Academic Journal of Science and Technology 
ISSN: 2771-3032 | Vol. 2, No. 1, 2022 

 

82 

A	Quantitative	Trading	Strategy	Based	on	A	Position	
Management	Model	
Tengfei Xu1, a, Zhichao Wang1, b, Zhimin Han1, c, Mengdi Zhang1, d, Jingcheng Liu1, e,  
Shixiao Chang1, f 

1Tianjin University of Science and Technology, Tianjin, China 
atust_xu2002@163.com, bwzc216254@163.com, ctust_xiaohan1209, d1536639857@qq.com, e2637660030@qq.com, fcsx_511@163.com 

 

Abstract: With the rapid development of economic globalization, various financial products have appeared in the domestic 
and foreign financial markets. How to adopt ideal trading strategies to meet the demands of market traders for high returns and 
low risks has become the focus of investors and the research goal of scholars. In order to solve this problem, this paper establishes 
a quantitative trading strategy based on the position management model.First, we performed price forecasting for gold and bitcoin 
based on the Time-series ARIMA method. A differential autoregressive moving average model was developed for the price data 
of gold and bitcoin at different cycle times, respectively, and used to predict the price of the next trading day. Error analysis was 
performed with the actual prices, and it was found that the prediction error was the smallest when the data was 60 days, and the 
relative error of the average prediction value (APV) could be controlled at 0.003016.Then, we establish a quantitative trading 
strategy based on the position management model. We use the Apriori algorithm of Association rules to study the rising and 
falling rules of gold and bitcoin assets. According to the rule of " High throw bargain - hunting" in the investment market, we 
established a position management model and achieved dynamic and stable returns. After the model is established, we continue 
to introduce the evaluation indexes of the investment value of financial assets, among which the first-class indexes are 
profitability and safety. We use the Analytic Hierarchy Process (AHP) to determine the weight of each evaluation index, and 
allocate the daily trading investment through the ratio of two asset evaluation indexes. This quantitative trading strategy, based 
on the position management model of AHP, can not only stabilize the income but also avoid risks, reaching a quantitative trading 
strategy with an annualized rate of return of 25%. On September 10th, 2021, the accumulated income could reach 223,640.58 
USD.Further, we evaluate the profitability and risk resistance of the strategy using Principal component analysis. Model 
validation was performed by varying the parameter values and selecting the parameters that yielded a locally optimal solution, 
which was found to be consistent with our initial parameters and was proof of the optimal solution of the model.Finally, we 
conducted a sensitivity analysis of the model. The two variable parameters of initial commission fluctuation and investment 
principal of gold and bitcoin are varied up and down respectively, and the results show that as the initial commission increases 
or the principal decreases, the number of trades under this strategy gradually decreases and the trading return gradually decreases, 
and the sensitivity curve shows that the model is sensitive and meets expectations. 

Keywords: Time-series Analysis, ARIMA, Position management model, Quantitative trading. 
 

1. Introduction 
With the rapid development of economic globalization, 

financial markets, at home and abroad, have been improving, 
and a variety of financial products appear. On January 3, 2009, 
Bitcoin, an online virtual currency, was created. Subsequently, 
more and more virtual currencies, Wright Currency (LTC) for 
instance, appear in the mainstream media; and promote the 
development of financial markets remarkably. As traders 
make decisions by choosing stocks, bonds, and currencies in 
financial markets, they find those traditional schemes for 
market analysis, investment and prediction fail to meet their 
urgent demand for investment. [1] 

Market traders need to make decisions in an extremely 
complex market, where both profit and risk are uncontrollable 
factors. [2]How to allocate capital in these financial products 
to meet market traders' requirements for high returns and low 
risks has become the focus of market traders and the target of 
scholars' research. 

1.1. Problem Background 

1.2. Restatement of the Problem 
In financial markets, market traders often buy and sell 

assets whose value fluctuates greatly to achieve the goal of 
maximizing total returns. Our team needed to use two pricing 
data files to build a model to determine whether traders should 
buy, hold, or sell their portfolios each day to give the optimal 
daily trading strategy. From September 11, 2016, to 
September 11, 2021, we will start at $1,000 for a total of five 
years of trading. Traders' portfolios will include cash, gold, 
and bitcoin. Given that the initial state is [1000,0,0]; Gold and 
Bitcoin are paid 1% and 2% of each transaction, respectively. 

