7069788.pdf - Hindawi.com

Research ArticleConstruction of a Fundamental Quantitative Evaluation Model ofthe A-Share Listed Companies Based on the BP Neural Network

Yankai Sheng Kui Fu and Jing Liang

School of Economics Wuhan University of Technology Wuhan 430070 China

Correspondence should be addressed to Yankai Sheng 298671whuteducn

Received 6 January 2022 Revised 1 February 2022 Accepted 7 February 2022 Published 1 March 2022

Academic Editor Shahid Mumtaz

Copyright copy 2022 Yankai Sheng et alis is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Quantitative investment has attracted much attention along with the vigorous development of Fintech Fundamentals are one ofthe most important reference factors for investment Before quantitative trading evaluation of fundamentals may have been moredependent on personal experience While artificial intelligence evaluation models can provide good investment suggestions andselect stocks with better fundamentals From the four angles of solvency growth ability operation ability and profitability thisresearch selects 13 financial indicators to build a fundamental evaluation system through correlation coefficient analysis ecorporate life cycle assessment indicator is innovatively added so that the fundamental improvement expectation is put into theevaluation system Four different kinds of scoring methods are applied to obtain a more rational and comprehensive evaluation ofindicators en grey relational analysis is adopted to determine the initial weight to calculate the expected output Finally BPneural network (back propagation) is used for training and testing to realize weight optimization It is concluded that the model issuitable for quantitative scoring of the fundamentals of listed companies and can effectively reflect their value of them

1 Introduction

Since Benjamin Graham founded the theory of value in-vestment [1] the fundamentals of stocks have become anindispensable factor of investment A large number ofeconometric models used to evaluate the intrinsic value ofcompanies have emerged such as Tobinrsquos Q theory [2 3] andfive-factor asset pricing model [4] which pushed the de-velopment of value investment However traditionaleconometric models are limited due to a lack of proof todemonstrate that the result is optimal

With the blossom of FinTech and quantitative invest-ment artificial intelligence algorithms are widely used tooptimize the results of traditional models and constructbetter portfolios For example a classifier model called SVM(support vector machines) has been shown to perform wellin stock price forecasting [5] and financial distress [6]

Moreover researchers have conducted a vast investigationon artificial neural networks used in the financial field suchas BP (back propagation) neural network [7] LSTM (longshort-term memory) neural network [8] and NARX(nonlinear autoregression exogenous) neural network [9]

Motivated by the above observations this paper focuseson the construction of a fundamental quantitative evaluationmodel of the A-share listed companies based on the BPneural network In section 2 the theoretical basis of themodel is provided in order to explain the backgroundconcepts and related technologies In section 3 this paperprovides the detailed process of model construction in-cluding the determination of expected output and parametersetting In section 4 the results of the model are analyzed Insection 5 the conclusion and outlook are summarized

e contribution of this research can be concluded asfollows

HindawiComputational Intelligence and NeuroscienceVolume 2022 Article ID 7069788 12 pageshttpsdoiorg10115520227069788

(1) When the indicator is under a dimensionless processa variety of evaluation methods are innovativelycombined so that the indicator score is more rea-sonable and comprehensive

(2) With the corporate life cycle evaluation indicatoradded the companies with fundamental improve-ment expectations can be identified Since the cashflow statement is included in the evaluation systemall the important data from the three statements areincluded in the evaluation model for the first time

(3) Grey relational analysis and BP neural networktraining simulation are used to optimize the weightof indicators so as to obtain the fundamental scoringsystem of all-industry listed companies

2 Theoretical Basis of Model Construction

21 Grey Relational Analysis Grey relational analysis is aquantitative description and comparison of the developmentstate of objects so as to analyze and determine the influencedegree between the contribution measure of factors on themain behavior e Grey correlation degree is a measure ofthe correlation between two systems If the relative changesituation of the two factors is consistent in the developmentprocess the grey correlation degree of the two factors islarge otherwise it is small Grey relational analysis is usuallyused to address the comprehensive evaluation problem [10]Chen et al [11] evaluated the growth of small and medium-sized listed companies through grey relational analysis andconcluded that the obtained information can be fully usedwith this method Zhang et al [12] established a quantitativeevaluation model of strategic emerging industries throughgrey relational analysis which made the evaluation systemmore reasonable Delcea et al [13] put forward suggestionson corporate development and extreme situation responseby using the quantitative results obtained with grey rela-tional analysis erefore it is concluded that grey relationalanalysis is suitable for quantitative scoring of the funda-mentals of listed companies and it is reasonable to take it asthe expected output of the BP neural network e progressof grey relational analysis is shown as follows

Step1 e data after standardization is used to gain the greycorrelation degree coefficients ζ i(k) of the reference seriesand comparison series with formula (1)

ζ i(k)

mini

mink

g0(k) minus gi(k)1113868111386811138681113868

1113868111386811138681113868 + ρmaxi

maxk

g0(k) minus gi(k)1113868111386811138681113868

1113868111386811138681113868

g0(k) minus gi(k)1113868111386811138681113868

1113868111386811138681113868 + ρmaxi

maxk

g0(k) minus gi(k)1113868111386811138681113868

1113868111386811138681113868

(1)

In formula (1) gi(k) is the score of indicators of listedcompanies and ρ is the resolution coefficient e usualpractice is adopted in this paper where ρ 05

Step2 When the calculated grey relational degree coeffi-cient is obtained the weight calculation formula is used toobtain the weight of the indicators

ωi 1113936

nk1 gi(k)

1113936nk1 1113936

mi1 gi(k)

(2)

In formula (2) n is the number of sample companies andm is the number of selected indicators

Step3 According to the weights gained the scores of listedcompanies are calculated by formula (3) as the expectedoutput of the neural network model

G(k) 1113944m

i1ωi middot gi(k) (3)

In the formula G(k) means the score of the company

22 BP Neural Network e BP (back propagation) neuralnetwork proposed by scientists led by Rumelhart andMcClelland in 1986 is a gradient descent method being ableto optimize the weight and threshold through error[14]Zhang Xuemin et al [15] have combined an analytic hier-archy process and BP neural network to build an earlywarning evaluation system in order to keep poverty-strickenpeople from returning to poverty in specific areas ZhangZhengang et al [16] used information entropy theory todetermine the index weight and then used BP neural net-work training and simulation to construct the performanceevaluation system of listed white household appliancescompanies in China anks to its advantages in weightoptimization the BP neural network can be effectively ap-plied to the weight selection of the quantitative evaluationsystem constructed in this paper e structure of the BPneural network is shown in Figure 1

Δwij minus ηzError

zwij

i 1 2 m j 1 2 n (6)

Δvjk minus ηzError

zvjk

k 1 2 n j 1 2 n (7)

