Effective components on the forecast of companies dividends using hybrid neural network and binary algorithm model

Transcription

1 Idia Joural of Sciece ad Techology Vol. 5 No. 9 (Sep. ) ISSN: Effective compoets o the forecast of compaies divideds usig hybrid eural etwork ad biary algorithm model Mahdi Salehi *, Behad Karda ad Zohresh Amiifard 3 *Departmet of Accoutig, Ferdowsi Uiversity of Mashhad, Ira Departmet of Accoutig, Ferdowsi Uiversity of Mashhad, Ira 3 Departmet of Accoutig, Sciece ad Research of Hormozga Brach, Islamic Azad Uiversity, Hormozga, Ira *mahdi_salehi54@yahoo.com Abstract The issue of accoutig profit has bee oticed from log time by ivestors, maagers, fiacial aalysts ad creditors. Due to the importace of divided per share is disclosed by compaies ad the role of divided i decisios ad because the most importat source of iformatio for ivestors ad maagers ad other users i the stock, is the forecasted divided by compaies, this study follows to recogize the affectig factors o 3 chemical compaies i the Tehra Stock Exchage divided usig geetic algorithms combied with artificial eural etwork. Fially, the variables affectig the output are used to predict divideds i the model that is by eural etwork desiged. The error is calculated ad be the basis of compariso with other methods. The study icluded chemical compaies accepted i Tehra Stock Exchage durig 6-. The idepedet variables i this study are accoutig ratios ad stock cash divided is depedet variable Keywords: Predictio, Divideds, Neural etwork, biary algorithm Itroductio Cash retur ad cash stock divideds, due to the objectivity ad tagible, has a special cosideratio amog some of stakeholders. I fact, actual ad potetial users of fiacial iformatio have bee eager to iform about the ability to create cash ad sometimes its distributio amog the stakeholders of the compay. Because this iformatio ot oly offers a clear picture of the curret situatio of compay, but it ca also provide the estimates of the future status that certaily is importat i their decisio makig process. This issue has a serious importace for corporate maagers, to use iformatio due to the use of iformatio obtaied i the maagemet process of compay ad market evaluatios of their performace. Therefore part of the attetio of corporate maagers paid to the subject that is titled "divided policy". But more importat tha the divided policy, is the root of reasos to adopt a specified divided policy by the compay. This may help i the importat ecoomic decisios for differet groups of stakeholders, particularly the ivestors. The reasos ad factors obtaied from this study, ot oly help to explai the behavior of compaies i the past, but rather provides a tool to predict the movemet ad its future directio (Jahakhai & Ghorbai, 6). I geeral, the compay's attractiveess to ivestors, i additio to the curret situatio, depeds o its potetial of creatig the future icome. The compay which is curretly profitable ad its profitability cotiuig i the future are more attractive tha the compay whose potetial beefits will be reduced soo. It ca be claimed that the most importat criteria for ivestmet decisios, is the curret ad future profitability of compaies, so that ivestors i differet ivestmet strategies, maily decide due to profit makig of compaies ad oe of the most importat decisios of fiacial maagers, is decisios related to stock cash For this purpose predictio to the 79 compaies 33 divideds (Nouraii, 6; Hghighat et al., ). Due to the importace of decisios related to stock cash divideds there are always a lot of models for predictig these factors that it ca be poited to oe of the strogest of them, the eural etwork. I the proposed model i this study, usig a combiatio of two methods (biary geetic algorithms ad eural etworks), have bee tried to idetify the factors that mostly ifluece o the chemical compay's stock cash divideds. Gouopoulos (3) evaluates the forecast accuracy ad the factors affectig the predicted accuracy of compaies that their shares were public offered for the first time i the Athes Stock Exchage durig The results show existece of a iverse relatioship betwee firm size ad the predictio error. While the more the time period, it causes the predictio error to decrease. Fially, this research did ot show a positive relatioship betwee the level of fiacial leverage, shareholdig structure ad macro-ecoomic coditios ad the predictio error. Kato et al., (9) ivestigated the optioal forecast of divideds by maagemet i listed compaies i Japa durig The results showed that most of the divided forecast is higher tha actual divided, but forecasts adjusted ad reduced durig the year. While the predictios have average iformatio, the aticipated divided of compaies with poor performace ad with maagers that had poor ad optimistic predictios, have little ad low reliability iformatio. Savov ad Weber (6) study sample cosisted of Germa compaies durig Results show that the market situatio, leads the decisio to icrease divided. I Ira the followig study had bee carried out. Mashayekh ad Shahrokhi (7) compared the accuracy of maager s predictio to future earigs per share, with predictios based o the radom walk model.