Problem 1 Build a model based on the price data up to the 
day, which can give the best daily trading strategy. Using the 
model and strategy, what is the investment value of the initial 
$1,000 on September 10, 2021? 

Problem 2 Provides data to prove that the model 
established in the first question provides the optimal strategy. 

Problem 3 Prove the sensitivity of our strategy to the 
transaction price and ask how our strategy and results are 
affected by the transaction price. 

Problem 4 Send our strategy, model, and results to the 
trader in a memo of no more than two pages. 

 
 
 



 

83 

1.3. Our Work 

 
Figure 1. The Flow Chart in This Paper 

 
First, according to the market emotional theory, it is better 

to use less data for short-term prediction. Therefore we 
selected 100 days of price data. 

Next, we established an ARIMA prediction model using 
time series analysis. Predict the rise and fall of the next 
trading day. Since the period of the data has a greater impact 
on the forecast results, we further analyze the forecast results 
of different periods and achieve an average. After inspection, 
the average prediction result is smaller. We finally got an 
accurate predictive model solution. 

Further, on the basis of a more accurate prediction, based 
on the investment law of the financial market, it is sold when 
the price of financial products has increased, and it is bought 
when the price of financial products fell. After analyzing the 
routine of Bitcoin and gold, we have established a trading 
strategy model for position management. 

Secondly, we have established an evaluation index of the 
extent of the trading strategy, and the hierarchical analysis is 
used to make a simple evaluation, which proves that our 
trading strategy model is the local optimal solution. We 
analyze how the transaction strategy model is sensitive to 
transaction costs and how transaction costs affect strategies. 

Finally, we analyzed the advantages and disadvantages of 
the model and wrote a memorandum to deal with our strategy, 
model, and results. 

2. Assumptions and Justifications 
1.Assume that the price data of Bitcoin and the price

 is not affected by the banker control of the dealer. 
The banker can take advantage of huge trading volumes 

and capital control over investment transactions.[3]Because 
of human subjective initiative, we cannot predict the 
operation of bankers to compare bitcoin and gold. So we 
assume that the man is a rational investor, and assume that 
there is no malicious control. 

2. Suppose we have a certain risk of risk during the 

investment process. 
Investors' personality determines their ability to accept 

risks, and at the same time, investors have certain limits to 
accept risks. 

3. Suppose our transaction operation is fully based on 
model planning. 

Investment traders are prone to various factors such a
s market sentiment in the transaction. In reality, the tra
der's mentality cannot be ignored, but we assume that t
he trader fully follows the trading strategy model to ver
ify the model. 

4. Suppose the Trading strategies cannot achieve the 
absolute optimal value. 

In addition to the operation of the bankers, we still can't 
fully accurately predict market activities. In fact, our best 
trading strategy should only be able to achieve the relative 
optimal value. 

3. Notations 
The key mathematical notations used in this paper are listed 

in Table 1. 
 

Table 1. Notations used in this paper 
Symbol Description 

Av Actual value 
Pv Predicted value 
h Number of consecutive increases 
k Number of consecutive declines 

APv Average predicted value 
P Position amount 

4. A Quantitative Trading Strategy 
Based on A Position Management 
Model 

 



 

84 

 
Figure 2. Trading strategy flow chart 

 
For problem 1, the best daily trading strategy is given based 

only on the price data up to that date and the value is found 
on 2021.9.10. We first pre-process the existing data and build 
an ARIMA model based on it to forecast the price of the 
financial product for the next trading day. Based on the 
forecast data, we build a trading strategy model based on 
position management planning and develop a trading strategy 
for that day based on the principle of buying low and selling 
high. 

4.1. Data Description 
Taking Bitcoin as an example, we select the first 100 days 

of Bitcoin price data from the database given in the question, 
exclude other data to get the following data, and plot a line 
graph of Bitcoin price change over 100 days. We can see that 
the price of bitcoin is trending upward. At the same time, there 
is some volatility as the price of bitcoin continues to rise.  

