper grids nu [P0 × 100/(𝑚𝑎𝑥𝑖𝑚𝑢𝑚 P𝑥 × 1.3)] − 10 10
number of lower grids nl [P0 × 100/(𝑚𝑎𝑥𝑖𝑚𝑢𝑚 P𝑥 × 1.3)] − 10 10

The solution encoding in this study please refer to following Figure 6, which x1, x2, x3, and x4 are set
as the grid upper bound Gul, the grid lower bound Gll, the number of upper grids nu, and the number of
lower grids nl, respectively.

Gul Gll nu nl

X x1 x2 x3 x4

Figure 6. Structure of solution encoding

For example, when X = (150, 100, 10, 30), it means the grid upper bound is 150, the grid lower bound
is 100, the number of upper grids is 10, the number of lower grids is 30, and the total number of grids is
40.

3.3.3. Parameter setting and scope of update mechanism

This study adopts SSO algorithm to solve the studied problem, where ρ is a random number subjects
to uniform distribution between [0, 1], and the solution of the problem is updated according to this random

number: when ρ is between [0, Cg), the variable xi,jt+1 is replaced by the global best solution gj that is the

best performing solution among all the solutions at present; when ρ is between [Cg, Cp), it is replaced by
the best solution in the region pi,j that is the optimal solution in the past generations of the variable; when

is between [Cp, Cw), it maintains the solution of the previous generation xi,jt ; when is between [Cw, 1), it

16
is replaced by x that is a random number generated in the upper and lower bounds of the variable. The
update mechanism can refer to Eq. (2).
The symbols and definitions used in the SSO operation are shown as following Table 2.
Table 2. Symbols and definitions of SSO
Symbols Definitions
𝑁𝑣𝑎𝑟 The number of variables: the grid upper bound Gul, grid lower bound Gll
and total number of grids n in this study.
𝑁sol Total number of solutions.

xit xit  (Guli

t
, Gllit , nit ) represents the ith solution in the tth generation,

where t=1, 2,…, Ngen, i=1, 2, …, 𝑁sol.

p ,G
g ,G
p ,G
g , np , ng pbest and gbest of each variable during the update process.
Gul ul ll ll

Cg, Cp, Cw The three key parameters used to determine the update value in SSO Can
be adjusted according to different situations.
UB UB = (ub1, ub2, …, ubNvar) is the upper bound of each variable, i.e.,

x  ub j .
LB
LB = (lb1, lb2, …, lbNvar) is the lower bound of each variable, i.e., x  lb j .

3.4. Training ANN to automatically adjust flexible grid parameters

After using SSO to obtain the optimal grid configuration under various market conditions, the market
conditions and the calculated grid parameters are used as the questions and answers for training ANN
training. The market situation is interpreted in terms of the following values and input into the ANN,
namely, the highest price during the period, the lowest price during the period, the average market price,
the average trading quantity, the price change (initial price minus final price), the trading quantity change
(initial quantity minus final quantity), price standard deviation, and trading quantity standard deviation.
The output of the ANN outputs the relevant parameters required to construct a grid trading model
including upper bound of the grid, lower bound of the grid, the number of upper grids, and the number of
lower grids. The specific ANN architecture can refer to the following Figure 7.

17
 the highest price during the period

upper bound of the grid

the lowest price during the period
...
the average market price
lower bound of the grid

the average trading quantity

the price change . ..

. .
. . .
. .
.
. the number of upper grids

the trading quantity change

the number of lower grids

price standard deviation

trading quantity standard deviation

input layer hidden layer output layer

Figure 7. ANN architecture

In the neural network training process, the weights in the ANN nodes continuously adjust themselves
through the error value calculated by the loss function until the training is completed when it converges to
a state with the smallest error. Afterwards, through this trained neural network, the optimal parameters of
the grid trading model can be generated by inputting the recent market state and trend, and a flexible grid
model can be automatically constructed for market trading activities.
In this study, two kinds of neural networks are used for training, namely, Fully-connect Neural
Network (FNN) and Long Shout-Term Memory (LSTM) neural network, which is often used for time
series prediction in recent years, to observe and compare which neural networks perform better in learning
grid parameters.

4. Experimental results
The data sets used for validation and comparison in this study are Standard & Poor's 500 (S&P 500),
NASDAQ Composite, Dow Jones Industrial Average (DJIA), Euro Stoxx 50, and Shanghai Composite, a total
of five large-cap indices from 2011 to 2022.