2 Idia Joural of Sciece ad Techology Vol. 5 No. 9 (Sep. ) ISSN: icludig 639 observed were aalyzed durig -4 usig mea differece. The fidigs of this research show, there are sigificat differeces betwee the maager s predictio error ad the error of predictio based o radom walk. Furthermore, the compariso of mea differece of two models, idicatig a higher predictio accuracy of maagers tha predictio based o radom walk. The other results of study assumptios show that predictios of maagers have optimistic deviatios ad predictio accurately is differet due to the size of the compay, profitable or detrimetal ad type of the idustry. Sarebaha ad Ashtab (8) look at factors affectig the predictio of error of 7 listed compaies i Tehra Stock exchage durig ad cocluded, the correlatio betwee compay size, compay life, time horizo forecastig, profitability ratios, leverage ratio ad auditor's valid oly profitability ratios factor affectig divided forecast errors, that are iversely associated with the predictio error. Azad (4) studied 58 compaies listed o the Stock Exchage. The results showed that the divided aticipated by the compaies ad the real profits have also a sigificat relatioship. Also, the relatio betwee the beefits aticipated by the compay ad stock returs are sigificat. Samadzadeh (993) evaluates the divided strategies ad their impact o compaies stock value. The study results show the divided policies i the stock market, are ot kow for corporate maagers ad shareholders do ot otice to cash divideds as a adjustmet that have message. Research methodology The choice of affectig factors o divided forecasts usig eural etworks ad biary geetic algorithms hybrid model ca correctly reduce the divided predict error of chemical compaies active i the stock exchage. This research is applicable ad is based o field research, based o iformatio collected from the Tehra Stock Exchage, research hypotheses ad the test results ca be geeralized to the whole populatio. Model, is a combiatio of, biary geetic algorithms, ad eural etwork that geetic algorithm is kow mostly as a way to optimize the fuctio. Implemetatio of geetic algorithm begis with a iitial populatio of chromosomes. The iitial solutio is the evaluated accordig to their level of competece to be give the opportuity to reproduce. To be made the best geeratio at the ed. I the first stage the variables that are likely to affect the divided will be etered ito the biary geetic algorithms. The fitess fuctio of geetic algorithm is a eural etwork, which its iverse error criterio is used to fit chromosomes i biary geetic algorithms. After the geetic algorithm biary, biary, chose the best combiatio of iput variables, the variables are etered ito artificial eural etwork that desiged to suit ad predict selected variables, after that data variables beig elected the previous sectio (geetic algorithm), traied eural etwork, the etwork is tested with evaluatio data ad thus the predictio error will be measurable. To achieve the desired objectives of research ad to reduce the effect of other factors such as type of idustry ad circumstaces prevailig i the idustry, from various idustries active i the Tehra Stock Exchage, o divideds, The chemical idustry which was oe of most active compaies i the stock ad its data are available for 6 years to is chose as the statistical commuity, as a result, 3 compaies are selected from 3 chemical compay active i the stock market. Regard to the specificatios of eural etworks that higher the umber of test data, results a better respose from the etwork (Azar & Karimi, 9), all compaies i the survey were selected as sample ad samplig is ot doe. Artificial eural etworks This area is oe of the most dyamic areas of cotemporary research that has attracted may researchers from various scietific fields. Usig eural etworks ad geetic algorithms has icreased more ad more these days to solve complex applied problems. The importace of this model is that it ca describe ad examie processes that deped o several parameters ad with differet degrees of importace ad the provide a satisfactory aswer. Way of dealig with