 
Table 2. One-day price of Bitcoin in 100 days 

 
 



 

85 

 
Figure 3. Bitcoin price changes in 100 days 

 

4.2. Short-term Price Forecasting Based on 
Time Series ARIMA Model 

Time series, also known as dynamic series, is a series of 
values of indicators of a phenomenon arranged in time order. 
ARIMA (Autoregressive Integrated Moving Average Model) 
is one of the most commonly used forecasting methods for 
time series. Because ARIMA models are very capable of 
capturing information from linear data and require only 
endogenous variables without resorting to other exogenous 
variables, they are particularly successful in statistical and 
forecasting stock price changes and are commonly used in 
financial forecasting. In this paper, we build a differential 
autoregressive moving average model ARIMA(p,d,q) with 
100 days of Bitcoin price change data based on the theory 
related to time series analysis. 

 

𝑦 𝛼 ∑   𝛼 𝑦 𝜀 ∑   𝛽 𝜀       (1) 
 
and  𝑦 Δ 𝑦 1 𝐿 𝑦 . 
 
 

1 ∑   𝛼 𝐿 1 𝐿 𝑦 𝛼 1 ∑   𝛽 𝐿 𝜀    (2) 

 
where p is the number of autoregressive terms; d denotes 

the order of the difference; and q is the number of sliding 
average terms. 

4.2.1. Checking the Smoothness of the Time Series 
For the ARIMA model, it is required that the time series 

data satisfy smoothness. View the results of the ADF test to 
determine whether it can significantly reject the hypothesis 
that the series is not smooth (p<0.05 or 0.01), based on the 
analyzed t-values. 

 
Table 3. ADF test 

ADF test 
 

variable 
Order of 

difference t p AIC 
critical value 

  1% 5% 10% 

 
Value 

0 0.988 0.994 369.969 -3.5 -2.892 -2.583 
1 -7.693 0.000*** 360.538 -3.5 -2.892 -2.583 
2 -7.306 0.000*** 364.595 -3.507 -2.895 -2.585 

Note: * * *, * * and * represent the significance levels of 1%, 5% and 10% respectively 
 

At the difference of order 0, the significance P-value is 
0.971, which does not present significance at the level, and 
the original hypothesis cannot be rejected, and the series is an 
unsteady time series. 

At the 1st order of difference, the significance P-value is 
0.000***, which is significant at the level and the original 

hypothesis is rejected, and the series is a smooth time series. 
At the difference of order 2, the significance P-value is 

0.000***, which presents significance at the level, and the 
original hypothesis is rejected, and the series is a smooth time 
series. 

 



 

86 

 
Figure 4. Optimal difference sequence diagram 

 
The original data do not fluctuate much up and down after 

first-order differencing, and the series satisfies smoothness. 
4.2.2. Selected Order 

 

 
Figure 5. Partial Autocorrelation Function (PACF) 

 

 
Figure 6. Autocorrelation Function (ACF) 

 



 

87 

 
Within the confidence interval, ACF or PACF fluctuates 

around 0. Both autocorrelation and partial autocorrelation 
plots dragged the tails, and the most significant order 

(minimum) in the combined PACF and ACF plots estimated 
p and q values of 1. 

 
Table 4. Model parameters table 

Model parameters table 
 Coefficient Standard deviation t p>|t| 0.025 0.975 

Constants 1.773 0.509 3.482 0 0.775 2.771 
ar.L1.D.Value 0.445 0.222 2.002 0.045 0.009 0.881 
ma.L1.D.Vale -0.713 0.176 -4.049 0 -1.058 -0.368 

 
The results of the model are the ARIMA model (1,1,1) test 

table and based on 1 difference data, the model formula is as 
follows: 

 
𝑦 𝑡 1.773 0.445 𝑦 𝑡 1 0.713 𝜀 𝑡 1  (3) 
 

(2) 

4.2.3. Test of the Model 
Table 5. Q Statistics 

Q 
Statistics 

Q6(P-value) 0.002(0.963) 

Q12(P-value) 0.296(1.000) 

Q18(P-value) 7.648(0.812) 

Q24(P-value) 16.309(0.571) 

Q30(P-value) 20.066(0.693) 
 
From the analysis of the Q-statistic results, it can be 

obtained that Q6 does not present significance at the level and 
the hypothesis that the residuals of the model are white noise 
series cannot be rejected. 

 

𝑅 1 1
∑

∑
        (4) 

 
The goodness-of-fit R^2 of the model is 0.977, the model 

performs well, and the model basically meets the 
requirements.  

4.2.4. Solving of the Mode 
Select the first 100 days, 60 days, 30 days, and 10 days of 

price data to make the forecast respectively. Calculate the 
relative error of the forecast results.  

Actual value(Av): 721.5. 
 

