ࡱ> #% !"U@ R]bjbjIvvvvLx[ {(|||,|       $RST ]șșș |,| ZZZș&|,| Zș Z>ZJ\~,|{ `G{\Uvȝt*B+ 0[ l<~T~dՖZE  oNv>NvThe Predictive Power of Short-term Exchange Rate based on ARIMA and Hybrid Models Syouching Lai, Hungchih Li*, Manling Lee Department of Accounting, National Cheng Kung University Tsung-yueh Yang Cheng Shin Company *Corresponding author. Tel.: 886-6-2757575ext53427; fax:886-6-2744104 E-mail: hcli@mail.ncku.edu.tw The Predictive Power of Short-term Exchange Rate based on ARIMA and Hybrid Models Abstracts Since the collapse of Bretton Woods Agreement in 1973, all countries in the world started to accept flexible exchange rates. It is possible that the exchange rate, the value of domestic currency against foreign currency, often moves drastically because of the demand and (or) supply side in the foreign exchange market, which will affect firms profit when firms involve international business activities. Therefore, how to predict future exchange rate correctly is an important mission for multinational corporations. In this study, the Autoregressive Integrated Moving Average (ARIMA) is used to predict short-term exchange rate. In addition, due to the rapid advancement of computer technology, this study also uses Genetic Algorithm (GA) and Back-Propagation Network (BPN) in order to see whether they can help raise predictive power of traditional time series model as suggested by Hu et al. (1999). By analyzing their precision and validity, this study can find which model, traditional time series or hybrid models, time series model in combination with BPN (called ARBPN) or with GA (called ARGA), is the best. The empirical results show that except for pound against US dollar ARGA is better than ARIMA and except for POUND and SF against US dollar ARGA is better than BPN in predictive power based on MAPE. But ARIMA model is not better than BPN except for YEN against US dollar when MAPE is used to measure precision of each model. However, the validity in predicting moving direction of the future exchange rate for ARBPN model is better than that for ARIMA model, although not significantly. In addition, the results show that at significant level 10%, the validity of ARGA model is higher than that of ARBPN model. Keywords: Genetic Algorithm, ARIMA, Back-Propagation Neural Network, Exchange rate Forecasting 1. Introduction Since the collapse of Bretton Woods Agreement in 1973, all countries in the world started to accept flexible exchange rates. Therefore, exchange rate often moves more drastically than before. How to predict future exchange rate correctly is thus an important mission for multinational corporations. This study mainly focuses in forecasting short term exchange rate. The traditional Box-Jenkins Autoregressive Integrated Moving Average (ARIMA) is used to predict short-term exchange rate. Considering rapid advancement of computer technology, this study also use hybrid model by combining ARIMA with Genetic Algorithm (called ARGA) or with Back-Propagation Network (called ARBPN) in order to see whether they can help raise predictive power of traditional time series model as suggested by Hu et al. (1999). The Autoregressive Integrated Moving Average (ARIMA) model considers that the future movement of time series might be affected by its past performance and then analyst can predict the future exchange rates based on the influence of its past exchange rate. The Back-Propagation Network (BPN), one of Artificial Neural Network, is used in this study because Borisov and Pavlov (1995) find BPN performed the best among two neural net work models and two exponential smoothing models. It is a simple simulation of a creature neuron and it can get information from external environment or other neurons and export its result to the external environment or other neurons. The Back- Propagation Network (BPN) doesnt need any assumption, which fits the real world better. As to Genetic Algorithms (GA), they simulate the naturally progressive rules, choose the better parents of species and exchange the genetic data of each other randomly in order to produce the better offspring and finally get the global optimum (Holland, 1975). This study uses four exchange rats including Yen/Dollar, Pound/Dollar, Swiss Franc/Dollar and NT/Dollar. The daily data cover the period from 1990 to 2001. Considering mixed results of previous studies which may result from differences across time periods and the number of observations in training sample, we utilize a moving cross-validation scheme as suggested by Hu et al. (1999). First, a moving cross-validation method with 12 test sets is utilized. This walk-forward testing procedure uses training set based on each year excluding the last two weeks and uses test sets based on the last two weeks of each year. The length of the in-sample period is the same across the 12 training sets. We use each year from 1990~2001 as the in-sample period and last two weeks of each year for the test period. Daily observations for each year from 1990 through 2001 are used as in-sample data in the first validation set. One-step-ahead predictions are made for a period of 10 days (last two weeks of each year). This cross-validation procedure may allow us to see which model, ARIMA, ARGA or ARBPN can adapt to the changing condition of the market quickly. Results from the cross-validation analysis will provide valuable insights on the reliability or robustness of each model with respect to sampling variation. A moving validation scheme with moving windows of fixed length provides an opportunity to investigate the effect of structural changes in a series on the performance of each model. The sample period is divided into 12 in-sample periods in order to examine the predictive power and validity of the predicted changing direction of the future exchange rate for each model. The procedure of rolling Regression is used to forecast the exchange rate. Finally, the Mean Absolute Percentage Error (MAPE) is used to measure the accuracy and the paired t test is used to evaluate the performance of predictive power and validity of chosen models. 