A routine BP neural network has 3 layers input layerhidden layer and output layer In Figure 1 x is the inputdata and y is the output data Assume the weight of neuronsin the hidden layer is w and the weight of neurons in theoutput layer is v Assume d is the expected output then theerror of the model Error can be defined by formula (4) asfollows

Error 12

1113944

n

k1dk minus yk( 1113857

2 (4)

In formula (4) yk can be denoted by formula (5) where fis the activation function

yk f 1113944n

j1vjkf 1113944

m

i1wijxi

⎛⎝ ⎞⎠⎛⎝ ⎞⎠ (5)

If Error does not reach the target the neural network willadjust w and v e adjustment amount Δwij and Δvjk are

2 Computational Intelligence and Neuroscience

defined by formula (6) and formula (7) where η is learningrate

After the neural network calculates for a certain timeuntil the Error reaches the target the iteration will stop

23 Corporate Life Cycle 0eory As early as the 1950sMason et al [17] put forward that the development stage of acompany can be viewed from the perspective of its life cycleSince then researchers have proposed a large number of lifecycle stage division models which are mainly based on theaccounting indicators related to corporate organizationalbehavior and corporate value where the life stages of acompany are divided into segments from 3 to 10 [18] epurpose of this paper is to establish a quantitative model soas to find a suitable method from the perspective of cor-porate accounting indicators Ye [19] has put forward arevised model of the corporate life cycle based on sales anddivided the corporate development into four stages Zhanget al [20] used the main business income growth rateretained yield capital expenditure rate corporate age andother indicators to divide the corporate life cycle and analyzeits impact on the overinvestment of the company Cao et al[21] divided the corporate development stages into the initialstage growth stage maturity stage and recession stage withthe method of combination of cash flow components anddiscussed the corporate financial distress in different de-velopment stages respectively

Being the most widely used method the cash flowcombination method divides the development of companiesinto four stages the start-up stage the growth stage themature stage and the recession stage which is the methodused in this paper e division is based on Table 1

3 Model Construction

31 Selection of Indicators Referring to the research of YaoHui and Zhang Hu et al [202122] 14 indicators are selectedfor analysis from four dimensions as shown in Table 2solvency growth ability operation ability and profitabilityas well as corporate life cycle judgment

32 Selection of Evaluation Methods According to thecharacteristics of different indicators different evaluationmethods are selected to turn raw data into a dimensionlessscore

321 Standardized Evaluation e indicators are stan-dardized by formula (8) to map the data to the interval of [0100] to complete the scoring of a single index directlyStandardization evaluation consists of forward standardi-zation evaluation and inverse standardization evaluationshown as follows

g x minus min(X)

max(X) minus min(X)times 100(forward standardization evaluation)

g max(X) minus x

max(X) minus min(X)times 100(inverse standardization evaluation)

(8)

In formula (8) x represents the original indicator data gmeans the indicator score calculated after standardizationand X is the dataset of the indicators of all listed companiesStandardized evaluation retains the relationship of the

original data and makes it dimensionless Forward stan-dardization is suitable for dealing with the indicators of ldquothebigger the betterrdquo while inverse standardization is suitablefor the opposite [23]

x1

x2

xi

xn

y1

y2

yi

yn

Input Layer Hidden Layer Output Layer

Figure 1 e structure of the BP neural network

Computational Intelligence and Neuroscience 3

322 Evaluation with Ranking Method Considering thegrowth rate of operating income and net profit the mea-surement standard should not be ldquothe bigger the betterrdquo Formature companies their financial situation may be healthybut the growth rate of operating income and net profit isrelatively low compared with those companies in a growthperiod erefore the ranking method is adopted in thisresearch By sorting these two indicators the first place in theranking is assigned 100 points the last place in the ranking isassigned 0 points and the middle-level companies areassigned equal and decreasing points so as to partially offsetthe phenomenon that the scores are too low or too con-centrated caused by individual outliers [24] e scoringcriteria are shown in the following formula

grank 100 minus100n

middot (rank minus 1) (9)

In formula (9) grank is the score after the rankingmethod n is the total number of samples and rank is theindicator ranking

323 Evaluation by Moderate Analysis Method Becausecompanies need to ensure the appropriate ratio of assets andliabilities to cope with the debt repayment crisis and a certainleverage ratio to help themselves develop the asset-liabilityratio is generally considered to be more appropriate in theinterval of (50 to 60) Outside the interval the more itdeviates from the interval the lower the score e scoringcriteria are shown in the following formula

glev

100 50le levle 60

100 minus 5 middot (lev minus 60) times 100 60le levle 80

100 minus 5 middot (50 minus lev) times 100 30le levle 50

0 levlt 30 or levgt 80

⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

(10)

In formula (10) glev is the score of asset-liability ratio andlev is the asset-liability ratio of the company

324 Life Cycle Assessment Listed companies have gener-ally passed the start-up period Because companies oftenadopt the expansion strategy in the growth period theirmarket share normally increases rapidly and their organi-zational structure would be constantly improved [25] wherethe fundamental improvement expectation would be thehighest However after the listed companies enter themature stage their market shares tend to be saturated theirorganizational structure gradually matures and their busi-ness activities remain in a stable stage [26] At this time thefundamentals of companies have entered a stable stage andthe expectation of their improvement is relatively weakWhen companies run into a recession their market sharewill be greatly reduced their organizational structure will beredundant and their fundamentals will gradually deterio-rate It can also be found that the allocation of fund com-panies is more inclined to companies in a growth stageAccording to the cash flow portfolio method the allocationratios of Huaxia fund in the growth stage mature stage and

Table 1 Corporate life cycle judgment by the combination of cash flow components

Start-up Growth Mature RecessionNet cash flow from operating activities minus + + minus + + minus

Net cash flow from investment activities minus minus minus minus + + +Net cash flow from financing activities + + minus minus + minus minus

Table 2 Selection of indicators

Level 1 indicator Level 2 indicator Evaluation method Symbol

Solvency

Asset-liability ratio Moderate analysis method evaluation X1Current ratio Forward standardization evaluation X2Quick ratio Forward standardization evaluation X3

Cash flow ratio Forward standardization evaluation X4

Growth ability Growth rate of operating income Ranking method evaluation X5Growth rate of net profit Ranking method evaluation X6

Operation abilityInventory turnover rate Forward standardization evaluation X7

Turnover rate of accounts receivable Forward standardization evaluation X8Turnover rate of total assets Forward standardization evaluation X9

Profitability

Return on net assets Forward standardization evaluation X10Net profit rate of sales Forward standardization evaluation X11

Expense rate during sales period Inverse standardization evaluation X12Operating cost rate Inverse standardization evaluation X13