𝑅𝐸
| |

                (5) 

 
According to the error analysis, the most accurate forecast 

result is obtained when the chosen forecast data is 60 days:  
 

𝑅𝐸 0.003454              (6) 
 

Table 6. Table of relative error of forecast for different days 
Relative error table 

Time Relative Error 
100 Days 0.00365 
60 Days 0.00345 
30 Days 0.00428 
10 Days 0.01991 

 
Table 7. Predicted value 

Predicted value 
Order (time) Predicted value 

1 719.008 
 

Because the forecast result is closely related to the period, 
we choose the price data solution model with different periods 

of 60 days. And the average value is taken to obtain a more 
accurate prediction result. 

 
Table 8. Predicted value 

Time Period Separate predicted values 
1 719.008 
2 724.767 
3 714.196 

 
Average predicted value (Apv): 719.324 𝑅𝐸 : 0.003016 
 

𝑅𝐸 𝑅𝐸              (7) 
 

Therefore, the average prediction results are more accurate. 
Similarly, for gold, we choose 60 days of different time 

periods of price data to solve the model. And take its average 
value, we get a more accurate prediction result: 1174.25, the 

model performance is excellent and basically meets the 
requirements. 

4.3. Position Management Model 
Fund management is a very important idea in transactions. 

We often control the risk of transactions by controlling the use 
of funds and placing orders based on a fixed percentage of 
funds. These strategies can be realized by writing the model 



 

88 

of adding and reducing warehouses. The model of adding and 
reducing positions refers to allowing continuous opening 
signals or continuous closing signals to add and reduce 
positions. For traders who need to add or subtract positions, 
when the strategy is implemented, they hope that the next 
action after opening the position can still be opening the 
position, or they can continue to close positions in batches. If 
it is a one-opening-one-leveling signal filtering model, the 
opening can only be the corresponding closing operation, thus 
the strategy of adding and reducing the positions cannot be 
realized. The model of adding and reducing positions allows 
continuous opening signals or continuous closing signals to 
solve this problem. 

According to the time series ARIMA model above, we first 
forecast the future trend of the short and medium-term and 
judge whether to trade. Then forecast the price of any future 
trading day and judge the price trend of a future trading day. 
After that, we control the trading risk by using the plus-minus 
model and establish a quantitative trading strategy. Taking 
into account the fluctuation of the positions held before the 
trading day, the total invested capital is guaranteed to be in a 
non-loss state by adding or reducing the positions. That is, 
when the risk is lowest, a trading strategy with relatively 
stable returns is obtained. The method avoids that the 
transaction is at extremely high risk due to a single search for 
the optimal solution when the transaction yield is the highest. 

4.3.1. Analysis of the Law of Rising and Fall 
First of all, we analyze the fluctuation rule of historical 

transaction price data of gold and bitcoin respectively. 
Step 1:  Separately calculate the rise and fall rates of gold 

and bitcoin for each trading day before the trading day. The 
price increase rate t_n on day n is {closing price on day n–
closing price on day (n-1)}/closing price on the day (n-1), i.e. 

 

𝑡                (8) 

 

 
Step 2: Get the median of all the gains/losses in gold and 

bitcoin respectively. 
Step 3: According to the association rule Apriori algorithm, 

count the number of subsets in the transaction data for 2 
consecutive times, 3 consecutive times, 4 consecutive times 
and 5 consecutive times of rise/fall: up {h2,h3,h4,h5}, down 
{k1,k2,k3,k4,k5}. 

Step 4: Obtaining that the number of consecutive times of 
more than 90% of the subsets in all continuously rising/falling 
subsets is greater than/less than u1 times. 

Step 5: In the historical transaction data, for every U price, 
increase and decrease, get M9 of all price increases and M1 
of all price decreases. Among them, M9 is the ninetieth digit 
in all the increased data, indicating that 90% of the data fall 
under M9; M1 is the first decile of all the increased data, 
indicating that 10% of the data falls under M1. 

Based on the above statistics, it can be analyzed that under 
ideal conditions, if there is no significant difference between 
each increase and decrease when the price falls u1 times, it 
just falls M1. If we add positions every time the price goes 
down, then we will get a profit the next time the price goes up 
by M5. 