4.1. Verification of flexible grid performance with fixed parameters

First of all, the performance of flexible grid is compared with equal-distance grid and equal-ratio grid
based on the same grid number and upper and lower bounds. The results can be found in Tables 3-5. The
number of grids is calculated by Eq. (20) to be close to the real investment situation and to ensure the
adequacy of the use of funds. According to the upper and lower bound conditions, each data set is divided

18
into two groups for experiments. The first group uses 1.3 times the initial price as the upper bound of the
grid, and 0.7 times the initial price as the lower bound of grid; while the second group uses 1.5 times the
initial price as the upper bound of the grid, and 0.5 times the initial price as the lower bound of grid.
From the results in Tables 3-5, it can be verified that the flexible grid obviously obtains the best ROI,
accumulated wealth and Sharpe ratio through trading with the same grid parameters under ten-year
fluctuations of five different composite indices. Overall, flexible grid appropriately delays the entry and
exit timing due to its trading structure, and obtains a relatively high ROI and Sharpe ratio.

Table 3. ROI obtained by flexible grid, equal-distance grid, and equal-ratio grid
Grid type Flexible grid Equal-distance Equal-ratio
S&P 500
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 31.692 % 27.934 % 22.234 %
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 43.727 % 39.556 % 30.174 %
Nasdaq 100
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 53.795 % 49.261 % 40.369 %
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 71.888 % 66.024 % 51.180 %
Dow Jones Industrial Average
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 21.935 % 18.332 % 13.629 %
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 33.541 % 29.741 % 21.648 %
Euro Stoxx 50
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 0.601 % -4.122 % -7.284 %
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 17.499 % 12.520 % 7.087 %
Shanghai Composite
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) -33.904 % -38.944 % -38.534 %
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) -12.955 % -19.936 % -18.290 %

Table 4. Accumulated wealth obtained by flexible grid, equal-distance grid, and equal-ratio grid
Grid type Flexible grid Equal-distance Equal-ratio
S&P 500
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 13169 12793 12223
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 14373 13956 13017
Nasdaq 100
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 15380 14926 14037
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 17189 16602 15118
Dow Jones Industrial Average
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 12194 11833 11363
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 13354 12974 12165
Euro Stoxx 50
19
 (𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 10060 9588 9272
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 11750 11252 10709
Shanghai Composite
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 6610 6106 6147
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 8705 8006 8171

Table 5. Sharpe ratio obtained by flexible grid, equal-distance grid, and equal-ratio grid
Grid type Flexible grid Equal-distance Equal-ratio
S&P 500
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 0.118 0.103 0.092
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 0.160 0.145 0.137
Nasdaq 100
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 0.187 0.172 0.161
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 0.234 0.219 0.215
Dow Jones Industrial Average
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 0.079 0.065 0.054
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 0.119 0.105 0.095
Euro Stoxx 50
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 0.002 -0.011 -0.021
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 0.046 0.033 0.023
Shanghai Composite
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) -0.082 -0.095 -0.106
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) -0.030 -0.047 -0.053

4.2. SSO parameter setting

Part of the dataset is then used to select appropriate SSO parameters for all subsequent experiments.
In the first set of experiments, Cg, Cp, Cw are configured as shown in Table 6. The meaning is to divide
the range of random number ρ into four parts according to 7:1:1:1, and the part with ration 7 is assigned

to gbest, pbest, xi,jt+1 and a new solution in turn in the experiment to detect what kind of the solution has

more critical influence on producing better quality solutions.

Table 6. Best and worst solutions under different parameter combinations (Experiment 1)
(Cg, Cp, Cw) Maximum scale solution ROI (max) ROI (min)
(0.7,0.8,0.9) gbest 69.74% 66.83%
(0.1,0.8,0.9) pbest 69.62% 67.18%

20
 (0.1,0.2,0.9) 69.69% 66.01%
xi,jt+1
(0.1,0.2,0.3) new solution 69.41% 66.69%

It can be found from Table 6 that it can produce a better quality solution when gbest is the largest,
and it is robust at the same time. Therefore, the probability of gbest is set to the maximum in the final
parameter configuration.
After it is determined that gbest is set as the maximum probability, the range of random number ρ is
then divided into four parts according to 5:3:1:1. Among them, the probability of taking gbest as the
solution is set to the maximum 0.5 because gbest has been determined in the previous step as the key factor

to generate a good quality solution, and the probability of 0.3 is assigned to pbest, xi,jt+1 and a new

solution in turn. According to the experimental results, which solution is the second key factor to produce
a better solution, the results are shown in Table 7.
It can be found from Table 7 that it can produce a better quality solution when the new random
solution is larger, and it is robust at the same time. Therefore, the probability of the new random solution
is set to the next largest in the final parameter configuration.