Periodic judgment Corporate cycle judgment Life cycle assessment X14


recession stage are 5441 28 and 441 respectivelywhich is similar to those of other fund companies [27]erefore the expected fundamental improvement broughtby the corporate life cycle is scored according to formula(11)

gT

100 Growth Stage

80 Mature Stage

60 Recession Stage

⎧⎪⎪⎨

⎪⎪⎩(11)

In formula (11) gT is the score given to the companyaccording to the corporate life cycle theory

33 Index Weight Calculation In this paper the annualreport data of listed companies in 2020 are exported throughOriental Fortune Choice financial terminal and someoutliers are eliminated where 3096 samples are eventuallyobtained

en through the correlation coefficient formula (12)the correlation degree between the indicators is calculated Ifthe indicator correlation degree |ρ|gt 08 then it is elimi-nated Table 3 shows the calculation result of the correlationcoefficient of the indicators

ρij

11138681113868111386811138681113868

11138681113868111386811138681113868 COV Xi Xj1113872 1113873D Xi( 1113857D Xj1113872 1113873

1113969

111386811138681113868111386811138681113868111386811138681113868111386811138681113868

111386811138681113868111386811138681113868111386811138681113868111386811138681113868

i 1 13 j 1 13 (12)

As can be seen from the above table the correlationbetween index X2 and index X9 is relatively high the rel-atively rarely used total asset turnover rate (X9) is elimi-nated and the current ratio (X2) is kept

In the following part the grey relational analysis will beadopted to calculate the initial weight and the expectedoutput of the BP (back propagation) neural network

rough the whole process of grey relational analysisshown in formulas (1) and (2) the results of the weight ofeach indicator are shown in Table 4 Using the weight andformula (3) the expected output can be obtained

34 Construction of BP Neural Network

341 Input Layer Neuron According to Table 3 13 sec-ondary indicators such as asset-liability ratio the growth rateof operating income inventory turnover rate and return onnet assets are selected as input layer neurons based on thefour dimensions of solvency growth ability operationability and profitability

342 Hidden Layer Neuron e value range of the hiddenlayer node is calculated according to the empirical formula

h p + q

1113968+ a a isin [1 10] (13)

In formula (13) h represents the number of hidden layernodes p is the number of input layer neurons and q is thenumber of output layer neurons In this research

p 13 and q 1e number of neurons of the hidden layercan be obtained in the range of [313]

343 Output Layer Neuron e number of output layerneurons is 1 which outputs the fundamental scores of listedcompanies with a value range of [0 100] e higher theoutput score the better the fundamentals of the company

Determination of initial values of weights and thresholdsIn order to improve the learning speed of neural networksthe weights are set within the range of [0 1] according to theusual practice and random numbers in the range of [0 1] areused as initial values of weights and thresholds

35 Network Model Training In the abovementioned 3096samples 2168 samples are randomly selected as trainingsamples which are recorded as TRi (i 1 2 2168) andthe predicted value is obtained

Since the number of neurons of the hidden layer is in acertain interval comparative experiments are conducted inorder to optimize the parameter to obtain the smallest errore results are shown in Figure 2

From the figure the conclusion is obvious that thesmallest error can be gained when the number of neurons ofthe hidden layer is 10 us the BP neural network designedin this paper sets the number of neurons in the input layerhidden layer and output layer as 13 10 and 1 respectivelye structure of the BP neural network is shown in Figure 3After 1000 times of training the error is reduced to101times 10minus 10 According to the error trend chart in Figure 4 aquantitative evaluation model of the fundamentals ofA-share listed companies in all industries

A model based on BP neural network is established andpart of the training results of the BP neural network is shownin Table 5

36 Simulation Result Analysis It is necessary to test thenetwork model in order to verify whether the BP neuralnetwork model is of the generalization ability 604 samplesare randomly selected from the remaining 1208 samples andrecorded as TEj (j 1 2 604) for testing and the testresults are shown in Table 6 e error distribution betweentraining results and test results is shown in Figure 3

According to Figure 5 the absolute errors of the trainingset and test set are mostly distributed in the interval of[-13times10minus 5 96times10minus 6] where the errors are extremely smallerefore the BP neural network model designed in thispaper is of good generalization ability and can be applied tothe fundamental evaluation and analysis of A-share listedcompanies in all industries

37 Robustness Test In order to test the robustness of themodel this paper changes the evaluation interval of theasset-liability ratio from formula (10) to formula (14) asfollows


Table 3 Calculation result of correlation coefficient of indicators

X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14X1 1000 0461 0138 0111 0058 0033 0070 0004 0444 0002 0206 0180 0021 0027X2 0461 1000 0176 0080 0043 0031 0018 0064 0947 0041 0277 0354 0062 0012X3 0138 0176 1000 0038 0222 0244 0094 0022 0160 0577 0389 0157 0307 0075X4 0111 0080 0038 1000 0238 0145 0361 0312 0064 0195 0218 0070 0279 0049X5 0058 0043 0222 0238 1000 0479 0061 0097 0047 0178 0086 0083 0241 0020X6 0033 0031 0244 0145 0479 1000 0027 0054 0035 0247 0105 0084 0127 0012X7 0070 0018 0094 0361 0061 0027 1000 0212 0075 0012 0209 0011 0101 0044X8 0004 0064 0022 0312 0097 0054 0212 1000 0050 0075 0086 0146 0055 0053X9 0444 0947 0160 0064 0047 0035 0075 0050 1000 0027 0263 0358 0059 0005X10 0002 0041 0577 0195 0178 0247 0012 0075 0027 1000 0192 0092 0286 0082X11 0206 0277 0389 0218 0086 0105 0209 0086 0263 0192 1000 0238 0320 0016X12 0180 0354 0157 0070 0083 0084 0011 0146 0358 0092 0238 1000 0016 0138X13 0021 0062 0307 0279 0241 0127 0101 0055 0059 0286 0320 0016 1000 0050X14 0027 0012 0075 0049 0020 0012 0044 0053 0005 0082 0016 0138 0050 1000

Table 4 Calculation results of the weight of each indicator

Level 1 indicator Level 2 indicator Symbol Initial weight

Solvency

Asset-liability ratio X1 679241Current ratio X2 708412Quick ratio X3 828635

Cash flow ratio X4 745493

Growth ability Growth rate of operating income X5 758143Growth rate of net profit X6 761459

Operation ability Inventory turnover rate X7 704270Turnover rate of accounts receivable X8 698549