2 Literatures Review 2.1 Literatures review about Back Propagation network Artificial neural network is an information system which imitates biological neural network. There are many different kinds of artificial neural network, but the most common used one is Back-Propagation network (BPN). BPN can minimizes Energy function to supervise the adjustment of weighted values in network learning process and set up network structure which can translate an input value into a presumed output value very close to a real output value (Borisov and Pavlov, 1995). Therefore, in this study, BPN is used to forecast exchange rate. Borisov and Pavlov (1995) applied neural networks to forecast the Russian ruble exchange rate. Two neural network models and two exponential smoothing models are used to predict the exchange rate. A backpropagation-based neural network performs the best in all cases although it consumes more time to get the results. Wu (1995) compares neural networks and ARIMA models in forecasting the Taiwan/U.S. dollar exchange rate and finds Neural networks perform significantly better than the best ARIMA models in both one-step-ahead and six-step-ahead forecasting. Zhang and Hutchinson (1994) forecast the direction of change in exchange rate by employing a coding system of +1 (appreciation), -1 (depreciation), and 0 (no change). They find mixed results for neural networks in comparison with those from the random walk model. Verkooijen (1996) reports the results for U.S. dollar/Deutsche mark exchange rate forecasting by using neural networks and linear models. Using monthly data, he finds that the neural network perform closely to the linear models in out-of-sample forecasting. However, neural networks are better than linear models and random walk models in terms of the percentage of correctly predicted signs. Hann and Steurer (1996) compare neural network models with linear monetary models in forecasting the U.S. dollar/Deutsch mark exchange rate. Based on the out-of-sample results, they find that for weekly data but not for monthly data, neural networks are much better than linear models, which might result from the fact that weekly data contain nonlinearities whereas monthly data do not. The mixed results about the performance of the neural network based on out-of-sample might be due to several possible explanations. One reason is likely to be a result of variation in the time frame and the number of observations used. The other reason might be due to the Differences in the length of forecast horizon. Also different measures such as absolute and relative performance are used, which might be another reason explaining the mixed results found in previous studies. 2.2 Literatures review about Genetic Algorithm Genetic Algorithms (GA) proposed by Holland(1975) are the best seeking mechanism during natural choosing process. Basic spirit of GA is to simulate the natural progressive rule of biosphere. It can choose parents which have better characteristics among all species and interchange randomly mutual genetic information so as to product better offspring than its parents. The above process will be repeated continuously in order to product the best species. Neely et al. (1997) use GA to seek for the best technical analysis. They adopted DM/JPY, pound/Swiss franc, U.S. dollar/DM, U.S. dollar/JPY, U.S. dollar/Swiss franc and U.S. dollar/pound with data of 1981 to 1995. The result was that the strategy acquired by using GA had better performance in most foreign currency market. Neely and Weller (2002) adopt exchange rates such as U.S. dollar/DM and U.S. dollar/JPY with GA,GARCH and RiskMetrics model to predict the volatility of foreign currency markets. The judgment standards were MSE, MAE and R2. The result showed that GA had better performance than GARCH and RiskMetrics. Leigh et al. (2002) compare neural network configuration found by the genetic algorithms search with neural network in predicting exchange rates using data of 1981 to 1997. The procedure of the neural network configuration found by GAs search model is that first, they used GA in order to lessen the input variables and then used the remaining variables as input variables of neural network. Their result showed that the performance of the neural network configuration found by GAs search model was better than that of neural network model. 3. Data and Methodology In this study, Three models including the Autoregressive Integrated Moving Average (ARIMA ), ARIMA combining with Back Propagation network (called ARBPN) and ARIMA combining with Genetic Algorithms (called ARGA) are used to forecast the futures exchange rate. We focus on one-step-ahead forecasts as in Diebold and Nason (1990). The exchange rate values are forecasted one step ahead of time and the actual rather than the forecasted values are then used for the next prediction in a forecasting horizon (see figure 1 for a schematic diagram). One-step-ahead forecasting is useful for evaluating the robustness of a forecasting technique.  SHAPE \* MERGEFORMAT  3.1 Autoregressive Integrated Moving Average (ARIMA) ARIMA, proposed by Box and Jenkins in 1970, is a forecasting model of time series. A complete ARIMA model includes 3 parts, Autoregressive terms (AR), Integrated (I) and Moving average terms (MA). When analyzing time series of a set of data, this study uses Box-Jenkins method to get p, d and q, which includes three steps: Identification, Estimation and Diagnosis. A. Identification: This step is to estimate a model which data set are likely to be and to decide the number of p, d and q. At first, this study tests whether data set is stationary or not. If not, this study should take difference (starting from first order difference) of the data set until the data set becomes stationary. The order of integrated degree, d, is regarded as 1 if the nonstationary series become stationary after first order difference. Two Unit Root tests, DF and ADF, proposed by Engle and Yoo (1987) are used to test whether the data is stationary or not. This study uses q as the lag period of moving average of error term and uses p as the lag period of autocorrelation and uses ACF and PACF charts to seek for the order of p and q. Estimation: After deciding the order of p and q, parameters are then estimated by using MLE method. Diagnostic Checking: After identification and estimation, this study continues to check whether the error terms still have serial correlations. If not, ARIMA model is desirable and can be used to predict exchange rates. On the contrary, the model should be re-estimated. This research uses Q statistic proposed by Box and Pierce (1970) to check whether the error terms of this model fit white noise. If Q > X, then the model is not desirable and should be re-estimated; otherwise, the model is accurate and can be used to predict future exchange rates. 