Analysis of Capital Flow in Commodity Futures Market Based on SVM

Size: px

Start display at page:

Download "Analysis of Capital Flow in Commodity Futures Market Based on SVM"

Veronica Hunter
5 years ago
Views:

1 Iteratioal Joural of Ecoomics ad Fiace; Vol. 10, No. 8; 2018 ISSN X E-ISSN Published by Caadia Ceter of Sciece ad Educatio Aalysis of Capital Flow i Commodity Futures Market Based o SVM Zi-ag Li 1, Shaozhe Che 2, Hogtao Liag 1 & Hog Zhag 1 1 Fiacial Maagemet of Iteratioal Busiess School, Jia Uiversity, Zhuhai, Guagdog Provice,Chia 2 Fiace Departmet of Iteratioal Busiess School, Jia Uiversity, Zhuhai, Guagdog Provice, Chia Correspodece: Shaozhe Che, Fiace Departmet of Iteratioal Busiess School, Jia Uiversity, Qiasha Road 206#, Zhuhai City, Guagdog Provice, Post No , Chia. Tel: @qq.com Received: May 17, 2018 Accepted: Jue 5, 2018 Olie Published: Jue 28, 2018 doi: /ijef.v108p28 URL: Abstract Commodity futures are futures cotracts based o the physical commodities. Ulike commodity stocks, which must be bought first ad the sold, commodity futures ca also be sold first ad the bought. Therefore, it is ot possible to directly use the formula of capital flow i the stock market to characterize the capital flow i futures cotracts. I this paper, the pricipal compoet aalysis method is used to costruct the pricipal compoet factors based o the K-lie basic market data ad oe based o the K-lie idex data. The the factors metioed above are cross-validated usig the Holdout verificatio form to geerate the traiig set ad test of the support vector machie. The, this paper applies geetic algorithm to optimize the pealty parameters ad kerel fuctios of SVM, ad obtais the parameters with the highest accuracy of classificatio ad predictio of capital flow. Fially, this paper uses the traversal algorithm to fid the time widow with the highest accuracy of the SVM classificatio to predict the capital flow. The research results of this paper show that the SVM-based classificatio of capital flow i commodity futures market is highly accurate. Keywords: commodity futures, capital flow, pricipal compoet aalysis, support vector machie, geetic algorithm, traversig algorithm 1. Itroductio Commodity futures are the futures cotracts based o the physical commodities. Commodity futures have a log history ad a wide rage of types. As early as the aciet Greek period i aciet Greece, there had bee a few cetral tradig veues. Chia s commodity futures market was bor i October With more the developmet of more tha 20 years, the role of commodity futures has cotiued to icrease i the atioal ecoomy, ad its tradig system ad related laws have also bee costatly improved. At preset, there are three futures exchages i Chia: Shaghai Futures Exchage, Dalia Commodity Exchage ad Zhegzhou Commodity Exchage. At preset, Chia s futures market is facig a golde period of developmet, ad commodity futures ivestmet has attracted more ad more ivestors attetio. As of the ed of November 2016, the cumulative volume of the atioal futures market was 3.85 billio lots throughout the year, ad the cumulative turover was trillio yua, which icreased 32.16% ad 31.67% over the same period of The variety of futures trade has also evolved from the first few varieties to about 20 species ow, icludig agricultural products, metal products, chemical products, ad forestry products. The futures market has played two extremely importat fuctios, amely price discovery ad risk avoidace. It goes without sayig that the correct grasp of commodity futures ad the forecast of the price rise ad fall of commodity futures will help us to take advatage of these two fuctios to seize the opportuities ad avoid disadvatages. However, studies at home ad abroad maily focus o the predictio of stock idex futures prices, ad there are still ot eough studies o commodity futures. I this cotext, this paper selects Shaghai Silver as a example, adopts a quatitative approach, aalyzes the law of capital flow of commodity futures, ad establishes a capital flow model of commodity futures market based o SVM method. This will provide ivestors with certai refereces ad suggestios. It will also help the coutry to formulate relevat ecoomic 28