Profitability

Return on net assets X10 822696Net profit rate of sales X11 775891

Expense rate during sales period X12 809567Operating cost rate X13 841013

Periodic judgment Corporate cycle judgment X14 866630

7

8times 10ndash8

6

Erro

r

5

4

3

2

1

02 4 6 8

Number of Neruons of Hidden Layer10 12 14

Figure 2 Contrast of error under different parameter settings


Figure 3 Structure diagram of BP neural network

100

10ndash5

10ndash100 200 400 600

1000 Epochs

Best Validation Performance is 18288endash10 at epoch 1000

800 1000

Mea

n Sq

uare

d Er

ror (

mse

)

TrainValidation

TestBest

Figure 4 Training error diagram of BP neural network

Table 5 Part of training results of BP neural network

Sample Expected output Actual output Relative error ()TR1 5373083 5373004 000147TR2 5000218 5000140 000157TR3 4679068 4679015 000113TR4 5167963 5167902 000119TR5 3312191 3312213 000064TR6 4996475 4996402 000145TR7 5376107 5376060 000088TR8 4272477 4272418 000138TR9 4414657 4414600 000130TR10 5130777 5130702 000146TR11 4705807 4705735 000154TR12 3852720 3852665 000144TR13 4956071 4955982 000179TR14 5586725 5586679 000083


Table 5 Continued

Sample Expected output Actual output Relative error ()TR15 4876789 4876711 000159TR16 3599980 3600023 000119TR17 6100887 6100867 000033TR18 5459437 5459362 000137TR19 4813704 4813696 000018TR20 5771973 5771944 000052

Table 6 Part of test results of BP neural network

Sample Expected output Actual output Relative error ()TE1 4936534 4936480 000109TE2 5805768 5805710 000100TE3 5253170 5253105 000125TE4 4480344 4480370 000058TE5 5188164 5188119 000087TE6 4713900 4713851 000105TE7 5508644 5508563 000147TE8 6244150 6244073 000122TE9 5611946 5611873 000129TE10 4480488 4480430 000129TE11 5430785 5430721 000118TE12 5628334 5628285 000087TE13 5543343 5543308 000064TE14 5317385 5317326 000111TE15 4763893 4763834 000124TE16 4567614 4567584 000066TE17 6113550 6113490 000098TE18 5573245 5573178 000121TE19 5139826 5139782 000084TE20 5522763 5522727 000064

1800

1600

1400

1200

Inst

ance

s

1000

800

600

400

200

0

Error = Targets ndash Outputs

Error Histogram with 20 Bins

ndash00

0012

ndash00

0011

ndash94

endash05

ndash82

endash05

ndash71

endash05

ndash59

endash05

ndash48

endash05

ndash36

endash05

ndash25

endash05

ndash13

endash05

ndash19

endash06

963

endash06

211

endash05

327

endash05

442

endash05

557

endash05

672

endash05

787

endash05

902

endash05

000

0102

TrainingValidation

TestZero error

Figure 5 Error distribution diagram of BP neural network


glev

100 45le levle 65




⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

(14)

e new score is substituted into the model of the BPneural network and the training error diagram shown inFigure 6 and the error distribution diagram shown in Fig-ure 7 are obtained It can be found that after 1000 iterationsthe error is reduced to 20664times10minus 8 and the absolute errorsare mostly distributed in [-46times10minus 436times10minus 4] ereforethe BP neural network model established in this paper isrobust

4 Result Analysis

e author calculates the initial weight of each indicator offundamental evaluation of listed companies through greyrelational analysis and optimizes the weight through BP(back propagation) neural network According to the cal-culation results of the training set data the model can beestablished According to the calculation results of the testset data it can be concluded that the model has general-ization ability With this model the fundamentals of A-sharelisted companies in China can be summarized to put forwardsome suggestions for fundamental evaluation

41 Weight Analysis e weights optimized by the BPneural network are shown in Table 7 Among them

100

10ndash5

0 200 400 6001000 Epochs


800 1000

Mea

n Sq

uare

d Er

ror (

mse

)

TrainValidation

TestBest

Figure 6 Training error diagram of robustness test

3000

2500

2000

1500

Inst

ance

s

1000

500

0



ndash00

0335

ndash00

0293

ndash00

0252

ndash00

0211

ndash00

017

ndash00

0129

ndash00

0087

ndash00

0046

ndash49

endash05

000

0363

000

0775

000

1187

000

1599

000

2011

000

2423

000

2836

000

3248

000

366

000

4072

000

4484

TrainingValidation

TestZero error

Figure 7 Error distribution diagram of robustness test


solvency growth ability operation ability profitability andcycle judgment account for 2961919 15195871402812 3248885 and 866757 respectively Profit-ability accounts for the highest proportion followed bysolvency and there is little difference between growth abilityand operational ability It can be explained that profitabilityreflects the income quality of listed companies and is themost important part in the fundamental evaluation of listedcompanies Solvency reflects the risk control of listedcompanies and is therefore of great significance in funda-mental evaluation

42 Scoring Results Analysis e top 10 listed companies inthe scoring order are selected on the basis of the scoringresults where it can be found in Table 8 that listed com-panies with higher scores have higher scores in return on netassets quick ratio and cash flow ratioese three indicatorsreflect the efficiency of shareholdersrsquo funds the solvency andliquidity of assets Companies with good fundamentals havecertain advantages in these aspects

According to the descriptive statistics of each indicatorshown in Table 9 Chinarsquos listed companies have higher

average scores in operating cost ratio and quick ratio butlower scores in inventory turnover rate and accounts re-ceivable turnover rate which reflects that the listed com-panies have relatively strong profitability and solvency whiletheir operational ability is relatively weak When making astock selection the important indicator of operational abilitycan be put into consideration to see if the company is leadingin the industry

5 Conclusion

is paper explores the application of BP (back propagation)neural networks in fundamental evaluation Firstly raw dataof Chinarsquos listed companies are imported and standardizedby different scoring methods which ensures rationality andcomprehensiveness of the original score Secondly greyrelational analysis is conducted in order to obtain the initialweight and expected output en a 3-layer BP neuralnetwork is constructed after parameter optimization eresult is tested and validated Finally the result was analyzedfrom 2 aspects weight analysis and scoring result analysiswhere relative suggestions are provided for investors

Table 7 Optimization results of the weight of each indicator

Level 1 indicator Level 2 indicator Symbol Initial weight Optimized weight

Solvency

Asset-liability ratio X1 679241 679240Current ratio X2 708412 708433Quick ratio X3 828635 828747

Cash flow ratio X4 745493 745499

Growth ability Growth rate of operating income X5 758143 758136Growth rate of net profit X6 761459 761451

Operation ability Inventory turnover rate X7 704270 704273Turnover rate of accounts receivable X8 698549 698539

Profitability

Return on net assets X10 822696 822687Net profit rate of sales X11 775891 775800

Expense rate during sales period X12 809567 809591Operating cost rate X13 841013 840807