computatioal methods of eural etworks is capturig the fudametal strategic priciples of brai processes ad their applicatio i computer systems. Artificial eural etworks are itelliget dyamic free-model systems that with the processig of experimetal data trasmit the kowledge, or the law behid the data, to the etwork structure. These systems based o computatioal itelligece, are tryig to model the huma eurosyaptic brai structure. Major compoets of computatioal itelligece or soft computig are eural etworks (euroal computatios), fuzzy logic (approximate calculatio) ad geetic algorithms (geetic aalysis) that each oe have a kid of brai modelig. Neural etworks, ispired by the extesive distributio ad parallel computig i the huma brai, the biological ad learig eural etworks i these systems, model the syaptic coectios ad euroal structure. A eural etwork has bee composed of a large umber of odes ad some directioal lies that are coected to odes, the odes i the iput layer are sesory odes ad the oes i output layer odes are called respodets, betwee iput ad output euros, hidde euros are located. Iformatio eters from iput odes to the etwork ad the coects to hidde layers from coectios ad fially the etwork output of output layer odes is achieved (Bill & Jackso, 7; Nikbakht & Sharifi, ) (Fig.).

3 Idia Joural of Sciece ad Techology Vol. 5 No. 9 (Sep. ) ISSN: X X X Fig.. A example of a eural etwork structure 3 Iput layer Middle layer Output layer Biary geetic algorithm Geetic Algorithm is a optimizatio method ispired by the livig ature that ca be classified as a umerical method, radom ad direct search. This algorithm is based o the repetitio that its basic priciples adapted of geetics ad has bee iveted duplicatio of some of the processes observed i atural evolutio, effectively uses the old kowledge of a populatio, to provide ew ad improved solutios. Fig.. Coceptual framework of the system I the begiig, the iitial populatio that shows the aswer will be radomly selected. Each of the members of these populatio are called chromosomes is oe of the aswers of problem. Each of these chromosomes is selected of equal-legth strig of umbers, which each of these umbers is called a gee. Geetic algorithm based o repeated acts. The populatio i each stage is called geeratio. Each of the members of this geeratio is evaluated based o the value fuctio. I these algorithms, the ew geeratio is tryig to allocate more value from the value fuctio ad with this performace to 6 Y Y 333 be closer to target fuctio. At each step of the replicatio, each of the chromosomes, with certai probability, cross with others, or married so that the outcome oe or more ew chromosome is called a child. I these childre may have geetic mutatios at certai probability, as this will chage the amout of oe or more gees of chromosome (Akhbari, 8). I the fial stage, the childre evaluated based o their value fuctio ad ew geeratio will be produced accordig to their value ad the worth of their parets, the first geeratio. These processes are repeated util the preset geeratio will coverge to the optimal solutio or oe of optimal subsolutios. There are differet methods for cross ad mutatio operators that due to the complexity of the problem oe of them is selected. Divided forecasts usig eural etwork ad biary algorithm Fig. ca be cosidered as a coceptual framework i which, X is the vector of iput variables (here, variables ifluecig the predictio of a stock divided) ad Y is the vector of output variables (i this case, the stock divided) ( Makvadi et al., 8). The mai purpose of this paper is to fid the optimum combiatio of iput variables of the system ad i this case, biary geetic algorithm (collective motio of particles algorithm) is used. I fact, we fid out what ratios ad variables must be extracted from fiacial statemets ad etered their values ito the predictio model to predict the stock divideds. I this sectio, the desig ad implemetatio of the proposed hybrid approach, is expressed. Scheme shows the complete process of the proposed method. There are 5 processes (level) i the proposed method that i order are: the data