2 ijef.ccseet.org Iteratioal Joural of Ecoomics ad Fiace Vol. 10, No. 8; 2018 policies. 1.2 Relevat Scholarship Research o commodity futures first bega abroad. Workig (1949) has the earliest metio of futures arbitrage o the futures storage. It is because of the existece of arbitrage that futures prices are more complex tha other fiacial products. Workig believes that the core of hedgig is whether or ot to fid the chages of basis to seek profit, that is, the chages i the price differece betwee the futures market ad the spot market to fid opportuities for hedgig. Black (1976) explicitly poits out that the futures market is forward-lookig, ad futures prices iclude future expectatios of spot prices. I terms of capital flow, James ad Richard (2001) foud strog evidece of moey flow mometum, i that lagged moey flows ca be used to predict future moey flows. Furthermore, they foud that moey flows appeared to predict cross-sectioal variatio i future returs. Adrea ad Owe (2007) used mutual fud flows as a measure of idividual ivestor setimet for differet stocks, ad foud that high setimet predicts low future returs. Domestic research o commodity futures maily focuses o quatitative tradig strategies. Ahua (2005) makes a empirical study of the above three hedgig tradig strategies by usig the historical trasactio data of Chia s soybea futures,ad fially foud that hedgig strategy based o traditioal hedgig strategies was the least effective while the oe based o HKM strategy was the most effective. Miimum variace strategy ad HKM strategy were both more effective tha the traditioal strategy to reduce the risk of hedgig. Yiqia (2013) foud that the beefits of the price mometum strategy came from the large tred of large commodity bull ad bear market ad the plate tred caused by seasoal factors. Idustrial products are closely liked to the macro-ecoomy, ad the cycle of price fluctuatios is relatively log. The large tred of more tha 1 year ofte appear, but such pheomea i the plate tred are fewer. Zhihog (2017) costructed a short-term quatitative ivestmet strategy based o the Hurst Idex, Turtle Rules, ad Bolliger Bads, empirically showig that the strategy ca obtai excess returs. Xiaojia ad Qiaqia (2018) costructed a paired tradig strategy based o the OU process. By seamlessly splicig the data of differet mai cotract data of commodity futures, a combiatio of selectios ca achieve a higher rate of retur; whether i success rate, profitability or the possible loss of ivestmet, the correlatio pairig method is better tha the SSD pairig method. I geeral, domestic ad foreig scholars research maily focuses o the arbitrage of commodity futures. The aalysis of capital flow is also limited to the stock market, ad the method of machie learig is rarely used to aalyze capital flow. This paper maily uses SVM to study capital flow of commodity futures market. Firstly, the pricipal compoet aalysis method is used to costruct the pricipal compoet factor based o the K-lie basic market data ad the K-lie idex data. The the above factors are cross-validated usig the Holdout verificatio form to geerate the traiig set ad test set of the support vector machie. The, we use the geetic algorithm to optimize the pealty parameters ad kerel fuctios of SVM ad obtai the highest accuracy of classified predictio of capital flow. Fially, this paper uses the traversal algorithm to fid the time widow with the highest accuracy of SVM classified predictio of capital flow. 2. Theoretical Models 2.1 Techical Aalysis The techical aalysis method is a method used to aalyze the tred accordig to the chage rule of the price chart ad the K-lie techical idicators. I this paper, the support vector machies method are used to aalyze the capital flow of commodity futures. What s more, the K-lie techical idicators of the techical aalysis method are itroduced to costruct a set of ew iput vectors. The selected techical idicators ad the relevat calculatio formulas are show i Table 1: Table 1. Idicators ad descriptio Idicator Formula Termiology C MA 1 MA MACD MACD EMA12 EMA26 EMA ' ( 1) C 2 EMA ( 1) ( 1) MA is the simple average of the closig price of the day; C is the closig price of the day; is the umber of tradig days. EMA is The movig average of the idex of the days; ' EMA is the movig average of the idex of the (-1) day; C is the closig price of the day. 29

