WO2019119478A1

WO2019119478A1 - Exchange rate analysis method based on genetic algorithm

Info

Publication number: WO2019119478A1
Application number: PCT/CN2017/118527
Authority: WO
Inventors: 陈益; 李耘; 于洪年
Original assignee: 东莞理工学院; 陈益; 李耘; 于洪年
Priority date: 2017-12-20
Filing date: 2017-12-26
Publication date: 2019-06-27
Also published as: CN108090828A

Abstract

Disclosed is an exchange rate analysis method based on a genetic algorithm. The method comprises the following steps: Step one: pre-processing exchange rate price data; Step two: inputting the pre-processed exchange rate price data into an exchange rate model and optimizing parameters of the exchange rate model using a fitness function; Step three: verifying the results after optimization, the data results passing the verification; if the verification fails, repeating the second and third steps after Mendelian genetic algorithm processing, until the optimized results pass the verification, and outputting the results. The method of the invention has better search capabilities than the standard genetic algorithm and the conventional least squares method.

Description

A Method of Exchange Rate Analysis Based on Genetic Algorithm

Technical field

The invention relates to computer software technology, in particular to a method for analyzing an exchange rate based on a genetic algorithm.

Background technique

The exchange rate movement model is a difficult task in the international financial arena. A strong consensus in academic research is that macroeconomic fundamentals have no explanatory power in terms of short-term exchange rate fluctuations [1, 2]. Instead, the microstructural approach focuses on information and macro fundamentals, non-basic factors and their transmission in the foreign exchange market, as well as the impact on exchange rate fluctuations. Empirical evidence suggests a positive link between the exchange rate and its corresponding concurrent order flow, defined as the net value between the buyer-initiated trade and the seller-initiated trade [3-5].

Other evolutionary computation (EC) methods and other financial studies have also proposed research on exchange rate analysis methods. In 1996, Hann and Steurer [6] analyzed the effect of data frequency on the US dollar (Mark prediction) through artificial neural networks (ANNs). Studies have shown that when monthly data is applied, ANN does not greatly improve the accuracy of the prediction. In 2003, Qi Hewu [7] proposed a multi-level feedforward network to predict the exchange rate. The calculation results concluded that the artificial neural network could not be effectively executed in the out-of-sample prediction accuracy. In 2007, Yadav et al. [8] used a standard multi-layer neural network (SMN) to predict a series of time series data in order to predict the exchange rate between 2002 and 2004.

In 2005, Rimcharoen et al. [9] proposed an adaptive evolutionary strategy (ESs) method for predicting stock trading in Thailand, in which the GA method and the ES method were combined. Since then, there has been no further study of the ES method used to predict exchange rates.

Worasucheep and Chongstitvatana [10] combined the advantages of multiple strategies in the differential evolution algorithm (DE) proposed in 2009, but did not further study the use of DE or DE related methods to analyze exchange rates.

The Particle Swarm Optimization (PSO) method is a group intelligence algorithm, which is a crowd-based search algorithm that moves between a multi-dimensional search space based on individual social behavior (particles). In 2004, Zhao and Yang [12] used both PSO and artificial neural networks to predict the stock market in 2009, but no research has been conducted on the use of PSO to predict exchange rates.

In the 1970s, genetic algorithms were introduced to the University of Michigan by Holland [13]. Inspired by Darwin's theory of evolution, the university applies three basic genetic operators—selection, crossover, and mutation—to population problems. The characteristics of actual problems are often imposed by some disproportionate performance and competitive measures or targets that impose some limiting decision variables. Choosing a suitable compromise solution from all non-inferiority choices is not the only problem that depends on it. It usually also depends on the subjective preferences of the decision maker. Therefore, the ultimate solution to the problem is the optimization and decision-making process.

In recent years, many fields of literature have proposed the application of genetic algorithms using Mendel's law [14-19]. As shown in Figure 1. In previous studies [16-19], the Mendelian operator was often inserted after the mutation operator. Based on the mutation probability Pm, genetic mutations may produce unstable populations, which will result in contamination of the entire genetic process output [20, 21]. Typically, mutation operators are randomly generated by changing a binary bit from '0' to '1', and vice versa. The basic method of mutation is the ability to recombine the improved solution at a given probability, but it may also damage the dominant population and lose good solutions and convergence [22-24]. The Mendelian operator will amplify this unstable community from the perspective of micro-evolution through its local search capabilities.