Periodic judgment Corporate cycle judgment X14 866630 866757

Table 8 Scores of each indicator of the 10 listed companies with the highest scores

Sample X1 X2 X3 X4 X5 X6 X7 X8 X10 X11 X12 X13 X14 ScoreS1 100000 34503 100000 100000 15700 34400 100000 100000 100000 81108 67500 100000 60000 76410S2 42100 100000 100000 100000 88300 2800 100000 95177 100000 81296 0000 100000 60000 74428S3 72250 70000 92482 100000 28400 13100 85031 100000 89890 50455 81165 76081 100000 74056S4 24700 100000 100000 100000 77500 83500 0000 100000 100000 100000 100000 0000 60000 73005S5 67950 55135 81812 100000 62700 75000 58710 100000 100000 83214 0000 69767 60000 70090S6 0000 19434 94273 100000 74900 91800 38377 100000 88921 61046 68471 92643 60000 69586S7 100000 7715 90124 100000 31000 60100 84081 20917 73218 66218 59180 98857 100000 69504S8 100000 46341 100000 97727 19800 25200 75463 68541 100000 65569 100000 43964 60000 69478S9 80250 18387 91789 100000 10100 47100 100000 100000 93759 66331 32890 100000 60000 69326S10 77350 13430 88656 51429 93200 88600 100000 9622 87519 59887 69430 80162 60000 68322

Table 9 Descriptive statistics of scores of all indicators

X1 X2 X3 X4 X5 X6 X7 X8 X10 X11 X12 X13 X14 ScoreMax 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 60000 18372Min 100000 100000 100000 100000 100000 100000 100000 100000 100000 100000 100000 100000 100000 76410Average 43854 22074 68523 30535 49972 49972 21364 20933 65064 47595 61820 74441 64289 48940


Based on the experiment a BP neural network used toevaluate the fundamentals of companies of all industries wasconstructed is experiment proposed a new system ofevaluation methods for all indicators that uses the importantfinancial indexes of three statements for the first time Inaddition with the use of the corporate life cycle evaluationindicator the companies with fundamental improvementexpectations can be identified which is an innovativeattempt

Although a BP neural network model has been con-structed successfully there are still some points that need tobe improved

(1) e scores shown in Table 9may not be rational sincemost of the samples seem to be in a recession stagewhich disobeys common sense erefore moreprecise algorithms should be applied to score thestage of companies

(2) ere is still room for continued optimization in theinternal structure of grey relational analysis and BPneural network which can lead to more preciseresults

(3) Other advanced artificial neural networks can beused to compare with BP neural network which isthe direction of future efforts

Data Availability

e data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

e authors declare that there are no conflicts of interest

References

[1] B Graham and D L Dodd Security Analysis [M] WhittleseyHouse New York 1934

[2] B Julio and Y Yook ldquoPolitical uncertainty and corporateinvestment cycles [J]rdquo 0e Journal of Finance no 67pp 45ndash83 2012

[3] S Chen Z Sun S Tang and D Wu ldquoGovernment inter-vention and investment efficiency Evidence from China [J]rdquoJournal of Corporate Finance vol 17 no 2 pp 259ndash271 2011

[4] E F Fama and K R French ldquoA five-factor asset pricingmodel[J]rdquo Journal of Financial Economics vol 116 no 1 pp 1ndash222015

[5] J Patel S Shah Pakkar and K Kotecha ldquoPredicting stockand stock price index movement using Trend DeterministicData Preparation and machine learning techniques [J]rdquo Ex-pert Systems with Applications vol 42 no 1 pp 259ndash2682015

[6] J Sun H Li H Fujita B Fu and W Ai ldquoClass-imbalanceddynamic financial distress prediction based on Adaboost-SVM ensemble combined with SMOTE and time weightingrdquoInformation Fusion vol 54 no 54 pp 128ndash144 2020

[7] X Jin Q Liu and H Long ldquoImpact of cost-benefit analysis onfinancial benefit evaluation of investment projects under backpropagation neural network [J]rdquo Journal of Computationaland Applied Mathematics vol 384 no 2 p 113172 2021

[8] T Fischer and C Krauss ldquoDeep learning with long short-termmemory networks for financial market predictions[J]rdquo Eu-ropean Journal of Operational Research vol 270 no 2pp 654ndash669 2018

[9] A Amo Baffour J Feng and E K Taylor ldquoA hybrid artificialneural network-GJR modeling approach to forecasting cur-rency exchange rate volatilityrdquo Neurocomputing vol 365no 365 pp 285ndash301 2019

[10] Q Wang and Li Cheng ldquoGrey relational analysis of tech-nological innovation and economic performance of enter-prises [J]rdquo Forum on Science and Technology in China no 12pp 64ndash68 2009

[11] X Chen S Jian and X Zou ldquoComparative study on growthevaluation methods of small and medium-sized listed com-panies [J]rdquo Science ResearchManagement no 01 pp 145ndash1512006

[12] L Zhang Z He and Y Wu ldquoEvaluation of strategic emergingindustries based on grey relational analysis ndash taking bio-medicine as an example [J]rdquo Journal of Quantitative Eco-nomics vol 27 no 03 pp 79ndash84 2010

[13] C Delcea E Scarlat and L A Cotfas ldquoCompaniesrsquo qualitycharacteristics vs their performancerdquo Grey Systems 0eoryand Application vol 3 no 2 pp 129ndash141 2013

[14] S Ding C Su and J Yu ldquoAn optimizing BP neural networkalgorithm based on genetic algorithm [J]rdquo Artificial Intelli-gence Review vol 36 pp 153ndash162 2011

[15] X Zhang L Shi X Yan and L Xinfa ldquoConstruction anddemonstration of early warning evaluation index system forkeep from returning to poverty in the perspective of ruralrevitalization [J]rdquo Statistics amp Decisions vol 37 no 13pp 58ndash62 2021

[16] Z Zhang M Xie and D Lin ldquoPerformance evaluation systemof Chinese manufacturing companies based on BP neuralnetwork ndash taking listed white household appliancesmanufacturing companies as an example [J]rdquo Science andTechnology Management Research vol 40 no 15 pp 217ndash223 2020

[17] M Haire ldquoBiologic models and empirical histories in thegrowth of organizations [A]rdquo inModern Organization0eoryA Symposium [M] M Haire Ed John Wiley amp Sons NewYork 1959

[18] Z Zhu Research on Financial Crisis Early Warning from thePerspective of Corporate Lifecycle [D] Southeast University2016

[19] Li Ye ldquoRevised model and thinking of corporate life cycle [J]rdquoSouthern China Journal of Economics no 02 pp 47ndash50 2000