3.2 Back-Propagation Network Artificial neural network uses large but simple inter-connected neurons to simulate the ability of the organism neural network. An artificial neuron receives and calculates the data which it collected from other artificial neurons or external environment and then outputs results to other artificial neurons or external environment. The fundamental factors of an artificial neural network are processing element, layer, network connection, network, which are introduced as follows.  SHAPE \* MERGEFORMAT  Feed-forward neural network with back-propagation learning is the most conventional sort of neural network. The feed-forward neural network computes input-to-output mappings on the basis of calculations occurring in a system of interconnected nodes, arranged in the form of layers. The output of each node is calculated as a nonlinear function of the weighted sum of inputs from the nodes in a layer which precedes it in computation order. Back-propagation employs a gradient-descent search method to find weights that minimize the global error from the error function. The error signal from the error function is propagated back through the network from output layers to make adjustments on connection weights that are proportional to the error. The process limits overreaction to any single and potentially inconsistent data item by making small shifts in the weights. In this study, Back-Propagation Network (BPN) with 1 hidden layer was applied. The TanH function is the appropriate transfer function because its value range is from -1 to 1 as suggested by Coakley and Brown (1999). In addition, the Norm-Cum Delta Rule is adopted as learning rule. 3.3 The Genetic Algorithms Model GA are invented by Holland (1975) to mimic some of the processes of natural evolution and selection. GA are implementations of various search paradigms inspired by natural evolution. GA involves a two-step process. It starts with a current population. Selection is applied to the current population to create an intermediate population. The next population is a result of genetic manipulation through crossover and mutation. The process of going from the current population to the next population represents a generation execution of GA. Crossover occurs when there is an exchange of genes between chromosomes. Mutation consists of randomly replacing some of the chromosomes' genes with genes that are not represented in the chromosomes and its role is to restore the lost genetic material. The chromosomes are evaluated after each cycle using the fitness function. Generation of new populations is repeated until a satisfactory solution is identified, or a specific termination criterion is met. The GA consist of the following main components: (1) Chromosomal representation. Each chromosome stands for a legal solution to the problem and is composed of a string of genes. The Binary alphabet {0,1} is often used to represent these genes, but sometimes integers or real number are used according to the application. (2) Initial population. Initial population can be created randomly or by using specialized problem-specific information. From empirical studies over a wide range of function optimization problems, a population size of between 30 and 100 is usually recommended. (3) Fitness evaluation. Chromosomes are tested for suitability in order to satisfy the fitness function. As the algorithm proceeds, we would expect the individual fitness of the best chromosome to increase as well as the total fitness of the population as a whole. The fitness function in this study is set by minimizing MAPE (Mean Absolute Percentage Error) as follows:  Min MAPE = where  is the t-th period predicted exchange rate St is the t-th period exchange rate. N is number of sample (4) Selection. It is necessary to select a chromosome from the current population for reproduction. If the size of the population is 2n and n is some positive integer value, through the selection procedure two parent chromosomes are picked to produce two offspring for a new population based on their fitness values. The higher the fitness value is the higher probability of that chromosome being selected for reproduction. (5) Cross-over and mutation. Once a pair of chromosomes have been selected, cross-over can take place to produce offspring. A cross-over probability of 1.0 indicates that all selected chromosomes are used in reproduction. The empirical studies have shown that better results are achieved by a cross-over probability between 0.65 and 0.85. If only the cross-over operator is used to produce offspring, one potential problem may arise. The problem is that if all the chromosomes in the initial population have the same value at a particular position, then all future offspring will have this same value at that position. To combat this undesirable situation, a mutation operator is used. This attempts to introduce some random alteration of the genes. The cross-over probability is 0.9 and the mutation probability is 0.05 in this study. Procedures mentioned above complete one cycle of GA and are repeated until either an optimal, or suitable suboptimal has been found or the maximum number of generations has been exceeded. 3.4 The t Test Except using MAPE to examine forecasting performance, this study also uses the t test instead of the Z test to examine whether overall forecasting performance of one model is better than that of the other model over 12 test periods. The null and alternative hypotheses are as below: H0MAPE of model 1 is not significantly different from that of model 2 H1MAPE of model 1 is significantly different from that of model 2 In addition, this study also uses the t test to examine the validity of each model in predicting the direction of future exchange rate over 12 test periods. The null hypothesis and alternative hypothesis are as below H0There is no significantly different between two models in predicting the direction of future exchange rate H1There is significantly different between two models in predicting the direction of future exchange rate 4. Empirical Results To evaluate the predictive power and the validity in predicting the direction of future exchange rate among three models, ARIMA, Genetic algorithms (GA) and Artificial neural network (ANN), we first introduce the data source and disposal