Summary of the invention

The determination of exchange rate values has long been regarded as the most challenging of the high-frequency time series trading applications [3-5, 29, 30]. In order to provide investors and researchers with more accurate predictions, some different models are described as the phenomenon that prices follow random walks, which is suitable for genetic algorithms with random and nonlinear search capabilities. The Mendelian genetic algorithm will be applied to the empirical analysis of exchange rate values, which can provide an evolutionary and computational method to solve the problem of exchange rate value determination. Therefore, the object of the present invention is to provide a method for analyzing exchange rate based on genetic algorithm, in particular, an exchange rate analysis method based on Mendelian genetic algorithm.

The technical solution of the present invention is: a method for analyzing an exchange rate based on a genetic algorithm, comprising the following steps:

The first step is to pre-process the exchange rate price data; the pre-processing includes removing the data away from the mean, normalization, etc.;

In the second step, the pre-processed exchange rate price data is input into the exchange rate model, and the fitness function is used to optimize the exchange rate model:

The third step is to verify the optimized result, verify that the output data result is verified, the verification fails, and the second and third steps are repeated after the Mendelian genetic algorithm is processed, until the optimized result is verified and the result is output.

In the second step, the parameters of the exchange rate model are optimized by the fitness function as follows:

Determine the exchange rate model and use the order flow to determine the exchange rate, as in (2)

Where Δ is the first-order difference of the sequence; S _t+1 is the exchange rate of t+1 time, and S _t is the spot exchange rate, which is defined as the domestic price of the foreign currency, see equation (1); ε _t+1 As the interference term, as shown in equation (3), it indicates that the future exchange rate change is a function of the gap between the spot exchange rate and the expected basic factor; b is the discount factor; f(x _t ) is the basic factor at time t ;x _t is a simultaneous order flow;

q is the future periodic factor,

For the market decision maker's expectation of the basic factors of the q-time at t time, there is an information condition at time t, and t+q is a time period, for example, t = 1 second, q = 5 seconds.

Corresponding to the time t+1,

Corresponding to the time t, the x _{t+q+1 in the} brackets corresponds to the t+q+1 moment of x, and x _t+q corresponds to the t+q moment of x, which is actually the sequence of x. At time t, t+q+1 data is taken backward to calculate.

Iterative calculation of equation (2) is performed using a genetic algorithm based on Mendelian operator:

A, initialize each parameter

B. Encode all variables;

C. Perform a selection operation of the genetic algorithm;

D, using the Mendelian operator for calculation;

E. Perform cross operation of the genetic algorithm;

F. performing a mutation operation of the genetic algorithm;

H. Decoding: Repeat the iterative AH step to obtain the optimal solution, and use the adaptive function to evaluate whether the optimal solution after decoding meets the requirements of the optimization calculation. If the requirements are met, the final optimized solution set is obtained; otherwise, the operating parameters are modified and recalculated. Until the final optimization solution set is obtained.

The encoding is a binary encoding.

The fitness function for evaluating the decoded optimal solution is as shown in equations (10) and (11). When the obtained optimal solution can make R ² close to the maximum, the requirements of the optimization calculation are satisfied:

Where β ₁ , β ₂ and β ₃ are 3 coefficients, respectively, xi is defined as a COF value observed in foreign exchange (cumulative order flow), and yi is defined as the foreign exchange and local currency obtained by log(/). proportion. The national currency is as follows:

Y _i =log (foreign exchange / local currency) × 10000

Δx _i (t) and Δy _i (t-1) are differential data sequences obtained during data preprocessing at times t and t-1, as shown in equations (14) and (15).

Δx _i (t)=x _i (t)x _i (t-1) (14)

Δy _i (t-1)=y _i (t-1)-y _i (t-2) ( ¹⁵ )

The calculation method of the Mendelian operator is as follows;

Each chromosomal position in the parent population is assigned an attribute, and there are three attribute types: D attribute represents dominant, pure and dominant genes, and R attributes represent recessive, pure and recessive genes and H Attributes represent hybrid, hybrid genes;

The two chromosomes in the optional parent population are cross-tested as the parental chromosome to obtain the progeny chromosome. If the attribute of the parental chromosomal location is D, the attribute of the chromosomal position of the offspring is D; if the attributes of the parental chromosomal location are R The attribute of the chromosomal position of the progeny is R; if the attribute of one parent chromosomal position is R and the attribute of the other parent chromosomal position is D, the attribute of the chromosomal position of the progeny is H; if the attribute of a parent chromosomal position is D The attribute of another parental chromosomal location is H, then the chromosomal position of the offspring is 50% D, 50% is H; if the attribute of one parent chromosomal location is R and the attribute of the other parent chromosomal location is H, then the daughter chromosome The position 50% is R, 50% is H; if the parental chromosomal location is H and the other parental chromosomal location is H, then the progeny chromosomal location is 25% R, 25% D, 50% H . And Table 1 represents the algorithm of the Mendelian genetic operator.