[20] Hu Zhang H Shen and Y Liu ldquoResearch on multi-factorquantitative stock selection based on self-attention neuralnetwork [J]rdquo Journal of Applied Sport Management vol 39no 03 pp 556ndash570 2020

[21] Yu Cao and X-hong Chen ldquoAn agent-based simulationmodel of enterprises financial distress for the enterprise ofdifferent life cycle stage [J]rdquo Simulation Modelling Practiceand 0eory vol 20 no 1 2011

[22] H Yao and T Wu ldquoAn empirical study on value inves-tingstrategy considering fundamentals and valuation indexesndash empirical data from Chinarsquos Shanghai and ShenzhenA-share markets from 2000 to 2013 [J]rdquo Review of InvestmentStudies vol 33 no 11 pp 123ndash138 2014

[23] Y Yang and D Zhang ldquoCorrelation analysis between com-prehensive score and stock price based on improved Wallmotion score index [J]rdquo China Township Enterprises Ac-counting no 02 pp 45ndash48 2021


[24] Q Xue and J Zhao ldquoWall motion score index and operationevaluation of enterprise productivity promotion center [J]rdquoJournal of Shanxi University(Philosophy and Social ScienceEdition) vol 37 no 03 pp 140ndash144 2014

[25] I M Jawahar and G L McLaughlin ldquoToward a descriptiveStakeholder theory an organizational life cycle approachrdquoAcademy of Management Review vol 26 no 3 pp 397ndash4142001

[26] X Cheng Q Wu M Liu and Yu Cheng ldquoCash distributionmanagement incentive and capital allocation efficiency ofenterprise groups [J]rdquo Journal of Financial Research no 02pp 91ndash108 2020

[27] S Chang and S Liu ldquoResearch on the division and mea-surement of Chinese corporate life cycle stages [J]rdquo Com-mercial Research no 01 pp 1ndash10 2011


(1) When the indicator is under a dimensionless processa variety of evaluation methods are innovativelycombined so that the indicator score is more rea-sonable and comprehensive

(2) With the corporate life cycle evaluation indicatoradded the companies with fundamental improve-ment expectations can be identified Since the cashflow statement is included in the evaluation systemall the important data from the three statements areincluded in the evaluation model for the first time

(3) Grey relational analysis and BP neural networktraining simulation are used to optimize the weightof indicators so as to obtain the fundamental scoringsystem of all-industry listed companies

2 Theoretical Basis of Model Construction

21 Grey Relational Analysis Grey relational analysis is aquantitative description and comparison of the developmentstate of objects so as to analyze and determine the influencedegree between the contribution measure of factors on themain behavior e Grey correlation degree is a measure ofthe correlation between two systems If the relative changesituation of the two factors is consistent in the developmentprocess the grey correlation degree of the two factors islarge otherwise it is small Grey relational analysis is usuallyused to address the comprehensive evaluation problem [10]Chen et al [11] evaluated the growth of small and medium-sized listed companies through grey relational analysis andconcluded that the obtained information can be fully usedwith this method Zhang et al [12] established a quantitativeevaluation model of strategic emerging industries throughgrey relational analysis which made the evaluation systemmore reasonable Delcea et al [13] put forward suggestionson corporate development and extreme situation responseby using the quantitative results obtained with grey rela-tional analysis erefore it is concluded that grey relationalanalysis is suitable for quantitative scoring of the funda-mentals of listed companies and it is reasonable to take it asthe expected output of the BP neural network e progressof grey relational analysis is shown as follows

Step1 e data after standardization is used to gain the greycorrelation degree coefficients ζ i(k) of the reference seriesand comparison series with formula (1)

ζ i(k)

mini

mink

g0(k) minus gi(k)1113868111386811138681113868

1113868111386811138681113868 + ρmaxi

maxk

g0(k) minus gi(k)1113868111386811138681113868

1113868111386811138681113868

g0(k) minus gi(k)1113868111386811138681113868

1113868111386811138681113868 + ρmaxi

maxk

g0(k) minus gi(k)1113868111386811138681113868

1113868111386811138681113868

(1)

In formula (1) gi(k) is the score of indicators of listedcompanies and ρ is the resolution coefficient e usualpractice is adopted in this paper where ρ 05

Step2 When the calculated grey relational degree coeffi-cient is obtained the weight calculation formula is used toobtain the weight of the indicators

ωi 1113936

nk1 gi(k)

1113936nk1 1113936

mi1 gi(k)

(2)

In formula (2) n is the number of sample companies andm is the number of selected indicators

Step3 According to the weights gained the scores of listedcompanies are calculated by formula (3) as the expectedoutput of the neural network model

G(k) 1113944m

i1ωi middot gi(k) (3)

In the formula G(k) means the score of the company

22 BP Neural Network e BP (back propagation) neuralnetwork proposed by scientists led by Rumelhart andMcClelland in 1986 is a gradient descent method being ableto optimize the weight and threshold through error[14]Zhang Xuemin et al [15] have combined an analytic hier-archy process and BP neural network to build an earlywarning evaluation system in order to keep poverty-strickenpeople from returning to poverty in specific areas ZhangZhengang et al [16] used information entropy theory todetermine the index weight and then used BP neural net-work training and simulation to construct the performanceevaluation system of listed white household appliancescompanies in China anks to its advantages in weightoptimization the BP neural network can be effectively ap-plied to the weight selection of the quantitative evaluationsystem constructed in this paper e structure of the BPneural network is shown in Figure 1

Δwij minus ηzError

zwij

i 1 2 m j 1 2 n (6)

Δvjk minus ηzError

zvjk

k 1 2 n j 1 2 n (7)

A routine BP neural network has 3 layers input layerhidden layer and output layer In Figure 1 x is the inputdata and y is the output data Assume the weight of neuronsin the hidden layer is w and the weight of neurons in theoutput layer is v Assume d is the expected output then theerror of the model Error can be defined by formula (4) asfollows

Error 12

1113944

n

k1dk minus yk( 1113857

2 (4)

In formula (4) yk can be denoted by formula (5) where fis the activation function

yk f 1113944n

j1vjkf 1113944

m

i1wijxi

⎛⎝ ⎞⎠⎛⎝ ⎞⎠ (5)

If Error does not reach the target the neural network willadjust w and v e adjustment amount Δwij and Δvjk are










g x minus min(X)


g max(X) minus x


(8)



x1

x2

xi

xn

y1

y2

yi

yn





grank 100 minus100n




glev

100 50le levle 60




⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

(10)








Solvency






Profitability






gT

100 Growth Stage

80 Mature Stage

60 Recession Stage

⎧⎪⎪⎨

⎪⎪⎩(11)




ρij

11138681113868111386811138681113868

11138681113868111386811138681113868 COV Xi Xj1113872 1113873D Xi( 1113857D Xj1113872 1113873

1113969

111386811138681113868111386811138681113868111386811138681113868111386811138681113868