process as follows. Table 1 Data source and disposal process Exchange RateData SourceOut-of-the sample PeriodNT/USDAREMOSThe last 10 days of each yearYEN/USDBoard of Governors of the Federal Reserve SystemThe last 10 days of each yearPOUND/USDBoard of Governors of the Federal Reserve SystemThe last 10 days of each yearSF*/USDBoard of Governors of the Federal Reserve SystemThe last 10 days of each year*SF is Swiss Franc. The ARIMA model When building up ARIMA (p,d,q) model, this study has to confirm whether the time series is stationary or not at first. If not, it should take difference of the time series until it becomes stationary. In order to avoid overdifferencing problem, it also uses two Unit Root test(called DF and ADF) proposed by Engle and Yoo (1987) to examine the stationarity of the exchange rates. All exchange rates are stationary after taking first order difference. After checking DF and ADF, this study also uses ACF and PACF to find out p and d of ARIMA (p,d,q). The ARIMA models of each exchange rate for each year are listed in table 2. The predicting results of ARIMA, ARIMA in combination with Genetic Algorithms (called ARGA) and Back-propagation Network (called ARBPN) This study finds that when out-of-the-sample of 10, 20 and 40 days are examined, 10 days out-of-the-sample can have the best forecasting results, which is similar as the findings by Hu et al. (1999) that the longer the forecast horizon is, the worse the chosen models perform. This study thus use 10 days out-of-the-sample. The MAPE results using the chosen ARIMA models for each year from 1990 to 2002 are listed in table3. Table 2 ARIMA(p,d,q) model of exchange rates NT/USDYEN/USDPOUND/USDSF/USD1990Random walkRandom walk(1,1,0)Random walk1991(1,1,0)Random walkRandom walkRandom walk1992Random walkRandom walk(1,1,0)Random walk1993(1,1,0)Random walkRandom walkRandom walk1994Random walk(2,1,0)Random walkRandom walk1995(1,1,0)Random walk(2,1,0)Random walk1996(2,1,0)Random walkRandom walkRandom walk1997(2,1,0)Random walkRandom walkRandom walk1998(1,1,0)Random walk(2,1,0)Random walk1999Random walkRandom walkRandom walkRandom walk2000(2,1,0)Random walk(2,1,0)Random walk2001Random walkRandom walkRandom walkRandom walkNote: SF is Swiss Franc Table 3 MAPE of ARIMA periodNT/USPOUND/USSF/USYEN/US19900.04 0.730.700.4119910.020.500.630.4919920.030.380.530.2319930.130.420.710.3319940.130.280.330.2019950.050.440.350.3419960.020.540.430.3219970.280.470.460.6719980.040.260.550.5819990.080.210.340.4620000.090.450.780.3820010.070.370.610.36AVG0.08170.42080.53500.3975Note: SF is Swiss Franc, unit:% Then, using the parameter of chosen ARIMA model of each year for each exchange rate, we continue using GA (called ARGA) and BPN (called ARBPN) models to forecast future exchange rate to see if the predictive power can be raised up or not. The results of GA and ANN are listed in table 4 and table 5 respectively. Table 4 MAPE of ARGA NT/USPOUND/USSF/USYEN/US19900.050.710.700.3819910.020.490.620.4619920.030.410.430.2319930.130.410.690.3319940.130.280.330.1919950.030.430.340.3319960.020.530.410.3319970.220.470.450.6619980.040.250.550.5719990.080.210.330.4320000.070.440.750.3620010.060.370.590.26AVG0.07330.41670.51580.3775Note: SF is Swiss Franc, unit:% Table 5 MAPE of ARBPN NT/USPOUND/USSF/USYEN/US19900.060.770.770.6919910.010.500.660.5919920.050.620.340.2619930.170.410.700.3719940.130.300.340.1519950.030.370.340.4319960.020.520.520.3319970.180.470.450.7319980.130.230.530.6519990.120.210.440.6220000.130.500.770.6020010.120.350.580.57AVG0.09580.43750.53670.4992Note: SF is Swiss Franc, unit:% To examine whether there exists significant difference among ARIMA, ARGA and ARBPN in predictive power, we use paired t test. The results in table 6 indicate that except for two exchange rates, NT/US and POUND/US, ARGA has higher predictive power than ARIMA in other two exchange rates, SF/US and YEN/US. Table 6 The Paired t test of ARIMA and ARGA NT/USPOUND/USSF/USYEN/USt1.5600*1.16392.4480**2.4494**Note: **significance at 5%, *significance at 10% Based on the results in table 7, we find that except for the exchange rate, YEN/US, the predictive power of ARBPN is not much different from that of ARIMA. Table 7 The Paired t test of ARIMA and ARBPN NT/USPOUND/USSF/USYEN/USt-0.0443-0.7483-0.0770-1.5304*Note: **significance at 5%, *significance at 10% We further compare the MAPE of ARGA to that of ARBPN for each exchange rate and find that except for two exchange rates, NT/US and YEN/US, ARGA perform significantly better than ARBPN in predictive power for the other two exchange rates as listed in table 8. Table 4-8 The Paired t test of ARGA and ARBPN NT/USPOUND/USSF/USYEN/USt2.1375**1.05891.28503.6024**Note: **significance at 5%, *significance at 10% Overall, ARGA performs the best in predictive power for the four chosen exchange rates, especially for the two exchange rates NT/US and YEN/US. (2)Validity Since higher performance in the prediction about the direction of future exchange rate is more important than higher performance in predictive accuracy of future exchange rate, this study list the results in table 9 for ARIMA, table 10 for GA and table 11 for ANN to see whether the predicted direction of the future exchange rate is the same as the true one. Table 9 Validity of ARIMA model NT/USPOUND/USSF/USYEN/US1990*50%**1991*****1992*70%**199340%***1994***60%199510%40%**199620%***199750%***199840%60%**1999****200050%40%**2001****AVG35%52%**** Note: * random walk model is not predictable. ** There is no change in actual exchange rate for last two weeks. *** We do not calculate AVG since only one year can be estimated Table 10 Validity of Genetic Algorithm model NT/USPOUND/USSF/USYEN/US199030%70%60%70%1991**60%80%70%199240%80%90%70%199350%60%60%40%199460%60%60%60%199550%60%60%70%199650%70%70%60%199770%50%60%60%199840%70%50%60%199960%50%60%60%200060%60%60%70%200160%50%60%80%AVG51.82%61.67%64.17%64.17%Note** There is no change in exchange rate for last two weeks Table 11 Validity of BPN model NT/USPOUND/USSF/USYEN/US199030%50%60%20%1991**70%50%30%199240%60%90%40%199340%60%60%40%199460%30%60%70%199540%60%50%30%199630%70%30%70%199770%50%60%40%199850%70%50%40%199950%60%40%40%200060%40%60%30%200140%50%70%20%AVG46.36%55.83%56.67%39.17%Note** There is no change in exchange rate for last two weeks This study further uses paired t test to compare validity of ARGA and ARBPN to know which model has the best validity as listed on