selectio, data cleaig ad preparatio ad, factors affectig choice of cash forecastig profits of stock, multi-layer perceptro eural etwork traiig process based o selected compoets ad fially evaluatio of models traied with test data that has ot bee observed so far by the algorithm. The MATLAB software versio 7.6 is used for algorithms implemetatio. This software package provides may ready tools to work with that are makes it very comfortable. Table. The idepedet variables of the study No Name of idepedet variable No Name of idepedet variable Capital 3 Quick Ratio Curret Ratio 4 Retur o capital 3 Divided yield 5 Retur o equity% 4 Earigs per share 6 Retur o sales% 5 Fixed asset turover 7 Sales to ed capital 6 gross icome to sales% 8 Sales to market value 7 Ivetory to assets Ratio% 9 Stock price before the aual geeral meetig 8 Ivetory turover Total asset turover 9 Net sales Total debt Operatig icome total debt to total assets ratio% Price to earigs Ratio 3 year Profit after tax

4 Idia Joural of Sciece ad Techology Vol. 5 No. 9 (Sep. ) ISSN: Data selectio Fiacial data for 3 differet chemical compaies i Tehra Stock Exchage with 3 idepedet variables ad a depedet variable (stock cash divideds) of 6 to years were collected from various sources. A total of 8 samples were collected. Idepedet variables are show i alphabetical order i Table. Data cleaig ad preparatio The secod stage is data cleaig. At this stage, the data that their idepedet variables iformatio were icomplete or were ot calculated are removed. Oe of the criteria used to evaluate a classificatio, is the error rate, which are of various types. I geeral we ca t properly judge about abilities of algorithm by comparig the errors that is calculated o the learig data. The error rate o learig data is usually lower tha the error rate o that have ot bee see i the learig process. With this reasoig, learig error ca t be used to compare the two algorithms. This is because for more complex models, classificatios that usually have more parameters, the boudaries are more complicated. This reduces the error o the learig data compared with simpler models (Abbas Kia, 9). So i additio to learig data sets, a set of data for evaluatio is required. Learig data is used for the model traiig ad evaluatio data is used for calculate the algorithm error rate o data that has ot see. I this paper, a radom 7% data as traiig data ad the remaiig 3% was cosidered as evaluatio data. I order to prepare data for traiig ad evaluatio with estimates, iitially each of the variables is ormalized usig the followig equatio to reduce the impact of large umbers: ~ Si S mi Si, i,,9 () S S max mi Scheme. The proposed algorithm Selectio Data cleaig ad preparatio Choice of ifluecig factors o predict stock cash divideds Traiig data Multi-Layer Perceptro eural etwork traiig based o selected compoets Test data Evaluatio multi-layer perceptro eural traied etwork That Smi ad S max shows the miimum ad maximum value of the variable ad S ~ i shows the ormalized value of i S. 334 Choice of ifluecig factors o predict stock cash divideds This sectio itroduces the approach to choice of effective compoets of predictio of stock cash divideds, usig a combiatio of multi-layer perceptro eural etwork with biary geetic algorithm. Due to the eural etwork algorithm beig used, at first it is discussed how to implemet a eural etwork ad the the method to predict the factors affectig stock cash divideds usig biary geetic algorithm, will be discussed. Multi-layer perceptro eural etwork How to code variables: the mai purpose of this review is to provide a model to idetify the variables that have more ifluece o decisio-makig, through the elimiatio of these low-impact ad o impact variables o the output, to improve the forecastig process. Variables should be coded so that they are cosidered idividually ad combied (Makvadi et al., 8). The umber of iput variables i the preset aalysis is 3. These variables are coded i biary method ito a strig. To achieve the desired result, we perform the followig way: A -bit strig ( is the umber of variables) ca be prepared. I this strig, each