3 ijef.ccseet.org Iteratioal Joural of Ecoomics ad Fiace Vol. 10, No. 8; 2018 BIAS OBV CCI RSI KDJ ( C MA ) BIAS MA OBV OBV ' sgv H L C ( MA ) CCI 3 ( MA C) 1 ( ) / C is the closig price of the day. MA C C C [ / ( )] 100 is the simple average of the closig price of the day. OBV is the value of OBV of the previous day; sg determies a value is positive or egative. V is the volume of futures traded today (the umber of futures cotracts) Whe sg=+1, today s closig price is larger tha yesterday s closig price. Whe sg=+1, today s closig price is less tha yesterday s closig price. H is the highest price of the day; L is the lowest price of the day; C is the closig price of the day; φ is a coefficiet equal to C is the sum of the risig price chages i the day; RSI C is the sum of the fallig price chages i the day. C L RSV ( ) 100 H L K 2 / 3 K ' 1/ 3RSV D 2 / 3 D ' 1/ 3K J 3D 2K 1 C is the closig price of the day. L is the lowest price of the day. H is the highest price of the day. D ' is the value of D of the previous day. 2.2 Support Vector Machie I support vector machie, the four most commoly used fuctios at preset are Sigmid kerel fuctio, liear kerel fuctio, polyomial kerel fuctio, ad Gaussia radial basis kerel fuctio. Because differet kerel fuctios have differet characteristics, the performace of classificatio tests usig differet kerel fuctios may be differet. (1) Sigmid kerel fuctio: K( x, x') tah(g (x,x') c)g 0, ( x) 0,c 0 (2) liear kerel fuctio: K( x, x') x T x ' (3) polyomial kerel fuctio: x,x')=(gx T d K( x' c), g 0, c 0 (4) Gaussia radial basis kerel fuctio: K(x, x') exp g x x' 2 This paper compares these four kerel fuctios ad chooses a support vector machie model based o Gaussia radial basis kerel fuctio. 2.3 Geetic Algorithm Geetic algorithm is a simulated evolutioary algorithm ad it is a effective search algorithm to solve the optimizatio problem. I the optimizatio process, because the geetic algorithm basically does ot eed other auxiliary coditio iformatio ad the kowledge of the search space, it has the advatages of self-adaptatio ad self-learig. So compared with may optimizatio algorithms, the geetic algorithm ca solve more optimizatio problems, such as No-differetiable, discotiuous, stochastic optimizatio ad other issues. The basic flow of geetic algorithms is show i Figure 1. 30

4 ijef.ccseet.org Iteratioal Joural of Ecoomics ad Fiace Vol. 10, No. 8; 2018 Start Geerate iitial radom populatio Calculate fitess of idividuals Satisfy stop criterio? Yes No Selectio of the idividuals Ed Selectio Geetic Opeatior Cross Operator Mutatio Operator Figure 1. Flowchart of geetic algorithm Step 1: Iitializatio: The radom method geerates the iitial populatio ad geetically ecodes the populatio. Step 2: Fitess evaluatio: The fitess of each idividual is calculated. Whe the fitess of the idividual is stable, the chromosomes (solutios) carried by idividuals i the curret species populatio meet our requiremets for optimal solutios. Step 3: Idividual selectio: The idividual choices are adaptive value proportioal selectio method, rakig selectio method, ad league selectio method. The adaptive value proportioal selectio method is the earest selectio method i idividual selectio. Therefore, this method is selected by this method ad its formula is as follows: f( xi ) i i 1 f( xi) Step 4: Crossover operatio: By radomly selectig two idividuals i the populatio, the chromosomes are assiged positios i accordace with a certai probability, so that chromosomes carried by the ext geeratio of idividuals become more excellet. Step 5: Mutatio operatio: by usig a certai probability to replace the codig of certai positios o the chromosome codig of a sigle idividual with other codes to simulate gee mutatios i the atural world, ad to improve the ability of the geetic algorithm i local optimizatio. 3. Empirical Aalysis 3.1 Data Resources This paper uses the mai cotract of silver i Shaghai Futures Exchage as the aalysis object, usig the commodity futures miute data form May 10, 2012 to December 31, 2013 to costruct the capital flow model of commodity futures based o SVM. I this paper, the basic market data of K-lie are selected, icludig the opeig price (O), closig price (C), highest price (H), lowest price (L), exchage volume (V), turover (T), ope iterest (P). the techical idicators data are selected, icludig MA, MACD, BIAS, OBV, etc. The data are from the fifth Teddy Cup data Miig Cotest. (1) 31

ijef.ccseet.org Iteratioal Joural of Ecoomics ad Fiace Vol. 10, No. 8; 2018 3.