table 12. The results on table 12 indicate that ARGA perform significantly better than ARBPN in the validity of predicted change direction for all the four exchange rates. ARBPN shows higher validity of predicted change direction than ARIMA for the two exchange rates, NT/US and POUND/US when average percentage of predicted change direction of ARBPN is compared to that of ARIMA based on the results of table 9 and table 11. Table 12 The Validity paired t test Between ARGA and ARBPN NT/USPOUND/USSF/USYEN/USt1.9364**1.5409*1.7498*3.8044**Note: **significance at 5%,*significance at 10% Conclusion This study compares the forecasting ability of ARIMA model with two hybrid models, ARGA and ARBPN. Using paired t test, we find that at significant level 10%, ARGA is better than ARIMA for three exchange rates, NT/US, SF/US and YEN/US, and better than ARBPN for two exchange rates, NT/US and YEN/US. Also, we find that ARIMA is better than BPN only for YEN/US at 10% significant level. Therefore, we may conclude that ARIMA in combination with Genetic Algorithm (called ARGA) are the best model and ARBPN is no much difference from ARIMA. There are two possible reasons explaining why BPN does not show higher performance than ARIMA. One reason is that there are many random walk existed in 12 in-sample periods and in order to compare with forecasting ability of ARIMA, BPN uses the same data as ARIMA, which restrains its ability from forecasting exchange rates. The other reason is that the development of BPN is not mature. Researcher should use trial and error method to find out the number of hidden layer, the learning rule and transfer function, and the result may not be the best one. In fact, there is not any function which can get proper number of hidden layer and other relative parameters, so researchers should uses trial and error method to get better performance but may be not the best one. Meanwhile, the validity is further used to examine whether the predicted changing direction of the future exchange rates is the same as actual one. By using paired t test, we find that at significance level 10%, the validity of ARGA is the best among three chosen models. The validity of ARBPN is higher that that of ARIMA, although the predictive power of ARBPN is inferior to that of ARIMA. Therefore, GA shows higher performance not only in predicting financially distressed firms as found by Huang et al. (1994) and Lensberg et al. (2004) but also in predicting the accuracy and the changing direction of the future exchange rates when GA is in combination with linear time series model such as ARIMA. Hopefully, GA can be further applied in financial fields relating to the validity in predictive power. References Akike, Hizotogu, fitting Autoregressive Model for Prediction, Annals of Institute of Statistical Mathematics, January 1969, pp.243-247 A Tutorial for NeuralWorks Professional II/PLUS, Neuralware, October 2001 Anderas, S. and Reou, Exchange Rates Forecasting: A Hybrid Algorithm Based on Genetically Optimized Adaptive Neural Networks, Computational Economics, 2002, pp.191-210. Borisov, A. N., and Pavlov, V A.Prediction of a continuous function with the aid of neural networks, Automatic Control and Computer Sciences, 29(5), 1995, pp39-50. Chen, An-Sing and Mark T. Leung, Regression neural network for error correction in foreign exchange forecasting and trading, Computers & Operations Research, 2003. Coakley, James R. and Carol E. Brown,Artificial Neural Networks in Accounting and Finance: Modeling Issues. International Journal of Intelligent Systems in Accounting, Finance & Management 9: 119-144. Diebold, F. X., and Nason, J. A., Nonparametric exchange rate prediction?, Journal of International Economics, 28(3-4), 1990, pp 315-332. Engle, R. F., and Byung S. Yoo, Forecasting and Testing in Co-integrated Systems, Journal of Econometrics, 35, May 1987, pp143-159. Fang, H. and K.K. Kwong, Forecasting Foreign Exchange Rate, The Journal of Business Forecasting, Winter 1991, pp.16-19 Grudnitski, B. and L. Osburn, Forecasting S&P and Gold Futures Prices : An Application of Neural Network, September 1993, pp.631-643. Hann, T. H., and Steurer, E., Much ado about nothing? Exchange rate forecasting: Neural networks vs. linear models using monthly and weekly data, Neurocomputing, 10, 1996, pp323-339. Hsiao, C., Autoregressive Modeling and Money Income Causality Detection, Journal of Monetary Econometrics, January 1981, pp.85-106. Hu, M. Y., Guoqiang Zhang , Christine X. Jiang and B. Eddy Patuwo, A Cross-Validation Analysis of Neural Network Out-of-Sample Performance in Exchange Rate Forecasting, Decision sciences, Winter 1999, pp197-216. 14.Huang, CS, Dorsey RE, Boose MA,(1994) Life Insurer financial distress prediction : a neural network model. Journal of Insurance Regulation 131-167. 15.GIL-ALANA, L.A. and J. TORO, Estimation and Testing of AFRIMA models in the Real Exchange Rate, International Journal of Finance and Economics, 2002, pp. 279-292 16.Leigh, William , Russel Purvis and James M. Ragusa , Forecasting the NYSE composite index with technical analysis, pattern recognizer, neural network, and genetic algorithm: a case study in romantic decision support, Decision Support Systems, pp. 361-377. 17.Lensberg, Terje, Aasmund Eilifsen and Thomas E. Mckee (2004) Bankruptcy Theory Development and Classification via Genetic Programming. European Journal of Operational Research. 1-21. 18.Leung, M.T., A.S. Chen & H. D.atouk, Forecasting exchange rates using general regression neural networks, Computers & Operations Research, 2000, pp.1093-1110 19.Meese, R., What determines real exchange rate? The long and the short of it, Journal of International Financial Markets, Institutions and Money, 8, pp.117-153,1998 20.Meese,R. & K. Rogoff, Empirical Exchange Rate Models of the Seventies: Do they fit out of sample?, Journal of International Economics,1983, pp.3-24 21.Meese,R. & K. Rogoff, The Out-of-Sample Failure of Empirical Exchange Rate Models : Sampling Error or Model Misspecification?, In Exchange Rates and International Macroeconomics, ed. J. Frenkel. Chicago: University of Chicago Press, 1983, pp67-112 22.Mehran, J. & M. Shahrokhi, An application of four foreign currency forecasting models to the U.S. dollar and Mexican peso, Global Finance Journal, Fall 1997, pp.211-220 23.Neely, C., P. Weller & R. Dittmar, Is Technical Analysis in the Foreign Exchange Market Profitable? A Genetic Programming Approach, Journal of Financial and Quantitive Analysis, December 1997, pp.405-426 24.Neely, C.J. & P. Weller, Predicting Exchange Rate Volatility: Genetic Programming Versus GARCH and RiskMetrics, The Federal Reserve Bank of St. Louis, May/June, 2002,pp.43-54 25.Rauscher, F.A., Exchange Rate Forecasting: A Neural VEC Approach to Non-Linear Time Series Analysis, Journal of Time Series Analysis, 1997, pp.461-471 26.Verkooijen, W., A neural network approach to long-run exchange rate prediction, Computational Economics, 9(1), 1996, pp51-65. 