Table 7. Best and worst solutions under different parameter combinations (Experiment 2)
(Cg, Cp, Cw) Maximum scale solution ROI (max) ROI (min)
(0.5,0.8,0.9) pbest 69.77% 66.22%
(0.5,0.6,0.9) 69.77% 66.91%
xi,jt+1
(0.5,0.6,0.7) new solution 69.86% 67.83%

After it is determined that the new random solution is set as the next largest probability, the range of
random number ρ is then divided into four parts according to 3:3:3:1. The subsequent steps are analogous

to the first two steps, and Table 8 is obtained as follows. It can be found that xi,jt+1 and pbest have little

difference in the quality of the solution, thus, both probabilities are set to the minimum.

Table 8. Best and worst solutions under different parameter combinations (Experiment 3)
(Cg, Cp, Cw) Maximum scale solution ROI (max) ROI (min)
(0.3,0.6,0.7) pbest 69.85% 67.99%
(0.3,0.4,0.7) 69.90% 67.81%
xi,jt+1

21
4.3. Verification of flexible grid performance with parameters selected by SSO
After setting the SSO parameters, the flexible grid, the equal-distance grid and the equal-ratio grid
are connected to the parameters selected by SSO, respectively. And find solutions with 10 runs, 20
generations per run, and 100 sets of solutions per generation. Two sets of experiments, which is same as
the previous verification of flexible grid architecture, are also conducted with the upper and lower bounds
of different risk levels. The results are shown in Tables 9-11.

Table 9. ROI obtained by flexible grid, equal-distance grid, and equal-ratio grid with SSO parameters
Grid type Flexible grid Equal-distance Equal-ratio
S&P 500
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 82.859 % 66.384 % 58.394 %
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 90.269 % 77.198 % 60.774 %
Nasdaq 100
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 111.649 % 66.384 % 82.662 %
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 127.268 % 110.179 % 86.208 %
Dow Jones Industrial Average
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 74.509 % 60.591 % 51.893 %
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 80.559 % 70.063 % 54.760 %
Euro Stoxx 50
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 77.220 % 56.650 % 48.365 %
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 88.844 % 72.293 % 55.856 %
Shanghai Composite
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 58.389 % 39.178 % 33.363 %
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 80.954 % 58.934 % 43.639 %

Table 10. Accumulated wealth obtained by flexible grid, equal-distance grid, and equal-ratio grid with
SSO parameters
Grid type Flexible grid Equal-distance Equal-ratio
S&P 500
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 18286 16638 15839
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 19027 17720 16077
Nasdaq 100
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 21165 16638 18266
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 22727 21018 18621
Dow Jones Industrial Average
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 17451 16059 15189
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 18056 17006 15476
Euro Stoxx 50
22
 (𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 17722 15665 14837
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 18884 17229 15586
Shanghai Composite
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 15839 13918 13336
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 18095 15893 14364

Table 11. Sharpe ratio obtained by flexible grid, equal-distance grid, and equal-ratio grid with SSO
parameters
Grid type Flexible grid Equal-distance Equal-ratio
S&P 500
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 0.439 0.350 0.351
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 0.442 0.391 0.388
Nasdaq 100
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 0.498 0.350 0.437
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 0.524 0.464 0.462
Dow Jones Industrial Average
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 0.377 0.310 0.301
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 0.412 0.344 0.341
Euro Stoxx 50
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 0.300 0.219 0.211
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 0.331 0.265 0.257
Shanghai Composite
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟑,𝐏𝟎 𝟎.𝟕) 0.189 0.134 0.131
(𝑮𝒖𝒍,𝑮𝒍𝒍) = (𝐏𝟎 𝟏.𝟓,𝐏𝟎 𝟎.𝟓) 0.220 0.179 0.172

From the results in Tables 9-11, it can be verified that in the application of SSO to solve the grid
trading parameters, the flexible grid is still the best model in terms of ROI, accumulated wealth and Sharpe
ratio compared with the equal-distance grid and equal-ratio grid. In addition, comparing the grid trading
results connected with the SSO parameters and the fixed parameter version, it can be found that the ROI
has increased significantly. The overall ROI results, the flexible grid is the best, the equal-distance grid is
the second, and the equal-ratio grid is the worst.