Each chromosomal bit in the parent population is randomly assigned an attribute.

DRAWINGS

Figure 1 is an exchange rate analysis method based on a computational intelligent aided design framework;

Figure 2 is a working flow chart of Mendelian genetic algorithm

Figure 3 is a Mendelian operator: parent to child;

Figure 4 is a definition of the decision coefficient R ₂ ;

Figure 5 is a raw data preprocessing process and fitness function generation;

Figure 6 is the evolutionary fitting data of the yen against the US dollar;

Figure 7 is the sampling and modeling of data 1 (Mark vs. US dollar);

Figure 8 is the sampling and modeling of data 2 (yen against US dollars);

Figure 9 is a plot of β3 versus R ² (β ₃ VS.Δlog(DM/USD) when (β ₁ =-4.9297, β ₂ =0.2211))

Figure 10 is a plot of β3 vs. R ² (β ₃ VS.Δlog(JPY/USD) when (β ₁ =-4.9625, β ₂ =0.3185))

Detailed ways

In recent years, many fields of literature have proposed the application of genetic algorithms using Mendel's law [14-19]. This application proposes a new model based on Mendelian genetic algorithm, which differs from the standard genetic algorithm and previous research in the following aspects:

(1) In the present application, the Mendelian operator is inserted after selecting the operator, which can effectively utilize the advantage of the Mendelian operator's local search ability, as shown in FIG. In previous studies [16-19], the general Mendelian operator was often inserted after the mutation operator. Based on the mutation probability Pm, genetic mutations may produce unstable populations, which will result in contamination of the entire genetic process output [20, 21]. Typically, mutation operators are randomly generated by changing a binary bit from '0' to '1', and vice versa. The basic method of mutation is the ability to recombine the improved solution at a given probability, but it may also damage the dominant population and lose good solutions and convergence [22-24]. The Mendelian operator will amplify this unstable community from the perspective of micro-evolution through its local search capabilities.

(2) Mendelian genetic algorithm is characterized by Mendelian operators, which are easily synchronized with multi-objective GA processes, such as multi-objective genetic algorithm (MOGA) [25], niche genetic algorithm (NPGA) [26], Non-dominated sorting genetic algorithm (NSGA) [27] Non-dominated sorting genetic algorithm II (NSGAII) [28].

(3) The standard genetic algorithm is based on Darwin's theory, which is characterized by differential survival and successful reproduction. In Mendelian genetic algorithm, the equal gametes pointed out by Mendel's law are combined in the process of randomly forming equal-proportioned fertilized eggs and equal-probative reproductive plants, which also run through all stages of the life cycle.

As shown in Figure 1, the exchange rate analysis method based on Mendelian genetic algorithm has the following advantages: (1) mobilizing computing resources; (2) reducing computational costs.

The determination of exchange rate values has long been considered the most challenging of trading applications for high frequency time series. [3-5, 29, 30] In order to provide investors and researchers with more accurate predictions, some different models are described as the phenomenon that prices follow random walks, which is suitable for random and non-linear search capabilities. Genetic algorithm. The Mendelian genetic algorithm will be applied to the empirical analysis of exchange rate values, which can provide an evolutionary and computational method to solve the problem of exchange rate value determination. Specifically, attempts are made to compare Mendelian genetic algorithms with traditional estimation methods, such as ordinary least squares (OLS) or linear least squares (LS) estimates. OLS and LS are methods used to estimate unknown parameters in a linear regression model. These methods minimize the sum of the squared distances between the data sets in the observed response and predict the response of the linear approximation. Compared with OLS or LS, Mendelian genetic algorithm can process linear and nonlinear models through evolutionary process, and has high complexity, and it is elastic and can be used as a positive optimization solution method to carry out between models. Conversion.

1. Exchange rate determination model

In 2001, Killeen et al. [4] found a cointegration relationship between the exchange rate and the cumulative order flow (COF), which is the closest determinant of the price of all microstructure models. In 2003, Payne [5] used non-standard vector autocorrelation to examine the causal relationship between order flow and exchange rate.

The order flow is defined as the net buyer and seller's order flow currency transactions, which is used as a measure of net buying pressure [3]. According to previous reports [1, 4, 5, 31, 32], order flows are closely related to a broad set of current and expected macroeconomic fundamentals, and such order flows are used as a useful predictor of exchange rate movements. .