27.Wu, B., Model-free forecasting for nonlinear time series (with application to exchange rates). Computational Statistics & Data Analysis, 19, 1995, pp433-459. 28.Zhang, X., and Hutchinson, J., Simple architectures on fast machines: 1256:?DKQRSVpq~{o]L hXo hXo CJKHOJQJaJ"hXo hXo 5CJOJQJ\aJhXo 5OJQJ\o(h_<,5CJOJQJ\aJo(%h_<,h_<,5CJOJQJ\aJo(h_<,5CJOJQJ\aJo(%h_<,h_<,5CJOJQJ\aJo(hXo 5CJOJQJ\aJo(h,5CJ(OJQJ\aJ(o(h 5CJ(OJQJ\aJ(o(h 5CJ(OJQJ\aJ(RSTUV% C D E F G H I J K L r 0r 0r 0r 0r 0r 0r 0r 0r r r r r |r |r r r r r r r r r $x1$WD2`xa$gdXo $1$a$gdXo $a$$dpa$\% C L P [ } ~ 5    v w ʷ}pcpWpcpcph CJOJQJaJhpCJOJQJaJo(h CJOJQJaJo(h 5OJQJ\o(hXo hXo 5OJQJ\o(hXo 5CJ(OJQJ\aJ(o(hXo 5CJ(OJQJ\aJ(%hXo hXo 5OJQJ\mH o(sH (hXo hXo CJKHOJQJaJmH sH  hXo hXo CJKHOJQJaJhXo CJKHOJQJaJo(L jklr 0r r r r p>r |r |r r r r pr r r )r hr |r r $w^wa$w^wv0VDuWD^v`0gd, WD>^`0WD`0 0WD`0gd/W$a$ $dpa$gdXo Lbdlopwhiklؾ嘌uhhnCJOJQJaJo(h 5OJQJ\h 5OJQJ\o(hP5OJQJ\o(h OJQJo(h,h CJOJQJaJo(h,CJOJQJaJo(hJ;CJOJQJaJo(hPCJOJQJaJo(hpCJOJQJaJo(h CJOJQJaJo(hCJOJQJaJo($8/:d:?'(./L!T!Y!!̯̿||ph 5OJQJ\o(h /CJOJQJaJo(hrCCJOJQJaJo(h,CJOJQJaJo(h CJOJQJaJh 5CJOJQJ\aJo(h CJOJQJaJo("hrCh CJOJQJ\aJo(h,h CJOJQJaJo("h,h CJOJQJ\aJo(,!!""$'((((i)x)*%*8*9*:*;*=*>*j*6,=,R.S.....//0000teh CJH*OJQJaJo(hud5OJQJ\o(h 5OJQJ\o(hudCJOJQJaJo(hS|CJOJQJaJo(hd@CJOJQJaJo(hBCJOJQJaJo(hVHCJOJQJaJo(hb@CJOJQJaJo(hCJOJQJaJo(h CJOJQJaJo(h CJOJQJaJ""((:*;*j*1,u-.000Z3\3]3^3_3333#5$5r r 84Wr |r pr |r r jr pr pr dr |r r "r |r |r |r |r dr |r |r r | 0WD`0gd,v0VDuWD^v`0gd,w^wvVDuWD^v`gd,001Z3_3b3c3z3{3|3~33333x4y444$5'5555555555 66h6i6j6⿯rrerXXXXXXh)VCJOJQJaJo(h^CJOJQJaJo(hCJOJQJaJo(h 5CJOJQJ\aJ(jh 5OJQJU\mHnHuh CJOJQJaJjh CJOJQJUaJ*jh CJOJQJUaJmHnHuhudCJOJQJaJo(h CJOJQJaJo(h 5CJOJQJ\aJo("j6666 7+7b777f8g888886999<9=9?9@9:,:.:8::::;; ;P;;========GB̿د؟̟ykh CJOJPJQJaJjh CJOJQJUaJ*jh CJOJQJUaJmHnHujh CJOJQJUaJh 5CJOJQJ\aJo(h CJH*OJQJaJh CJOJQJaJh CJOJQJaJo(h)VCJOJQJaJo(h^CJOJQJaJo()$588|8}8P;;===-AHBiB CPFFF r 0r |r r |r 0r 0r PL r hCr hCr r hC r hCr hCr hC$dHa$d $dH`a$d dWD`gd,\dWDj^`\ & F & Fgd^w^wdpVD@WD^`gd^GBiBCFEFJJJ,J-JfJgJJJJOOOUPdPQRRRSTTȶ~o`PDh CJOJQJaJh 5CJOJQJ\aJo(h,CJOJPJQJaJo(h CJOJPJQJaJo(h CJPJaJh CJH*OJQJaJo(#j h CJEHOJQJUaJh CJOJQJaJo("jh CJUaJmHnHuh CJPJaJo(h /CJOJPJQJaJo(h CJOJPJQJaJh 5CJOJPJQJaJFGGHHIJJ%J&JYJJJJTLqLOOr hCr hCr hCr hCr hCr hCr r r r r hCr hCr hC r hCr hC $dH`a$d Hd WD,`H$d a$ d WD` d WD2` $dH^a$$dHa$$dH^`a$OUPdPQRSTU;VQVSWTWUWVWWWWWr hCr r r r r r r |r |r |r | | |  $$Ifa$$a$ 0WD`0gd,0d WD8^0` d WD2`$d a$ 0d WD`0gd,d $0dHWD^`0a$gd,TFTTTUU;VQVTWVWYY"Y)YYY[[[[[[ \ \\"]#]Q]R]^^^^.^/^:^@^H^I^T^U^`^a^孑xkxkxkkxkxkxh CJOJQJaJo(h CJOJQJaJh:CJOJQJaJo(h:5CJOJQJ\aJo(h CJOJQJaJh 5CJOJQJ\aJo(h /CJOJQJaJo(h 5OJQJ\o(h CJH*OJQJaJo(h CJOJQJaJo(hj}lCJOJQJaJo(*WWWWWtk |k |k  $$Ifa$kd3$$IfFw Z>    t06    44 laWWWX8Xtk |k v k  $$Ifa$kd$$IfFw Z>    t06    44 la8X9XCXtXXtk |k v k  $$Ifa$kd]$$IfFw Z>    t06    44 laXXXXXtk |k v k  $$Ifa$kd$$IfFw Z>    t06    44 laXXXYYZZ[[\\]trr |rr |mr |br br |rr v rr |mr v rr |br  0WD`0gd, & Fkd$$IfFw Z>    t06    44 la ]]]]]]^^r |r ||||C|| $$Ifa$$a$^ ^^^&^.^:^KB|B|B|BC|B| $$Ifa$kd$$Ifr ,G  t0644 la:^;^@^H^T^`^l^KB|B|B|BC|B| $$Ifa$kd$$Ifr ,G  t0644 laa^l^r^s^~^^^^^^^^^^^^^^^^^^^____$_%_0_6_>_?_J_K_V_W_b_h_p_q_|_}____________________````&`,`-`8`9`D`E`P`Q`\```%h CJKHOJPJQJ^JaJo(h CJOJQJaJh CJOJQJaJo(h CJOJQJaJo(Ml^m^r^~^^^^KB|B|B|BC|B| $$Ifa$kd$$Ifr ,G  t0644 la^^^^^^^KB|B|B|BC|B| $$Ifa$kdD$$Ifr ,G  t0644 la^^^^^^_KB|B|B|BC|B| $$Ifa$kd$$Ifr ,G  t0644 la_____$_0_KB|B|B|BC|B| $$Ifa$kd $$Ifr ,G  t0644 la0_1_6_>_J_V_b_KB|B|B|BC|B| $$Ifa$kdl $$Ifr ,G  t0644 lab_c_h_p_|___KB|B|B|BC|B| $$Ifa$kd$ $$Ifr ,G  t0644 la_______KB|B|B|BC|B| $$Ifa$kd $$Ifr ,G  t0644 la_______KB|B|B|BC|B| $$Ifa$kd $$Ifr ,G  t0644 la___```&`KB|B|B|BC|B| $$Ifa$kdL $$Ifr ,G  t0644 la&`'`,`8`D`P`\`KB|B|B|BC|B| $$Ifa$kd$$Ifr ,G  t0644 la\`]`u`````KIr |Ar |6|6|6T| $$1$Ifa$L WDR`L kd$$Ifr ,G  t0644 la```````````````````````````````````````````aaa a aaaaaaaaa"a#a'a(a,a-a1a2a3a7aʷʷʷʷʷʷ"h CJKHOJPJQJ^JaJ%h CJKHOJPJQJ^JaJo(h CJKHaJ(h 5CJKHOJPJQJ\^JaJ+h 5CJKHOJPJQJ\^JaJo(A`````````N||M||T|N||kdt$$IfTJr "T'8'8'2644 aT$?$1$IfVD?WDE^`?a$ $$1$Ifa$```````od|d|dT|dN|d| $$1$Ifa$kdw$$IfTJr "T'8'8'2644 aT```````od|d|dT|dN|d| $$1$Ifa$kdz$$IfTJr "T'8'8'2644 aT``a aaaaod|d|dT|dN|d| $$1$Ifa$kd}$$IfTJr "T'8'8'2644 aTaaa#a(a-a2aod|d|dT|dN|d| $$1$Ifa$kd$$IfTJr "T'8'8'2644 aT2a3a8a=aBaGaLaod|d|dT|dN|d| $$1$Ifa$kd$$IfTJr "T'8'8'2644 aT7a8afBfCfGfHfIfMfNfRfSfWfXf\f]fafbfh CJKHOJPJQJaJ%h CJKHOJPJQJ^JaJo("h CJKHOJPJQJ^JaJPeeeeeeeod|d|d|d|d| $$1$Ifa$kd,$$IfTJr 8p'8'8'('8'8644 aTeeeeeeeod|d|d|d|d| $$1$Ifa$kd-$$IfTJr 8p'8'8'('8'8644 aTeeeeeeeod|d|d|d|d| $$1$Ifa$kdv.$$IfTJr 8p'8'8'('8'8644 aTeeff fffod|d|d|d|d| $$1$Ifa$kdO/$$IfTJr 8p'8'8'('8'8644 aTffff$f)f.fod|d|d|d|d| $$1$Ifa$kd(0$$IfTJr 8p'8'8'('8'8644 aT.f/f4f9f>fCfHfod|d|d|d|d| $$1$Ifa$kd1$$IfTJr 8p'8'8'('8'8644 aTHfIfNfSfXf]fbfod|d|d|d|d| $$1$Ifa$kd1$$IfTJr 8p'8'8'('8'8644 aTbfcfhfmfrfwf|fod|d|d|d|d| $$1$Ifa$kd2$$IfTJr 8p'8'8'('8'8644 aTbfcfgfhflfmfqfrfvfwf{f|f}ffffffffffffffgg gg.g6g;g@g[g𹦗wj]j]j]j]j]jh:CJOJQJaJo(h CJOJQJaJo(h CJKHOJPJQJaJh CJOJPJQJ^JaJh CJOJPJQJaJo($h 5CJOJPJQJ\^JaJ#h 5CJOJPJQJ\aJo("h CJKHOJPJQJ^JaJ%h CJKHOJPJQJ^JaJo(h CJKHOJPJQJaJ"|f}ffffffof|f|f|f|f| $$Ifa$kd3$$IfTJr 8p'8'8'('8'8644 aTffffghh#h,hogr |er |Zr pUr |J|J|Ji| $$1$Ifa$$a$ 0WD`0gd,xWD`xkde4$$IfTJr 8p&8&8&(&8&8644 aT[g\gagggggggghhhh"h#h+h,h1h2h8h9h:h;hkFkRkXkYkpkkkkkkkkl lll>mAm{m|mmѾѾ{{k^^h /CJOJQJaJo(h 5CJOJQJ\aJo(hX\CJOJQJaJo(hM7CJOJQJaJo(h CJOJQJaJh CJOJQJaJo(h CJKHOJPJQJaJ%h CJKHOJPJQJ^JaJo(.h B*CJKHOJPJQJ^JaJo(ph+h B*CJKHOJPJQJ^JaJph =k>kokpkll lxmymzmskr |ir |^r v ir |ir |^r Vr |Vr |WD2` 0WD`0gd,WD^`kd9$$IfRre [&&U&&U&644 a zm{m|mmmmmmmmmmr |r |r ||P|||P|c|P|kdt:$$IfJr v&8&&(&8&644 a $$1$Ifa$$a$ mmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmnnnnnnó%h CJKHOJPJQJ^JaJo(h CJKHOJPJQJaJ(h 5CJKHOJPJQJ\^JaJ+h 5CJKHOJPJQJ\^JaJo("h CJKHOJPJQJ^JaJ?mmmmmmmmmm||P|h|P|||P|kdW;$$IfJr v&8&&(&8&644 a $$1$Ifa$ mmmmmmmsh|hP|h|h|hP| $$1$Ifa$kdH<$$IfJr v&8&&(&8&644 ammmmmmmsh|hP|h|h|hP| $$1$Ifa$kd9=$$IfJr v&8&&(&8&644 ammmnnnnsh|hP|h|h|hP| $$1$Ifa$kd*>$$IfJr v&8&&(&8&644 ann n nnnnnnnnnnnn n#n$n%n&n'n(n)n*n+n/n0n3n4n5n6n7n8n9n:n;n?n@nCnDnGnHnInJnKnLnMnQnRnSnTnUnVnWnXnYnZn[n_n`ncndngnhninjnknlnmnqnrnsntnunvnwnxnynzn{n%h CJKHOJPJQJ^JaJo(h CJKHOJPJQJaJ"h CJKHOJPJQJ^JaJOn nnnnnnsh|hP|h|h|hP| $$1$Ifa$kd?