111386811138681113868111386811138681113868111386811138681113868111386811138681113868

i 1 13 j 1 13 (12)







h p + q

1113968+ a a isin [1 10] (13)

















Solvency





Profitability




7

8times 10ndash8

6

Erro

r

5

4

3

2

1

02 4 6 8





100

10ndash5

10ndash100 200 400 600

1000 Epochs


800 1000

Mea

n Sq

uare

d Er

ror (

mse

)

TrainValidation

TestBest





Table 5 Continued




1800

1600

1400

1200

Inst

ance

s

1000

800

600

400

200

0



ndash00

0012

ndash00

0011

ndash94

endash05

ndash82

endash05

ndash71

endash05

ndash59

endash05

ndash48

endash05

ndash36

endash05

ndash25

endash05

ndash13

endash05

ndash19

endash06

963

endash06

211

endash05

327

endash05

442

endash05

557

endash05

672

endash05

787

endash05

902

endash05

000

0102

TrainingValidation

TestZero error



glev

100 45le levle 65




⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

(14)


4 Result Analysis



100

10ndash5

0 200 400 6001000 Epochs


800 1000

Mea

n Sq

uare

d Er

ror (

mse

)

TrainValidation

TestBest


3000

2500

2000

1500

Inst

ance

s

1000

500

0



ndash00

0335

ndash00

0293

ndash00

0252

ndash00

0211

ndash00

017

ndash00

0129

ndash00

0087

ndash00

0046

ndash49

endash05

000

0363

000

0775

000

1187

000

1599

000

2011

000

2423

000

2836

000

3248

000

366

000

4072

000

4484

TrainingValidation

TestZero error







5 Conclusion




Solvency





Profitability














Data Availability




References






































g x minus min(X)


g max(X) minus x


(8)



x1

x2

xi

xn

y1

y2

yi

yn





grank 100 minus100n




glev

100 50le levle 60




⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

(10)








Solvency






Profitability






gT

100 Growth Stage

80 Mature Stage

60 Recession Stage

⎧⎪⎪⎨

⎪⎪⎩(11)




ρij

11138681113868111386811138681113868

11138681113868111386811138681113868 COV Xi Xj1113872 1113873D Xi( 1113857D Xj1113872 1113873

1113969

111386811138681113868111386811138681113868111386811138681113868111386811138681113868

111386811138681113868111386811138681113868111386811138681113868111386811138681113868

i 1 13 j 1 13 (12)







h p + q

1113968+ a a isin [1 10] (13)

















Solvency





Profitability




7

8times 10ndash8

6

Erro

r

5

4

3

2

1

02 4 6 8





100

10ndash5

10ndash100 200 400 600

1000 Epochs


800 1000

Mea

n Sq

uare

d Er

ror (

mse

)

TrainValidation

TestBest





Table 5 Continued




1800

1600

1400

1200

Inst

ance

s

1000

800

600

400

200

0



ndash00

0012

ndash00

0011

ndash94

endash05

ndash82

endash05

ndash71

endash05

ndash59

endash05

ndash48

endash05

ndash36

endash05

ndash25

endash05

ndash13

endash05

ndash19

endash06

963

endash06

211

endash05

327

endash05

442

endash05

557

endash05

672

endash05

787

endash05

902

endash05

000

0102

TrainingValidation

TestZero error



glev

100 45le levle 65




⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

(14)


4 Result Analysis



100

10ndash5

0 200 400 6001000 Epochs


800 1000

Mea

n Sq

uare

d Er

ror (

mse

)

TrainValidation

TestBest


3000

2500

2000

1500

Inst

ance

s

1000

500

0



ndash00

0335

ndash00

0293

ndash00

0252

ndash00

0211

ndash00

017

ndash00

0129

ndash00

0087

ndash00

0046

ndash49

endash05

000

0363

000

0775

000

1187

000

1599

000

2011

000

2423

000

2836

000

3248

000

366

000

4072

000

4484

TrainingValidation

TestZero error







5 Conclusion




Solvency





Profitability














Data Availability




References































grank 100 minus100n




glev

100 50le levle 60




⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

(10)








Solvency






Profitability






gT

100 Growth Stage

80 Mature Stage

60 Recession Stage

⎧⎪⎪⎨

⎪⎪⎩(11)




ρij

11138681113868111386811138681113868

11138681113868111386811138681113868 COV Xi Xj1113872 1113873D Xi( 1113857D Xj1113872 1113873

1113969

111386811138681113868111386811138681113868111386811138681113868111386811138681113868

111386811138681113868111386811138681113868111386811138681113868111386811138681113868

i 1 13 j 1 13 (12)







h p + q

1113968+ a a isin [1 10] (13)

















Solvency





Profitability




7

8times 10ndash8

6

Erro

r

5

4

3

2

1

02 4 6 8





100

10ndash5

10ndash100 200 400 600

1000 Epochs


800 1000

Mea

n Sq

uare

d Er

ror (

mse

)

TrainValidation

TestBest





Table 5 Continued




1800

1600

1400

1200

Inst

ance

s

1000

800

600

400

200

0



ndash00

0012

ndash00

0011

ndash94

endash05

ndash82

endash05

ndash71

endash05

ndash59

endash05

ndash48

endash05

ndash36

endash05

ndash25

endash05

ndash13

endash05

ndash19

endash06

963

endash06

211

endash05

327

endash05

442

endash05

557

endash05

672

endash05

787

endash05

902

endash05

000

0102

TrainingValidation

TestZero error



glev

100 45le levle 65




⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

(14)


4 Result Analysis



100

10ndash5

0 200 400 6001000 Epochs


800 1000

Mea

n Sq

uare

d Er

ror (

mse

)

TrainValidation

TestBest


3000

2500

2000

1500

Inst

ance

s

1000

500

0



ndash00

0335

ndash00

0293

ndash00

0252

ndash00

0211

ndash00

017

ndash00

0129

ndash00

0087

ndash00

0046

ndash49

endash05

000

0363

000

0775

000

1187

000

1599

000

2011

000

2423

000

2836

000

3248

000

366

000

4072

000

4484

TrainingValidation

TestZero error







5 Conclusion




Solvency





Profitability














Data Availability




References































gT

100 Growth Stage

80 Mature Stage

60 Recession Stage

⎧⎪⎪⎨

⎪⎪⎩(11)




ρij

11138681113868111386811138681113868

11138681113868111386811138681113868 COV Xi Xj1113872 1113873D Xi( 1113857D Xj1113872 1113873

1113969

111386811138681113868111386811138681113868111386811138681113868111386811138681113868