4.4. Training ANN to automatically adjust flexible grid parameters

After confirming the excellent performance of flexible grid using SSO to search for trading
parameters, the flexible grid and SSO are adopted to record the best trading parameters of each index in
ten years, and obtain a set of trading parameters every 30 days. In order to expand the follow-up training
and data, the moving pace of the model is set to 5. In each index, about 500 pieces of parameter data are
obtained for the training and verification of ANN, in which the data of the first nine years is used as the

23
training set and the data of the last year is used as the validation set.
Because the data of the training set in this study has temporal continuity, the data is re-randomly
ordered before training in the FNN in order to avoid affecting the training results of the model. In the
LSTM neural network, the input training data must be sequential because the model design, thus, the
training data does not need to be re-ordered randomly.
The architecture of the FNN and the LSTM neural network has undergone many experiments, and
the final architecture and hyper-parameter settings are shown in Tables 12-14.
The input parameters include the highest price during the period, the lowest price during the period,
the average market price, the average trading quantity, the price change, the trading quantity change, price
standard deviation, and trading quantity standard deviation, which a total of eight variables to describe a
market situation. And the output variables are set to upper bound of the grid, lower bound of the grid, the
number of upper grids, and the number of lower grids for grid trading.
The number of hidden layers is set to three. The optimizer is set to adam, which is common in recent
years, has fast convergence speed, and has excellent performance in searching solution. The excitation
function uses sigmoid and relu. And the loss function adopts mean squared error, which is commonly used
in regression problems.
Table 12. FNN hyper-parameters and architecture related settings
Item Value set
Number of input variables 8
Number of hidden layers 3
Number of output variables 4
Number of hidden layer nodes 500
Optimizer adam
Excitation function sigmoid
Loss function mean squared error
Number of generations 300
Batch size 40
Table 13. LSTM neural network hyper-parameters and architecture related settings
Item Value set
Number of input variables 8
Number of hidden layers 3
Number of output variables 4
Number of hidden layer nodes 256, 128, 64
Optimizer adam
Excitation function relu
Loss function mean squared error
Number of generations 300
Batch size 32
24
 After the neural network training set above, the training and comparison results are presented in
Tables 14-15. From the results of the mean square error as shown in Eq. (20) in Table 14: Except for the
upper and lower bound of the grid in the Nasdaq 100 index, LSTM has a smaller mean square error. All
four output variables in the other indices have a smaller mean square error, which the mean square errors
of the four output variables in other indices are all FNN performs better.

1 N
MSE  
N i 1
( yi  y i )2 (20)

Table 14. Comparison of mean square error between FNN and LSTM models
Neural network type FNN LSTM
S&P 500
upper bound of the grid Gul 129510 186366
lower bound of the grid Gll 59787 108622
number of upper grids nu 207 252
number of lower grids nl 308 442
Nasdaq 100
upper bound of the grid Gul 4725772 518181
lower bound of the grid Gll 1681699 1162381
number of upper grids nu 133 296
number of lower grids nl 246 568
Dow Jones Industrial Average
upper bound of the grid Gul 6016533 24446812
lower bound of the grid Gll 2011354 19594848
number of upper grids nu 164 237
number of lower grids nl 319 402
Euro Stoxx 50
upper bound of the grid Gul 105024 334456
lower bound of the grid Gll 33988 106422
number of upper grids nu 132 202
number of lower grids nl 199 290
Shanghai Composite
upper bound of the grid Gul 65199 92869
lower bound of the grid Gll 33604 26189
number of upper grids nu 85 280
number of lower grids nl 214 459

25
 The coefficient of determination (R square) of the two models, which is an index to measure the
fitness of the regression model and can also be interpreted as the degree of interpretation of the model, as
shown in Table 15 is calculated based on Eq. (21).

SS
R  1  res 
2  i 1
( yi  fi )2
(21)
SStot  i 1
( yi  y )2

In Table 15, it can be found that ecept for the Nasdaq 100 index, most of the time, FNN has a better
fit for the four output variables.