The microstructural approach uses an order flow to proxy information that reflects exchange rate fluctuations. Obstfeld and Rogoff [1] conducted an in-depth study in 2000, indicating that the order flow contains information about exchange rate movements. From the traditional exchange rate theory, the exchange rate can be expressed as the discounted present value of the current and expected basic factors (see Equation 1). [1,4,5,32,33]

Where, S _t is the spot exchange rate, which is defined as the domestic price of the foreign currency, b is the discount factor; q is the future periodic factor, f(x _t ) is the basic factor at time t; x _t is simultaneously engraved Order flow

For the market decision makers' expectations of the basic factors of the q-time at t time, there are information conditions at time t.

Specifically, the relationship between the spot exchange rate S _t and the order flow x _t at the same time can be written as follows, by iterating (2).

among them,

Δ is the first-order difference of the sequence; ε _t is the interference term, as shown in equation (3) [34], which indicates that the future exchange rate change is a function of the gap between the spot exchange rate and the expected basic factor. This is also an expected change in the fundamental factors in the long run.

The increase in demand for foreign currency purchases for foreign currencies has led to an increase in foreign currency and a depreciation of the domestic currency, and the spot exchange rate has increased. We predict that the spot exchange rate is positively correlated with the corresponding order flow.

2. Genetic algorithm using the Mendelian principle

Mendel's principles (some biologists refer to Mendel's principles as rules) apply to genetic inheritance in traditional natural environments. Mendel's conclusions can be summarized in two principles for these pea experiments. [14,15,35]

(i) Principle of separation - In the somatic cells of living organisms, genetic factors controlling the same trait are paired and not fused; when gametes are formed, the paired genetic factors are separated, and the isolated genetic factors enter different In gametes, the gametes are inherited to the offspring.

(ii) Free combination principle - The separation and combination of genetic factors controlling different traits do not interfere with each other; in the formation of gametes, the paired genetic factors that determine the same trait are separated from each other, and the genetic factors that determine different traits are freely combined.

As shown in Figure 3, we have included the overall evolutionary framework of the Mendelian genetic law in GA, which can accurately describe the genetic process of cell division in the process of genetic algorithm from the perspective of micro-evolution. A Mendelian genetic algorithm can be defined as a Mendelian genetic operator introduced into the genetic algorithm, and then the optimization study and parameter estimation are applied to establish a regression model for the determination of exchange rate values.

3, Mendelian operator

In this paper, the encoding/decoding operation of the Mendelian genetic algorithm utilizes a binary encoding method. Normally, all bits (genes) of the chromosome can use two binary-encoded letters '1' and '0', where the gene represented by "1" is activated and '0' is invalid.

As shown in Figure 3, '0' and '1' are the binary values of the two bits, the length of the chromosome, or the binary number, which are all Mendelian genetic operators, known as the percentage of Mendelian age ( MP), where MP is a scale factor that includes the length of the balanced parent chromosome binary number. The Mendelian operator produces chromosomes Xo1 and Xo2 from the descendants of the parental chromosome based on the properties of each binary number, which is defined by Punnet-aquare [14, 15]. The two parental chromosomes Xp1 and Xp2 can generate Xo1 of a sub-chromosome, as shown in Figure 3. For the second progeny of the Xo2 chromosome obtained, the production process needs to be repeated.

As shown in Table 1, in Punnet-square for the Mendelian genetic operator, each chromosomal bit is assigned an attribute that includes the type of the corresponding character of the gene. There are three types of genes: D (dominant, pure and dominant), R (recessive, pure and recessive) and H (mixed, hybrid). attrP1 and attrP2 are attributes of a binary position of the parental chromosomes Xp1 and Xp2; attrO is a property of a binary bit of the descendant chromosome Xo1 or Xo2.

Table 1. Parallel table of chromosome 1

4. Definition of fitness function

4.1. Goodness of fit for measurement - decision coefficient

The coefficient of regression analysis is the study of the functional relationship between two or more variables, where the regression model used in the decision coefficient (R2) is a quantitative measurement of the AC count, passing the model in Figure 4 [30, 36] , and larger R2 values (close to uniform) are considered to be better variability data, as given by equation (4).

among them,

R ² is the two given data sequences y _i and

The decision coefficient, as shown in Figure 5, y _i is the implementation data,

For dotted data, y _i is the observation data for the exchange trade [37];

For the sample data, the estimation process from Mendel-GA;

Is the average data of the observed data y _i ; k is the sample size of two given data sequences; SS _tot total corrected square sum, as shown in equation (5).

among them,

SS _err is the sum of squared residuals, as shown in equation (6) [30, 36], which is the goal of traditional OLS estimation. The purpose of the OLS method is to minimize SS _err .