$$IfJr v&8&&(&8&644 ann n$n&n(n*nsh|hP|h|h|hP| $$1$Ifa$kd @$$IfJr v&8&&(&8&644 a*n+n0n4n6n8n:nsh|hP|h|h|hP| $$1$Ifa$kd@$$IfJr v&8&&(&8&644 a:n;n@nDnHnJnLnsh|hP|h|h|hP| $$1$Ifa$kdA$$IfJr v&8&&(&8&644 aLnMnRnTnVnXnZnsh|hP|h|h|hP| $$1$Ifa$kdB$$IfJr v&8&&(&8&644 aZn[n`ndnhnjnlnsh|hP|h|h|hP| $$1$Ifa$kdC$$IfJr v&8&&(&8&644 alnmnrntnvnxnznsh|hP|h|h|hP| $$1$Ifa$kdD$$IfJr v&8&&(&8&644 azn{nnnnnnsh|hP|h|h|hP| $$1$Ifa$kdE$$IfJr v&8&&(&8&644 a{n~nnnnnnnnnnnnnnnnnnnop`P`C`C`hX\CJOJQJaJo(hX\hX\CJOJQJaJo(hX\h CJOJQJaJo(hX\h CJOJQJaJh CJOJQJaJo(h CJKHOJPJQJaJ%hZ'CJKHOJPJQJ^JaJo("h CJKHOJPJQJ^JaJ%h CJKHOJPJQJ^JaJo((h 5CJKHOJPJQJ\^JaJ+h 5CJKHOJPJQJ\^JaJo(nnnohoiooooosqr |`r |`r |[r |[r |P|PP|P| $$1$Ifa$$a$VD,WD^`gdX\kdF$$IfJr v&8&&(&8&644 a ofohoooooooooooooooooooooooraQraQArh CJKHOJPJQJaJh CJKHOJPJQJaJ!h CJKHOJPJQJaJo("h CJKHOJPJQJ^JaJ%h CJKHOJPJQJ^JaJo($h 5CJKHOJPJQJ\aJ+h 5CJKHOJPJQJ\^JaJo((h 5CJKHOJPJQJ\^JaJh CJOJQJaJo(hX\h CJOJQJaJo(hZ'CJOJQJaJo(ooooooooo||h|P||||kdxG$$IfJrR z&8&&(&8& 644 a $$1$Ifa$ooooooosh|hP|h|h|h| $$1$Ifa$kd[H$$IfJrR z&8&&(&8& 644 aoooooooooooooooooooopp p p pppppp p!p%p&p)p*p,p.p1p2p5p6p7p;prBrFrJrsh|h|h|h|h| $$1$Ifa$kd\$$IfJr X (&8&8&(&8&8644 a<4r5r9r:r=r>rArBrErFrIrJrKrOrPrSrTrWrXr[r\r_r`rardrerkrlrrrsryrzrrrrrrrssCsDsvvih(CJOJQJaJo(hZ'CJOJQJaJo(h CJOJQJaJo(hZ'h CJOJQJaJo((h 5CJKHOJPJQJ\^JaJ+h 5CJKHOJPJQJ\^JaJo("h CJKHOJPJQJ^JaJ%h CJKHOJPJQJ^JaJo(h CJKHOJPJQJaJ)JrKrPrTrXr\r`rsh|h|h|h|h| $$1$Ifa$kd]$$IfJr X (&8&8&(&8&8644 a<`rarerlrsrzrrsh|h|h|h|h| $$1$Ifa$kd^$$IfJr X (&8&8&(&8&8644 a<rrrrrtt0u1ushr |fr |fr |[r dVr |fr |P|$If$a$ 0WD`0gd, WD`gdZ'kdM_$$IfJr X (&8&8&(&8&8644 a<DsNsOsPss ttttt/u0u1u6u7u?u@uEuFuLuMuNuPurusuwuuuuuuïÆwk^Qkh(CJOJQJaJo(h /CJOJQJaJo(h CJOJQJaJh(h CJOJQJaJ!h CJKHOJQJ^JaJo( h o($h 5CJKHOJQJ\^JaJ'h 5CJKHOJQJ\^JaJo(h CJaJo(h CJOJQJaJh(CJOJQJaJo(hZ'CJOJQJaJo(h CJOJQJaJo(1u7u@uFuMuNuPu{||||>5| $$Ifa$kd"`$$Ifr 6S t0644 laN $$1$Ifa$PuYuauiurusu{||||>kd`$$Ifr 6S t0644 laN $$1$Ifa$suuuuz} ~~~/׀--n˅dr |r |r |r hFBr )r r v r r v r v r v r r v r v r r v r v r v r r v $\WDj^`\a$gd  & Fgdyk & F 0WD`0gd, 0dWD`0gd & Fgd, $WD`a$gd /uuuuv v-v/v;vIvJvQvSvVvZvovvvvwwwwwwwwwwwwwx$x*x+x,x-x/xCxDxExjx~xy2yoyryzzzҸҸҸҸҸҫҫҞҞҫґґҞҞҞ҄ҫҫhCJOJQJaJo(h,CJOJQJaJo(h /CJOJQJaJo(h\CJOJQJaJo(hwNCJOJQJaJo(h(CJOJQJaJo(h CJOJQJaJo(h,5CJOJQJ\aJo(h 5CJOJQJ\aJo(2zzzzzzzzz{/{L{\{c{q{z{{{{{{{|| }}}} ~~~(~*~G~H~J~x~~~~~Z[]tyhU=CJOJQJaJo(h >*CJOJQJaJo(h?XCJOJQJaJo(h CJOJQJaJhQh OJQJ hQo(h? KCJOJQJaJo(hCJOJQJaJo(h\CJOJQJaJo(h CJOJQJaJo(hQCJOJQJaJo(-/6=@PQӀԀր׀DEāŁ-MN͂΂峤xlxlx]hykhCJOJQJaJhCJOJQJaJhCJOJQJaJo(hykhyk6CJOJQJaJhykhykCJOJQJaJhykCJOJPJQJaJo(hykCJOJQJaJo(h CJOJQJaJhU=CJOJQJaJo(hd4CJOJQJaJh CJOJQJaJo(hd4CJOJQJaJo($΂79;@KLӃԃFGnyz~ل578ʅ΅9򰤰򽗋~oh h CJOJQJaJh CJOJQJaJo(hSCJOJQJaJhSCJOJQJaJo(hd4CJOJQJaJhd4CJOJQJaJo(hykCJOJQJaJo(hU=CJOJQJaJo(h >*CJOJQJaJo(h CJOJQJaJh CJOJQJaJo('9adgpqrwƆdž CDʈˈΈ789Ynq|}0ȻȻȯȯբxȯȯiȯȯh >*CJOJQJaJo( h h CJOJPJQJaJh/CJOJQJaJo(hmPCJOJQJaJhmPCJOJQJaJo(h CJOJQJaJhU=CJOJQJaJo(h CJOJQJaJo(hI9CJOJQJaJo(h CJOJQJaJo(h h 6CJOJQJaJ)d ˈn[,}@A^_r v r r v r v r v r v r r v r r v r v r r v r pr r $\WDj^`\a$gd \WDj^`\gdI901|}Ɋʊ124ˋ̋*+[^lr,/HIHI}ώЎ ďŏ56ھھھ竜hXo 5CJOJQJ\aJ h o(UhW*CJOJQJaJhW*CJOJQJaJo(h >*CJOJQJaJo(hI9CJOJQJaJo(h CJOJQJaJo(h CJOJQJaJ;Practical issues in nonlinear time series prediction. In A. S. Weigend & N. A. Gershenfeld, Time series prediction: Forecasting the future and understanding the past. Reading, MA: Addison-Wesley, 1994, pp219-241. The first prediction period The first estimating period Figure 1 The concept of rolling regression The second estimating period The second prediction period Output Signals Input Signals Output layer Hidden layer Input layer Figure 2 The Fundamental Structure of An Artificial Neural Network  EMBED Equation.DSMT4    ?@AEKSV^_cijtuv} 89:̽ܵܬ̽̽̽̽̽ܟhXo 5CJOJQJ\aJo(hXo CJOJQJaJo(hXo OJQJo(hXo OJQJhXo 5CJOJQJ\aJhXo 5CJOJQJ\aJo(hXo hXo 5CJOJQJ\aJhXo 5CJOJQJ\aJo(5_v}~9:VWXYZ[\]r $a$:;RSTU[\] h o(h jahXo EHU$j8D hXo CJKHUVaJo(hXo jhXo U01h2P. A!"#$%S `!e>AF50, 8xڭTOAfvRm zP&z AcKi`IÖxllRCה^ AcR /Ą@OrMl[i.Q,ȿcoLs8Q56tΞ.&TjNi8դT7/Lj4~Mc*?^M̚uLu3 6h?{0ȗ w1+g1g1yңɂ5)l45eOU]A|7_o|E;}`_"C7sφpR%>#%RO9=!2qZ Hs)G| ;2N(D l 0AC8a5JLM$;ϴz9#zAFce{l(j$Ytqٰ՟P/H0Z\vIxcfl[EWx>9x~"k;/Ps0/?Dd D  3 @@"?Dd ~D  3 @@"?Dd @b  c $A? ?3"`?2]_jߑI¨E`JTb^9d`!1_jߑI¨E`JTb^  xcdd`` @c112BYL%bpu?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}~     )$'(*,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}~Root Entry  F{\U&Data AF5V`!e>AF50, 8xڭTOAfvRm zP&z AcKi`IÖxllRCה^ AcR /Ą@OrMl[i.Q,ȿcoLs8Q56tΞ.&TjNi8դT7/Lj4~Mc*?^M̚uLu3 6h?{0ȗ w1+g1g1yңɂ5)l45eOU]A|7_o|E;}`_"C7sφpR%>#%RO9=!2qZ Hs)G| ;2N(D l 0AC8a5JLM$;ϴz9#zAFce{l(j$Ytqٰ՟P/H0Z\vIxcfl[EWx>9x~"k;/Ps0/?P@P gQe1$,CJ KHOJPJQJ_HaJ mH nHsH tH$A@$ -k=W[WFi@F h>??@AA%A&AYAAAATCqCFFUGdGHH IISJJJKKKKLLL6L7L>LELcLdLlLLLLLLMMMOMmMnMMMMOWOPPPPUCUHUMUNUSUXU]UbUgUhUmUrUwU|UUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUVV VVVVVV#V(V-V2V7V8V=VBVGVLVQVRVWV\VaVfVkVlVpVwV~VVVVVWWWWX XXXXX$X)X.X3X4X9X>XCXHXMXNXSXXX]XbXgXhXmXrXwX|XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXYY YYYYYY#Y(Y-Y2Y7Y8Y=YBYGYLYQYRYVY]YdYkYrYsYYYYYYYYYYYYYYYYYYYYYYZZ ZZZZZ Z%Z*Z/Z0Z5Z:Z?ZDZIZJZOZTZYZ^ZcZdZiZnZsZxZ}Z~ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ[[ [[[ [![A[B[s\\\\\\\\\\\\\\]]]]]]]]]]^ ^^^^N^Q_R________________```aaaaab b&b/b5beBeIePeWe^e_eeeeeeeeeeeeeeeeeeeeefff ffffff#f'f+f/f3f4f9f=fAfEfIfJfOfSfWf[f_f`fefifmfqfufvf{ffffffffffffffffffffffffffffffffffggDgEgFgsitiiiiiiiiiiiiii&j/j:jWorrs`s ttZu%vv7ww8xxwyNzz{|N}}~30ހd%GHI_fgͅ΅()67CDņƆdžȆɆʆ͆08000x0x00000000000x000000000000000x000x00 0 0 0 0(0x0(0(0(0(0x00(0( 00x 0000000x00080808080808000x080800@00@0@0@00x0@0@0H00H0H00 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00 00H0 00P000 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00 0 0 0  0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0x0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00 0 0 0 0 0 0 0 0 0 0 0 00`00`000 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00`00 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 000`0`0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00000 0 0 0 0 0 0 0 0 0 0 0 0 00h 0p 0p 0p 0x 0p 0p 0p 0p 0p 0  0 p 0x 00000 ;00Z0 ;00Z0 ;00Z0 ;00Z00 ;0 0Z ;0 0Z ;00Z0 ;00Z0 ;00Z0 ;00Z0 ;00Z0 ;00Z0 ;00Z ;00Z0 ;00 ;00 ;00 ;00 ;00 ;00 ;00 ;0#0s`st͆;00xt;00;000Rt!0j6GBTa^`7aacpd!eebf[gQh^iikmn{nooZppq4rDsuz΂90:]HKMNOQRTW`ls{L $5FOWW8XXX]^:^l^^^_0_b____&`\`````a2aLafaaaaaa bcccccd2dLdfddddddDe^exeeeeeef.fHfbf|ff,h]huiik=kzmmmmmnn*n:nLnZnlnznnoooo p p6pLpbpxpppppRqoqqqqqqqrr4rJr`rr1uPusud_]ILPSUVXYZ[\]^_abcdefghijkmnopqrtuvwxyz|}~\Jb+z+}+444̆__`xz:RS?2$e&p#g2$3ļ*hx=2$e>AF52@;R~(    3 S%, 3  s"*?