bit represets a variable, the presece of these variable o the fial result meas bit= ad if it is ot preset the bit=. Type of etwork ad the learig rule: a artificial perceptro multilayer eural etwork used i the preset model with error back propagatio learig rule. The umber of euros ad layers: the etwork cosists of three layers with oe euro i the output layer ad the middle layer, has respectively ad 5 euros. Number of etwork layers ad middle layer euros is determied usig the method of trial ad error (Fig.3). Stimulus fuctios Sice it is ecessary to calculate the momet derivative we eed to use stimulus fuctios that ca be derived. I the preset cotext hyperbolic taget fuctio is the best fuctio. This fuctio is as follows (Makvadi et al., 8): Fig. 3. The umber of euros ad layers

5 Idia Joural of Sciece ad Techology Vol. 5 No. 9 (Sep. ) ISSN: Fig. 5. Sigle poit crossover 335 Error fuctio: error fuctios i eural etworks, i terms of meaig, do ot differ much from each other. Hece, for this case, mea square error is cosidered (Makvadi et al., 8). Fig. 6. A example of a mutatio That A I is the actual value ad F I is the predicted value. Neural etwork test: to test etwork performace ad determie fitess value, percetage of data that previously did ot participate i the traiig process, acted as test data, ad measures the predictio ad geeralizatio power of eural etwork (Makvadi et al., 8). Choice of ifluecig factors o stock cash divideds predict usig biary geetic algorithm As previously described, the aim of algorithm of the preset model is to idetify the combiatio of variables that mostly iflueces the predictio of the output variable (divided shares). I the process of implemetig this algorithm, the fitess value of idividual members of each geeratio is calculated, accordig to fitess values, the ext geeratios are produced by applicatio of 3 operator, selectio, trasplatatio ad mutatio. This radom search process, cotiues util the termiatio Fig. 4. The proposed algorithm for the selectio of ifluecig factors usig a biary geetic algorithm Creatig eural etwork proportioal to the umber of variables Learig the etwork Error criteria determiatio No First geeratio (radomly) Evaluatio Termiatio criteria satisfied? Yes Retur Best solutio criterio is achieved. The geeral algorithm of this process preseted i Fig. 4. Geetic Algorithm Implemetatio details Strig presetatio: I this study, the variables are biary ecoded with fixed legth strigs. Due to the discotiuousess ature of the variables used, each bit correspodig to each geeratio of chromosomes, represets oe of the used variables (Abbas Kia, 9). Fitess calculatio: For evaluatig the appropriate aswers ad a atural choice, the eed for a idex of the appropriate aswers from improper. This idex is a objective fuctio, which ca be selected based o a mathematical model or a computer simulatio or qualitative criteria, which better aswer ca be recogized by it. The fitess fuctio ca be the mea square error (MSE). Whatever the MSE be less for a chromosome or a aswer, the aswer is more appropriate. I this study, the fitess fuctio has bee equaled to reverse the errors of the eural etwork obtaied from the traiig eural etwork i ay strig (Akhbari, 8). Populatio size: populatio size is ofte oe of the affectig parameters the efficiecy of the algorithm. For example, if the populatio size is cosidered small, may lead to premature covergece. Ad if it is cosidered large, the algorithm executio time may be too much. I the used algorithm i this study, the populatio size of 5 has bee cosidered (Akhbari, 8). - Operators of geetic algorithm Selectio: paret chromosomes are selected radomly of iitial populatio, so that the aswers have higher fitess values, more selectio are possible. I other words, the better aswer is preferred to the worse aswer (Abbas Kia, 9). Crossover: The most importat operator i geetic algorithm is crossover operator. Crossover is process which the old chromosomes are combied ad mixed together to create a ew geeratio of chromosomes. The pairs that were cosidered as parets i selectio part exchage their gees ad create ew members.