4.3.2. Model Establishment and Solution 
When making the forecast based on the ARIMA model, we 

found that the error between the forecast result and the actual 
price is the smallest when the 60-day data is used for the 
forecast. Therefore, our cash, gold, and bitcoin [C, G, B] 
allocation values for the first 60 days remain in the initial state 
[1000, 0, 0]. When we forecast based on the ARIMA model 
when trading from the 61st day, we found that when we 
forecast using the 60-day data, the error between the forecast 
result and the actual price is minimal. Therefore, our cash, 
gold, and bitcoin [C, G, B] allocations for the first 60 days 
remain in their initial state [1000,0,0] and are traded from day 
61 onwards. 

 
Table 9. Statement of loss at trade date 

Trading 
Day 

Investment 
amount 

The previous day's trading 
day's increase or decrease 

service 
charge 

Actual loss 

Day 1 P1 -m 1% P1 1% P1 

Day 2 P2 -n 1% P2 1%( P1+ P2)+ P1×m 

Day 3 P3 -o 1%P3 1%( P1+P + P3)+ P1×m+( P1+P )×n 
… …   … 

Day n Pn -z 1%Pn 
1% 𝑃 𝑃 𝑚 

𝑃 P n ⋯ 
𝑃 P ⋯ P z 

 
Based on the above statistics, it can be analyzed that under 

ideal conditions, if there is no significant difference between 
each increase and decrease when the price falls u1 times, it 
just falls m1. If we add positions every time the price goes 
down, then we will get a profit the next time the price goes up 
by M5. 

Among them, the amount of the first addition is P1; 
The amount of the second addition is P2; 
The amount of the third position increase is P3; 
The amount of the nth additional position is Pn. 
Countable formula: 
 

 
𝑃 1 1% 𝑀 . 𝑃 1 1% . 𝑀 . 0      (9) 

 
𝑃 1 1% 𝑀 5 𝑃 1 𝑃 . 𝑀 , 𝑃 , M . 0    (10) 

 
By analogy, the conditions to be satisfied for the n-th 

addition are:  
 

1 1% 𝑃 𝑀 . 𝑃 . 𝑀 . ⋯ 𝑃 . 𝑀 . 0   11  

 
Among them, n also indicates our ability to increase 



 

89 

positions, which is related to the maximum number of ups and 
downs we counted. That is, if the maximum number of falls 
calculated for more than 90% on that day is 4, this means that 
the probability will only fall 4 times in a row. We will ensure 
that we have the ability to add positions 5 times to determine 
the number of positions for the first time. 

For example, when we were working on the 300-day 
bitcoin position on the same day, we knew that the 299-
day bitcoin fell by -0.0455. We first counted the number 
and frequency of consecutive decreases in the first 299 
days, as shown in the following table: 

 
Table 10. The number and frequency of consecutive declines in the previous 299 days 

K Frequency Probability 
K2 27 58.70% 
K3 11 23.91% 
K4 6 13.04% 
K5 2 4.35% 

 
From the data in the table: 
 

P(K2)+ P(K3)+ P(K4)=95.65%      (12) 
 

We calculated that the maximum number of consecutive 
falls of more than 90% is 4, i.e. only 4.35% is likely to have 
more than 4 consecutive falls. 

Calculated by MATLAB, the median decrease is M5: -
0.0047593, and the decile decrease is M1:-0.056583. At this 
point, we get the required parameters. 

At this time, we only need to consider that we can still close 
the position under the condition of 4 consecutive falls. 
Through the above formula, we can use MATLAB 
polynomial to calculate and solve P1=75.4158345. P1 is the 

amount we added when we first fell, i.e. we should at least 
make up the position of bitcoin at $75.416 on the 300th day 
to guarantee revenue. 

The above is the analysis and calculation of the position 
management strategy. Next, we establish an evaluation index 
for financial products and allocate the amount of investment 
in the two financial products in proportion when the second 
financial product needs to be added to the position 
simultaneously. 

We use the volatility of financial products to express their 
safety and use the average increase and the number of days of 
increase to express their profitability. We then use safety and 
profitability to calculate the allocation of funds when both 
need to be placed simultaneously. 

 

 
Figure 7. Evaluation Index System of Financial Products 

 
A blank line in the chart indicates that the day is a non-

trading day for gold, so there is no allocation. At the 
beginning of the trading period, the average increase and the 
number of days of increase are equal for both, so the amount 
of money allocated to both is equal. As bitcoin rises more, the 
amount invested is allocated more. But at the same time, the 

volatility of bitcoin becomes larger, and the maximum 60-day 
standard deviation of the neighboring bitcoin reaches 486.37, 
which makes it less safe. Therefore, the tendency of the 
bitcoin investment amount allocation to become larger is 
reduced, and this allocation strategy ensures both the safety 
and profitability of the fund allocation. 