`  c $X99?3 S%,42  u42  |u42  542   542   5|42   5`42    U#42    %U#ZB   S D 5ZB   S D U|5ZB  S D&5ZB  S D&5ZB  S DRU'5ZB  S D'5ZB  S DR5ZB  S DU5ZB  S D P ZB  S DP ZB  S DP ZB  S DP ZB  S DR ZB  S DR ZB  S D|R ZB  S DR' ZB  S DU#5%ZB  S DP5PuZB  S DRU#S5%ZB  S D5uh   3 "`/ b ! # "`P%U(  b " # "`u$U  b # # "`$  b $ # "`u!$U#  b % # "` (S%Z+  & ]S% &3  s"*?` ' c $X99?]S%ZB ( S Dol m | ZB ) S Dol < "= ZB * S Do""| ZB + S Do{ |  , lBCDEFu J@  l \< ZB - S Do  ZB . S Do' ) ZB / S Do   0 lBCDEFhp@ xx@  { < '  1 lBCDEFxZh@   '<  2 lBCDEFxZhXZ@  '<  n 3 C 3"` o n 4 C 4"` S% `B 5 c $Do\ `B 6 c $DԔ` a b D C D"` \ E 3 "` b C C C"` \ M 3 M"` n P s *A ? ?#" `B S  ?_+`+a+{+4AĂCstDdL!utE t" t&t"@ tX <tP@ [TM @t"t9T:i;,f<4u=DčCnCnDCnCnĎCn5CnyCnLCnCnDCnCnCnCnď""|9|9rrrzzAA77͆     ""99rrrzzJJAA͆  =*urn:schemas-microsoft-com:office:smarttags PlaceName=*urn:schemas-microsoft-com:office:smarttags PlaceType8*urn:schemas-microsoft-com:office:smarttagsCity9*urn:schemas-microsoft-com:office:smarttagsplace V_emsz%+-C " h o uwJT%%&& ****..!/$/2288887A9AqAsAHHHHIIIIUJVJMMoNNNNNNOOHQJQKRPR_qgqrrrr%s0sGsQs`sgspsts tt)t6tZuauyuu%v,v9v>vvvvvIwNwwwww8x1I1l9o999X=_=>>&A+AvCCGHMMNNOOPPQQERKRVVYY>[@[i_m_llnvpvw wwwPzVzzz{{||}G}H}M}P}V}}}~~~~5=fo $'-I_eMP͆:::::::::::::::::::::::::::::::::::::::::::::::::l1$>$&&H9i9AATCqCLL7LELdLlLLLMMnMMMMRVWsYY![\\\\]^__ bbbbbcd_eegii/j9jTqoqu$vIeg̅΅')57B͆I͆alanwinxpsp2USERUSERC<(J2W-r(zfs Qh"|au ~IIg$[ SljgF <(h~,jgS2*Qc"<%Em=~%%\,'\fx,*a,D_1i~3jgBrzCScE ^ OEmQ `gT|'S\Od-`iK0mJuX A| *hh^h`o(()  ^` hH0  ^` hH.  ^` hH. `  ` ^` ` hH0 @  @ ^@ ` hH.    ^ ` hH.  ^` hH0  ^` hH.z^`zo(()  ^` hH0  ^` hH.  ^` hH. `  ` ^` ` hH0 @  @ ^@ ` hH.    ^ ` hH.  ^` hH0  ^` hH.hh^h`o(()  ^` hH0  ^` hH.  ^` hH. `  ` ^` ` hH0 @  @ ^@ ` hH.    ^ ` hH.  ^` hH0  ^` hH.z^`zo(()HH^H`o(.  ^` hH.  ^` hH. `  ` ^` ` hH0 @  @ ^@ ` hH.    ^ ` hH.  ^` hH0  ^` hH.M^`Mo(0^`0o(-0^`0o(-.88^8`o(-.. 88^8`o( -... `^``o( -.... `^``o( -..... ^`o(-...... pp^p`o(-.......z^`zo(()QQ^Q`OJQJo(hH  ^` hH.  ^` hH. `  ` ^` ` hH0 @  @ ^@ ` hH.    ^ ` hH.  ^` hH0  ^` hH.hh^h`o(0^`0o(-0^`0o(-.88^8`o(-.. 88^8`o( -... `^``o( -.... `^``o( -..... ^`o(-...... pp^p`o(-.......ww^w`o(ww^w`o(-0^`0o(-.0^`0o(-.. 88^8`o( -... 88^8`o( -.... `^``o( -..... `^``o(-...... ^`o(-.......^`o(() t t^t` hH0 T T^T` hH. 4 4^4` hH.    ^ ` hH0    ^ ` hH.    ^ ` hH.  ^` hH0  ^` hH.hh^h`o(0^`0o(-0^`0o(-.88^8`o(-.. 88^8`o( -... `^``o( -.... `^``o( -..... ^`o(-...... pp^p`o(-.......ww^w`o(()  ^` hH0  ^` hH.  ^` hH. `  ` ^` ` hH0 @  @ ^@ ` hH.    ^ ` hH.  ^` hH0  ^` hH. M^`Mo(hH 0^`0o(hH4-2 0^`0o(hH-. 88^8`o(hH-.. 88^8`o(hH -... `^``o(hH -.... `^``o(hH -.....  ^`o(hH-......  pp^p`o(hH-.......z^`zo(()  ^` hH0  ^` hH.  ^` hH. `  ` ^` ` hH0 @  @ ^@ ` hH.    ^ ` hH.  ^` hH0  ^` hH.hh^h`o(()  ^` hH0  ^` hH.  ^` hH. `  ` ^` ` hH0 @  @ ^@ ` hH.    ^ ` hH.  ^` hH0  ^` hH.z^`zo(()  ^` hH0  ^` hH.  ^` hH. `  ` ^` ` hH0 @  @ ^@ ` hH.    ^ ` hH.  ^` hH0  ^` hH. G0G^G`0o(hH() 7 7^7` hH0  ^` hH.  ^` hH.    ^ ` hH0    ^ ` hH.  ^` hH. w w^w` hH0 W W^W` hH.^`o(. t t^t` hH0 T T^T` hH. 4 4^4` hH.    ^ ` hH0    ^ ` hH.    ^ ` hH.  ^` hH0  ^` hH.hh^h`o(0^`0o(-0^`0o(-.88^8`o(-.. 88^8`o( -... `^``o( -.... `^``o( -..... ^`o(-...... pp^p`o(-.......0^`05\o(hH()  ^` hH0  ^` hH.  ^` hH. `  ` ^` ` hH0 @  @ ^@ ` hH.    ^ ` hH.  ^` hH0  ^` hH.hh^h`o(.  ^` hH0  ^` hH.  ^` hH. `  ` ^` ` hH0 @  @ ^@ ` hH.    ^ ` hH.  ^` hH0  ^` hH.#0^`0o(sH 0  ^` hH0  ^` hH.  ^` hH. `  ` ^` ` hH0 @  @ ^@ ` hH.    ^ ` hH.  ^` hH0  ^` hH. M^`Mo(hH 0^`0o(hH4-2 0^`0o(hH-. 88^8`o(hH-.. 88^8`o(hH -... `^``o(hH -.... `^``o(hH -.....  ^`o(hH-......  pp^p`o(hH-.......hh^h`o(()  ^` hH0  ^` hH.  ^` hH. `  ` ^` ` hH0 @  @ ^@ ` hH.    ^ ` hH.  ^` hH0  ^` hH.^`o(. t t^t` hH0 T T^T` hH. 4 4^4` hH.    ^ ` hH0    ^ ` hH.    ^ ` hH.  ^` hH0  ^` hH. M^`Mo(hH 0^`0o(hH4-1 0^`0o(hH-. 88^8`o(hH-.. 88^8`o(hH -... `^``o(hH -.... `^``o(hH -.....  ^`o(hH-......  pp^p`o(hH-.......hh^h`o(hh^h`o(.0^`0o(..88^8`o(... 88^8`o( .... `^``o( ..... `^``o( ...... ^`o(....... pp^p`o(........hh^h`o(.  ^` hH0  ^` hH.  ^` hH. `  ` ^` ` hH0 @  @ ^@ ` hH.    ^ ` hH.  ^` hH0  ^` hH.hh^h`o(hh^h`o(.0^`0o(..88^8`o(... 88^8`o( .... `^``o( ..... `^``o( ...... ^`o(....... pp^p`o(........hh^h`o(.  ^` hH0  ^` hH.  ^` hH. `  ` ^` ` hH0 @  @ ^@ ` hH.    ^ ` hH.  ^` hH0  ^` hH.hh^h`o(.fzf^f`zo(()((^(`OJPJQJ^Jo(l  ^` hH. `  ` ^` ` hH0 @  @ ^@ ` hH.    ^ ` hH.  ^` hH0  ^` hH.Ba,S2<(C<(,',*Ig"<%|au 'S\^ O3 A|JuW-fsQ =~%icErzC`gTSl~,0md-`_1 >l        ,         Z        ,6}       ,       pTN        Nd        ,        ۞        @H        4        x^t        4        Ь        h\                \        Ь        Ь        Ьd_Td3      C6&#DC6&dw8{g#Dd4-tw8{gd w4-td10b@Xo :Z'_<, /?2d4M7 8J;U=d@VH? KwNmP)V\X\ykj}ln-asrC(/S|  ,ud?XI9`BQ^PW*/WSW pVKKLLL6L7L>LELcLdLlLLLLLLMMMOMmMnMjRkRrRzRRRRRRRRRRRRRRRRRS SS!S"S'S/S;SGSSSTSYSeSmSySSSSSSSSSSSSSSSSSS TTTT%T1T9TETFTKTWTcToT{T|TTTTTTTTTTTTTUUU$U*U1U2U7U>UCUHUMUNUSUXU]UbUgUhUmUrUwU|UUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUVV VVVVVV#V(V-V2V7V8V=VBVGVLVQVRVWV\VaVfVkVlVpVwV~VVVVWWX XXXXX$X)X.X3X4X9X>XCXHXMXNXSXXX]XbXgXhXmXrXwX|XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXYY YYYYYY#Y(Y-Y2Y7Y8Y=YBYGYLYQYRYVY]YdYkYrYsYYYYYYYYYYYYYYYYYYYYZZ ZZZZZ Z%Z*Z/Z0Z5Z:Z?ZDZIZJZOZTZYZ^ZcZdZiZnZsZxZ}Z~ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ[[ [[[ [![\\\\\\\\\\\\\]]]]]]]]^ ^^^^_____________b b&b26xMicrosoft Word 10.0@|dM@vtQU@nS@0h\Ueq՜.+,0 X`lt|  crD{   FMicrosoft Word MSWordDocWord.Document.89q/b5beBeIePeWe^e_eeeeeeeeeeeeeeeeefff ffffff#f'f+f/f3f4f9f=fAfEfIfJfOfSfWf[f_f`fefifmfqfufvf{ffffffffffffffffffffffffffffffffffggiiiiiiiiiiiii͆e0@VX23HIJŏp@p:px@pRpTp@p@UnknownGz Times New Roman5Symbol3& z ArialC.e0}fԚPMingLiU?5 .0}fԚMingLiU/E .jwiԚ;Wingdings qh Ɓ *eqDeqD!?!),.:;?]}    " % & ' 2 t%00 0 0 00000013468:<>@BDOPQRTUVWZ\^ \]d([{  5 0 0 00000579;=?ACY[][>d- 3qX(?_<, Abstractsalanwinxpsp2                         CompObjf