Table 15. Comparison of coefficient of determination between FNN and LSTM models
Neural network type FNN LSTM
S&P 500 90.490%
upper bound of the grid Gul 97.410% 90.490%
lower bound of the grid Gll 99.586% 97.240%
number of upper grids nu 94.972% 99.469%
number of lower grids nl 99.392% 94.892%
Nasdaq 100
upper bound of the grid Gul 96.415% 98.075%
lower bound of the grid Gll 97.088% 99.915%
number of upper grids nu 94.942% 99.635%
number of lower grids nl 94.737% 98.019%
Dow Jones Industrial Average
upper bound of the grid Gul 97.704% 95.970%
lower bound of the grid Gll 99.771% 93.429%
number of upper grids nu 99.855% 93.685%
number of lower grids nl 99.559% 91.285%
Euro Stoxx 50
upper bound of the grid Gul 96.183% 94.392%
lower bound of the grid Gll 99.324% 98.063%
number of upper grids nu 99.297% 96.858%
number of lower grids nl 98.405% 99.760%
Shanghai Composite
upper bound of the grid Gul 97.157% 96.577%
lower bound of the grid Gll 96.749% 95.545%
number of upper grids nu 93.165% 93.881%
number of lower grids nl 99.661% 99.003%

26
 Furthermore, the following four values of return of investment (ROI), maximum drawdown (MDD),
volatility, and Sharpe ratio are used to examine the performance of the model in this study. In terms of
methods include first buy and last sell (Buy and Sell, B&S), first sell and last buy (Sell and Buy, S&B),
grid trading system robot (GTSbot) [10], Ichimoku equilibrium Figure (IK) [12], Flexible Grid trained
with Fully-Connected Neural Network (FG-FNN), Flexible Grid trained with LSTM (FG-LSTM), equal-
distance grid, equal-ratio grid, and flexible grid, a total of 9 methods for comparison.
First of all, the formula of ROI can refer to Eq. (22), which is the most common indicator of
quantitative investment performance in investment sciences. The calculation of the ratio of investment
income to cost usually presents as an annualized return to total return. The trading costs including handling
fees have been included in the calculation results. The results please refer to Table 16.

ROI = (net profit – investment costs) × 100% (22)

Table 16. Comparison of ROI

S&P Nasdaq DJI Euro Stoxx Shanghai
B&S 14.392% 1.519% 9.880% 28.340% -15.581%
S&B -14.392% -1.519% -9.880% -28.340% 15.581%
GTSbot 6.408% 4.680% 6.594% 1.191% -0.466%
IWOC 24.111% 23.171% 7.143% 32.087% -18.758%
FG-FNN 11.520% 11.733% 8.849% 12.977% -3.125%
FG-LSTM 4.639% 1.823% 2.940% 7.734% -9.283%
Equal-distance 5.194% -2.972% 3.480% 10.748% -4.495%
Equal-ratio 4.536% -3.625% 3.276% 10.764% -4.892%
Flexible 5.589% -1.895% 4.270% 11.477% -4.309%

In terms of ROI, the FNN flexible grid achieves the top three returns in each index. In particular, the
FNN flexible grid has the most ability to control the overall loss in the downward trend of prices.
Second, the comparison is the maximum drawdown, i.e., the amount of income that has fallen sharply
in all periods, which is one of the indicators for evaluating investment risk. Please refer to Table 17 for the
results.

Table 17. Comparison of MDD

S&P Nasdaq DJI Euro Stoxx Shanghai
B&S 7.960% 12.570% 7.290% 4.029% 8.529%
S&B 7.491% 10.370% 6.580% 7.313% 3.103%
GTSbot 0.524% 5.335% 1.000% 0.510% 1.126%
IWOC 10.345% 12.421% 8.606% 5.467% 7.640%
FG-FNN 6.455% 5.596% 7.491% 2.034% 4.105%
27
 FG-LSTM 1.580% 49.031% 4.809% 2.966% 3.551%
Equal-distance 5.252% 8.594% 4.078% 1.513% 4.074%
Equal-ratio 5.243% 10.096% 4.057% 1.529% 4.059%
Flexible 4.737% 8.780% 4.163% 4.144% 4.208%

In terms of the maximum drawdown, it can be found that the method with a larger return on investment
has a relatively higher maximum drawdown, which verifies the theory of high risk and high return in
investment science. There are two reasons for the performance of the FNN flexible grid on the MDD: The risk
is relatively high in the case of a relatively high return on investment. In addition, the parameters of each period
are calculated by the neural network. If the parameters predicted on one period are particularly poor, it is likely
to cause a large drop in a single period. Whether this MDD condition can be regarded as a single condition of
outliers is yet to be verified by volatility.
In quantitative trading, risk and model stability are assessed through its volatility. The calculation method
refers to Eq. (23) and the results are shown in Table 18.