6 Idia Joural of Sciece ad Techology Vol. 5 No. 9 (Sep. ) ISSN: Table. Parameters of geetic algorithm applied o model Type/amout Parameters 5 Number of populatio Sigle poit Crossover type.8 Crossover probability. Mutatio probability Roulette wheel Selectio fuctio Crossover causes the fragmetatio or loss of geetic variatio is elimiated. Because it allows the appropriate gees fid each other. Here, the sigle poit crossover is used. I sigle poit crossover at first the pair of paret chromosome is cut at appropriate poit alog the strig the parts of the cut-off poit chage together. This creates two ew chromosomes that ay poit of it, iherit gees from paret chromosomes (Abbas Kia, 9) (Fig.5). Mutatio: mutatio is the third operator i geetic algorithms. I geetic algorithm, after the creatio of a ew member each of its gee may be mutated. I mutatio, a gee may be elimiated from the populatio, or to be added. The mutatio operator behavior is so that for every perso i the set mutatio probability, which is usually less tha oe percet, is examied. If the mutatio should be doe a bit of chromosome radomly selected ad its value is coverted from zero to oe, or vice versa (Abbas Kia, 9) (Fig.6). Covergece: I every iteratio algorithm, the obtaied populatio has to evaluated, i terms of covergece. For this purpose, a idex ca be defied as the ratio of miimum value of aswer to the average value of aswer i the populatio ad was compared with predefied value, ad icrease mutatio probability, if the populatio id coverged (Akhbari, 8). Research fidigs As previously described, 3 idepedet variables from 3 compaies durig years 6 to were chose i total, 8 samples were selected. I order to prepare the data for traiig ad evaluatio, at first each of the variables are ormalized usig equatio () to reduce the impact of large umbers. To evaluate the regressio models, two criteria s have bee used, mea square error (MSE) ad root mea square error (RMSE), that is calculated usig the followig relatios: MSE y i d i i () RMSE y i d i i (3) That d i ad y i are respectively real divideds ad predicted divided of samples by each algorithm ad is the umber of samples. The much MSE ad RMSE values are closer to zero; the predictio algorithm is closer to reality. Ifluecig factors o predict stock cash divideds ad results Table 5. Calculated error rates for traiig data of MLP algorithm combied with BGA MSE Trai RMSE Trai Table 6. The calculated error rate for evaluatio data of MLP algorithm combied with BGA MSE Test RMSE Test Table 7. Calculated error rates for traiig data of MLP algorithm MSE Trai RMSE Trai Table 8. The calculated error rate for evaluatio data of MLP algorithm MSE Test RMSE Test Geetic algorithm was executed to fid a set of affective factors i predictig stock cash divideds, with the parameters which preseted i Table. The fial coditio is the uchagig of the best strig i each geeratio, if the best strig remais uchaged for fifty successive geeratios, the algorithm termiates. After executio the geetic algorithm with the above parameters, results are obtaied for traiig data ad evaluatio data are show i Tables 3 ad 4 respectively. Iput variables after covergig were the followig strig. This strig is the fial aswer of the algorithm that itroduces the variables i the fial compositio that after decodig, the idepedet variables ifluecig the predictio of stock cash divideds are obtaied. After determiig the fial compositio, ad idetify variables that geetic algorithm PSO has chose them as variables to predict future stock divideds, we test the MLP Number Chromosome Table 4. Idepedet variables ifluecig the predictio of stocks cash divideds with geetic algorithm Fixed asset Total asset gross icome to Retur o Quick Ratio turover turover sales% sales% Ivetory turover Price to earigs Ratio Divided yield Operatig icome Table 3. The results of traiig data Stock price before the aual geeral meetig Profit after taxes Retur o capital Total debt Net sales