 



 

90 

 
Figure 8. Investment allocation rate 

 
The above is the analysis and calculation of the position 

management strategy. Next, we set up an evaluation index for 
financial products to allocate the investment amount in 
proportion. We use the volatility of financial products to 
express their safety and use the average increase and the 
number of days of increase to express their profitability. We 
then use safety and profitability to calculate the allocation of 
funds when both need to be placed simultaneously. At the 
beginning of the transaction, the average increase of the two 
companies is equal to the number of days of increase, so the 
allocation of funds between the two companies is equal. 
When the increase in bitcoin increases, the allocation of 
investment amount increases, but at the same time, the 
fluctuation degree of bitcoin increases, and the security 
decreases. Therefore, the trend of bitcoin investment amount 
allocation becoming larger decreases, and this allocation 
strategy ensures the safety and profitability of fund allocation. 

Based on the above calculation of transaction allocation for 
different financial products, combined with the judgment of 
whether to conduct transactions and the analysis and 
calculation of position management strategies, we have 
established a complete transaction strategy model. 

Based on our trading strategy model, we conduct daily 
trading and calculate the annualized trading rate for each 
trading day. The results are shown in the figure below. It can 
be seen that the annualized yield is generally on the rise. The 
annualized rate of return was negative for a few days in the 
beginning because the prior period was affected by the decline. 
With the increase in the number of transactions, the daily 
annualized rate of return rose rapidly in the first three months 
or so before the investment began, and finally stabilized at 
about 25%. Our trading achieved the trading objective of 
continuous and stable revenue and low risk. 

 

 
Figure 9. Earning rate 

 
 



 

91 

 

5. Proof of the Best Strategy 
A quantitative trading strategy requires a continuous 

updating and optimization process to adapt to the increasingly 
complex changes in the investment market. The use of 
computer-aided analysis and optimization processes is 
obviously more efficient than manual analysis. 

Through the continuous optimization of quantitative 
trading strategies, the operation of models, and the analysis of 
data, quantitative investment can better help people to 
manage capital, improve revenue, and control withdrawal. 

Through continuous optimization of trading strategies, we 
can guide people's trading behavior. 

Yield, revenue stability, and expected revenue are all 
evaluation indicators to measure the feasibility of this trading 
strategy. The yield can be used to measure the profitability of 
a model. At the same time, considering the risks, the model 
should have the ability to resist the risks. Therefore, a trading 
strategy evaluation index includes the ability to return and the 
ability to resist risks. The evaluation index of trading strategy 
is used to judge whether the trading strategy of a model is 
good or bad. The higher the value, the better the trading 
strategy of the model. 

 

 
Figure 10. Bitcoin and gold daily price rises and falls 

 
According to the test result of the trading strategy in 

Question 1, when using the existing data to simulate the 
strategy given by the model, the strategy can obtain higher 
revenue than buying and holding, maintain certain stability 
under various conditions, and obtain better revenue. Based on 
the trading strategy proposed above. We try our best to grasp 
the maximum range of each rising price trend and increase the 

overall yield under the premise of equal risk. Moreover, we 
can help investors to find statistically significant parts in 
highly random or non-random price fluctuations, so as to 
improve the security of transactions and obtain long-term 
stable profits. Therefore, the model in this paper gives the best 
strategy. 

 

 
Figure 11. Strategy evaluation index thermal diagram 

 
We change the forecast range of the trading strategy model 

and the starting time of trading to get the data of different 
situations. Evaluate their profitability and ability to resist 
risks by principal component analysis. And calculate the 



 

92 

evaluation index of trading strategy when these two data are 
different. The red part indicates that the value of the trading 
strategy evaluation index is high. 

6. Transaction Costs to The Sensitivity 
Analysis of The Strategy 

To address question three, we need to determine the 

sensitivity of the strategy to the transaction cost. By changing 
the transaction cost parameters, we make the commission 
fluctuate from 1% and 2% for gold and bitcoin respectively 
when trading. Using the quantitative trading strategy 
established in the previous step, the following results were 
obtained. 