Vol = σ[return] (23)

It can be found from Table 18 that the performance of FNN flexible grid on volatility is the top three in
each index, i.e., the performance of FNN flexible grid is very stable compared with other methods. It can also
be speculated that the poor performance on the MDD may be a single event, and it can be regarded as a robust
investment model in most cases. The LSTM flexible grid is the best among all models in the S&P and Nasdaq
data sets and is a less risky and more robust investment model.

Table 18. Comparison of volatility

S&P Nasdaq DJI Euro Stoxx Shanghai
B&S 0.06126 0.07671 0.03568 0.06221 0.01840
S&B 0.06126 0.07671 0.03568 0.06221 0.01840
GTSbot 0.01710 0.03589 0.01795 0.00267 0.00190
IWOC 0.07689 0.08398 0.04564 0.07337 0.03568
FG-FNN 0.01687 0.02076 0.02149 0.01730 0.01414
FG-LSTM 0.01672 0.01637 0.02203 0.01679 0.02030
Equal-distance 0.02450 0.03579 0.02129 0.01786 0.01590
Equal-ratio 0.02473 0.03629 0.02155 0.01828 0.01629
Flexible 0.02503 0.03665 0.02162 0.01862 0.01624

Finally, in terms of the Sharpe ratio, which is a model performance indicator that calculates the ratio
between investment return and risk. It can also be interpreted as the reward that can be exchanged for each
unit of risk. The calculation method refers to Eq. (24) and the results are as shown in Table 19.
28
 Sharpe = return /σ[return] (24)

Table 19. Comparison of Sharpe ratio

S&P Nasdaq DJI Euro Stoxx Shanghai
B&S 2.349 0.198 2.769 4.556 -8.466
S&B -2.349 -0.198 -2.769 -4.556 8.466
GTSbot 3.748 1.304 3.673 4.466 -2.446
IWOC 3.136 2.759 1.565 4.373 -5.258
FG-FNN 6.828 5.651 4.119 7.499 -2.210
FG-LSTM 2.774 1.114 1.335 4.606 -4.573
Equal-distance 2.120 -0.830 1.634 6.017 -2.827
Equal-ratio 1.834 -0.999 1.520 5.887 -3.004
Flexible 2.233 -0.517 1.975 6.165 -2.653

From Table 19, it can be found that the Sharpe ratio of FNN flexible grid is the best among all methods
on the four indices. In the falling market situation, in addition to short selling, its Sharpe ratio is also the
largest. From this, it can be found that the FNN flexible grid can obtain the highest reward for each unit
of risk.

5. Conclusions
In addition to proposing a new grid trading architecture, this study effectively improves the ROI of
the original model, improves the drawbacks of premature entry and exit, and utilizes SSO algorithm and
ANN to assist grid trading for parameter selection and providing the model the ability to adapt to the
market.
In terms of model performance, five major market indices are used as verification data, which cover
the United States, Europe and China. The FNN flexible grid performs very well in Sharpe ratio. It has
excellent investment return rate and model robustness, and can properly control the balance between risk
and return.
This study proves that the technology combined with AI in can bring breakthroughs to the original
grid trading model in quantitative trading, and can adapt to various rapidly changing market situations. In
the future, looking forward to more related extended research resulting that the grid trading model can be
more accurately close to human decision-making or even better than human judgment when adjusting
parameters to eliminate human impulses and market emotions, and make rational market trading decisions
to obtain better investment benefits.

29
Acknowledgments
This research was supported in part by the National Science and Technology Council, R.O.C (MOST
107-2221-E-007-072-MY3, MOST 110-2221-E-007-107-MY3, MOST 109-2221-E-424-002 and MOST
110-2511-H-130-002). This article was once submitted to arXiv as a temporary submission that was just
for reference and did not provide the copyright.