7 Idia Joural of Sciece ad Techology Vol. 5 No. 9 (Sep. ) ISSN: Fig. 7. MLP eural etwork with 4 variables selected for predictio Refereces. Abbas Kia M (9) Meta-heuristic search algorithms, geetic algorithms. Samat Publi., Tehra, Ira.. Akhbari M (8) The applicatio of geetic algorithms i structure of forecast of iflatio. Eco. Res. 85, Azad M (4) The iformatio cotet of divided forecasts of compaies. MS Thesis, Dept. Maagt. & Fig. 8. MLP eural etwork with 3 variables selected for predictio Accou., Allameh Tabatabai Uiv., Ira. 4. Azar A ad Karimi S (9 ) Forecastig stock returs usig accoutig ratios with a eural etwork approach. Fia. Res. (8), Bill A ad Jackso T (7) Itroductio to eural etworks. Ist. Sci. Publ. eural etwork with these variables. The eural etwork that 6. Gouopoulos D (3) Associatios betwee is desiged for this experimet ca be like as Fig. 7. Fig.8. maagemet forecast accuracy ad pricig of IPOs i presets MLP eural etwork with 3 variables selected for Athes stock exchage. Multiatioal Fia. J., predictio Now eural etwork with traiig data (4 idepedet variables i Table 3) is taught ad the evaluated data affected to the traied eural etwork. Table 5 ad 6 show the results of traiig ad assessmet data for this model. I the secod stage, all 3 idepedet variables listed i Forthcomig available at SSRN : 7. Hghighat H, Bakhtiari M ad Beheshti pour M () Settig the priority of the factors ifluecig the amout of accuracy of profit predictio of accepted compaies i Tehra Stock Exchage i the time of capital growth. Ira. Accou. Rev. 8(65), 4-6. Table are used to predict stock cash divideds. Now 8. Jahakhai A ad Ghorbai S (6) Idetifyig ad eural etwork with traiig data is taught ad the explaiig affectig factors o the divided policy i evaluated data affected to the traied eural etwork. Table 7 ad 8 show the results of traiig ad assessmet data for this model. Accordig to the results, the performace of model is better tha model. This suggests that a combiatio of geetic algorithms ad eural etwork model ca lead to a substatial icrease i explaatory power of the model. Thus the research hypothesis is accepted. Coclusio The overall aim of this study, is selectig the affectig compoets o forecast cash divideds of stock, usig eural etworks ad geetic algorithms. The results show that, combiig eural etworks ad geetic algorithms, i adopted compaies o the Tehra stock exchage. Fia. Res., Kato K, Skier D ad Kuimura M (9) Maagemet forecasts i Japa: A empirical study of forecasts that are effectively madated. The Accou. Rev. 84(5), Makvadi P, Jasbi A ad Alavi H (8) Selectig ifluecig elemets o forecast of future stock divideds usig a combiatio of geetic algorithms ad eural etworks. Eco. Res. 5(), Mashayekh Sh ad Shahrokhi Samae (7) Evaluatio of accuracy of divideds forecasts by maagemet ad its ifluecig factors. Studies of Accou. & Auditig. 4(5), order to select the optimal parameters, sigificatly. Nikbakht MR ad Sharifi M () Predictio of fiacial icreases the predictive power compared to the oly use of eural etworks. It should be oted that artificial eural etworks, are bakruptcy i the Tehra stock exchage compaies usig artificial eural etwork. Idust. Maage. (4), kow as the black box test, meaig that despite of the 3. Nouraii M (6) The role of earigs quality, i stregth of these models to recogize relatioships betwee variables, they do ot show the user the form of this relatioship (Reber et al., 5). predictig future earigs. MA Thesis, Maage. & Accou. Faculty, Shahid Beheshti Uiv., Ira. 4.Reber B, Berry B ad Toms T (5) Predictig mispricig Sice oly fourtee variables of twety-three of iitial public offerigs. Itel. Sys. Accout. Fia. idepedet variables i this study were selected by geetic algorithm for traiig eural etworks, it shows that the ot elected variables do ot have sigificat effects o cash divideds i studied sample. Maagt. 3, Samadzadeh M (993) Divided policies ad their effects o stock value at the Tehra stock exchage. MS Thesis, Uiv. Isfaha, Ira. Research suggestios 6. Sarebaha MR ad Ashtab A (8) Idetifyig We suggest the followig topics for future studies:. Other methods of artificial itelligece such as artificialfuzzy etworks, other geetic algorithms or other hybrid models ca be used to forecast divideds.. The research has doe i a active idustry i stock exchage, so it ca be doe i a larger level, or i all ifluecig factors o divideds forecast errors of ewly accepted compaies at the Tehra stock exchage. J. Soc. & Huma Sci., Eco. Sci., 6, Savov S ad Weber M (6) Fudametals or market movemets: What drives the divided decisio? Workig paper; available at: accepted compaies i stock exchage. 337