At the same time, R ² can also measure the variance ratio of independent variables [38], therefore, by comparing the explained variance

Total variance

Equation (4) can be rewritten as equation (7) from the perspective of variance.

among them,

As the explanatory variable, as a variable in the model prediction, as shown in equation (8).

For the total variable, as shown in equation (9).

4.2 The fitting function of the regression model

In order to evaluate the performance of the Mendelian genetic algorithm with existing exchange rate and COF functions, R ² is borrowed from the definition of the fitness function as an observation data sequence.

And a measurement statistic of the protocol data sequence between the samples y _i generated by the Mendelian genetic algorithm. Using the basic idea applied by R ² , the relationship between exchange rate and COF can be described by equation (10). The fitness function defined in this paper is shown in equation (11). The purpose is to find the best parameters by using Mendelian genetic algorithm. Bring R ² close to maximum.

Slope coefficient and historical data, where Xi is defined as the COF value observed for a foreign exchange, and Yi is defined as the ratio of foreign exchange to the US dollar through log(/). Equations (12) and (13) are the conversion of the German mark to the US dollar and the Japanese yen against the US dollar.

Δx _i (t)=x _i (t)-x _i (t-1) (14)

Δy _i (t-1)=y _i (t-1)-yi ₍ t-2) (15)

Can be seen by equation (4), numerical simulation passed

with

A small error is produced which may result in a loss of accuracy. For the protection of the calculation accuracy, the accuracy of the numerical error can be scaled to an acceptable range of values by the scale factor of the scaling. The scale factor can be selected one by one according to the specific situation. As shown in Figure 6, in order to reduce the possibility of numerical simulation errors, the data sequence y _{i is} always scaled multiple times to a quantity. In the case of DM versus USD and JPY against the US dollar, a scaling factor (constant) of 10,000. Then, a differential data sequence of Δx _i (t) and Δy _i (t) is generated, and the next step is to generate a fitting function by equation (11).

1. Empirical results and conclusions

Table 2. Empirical parameters of the Mendelian genetic algorithm

The original exchange rate data was collected by Evans and Lyons in 2002, with two sets of transaction data:

The empirical results use a specially designed MATLAB simulation toolbox based on Mendelian genetic algorithm (the simulation toolbox is the toolbox obtained by the Mendelian genetic algorithm based on the exchange rate analysis method in this application), and then unified by SGALAB. Description [39]. Unless otherwise stated, all results were generated by the parameters of the following genetic algorithms, as shown in Table 2, which used binary encoding/decoding, league selection, single point crossing and mutation in the Mendelian genetic evolution process. The results of the Mendelian genetic algorithm are also compared with those of the standard genetic algorithm and the OLS method.

As shown in Table 2, the total number of experiments is 1000. For the fitting results, the maximum, minimum, and average values of the fits are used to show the single-run performance of the Mendelian genetic algorithm (fitmaxi, fitmini, and fitmeani). The Mendelian genetic algorithm runs 1000 times of total results, and the average data for fitmaxi, fitmini and fitmeani for all experiments is shown in the figure.

At the same time, for all 1000 experiments of the Mendelian genetic algorithm, the E(*) and variance VAR(*) of the maximum fitness values of all experiments can be used as indicators to evaluate the overall performance of the Mendelian genetic algorithm, as shown in the table. Shown in 3-6.

Figures 5 and 6 depict the fitted values of Data-I (DM vs. US Dollar) and Data-II (Japanese Yen vs. US Dollar), respectively. Throughout evolution, time is defined as the maximum population algebra, and the fitness is set from a rapidly growing state to a stable state under the initial parameters set forth in Table 2. The solid lines in Figures 5 and 6 represent the average of the total fitness, the fitness value is marked as "+", the data at the top is the maximum of the average total fitness, and the data at the bottom is marked as '+'. The minimum value of the average total fitness.

The results of the mean and variance of Data-I (DM vs. USD) are shown in Tables 3 and 4. Specifically, in Table 3, E[β1], E[β2], and E[β3] in the Mendelian genetic algorithm, the standard GA, and the OLS algorithm are all similar. In the comparison of Mendelian genetic algorithm, standard genetic algorithm and OLS E[R ² ], the Mendelian genetic algorithm E[R ² ] is the largest, and the standard genetic algorithm E[R ² ] is the second largest. OLS has the smallest E[R ² ], which indicates that the Mendelian genetic algorithm is superior to the other two.