References
[1] J. IOSCO, Regulatory Issues Raised by the Impact of Technological Changes on Market Integrity
and Efficiency, Consultation Report, 2011.
[2] F. Allen, J. McAndrews, P. Strahan, E-finance: an introduction. Journal of financial services
research 22(1) (2002) 5-27.
[3] M. Shahrokhi, E‐finance: status, innovations, resources and future challenges, Managerial Finance,
2008.
[4] A.M. Turing, Computing machinery and intelligence, in Parsing the turing test, Springer (2009) 23-
65.
[5] M. Hutson, How researchers are teaching AI to learn like a child, Science 10, 2018.
[6] M. Minsky, S.A. Papert, Perceptrons: An introduction to computational geometry, MIT press, 2017.
[7] Silver, D., et al., Mastering the game of Go with deep neural networks and tree search. nature
529(7587) (2016) 484-489.
[8] I. Aldridge, High-frequency trading: a practical guide to algorithmic strategies and trading systems,
John Wiley & Sons 604, 2013.
[9] Z.R. Lai, P.Y. Yang, L. Fang, X. Wu, Reweighted price relative tracking system for automatic
portfolio optimization, IEEE Transactions on Systems, Man, and Cybernetics: Systems 50(11)
(2018) 4349-4361.
[10] F. Rundo, F. Trenta, A. L. d. Stallo, S. Battiato, Grid trading system robot (gtsbot): A novel
mathematical algorithm for trading fx market, Applied Sciences 9(9) (2019) 1796.
[11] Q. Wen, Z. Yang, Y. Song, P. Jia, Automatic stock decision support system based on box theory and
SVM algorithm, Expert systems with Applications 37(2) (2010) 1015-1022.
[12] S. Deng, A. Sakurai, Short-term foreign exchange rate trading based on the support/resistance level
of Ichimoku Kinkohyo, 2014 International Conference on Information Science, Electronics and
Electrical Engineering, IEEE, 2014.
[13] A. Ye, V. L. R. Chinthalapati, A. Serguieva, E. Tsang, Developing sustainable trading strategies
using directional changes with high frequency data, 2017 IEEE International Conference on Big
Data, 2017.
[14] A. DuPoly, The Expert4x, No stop, Hedged, Grid Trading System and The Hedged, Multi-
Currency, Forex Trading System, 2008.
[15] W. C. Yeh, S. C. Wei, Economic-based resource allocation for reliable Grid-computing service
based on Grid Bank, Future Generation Computer Systems 28 (7) (2012.) 989-1002.
[16] S. C. Wei, W. C. Yeh, Resource allocation decision model for dependable and cost-effective grid
30
 applications based on Grid Bank, Future Generation Computer Systems 77 (2017) 12-28.
[17] W. C. Yeh, A two-stage discrete particle swarm optimization for the problem of multiple multi-level
redundancy allocation in series systems, Expert Systems with Applications 36(5) (2009) 9192-
9200.
[18] W. C. Yeh, Novel swarm optimization for mining classification rules on thyroid gland data,
Information Sciences 197 (2012) 65-76.
[19] W. C. Yeh, W. W. Chang, Y. Y. Chung, A new hybrid approach for mining breast cancer pattern
using discrete particle swarm optimization and statistical method, Expert Systems with Applications
36(4) (2009) 8204-8211.
[20] W. C. Yeh, Z. Liu, Y.C. Yang, S.Y. Tan, Solving Dual-Channel Supply Chain Pricing Strategy
Problem with Multi-Level Programming Based on Improved Simplified Swarm Optimization,
Technologies2022(10) (2022) doi.org/10.3390/technologies10030073
[21] H.C.S. Lin, C.L. Huang, W.C. Yeh, A Novel Constraints Model of Credibility-Fuzzy for Reliability
Redundancy Allocation Problem by Simplified Swarm Optimization, Applied Sciences 11 (10765)
(2021) doi: 10.3390/app112210765
[22] W.C. Yeh, P.J. Lai, W.C. Lee, M.C. Chuang, Parallel-machine scheduling to minimize makespan
with fuzzy processing times and learning effects, Information Sciences 269 (2014) 142-158.
[23] W.C. Yeh, W. Zhu, S.Y. Tan, G.G. Wang, Y.H. Yeh, Novel General Active Reliability Redundancy
Allocation Problems and Algorithm, Reliability Engineering & System Safety, 218 (2021) 108167.
[24] W.C. Yeh, S.Y. Tan, Simplified Swarm Optimization for the Heterogeneous Fleet Vehicle Routing
Problem with Time-Varying Continuous Speed Function, Electronics, 2021 (10) (2021)
doi:10.3390/electronics10151775
[25] C. Bae, W.C. Yeh, N. Wahid, Y.Y. Chung, Y. Liu, A New Simplified Swarm Optimization (SSO)
using Exchange Local Search Scheme, International Journal of Innovative Computing, Information
and Control 8 (6) (2012) 4391-4406.
[26] W.C. Yeh, Y.Z. Su, X.Z. Gao, C.F. Hu, J. Wang, C.L. Huang, Simplified Swarm Optimization for
Bi-Objection Active Reliability Redundancy Allocation Problems, Applied Soft Computing 106
(2021) 107321.
[27] M. Wang, W.C. Yeh, T.C. Chu, X. Zhang, C.L. Huang, J. Yang, Solving multi-objective fuzzy
optimization in wireless smart sensor networks under uncertainty using a hybrid of IFR and SSO
algorithm, Energies 11 (9) (2018) 2385.
[28] W.C. Yeh, Simplified swarm optimization in disassembly sequencing problems with learning
effects, Computers & Operations Research 39 (9) (2012) 2168-2177.
[29] W.C. Yeh, A novel boundary swarm optimization method for reliability redundancy allocation
problems, Reliability Engineering & System Safety 192 (2019) 106060.
[30] W.C. Yeh, Orthogonal simplified swarm optimization for the series–parallel redundancy allocation
problem with a mix of components, Knowledge-Based Systems 64(2014) 1-12.
[31] W. Zhu, W.C. Yeh, L. Cao, Z. Zhu, D. Chen, J. Chen, A. Li, Y. Lin, Faster evolutionary