5 ijef.ccseet.org Iteratioal Joural of Ecoomics ad Fiace Vol. 10, No. 8; Pricipal Compoet Aalysis Accordig to the correlatio aalysis of K-lie basic market data ad K-lie techical idicators data, this paper fids that there exists a certai correlatio betwee data. Therefore, this paper uses the pricipal compoet aalysis method to reduce the iput vectors of the K-lie basic market data ad K-lie techical idicators data respectively, ad obtais the pricipal compoet factors. The sum of the variace cotributio rate of these factors is i coformity with the stadard requiremets. The variace cotributio rate is show i Figure 2 ad Figure 3. Figure 2. The pricipal compoet aalysis of the K-lie basic market data Figure 3. The pricipal compoet aalysis of K-lie techical idicator s data I this paper, based o the coditio that the variace cotributio rate is greater tha 95%, the origial data for the SVM iput vectors, we ca obtai the K-lie basic market pricipal compoet iput vectors ad the K-lie techical idicators pricipal compoet iput vectors. 3.3 Cross Validatio ad Maximum-Miimum Normalizatio The basic idea of cross validatio is to group data sets by some stadard, with oe part as a traiig set ad the other as a test set. The first step is to trai the classifier usig the traiig set, ad the utilize the test set to test the traiig model ad evaluate the classifier. This paper uses the Holdout verificatio to cross-validate the pricipal compoet iput vectors, geeratig the traiig set ad the test set. I additio, i order to elimiate the dimesioal impact betwee the idicators, the traiig set ad the test set are subjected to the maximum-miimizatio ormalizatio process. Take the iput compoets of the pricipal compoets of the K-lie techical idicators as a example. The ormalizatio process of the traiig set ad the test set is show i Table 2 ad Table 3. 32

734438 0.312558-0.168353-0.527510 0.413289-0.576738 0.529379 0.690191-0.655406-2.096482 1.811881-0.760243 0.543435-0.908992-1.323485 0.915296-0.453550 0.932429-0.834858-1.880137 1.404727-1.224413 0.

6 ijef.ccseet.org Iteratioal Joural of Ecoomics ad Fiace Vol. 10, No. 8; 2018 Table 2. Normalizatio of the traiig set of techical idicators pricipal compoet iput vectors Factor 1 Factor 2 Factor 3 Factor 4 Factor Table 3. Normalizatio of the test set of techical idicators pricipal compoet iput vectors Factor 1 Factor 2 Factor 3 Factor 4 Factor Costructio of the Capital Flow Model After the SVM iput vectors ad output vectors are obtaied, the SVM ca be traied ad tested. I order to improve the classificatio accuracy, the geetic algorithm is used to optimize the pealty parameter c ad kerel fuctio g i SVM. Thaks to MATLAB, the relevat parameters are worked out ad show i Table 4, the model fitess is show i Figure 4 ad Figure 5. Table 4. Compariso of capital flow models Model Capital Flow Model Based o K-Lie Basic data Capital Flow Model Based o K-Lie Techical Idicator s data c g Accuracy of the traiig set % % Accuracy of the test set % % Figure 4. The capital flow model based o K-lie basic data Figure 5. The capital flow model based o K-lie techical idicator s data 33