Table 4 shows the VAR[β1], VAR[β2] and VAR[β3] of the Mendelian genetic algorithm, the standard genetic algorithm and the OLS algorithm, which are slightly different. The VAR[R ² ] of the Mendelian genetic algorithm is smaller than the standard GA. Genetic algorithms, which means that the results of the Mendelian genetic algorithm stay in a smaller dispersion range than the standard genetic algorithm. The VAR[*] value of the OLS algorithm is zero because it is not solved by traditional numerical methods, and the results are the same in every experiment.

Similarly, the results of the mean and variance of Data-II (Japanese yen versus US dollar) shown in Tables 5 and 6, can be used to prove that Mendelian genetic algorithm is better than standard genetic algorithm and OLS method. Search ability.

As shown in Fig. 7 and Fig. 8, the data points represented by 'o' are from the sample data, and the data points represented by '*' are autoregressive models as given in equation (10), which all show how the regression model Estimate exchange rate trading behavior. Estimates for β1, β2 and β3 are given in Tables 3 and 5. According to the simulation results obtained by the Mendelian GA method, β1 and β2 are in a relatively stable state, and β3 shows how the historical data of Δy _i (t-1) is estimated.

Figures 9 and 10 show how β3 affects R ² , which measures the performance of the regression model. As shown in Tables 3 and 5, the coefficient of determination is improved compared to the results of Evans and Lyons [37]: where, for Mark/USD, the coefficient of determination is increased to 60% → 64%.

For the yen/dollar, the coefficient of determination has improved a little, from 40% to 40.73%.

Table 3. Average of the Mendelian genetic algorithm for data 1 (Mark vs. US dollar)

Table 4. Variance values under the Mendelian genetic algorithm for data 1 (Mark vs. US dollar)

Table 5. Average of the Mendelian genetic algorithm for data 2 (yen against the US dollar)

Table 6. Variance values under the Mendelian genetic algorithm for data 2 (Japanese yen versus US dollar)

Comparing the results of Mendelian genetic algorithm, standard genetic algorithm and OLS algorithm, it shows that Mendelian genetic algorithm is superior to standard genetic algorithm and OLS.

references

[1]M.Obstfeld, K. Rogoff, The six major puzzles in international macroeconomics: is there a common cause? , NBER Working Papers 7777, 2000.

[2] K. Rogoff, The purchasing power parity puzzle, Journal of Economic Literature 34 (1996) 647–668.

[3] R. Lyons, The Microstructure Approach to Exchange Rates, MIT Press, London, England, 2001.

[4] W. P. Killeen, R. K. Lyons, M. J. Moore, Fixed versus flexible: lessons from ems order flow, Journal of International Money and Finance 25 (2006) 551–579.

[5] R. Payne, Informed trade in spot foreign exchange markets: an empirical investigation, Journal of International Economics 61 (2003) 307–329.

[6]T.H.Hann, E.Steurer, Much ado about nothing? Exchange rate forecasting: neural networks vs. linear models using monthly and weekly data, Neurocomputing 10 (1996) 323–339.

[7] M. Qi, Y. Wu, Nonlinear prediction of exchange rates with monetary fundamentals, Journal of Empirical Finance 10 (2003) 623–640.

[8] R. N. Yadav, P. K. Kalra, J. John, Time series prediction with single multiplicative neuron model, Applied Soft Computing 7 (2007) 1157–1163.

[9] S. Rimcharoen, D. Sutivong, P. Chongstitvatana, Prediction of the stock exchange of thailand using adaptive evolution strategies, in: Proceedings of the 17th IEEE International Conference on Tools with Artificial Intelligence, IEEE Computer Society, Washington, DC, USA, 2005.

[10] C. Worasucheep, P. Chongstitvatana, A multi-strategy differential evolution algorithm for financial, Prediction with Single Multiplicative Neuron, ICONIP 2009, Part II, LNCS, vol. 5864, 2009, pp. 122–130.

[11] J. Nenortaite, R. Simutis, Stocks’ Trading System Based on the Particle Swarm Optimization Algorithm, Workshop on Computational Methods in Finance and Insurance, Lecture Notes in Computer Science, vol. 3039/2004, Springer, 2004.

[12] L. Zhao, Y. Yang, Expert systems with applications: pso-based single multiplicative neuron model for time series prediction, Expert Systems with Applications 36 (2009) 2805–2812

[13] J. H. Holland, Adaptation in Natural and Artificial Systems, The University of Michigan Press, 1975.

[14] G. Mendel, Experiments in Plant Hybridization, 1865. < www. mendelweb. org/ Mendel. html>.