31
 convolutional neural networks based on iSSO for lesion recognition in medical images, Basic &
Clinical Pharmacology & Toxicology 124 (2019) 329-329.
[32] W.C. Yeh, P. Lin, C.L. Huang, Simplified swarm optimisation for the solar cell models parameter
estimation problem, IET Renewable Power Generation 11 (8) (2017) 1166-1173.
[33] P. Lin, S. Cheng, W.C. Yeh, Z. Chen, L. Wu, Parameters extraction of solar cell models using a
modified simplified swarm optimization algorithm, Solar Energy 144 (2017) 594-603.
[34] W.C. Yeh, New Parameter-Free Simplified Swarm Optimization for Artificial Neural Network
Training and Its Application in the Prediction of Time Series, IEEE Transactions on Neural
Networks and Learning Systems 24 (4) (2013) 661-665.
[35] W. Huang, K. K. Lai, Y. Nakamori, S. Wang, L. Yu, Neural networks in finance and economics
forecasting, International Journal of Information Technology & Decision Making 6(01) (2007) 113-
140.
[36] S. Qi, K. Jin, B. Li, Y. Qian, The exploration of internet finance by using neural network, Journal of
Computational and Applied Mathematics 369 (2020) 112630.
[37] S. Siami-Namini, A.S. Namin, Forecasting economics and financial time series: ARIMA vs. LST,.
arXiv preprint arXiv:1803.06386, 2018.
[38] J. Cao, Z. Li, J. Li, Financial time series forecasting model based on CEEMDAN and LST, Physica
A: Statistical Mechanics and its Applications 519 (2019) 127-139.
[39] N. Jegadeesh, Evidence of predictable behavior of security returns, The Journal of finance 45(3)
(1990) 881-898.
[40] N. Jegadeesh, Seasonality in stock price mean reversion: Evidence from the US and the UK, The
Journal of Finance 46(4) (1991) 1427-1444.
[41] B. Li, P. Zhao, S. C. H. HOI, V. Gopalkrishnan, PAMR: Passive aggressive mean reversion strategy
for portfolio selection, Machine learning 87(2) (2012) 221-258.
[42] W.S. McCulloch, W. Pitts, A logical calculus of the ideas immanent in nervous activity, The bulletin
of mathematical biophysics 5(4) (1943) 115-133.
[43] F. Rosenblatt, The perceptron: a probabilistic model for information storage and organization in the
brain, Psychological review 65(6) (1958) 386.
[44] P. Werbos, Beyond regression:new tools for prediction and analysis in the behavioral sciences, Ph.
D. dissertation, Harvard University, 1974.
[45] D.E. Rumelhart, G.E. Hinton, R.J. Williams, Learning representations by back-propagating errors.
Nature 323(6088) (1986) 533-536.
[46] S. Ruder, An overview of gradient descent optimization algorithms, arXiv preprint
arXiv:1609.04747, 2016.
[47] J.M. Poterba, L.H. Summers, Mean reversion in stock prices: Evidence and implications, Journal of
financial economics 22(1) (1988) 27-59.
[48] B. Hurst, Y.H. Ooi, L.H. Pedersen, A century of evidence on trend-following investing, The Journal
of Portfolio Management 44(1) (2017) 15-29.

32