7 ijef.ccseet.org Iteratioal Joural of Ecoomics ad Fiace Vol. 10, No. 8; 2018 It is obvious that the capital flow model based o K-lie basic data ad the capital flow model based o K-lie techical idicators data both have higher predictio accuracy rates, that is to say, these two models ca well portray the capital flow of the futures market. 3.5 Optimizatio of Capital Flow Model The data of the capital flow models established i 4.2 has a large amout scale ad a log time period is also loger both i the traiig set ad the test set. Therefore, it takes more time to calculate ad costruct the strategy. What s more, i reality, the capital flow i the futures market chages rapidly, ad the applicability of the static capital flow models may be affected. Therefore, from the perspective of improvig the speed of model calculatio ad shorteig the cycle of model traiig data, traversal algorithms is used i this paper. The data cycle of the SVM-based capital flow model is optimized to fid the cycle with the highest accuracy. Takig as the iput scroll time widow legth of the iput vectors, the optimal scroll time widow legth based o the pricipal compoet iput vectors of K-lie basic data ad the K-lie techical idicators data ca be obtaied. The results are show i Figure 6 ad Figure 7. Figure 6. The optimal rollig time widow legth based o the priciple compoet iput vectors of K-Lie basic data Figure 7. The optimal rollig time widow legth based o the priciple compoet iput vectors of K-Lie techical idicators data It ca be see from Figure 6 that the pricipal compoet iput vectors of K-lie basic data has a maximum accuracy of 73.08% for classificatio predictio ad a optimal rollig widow legth of 5 days. The pricipal compoet iput vectors of K-lie techical idicators data has a maximum classificatio accuracy of 62.46% ad a optimal rollig widow legth of 1 day. 4. Coclusio The research of this paper has certai academic sigificace ad practical sigificace. 34

8 ijef.ccseet.org Iteratioal Joural of Ecoomics ad Fiace Vol. 10, No. 8; 2018 Academic sigificace: 1) Existig studies rarely use SVM to aalyze the flow of commodity futures fuds. The study i this paper has icreased the applicatio of this aspect. I this paper, the kerel fuctio is used to costruct the model of the capital flow of commodity futures, istead of simply usig a certai kid of kerel fuctio to build the model. Through the optimizatio of the parameters of the geetic algorithm ad the traversal algorithm widow optimizatio, to build a capital flow model with stroger applicability ad practicality; 2) Compared to the traditioal method of forecastig the value of the commodity futures closig price, this paper uses the advatage of statistical classificatio of the support vector machie to predict the ups ad dows of the closig price of commodity futures, which is ot limited to the depedece of traditioal methods o the accuracy of absolute umerical predictio. Practical sigificace: This paper uses a support vector machie, pricipal compoet aalysis, traversal algorithms, etc. to build a classificatio predictio model, based o which a model suitable for aalysis of the capital flow of commodity futures is costructed. The applicatio of this classificatio predictio model is ot limited to commodity futures, it ca also provide certai guidace for the productio ad busiess pla of the compay, govermet ecoomic policies, etc. It is beeficial for eterprises, self-employed idividuals, etc. to use the futures market to hedge, preserve value ad guide productio. Takig agricultural product futures market as a example, this model helps to comprehesively ad objectively grasp the supply chages, price chages ad future treds of agricultural products, scietifically ad ratioally study ad formulate agricultural developmet plas, guide agricultural stadardized productio ad idustrialized operatios, ad promote the moder developmet of agriculture. Refereces Ahua, Z. (2005). Empirical research o hedgig strategy of commodity futures. Wuha Uiversity, Beett, J. A., & Sias, R. W. (2001). Ca Moey Flows Predict Stock Returs? Fiacial Aalysts Joural, 57(6), Carmoa, R., & Durrlema, V. (2003). Pricig ad Hedgig Spread Optios. SIAM Review, 45(4), Frazzii, A., & Lamot, O. A. (2008). Dumb moey: Mutual fud flows ad the cross-sectio of stock returs. Joural of Fiacial Ecoomics, 88(2), Workig, H. (1949). The Theory of Price of Storage. The A-merica Ecoomic Review, 39(6), Xiaojia, Y., & Qiaqia, Z. (2018). Matchig Tradig Strategy of Commodity Futures Market Based o OU Process]. Souther Fiace, Yiqia, W. (2013). Research o Commodity Futures Tradig Strategy Based o Mometum Effect. Fuda Uiversity, Zhihog, L. (2017). Research o the short period quatitative ivestmet strategy of domestic commodity futures. Zhejiag Uiversity, Copyrights Copyright for this article is retaied by the author(s), with first publicatio rights grated to the joural. This is a ope-access article distributed uder the terms ad coditios of the Creative Commos Attributio licese ( 35

The Time Value of Money in Financial Management

The Time Value of Money in Financial Management The Time Value of Moey i Fiacial Maagemet Muteau Irea Ovidius Uiversity of Costata irea.muteau@yahoo.com Bacula Mariaa Traia Theoretical High School, Costata baculamariaa@yahoo.com Abstract The Time Value