[15] D. O' Neil, Mendel's Genetics, 2009. < http: //anthro. palomar. edu/mendel/ mendel_1.htm>.

[16] I. S. Song, H. W. Woo, M. J. Tahk, Agenetic algorithm with a Mendel operator for global minimization, in: Proceedings of the 1999 Congress on Evolutionary Computation, Washington, DC, USA, vol. 2, 1999, pp. 1521–1526.

[17] C. S. Park, H. Lee, H. C. Bang, M. J. Tahk, Modified mendel operation for multimodal function optimization, in: Proceedings of the 2001 Congress on Evolutionary Computation, Seoul, South Korea, vol 2, 2001, pp. 1388–1392.

[18] B. Anthony Kadrovach, S. R. Michaud, J. B. Zydallis, G. B. Lamont, B. Secrest, D. Strong, Extending the simple genetic algorithm into multi-objective problems via mendelian pressure, 2001 Genetic And Evolutionary Computation Conference, San Fran Mendelian Genetic Algorithm Mendelian Genetic Algorithms, California, 7–11 July, vol. 1, 2001, pp.181–188.

[19] BAK Adrovach, JBZydallis, GB Lamont, Use of mendelian pressure in a multi-objective genetic algorithm, in: Proceedings of the 2002 Congress on Evolutionary Computation, Piscataway, New Jersey, 12–17 May, vol. 1, 2002, pp .962–967.

[20] H. Ishibuchi, T. Yoshida, T. Murata, Balance between genetic search and local search in memetic algorithms for multiobjective permutation flowshop scheduling, IEEE Transactions on Evolutionary Computation 7 (2003) 204–223.

[21] Victoria Furió, Andrés Moya, Rafael Sanjuán, The cost of replication fidelity inan RNA virus, Proceedings of the National Academy of Sciences 102 (2005) 10233–10237.

[22] E. Zitzler, Evolutionary Algorithms for Multiobjective Optimization: Methods and Applications, Diss. ETH No. 13398, ETH Zurich, Switzerland, 1999.

[23] The National Academy of Sciences (NAS), Teaching about Evolution and the Nature of Science, National Academy of Sciences Press, 2004.

[24] R.L. Haupt, S.E. Haupt, Practical Genetic Algorithms, second ed., John Wiley & Sons, Inc., 2004.

[25] CMFonseca, PJFleming, Genetic algorithm for multi-objective optimisation: formulation, discussion and generalization, in: S. Forrest (Ed.), Genetic Algorithm: Proceedings of the Fifth International Conference, vol. 1, Morgan Kaufmann, San Mateo, CA, 1993, pp. 141–153.

[26] JNHorn, DENGoldberg, A niched pareto genetic algorithm for multi-objective optimisation, in: Proceedings of the First IEEE Conference on Evolutionary Computation, IEEE World Congress on Computational Intelligence, 27–29 June, vol. 1, 1994, pp .82–87.

[27] N. Srinivas, K. Deb, Multiobjective optimisation using nondominated sorting in genetic algorithm, Evolutionary Computation 2 (1994) 221–248.

[28] K. Deb, A. Pratap, S. Agarwal, T. Meyarivan, A fast and elitist multiobjective genetic algorithm: NSGAII, IEEE Transactions on Evolutionary Computation 6 (2002) 182–197.

[29] P. Tenti, Forecasting foreign exchange rates using recurrent neural networks, Applied Artificial Intelligence 10 (1996) 567–581.

[30] D. N. Gujarati, Basic Econometrics, fourth ed., McGraw-Hill Companies, 2004.

[31] S. Reitz, M. A. Schmidt, M. P. Taylor, End-user order flow and exchange rate dynamics, The European Journal of Finance 17 (2011) 153–168.

[32] D. Rime, L. Sarno, E. Sojli, Exchange rate forecasting, order flow, and macroeconomic information, CEPR Discussion Paper No. DP7225, 2009.Available at SSRN: < http: //ssrn. com/abstract= 1372545> .

[33] T. Andersen, T. Bollerslev, F. X. Diebold, C. Vega, Microeffects of macroannouncements: real-time price discovery in foreign exchange, American Economic Review 93 (2003) 38–62.

[34] P. Bacchetta, E. Wincoop, Can information heterogeneity explain the exchange rate determination puzzle? , American Economic Review 96 (2006) 552–576

[35] J. Marks, The construction of mendel’s laws, Evolutionary Anthropology 17 (2008) 250–253.

[36] D. C. Montgomery, G. C. Runger, Applied Statistics and Probability for Engineers, third ed., John Wiley & Sons, Inc., 2003.