111386811138681113868111386811138681113868111386811138681113868111386811138681113868

i 1 13 j 1 13 (12)







h p + q

1113968+ a a isin [1 10] (13)

















Solvency





Profitability




7

8times 10ndash8

6

Erro

r

5

4

3

2

1

02 4 6 8





100

10ndash5

10ndash100 200 400 600

1000 Epochs


800 1000

Mea

n Sq

uare

d Er

ror (

mse

)

TrainValidation

TestBest





Table 5 Continued




1800

1600

1400

1200

Inst

ance

s

1000

800

600

400

200

0



ndash00

0012

ndash00

0011

ndash94

endash05

ndash82

endash05

ndash71

endash05

ndash59

endash05

ndash48

endash05

ndash36

endash05

ndash25

endash05

ndash13

endash05

ndash19

endash06

963

endash06

211

endash05

327

endash05

442

endash05

557

endash05

672

endash05

787

endash05

902

endash05

000

0102

TrainingValidation

TestZero error



glev

100 45le levle 65




⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

(14)


4 Result Analysis



100

10ndash5

0 200 400 6001000 Epochs


800 1000

Mea

n Sq

uare

d Er

ror (

mse

)

TrainValidation

TestBest


3000

2500

2000

1500

Inst

ance

s

1000

500

0



ndash00

0335

ndash00

0293

ndash00

0252

ndash00

0211

ndash00

017

ndash00

0129

ndash00

0087

ndash00

0046

ndash49

endash05

000

0363

000

0775

000

1187

000

1599

000

2011

000

2423

000

2836

000

3248

000

366

000

4072

000

4484

TrainingValidation

TestZero error







5 Conclusion




Solvency





Profitability














Data Availability




References


































Solvency





Profitability




7

8times 10ndash8

6

Erro

r

5

4

3

2

1

02 4 6 8





100

10ndash5

10ndash100 200 400 600

1000 Epochs


800 1000

Mea

n Sq

uare

d Er

ror (

mse

)

TrainValidation

TestBest





Table 5 Continued




1800

1600

1400

1200

Inst

ance

s

1000

800

600

400

200

0



ndash00

0012

ndash00

0011

ndash94

endash05

ndash82

endash05

ndash71

endash05

ndash59

endash05

ndash48

endash05

ndash36

endash05

ndash25

endash05

ndash13

endash05

ndash19

endash06

963

endash06

211

endash05

327

endash05

442

endash05

557

endash05

672

endash05

787

endash05

902

endash05

000

0102

TrainingValidation

TestZero error



glev

100 45le levle 65




⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

(14)


4 Result Analysis



100

10ndash5

0 200 400 6001000 Epochs


800 1000

Mea

n Sq

uare

d Er

ror (

mse

)

TrainValidation

TestBest


3000

2500

2000

1500

Inst

ance

s

1000

500

0



ndash00

0335

ndash00

0293

ndash00

0252

ndash00

0211

ndash00

017

ndash00

0129

ndash00

0087

ndash00

0046

ndash49

endash05

000

0363

000

0775

000

1187

000

1599

000

2011

000

2423

000

2836

000

3248

000

366

000

4072

000

4484

TrainingValidation

TestZero error







5 Conclusion




Solvency





Profitability














Data Availability




References































100

10ndash5

10ndash100 200 400 600

1000 Epochs


800 1000

Mea

n Sq

uare

d Er

ror (

mse

)

TrainValidation

TestBest





Table 5 Continued




1800

1600

1400

1200

Inst

ance

s

1000

800

600

400

200

0



ndash00

0012

ndash00

0011

ndash94

endash05

ndash82

endash05

ndash71

endash05

ndash59

endash05

ndash48

endash05

ndash36

endash05

ndash25

endash05

ndash13

endash05

ndash19

endash06

963

endash06

211

endash05

327

endash05

442

endash05

557

endash05

672

endash05

787

endash05

902

endash05

000

0102

TrainingValidation

TestZero error



glev

100 45le levle 65




⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

(14)


4 Result Analysis



100

10ndash5

0 200 400 6001000 Epochs


800 1000

Mea

n Sq

uare

d Er

ror (

mse

)

TrainValidation

TestBest


3000

2500

2000

1500

Inst

ance

s

1000

500

0



ndash00

0335

ndash00

0293

ndash00

0252

ndash00

0211

ndash00

017

ndash00

0129

ndash00

0087

ndash00

0046

ndash49

endash05

000

0363

000

0775

000

1187

000

1599

000

2011

000

2423

000

2836

000

3248

000

366

000

4072

000

4484

TrainingValidation

TestZero error







5 Conclusion




Solvency





Profitability














Data Availability




References






























Table 5 Continued




1800

1600

1400

1200

Inst

ance

s

1000

800

600

400

200

0



ndash00

0012

ndash00

0011

ndash94

endash05

ndash82

endash05

ndash71

endash05

ndash59

endash05

ndash48

endash05

ndash36

endash05

ndash25

endash05

ndash13

endash05

ndash19

endash06

963

endash06

211

endash05

327

endash05

442

endash05

557

endash05

672

endash05

787

endash05

902

endash05

000

0102

TrainingValidation

TestZero error



glev

100 45le levle 65




⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

(14)


4 Result Analysis



100

10ndash5

0 200 400 6001000 Epochs


800 1000

Mea

n Sq

uare

d Er

ror (

mse

)

TrainValidation

TestBest


3000

2500

2000

1500

Inst

ance

s

1000

500

0



ndash00

0335

ndash00

0293

ndash00

0252

ndash00

0211

ndash00

017

ndash00

0129

ndash00

0087

ndash00

0046

ndash49

endash05

000

0363

000

0775

000

1187

000

1599

000

2011

000

2423

000

2836

000

3248

000

366

000

4072

000

4484

TrainingValidation

TestZero error







5 Conclusion




Solvency





Profitability














Data Availability




References






























glev

100 45le levle 65




⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

(14)


4 Result Analysis



100

10ndash5

0 200 400 6001000 Epochs


800 1000

Mea

n Sq

uare

d Er

ror (

mse

)

TrainValidation

TestBest


3000

2500

2000

1500

Inst

ance

s

1000

500

0



ndash00

0335

ndash00

0293

ndash00

0252

ndash00

0211

ndash00

017

ndash00

0129

ndash00

0087

ndash00

0046

ndash49

endash05

000

0363

000

0775

000

1187

000

1599

000

2011

000

2423

000

2836

000

3248

000

366

000

4072

000

4484

TrainingValidation

TestZero error







5 Conclusion




Solvency





Profitability














Data Availability




References


































5 Conclusion




Solvency





Profitability














Data Availability




References



































Data Availability




References






























7069788.pdf - Hindawi.com

Documents

Transcript of 7069788.pdf - Hindawi.com