 

 
Figure 12. Number of transactions and total revenue when transaction costs fluctuate 

 
The results show that when the trading commissions of 

gold and bitcoin are 1% and 2%, respectively, the total 
transaction income is 223,640.6 US dollars. When the trading 
commissions of gold and bitcoin are 2% and 3%, the total 
revenue of the transaction is 21,846.37 US dollars. When the 
trading commissions of gold and bitcoin are 3% and 4%, the 
total revenue of the transaction is 21,843.45 USD. When the 
trading commissions of gold and bitcoin are 4% and 5%, the 
total revenue of the transaction is 20156.89 USD. Since then, 
when the transaction cost becomes larger, the number of 
transactions gradually decreases, the decline is faster and 
faster, and the total income shows a downward trend. It can 
be seen that the quantitative trading strategy we established is 

highly sensitive to the transaction cost. 
When the transaction cost does not change much, for 

example, when the trading commissions of gold and bitcoin 
change from 2% and 3% to 3% and 4%, there is a slight 
increase, because both risks and benefits are affected. 
Because reducing the risk of transaction avoidance is greater 
than the loss. However, when the transaction cost becomes 
too large, the number of transactions will drop significantly, 
and the income will also drop significantly. 

7. Sensitivity Analysis 

 

 
Figure 13. The rate of return when you change the principal 



 

93 

 
The results show that when the transaction cost is reduced 

by 10%, the annual return rate is obviously increased. 
However, when transaction costs increased by 10%, the 
annual yield decreased. This is reasonable and understandable. 
This is because both risks and benefits are affected. The 
changing trend of the curve obtained from the sensitivity test 
is consistent with the actual situation. 

8. Model Evaluation and Further 
Discussion 

8.1. Strengths 
1. ARIMA model can be used as a short-term prediction 

model, and it can be seen from the fitting diagram that the 
fitting effect is good, which provides a certain reference for 
the formulation of trading strategies. The error of the result 
obtained by reasonable assumption is relatively small, 

2. When making trading strategies, we use the Apriori 
algorithm in association rules to produce candidate sets, 
which greatly reduces the size of frequent sets and achieves 
good performance. 

3. This strategy can achieve higher returns than buying and 
holding, and maintain certain stability in various situations to 
obtain better returns. 

8.2. Weaknesses 
Our model is only suitable for short-term prediction, and 

the prediction error of long-term results will be relatively 
large. If the external sound changes greatly, our model will 
have greatly deviated. 

8.3. Further Discussion 
Every strategy is formulated with certain timeliness. The 

trading model can bring investors excess returns in a specific 
time window and under a specific market situation, but it does 
not mean that it is a successful trading model at any time. 
Once it is imitated by a large number of traders in the market, 

the yield will inevitably decrease. A successful trading model 
relies on continuous optimization and improvement by using 
the latest data and algorithms. At the same time, in the process 
of trading, dynamic information should be monitored in real-
time, an early warning mechanism should be established, and 
strategies should be stopped when problems are found, to 
prevent the occurrence of major events. 

Most of the problems in the problem are idealized and 
simplified, and some information is covered up to a certain 
extent. Aiming at the problem, the number of idealized 
assumptions when the model is established can be reduced by 
combining with actual experiments, so that the model is more 
in line with the actual situation. 

In the process of the practical application of the model, 
multi-direction efforts are also needed to realize effective 
control of investment risks. 

References 
[1] Da Jieqiong. Research on Quantitative Investment Strategy of 

A-share Market based on RNN-LSTM Model [D]. Zhongnan 
University of Economics and Law, 2020.  

[2] Li Haoran. Quantitative Trading strategy design based on 
genetic programming algorithm [D]. Shanghai Normal 
University,2021. 

[3] Li Yi. Research on the application of Quantitative Trading 
Strategy Model [D]. Yunnan University of Finance and 
Economics, 2014. 

[4] Liu Lijun, Liang Guopeng, Wang Haitao. Research on 
quantitative Trading Strategy System Model [J]. Shanghai 
Commerce, 2021(09):158-159. 

[5] Fang quannan, Zhu Jianping, Xie Bangchang. Fund yield 
direction prediction and trading strategy based on random 
forest method[J]. Economic Journal,2010(02):61-65.  

[6] Rossi Stefano,Tinn Katrin. Rational quantitative trading in 
efficient markets[J]. Journal of Economic Theory,2021,191.