[37] M. Evans, R. Lyons, Order flow and exchange rate dynamics, Journal of Political Economy 110 (2002) 170–180.

[38] M. P. Allen, Understanding Regression Analysis, Springer, 2007.

[39] Y. Chen, Simple Genetic Algorithm Laboratory Toolbox for MATLAB, 2009.

< http: //www. mathworks. co. uk/matlabcentral/fileexchange/5882 >

Claims

A method for analyzing exchange rate based on genetic algorithm includes the following steps:

The first step is to preprocess the exchange rate price data;

In the second step, the pre-processed exchange rate price data is input into the exchange rate model, and the fitness function is used to optimize the exchange rate model:

The third step is to verify the optimized result, verify the data result, verify that it has not passed, and then perform the Mendelian genetic algorithm processing and then repeat the second step and the third step until the optimized result is verified and the result is output.
The genetic algorithm-based exchange rate analysis method according to claim 1, wherein the method for optimizing the exchange rate model by the fitness function in the second step is as follows:

Determine the exchange rate model and use the order flow to determine the exchange rate, as in (2)

Where Δ is the first-order difference of the sequence; S t+1 is the exchange rate of t+1 time, and S t is the spot exchange rate, which is defined as the domestic price of the foreign currency, see equation (1); ε t+1 is The interference term, as shown in equation (3), indicates that the future exchange rate change is a function of the gap between the spot exchange rate and the expected basic factor; b is the discount factor; f(x t ) is the basic factor at time t; x t is a simultaneous order flow;

q is the future periodic factor,
For the market decision maker's expectation of the basic factors of the q-time at t time, there is an information condition at time t, and t+q is a time period.
The genetic algorithm-based exchange rate analysis method according to claim 1, wherein the genetic algorithm based on the Mendelian operator is used to perform an iterative calculation on the formula (2):

A, initialize each parameter;

B. Encode all variables;

C. Perform a selection operation of the genetic algorithm;

D, using the Mendelian operator for calculation;

E. Perform cross operation of the genetic algorithm;

F. performing a mutation operation of the genetic algorithm;

H. Decoding: Repeat the iterative AH step to obtain the optimal solution, and use the adaptive function to evaluate whether the optimal solution after decoding meets the requirements of the optimization calculation. If the requirements are met, the final optimized solution set is obtained; otherwise, the operating parameters are modified and recalculated. Until the final optimization solution set is obtained.
The genetic algorithm-based exchange rate analysis method according to claim 3, wherein the encoding is a binary encoding.
The genetic algorithm-based exchange rate analysis method according to claim 3, wherein the fitness function for evaluating the decoded optimal solution is expressed by equations (10) and (11), and the obtained optimal solution enables R to be obtained. 2 When the maximum is reached, the requirements of the optimization calculation are met:

Where β 1 , β 2 and β 3 are 3 coefficients, respectively, xi is defined as a COF value observed in foreign exchange (cumulative order flow), and yi is defined as the foreign exchange and local currency obtained by log(/). proportion. The national currency is as follows:

Y i =log (foreign exchange / local currency) × 10000

Δx i (t) and Δy i (t-1) are differential data sequences obtained during data preprocessing at times t and t-1, as shown in equations (14) and (15).

Δx i (t)=x i (t)-x i (t-1) (14)

Δy i (t-1)=y i (t-1)-y i (t-2) (15)
The genetic algorithm-based exchange rate analysis method according to claim 3, wherein the calculation method of the Mendelian operator is as follows;

Each chromosomal position in the parent population is assigned an attribute, and there are three attribute types: D attribute represents dominant, pure and dominant genes, and R attribute represents recessive, pure and recessive genes, H Attributes represent hybrid, hybrid genes;

The two chromosomes in the optional parent population are cross-tested as the parental chromosome to obtain the progeny chromosome. If the attribute of the parental chromosomal location is D, the attribute of the chromosomal position of the offspring is D; if the attributes of the parental chromosomal location are R The attribute of the chromosomal position of the progeny is R; if the attribute of one parent chromosomal position is R and the attribute of the other parent chromosomal position is D, the attribute of the chromosomal position of the progeny is H; if the attribute of a parent chromosomal position is D The attribute of another parental chromosomal location is H, then the chromosomal position of the offspring is 50% D, 50% is H; if the attribute of one parent chromosomal location is R and the attribute of the other parent chromosomal location is H, then the daughter chromosome The position 50% is R, 50% is H; if the parental chromosomal location is H and the other parental chromosomal location is H, then the progeny chromosomal location is 25% R, 25% D, 50% H .