# Financial crisis short term forecasting method using artificial intelligence based classification algorithm.

1. IntroductionIn modern economic globalization process, with the rapid development of the capital markets, business competition has been more and fiercer. In modern society, each company may encounter financial crisis (Cao Wei, Cao Longbing, 2015; Audrino Francesco, 2014). If the company manager cannot tackle these problems in time, company will go into the bankruptcy state. Although we cannot completely avoid each company to be far from the financial crisis, we should forecast short term financial crisis for all companies (Kunze Frederik, Kramer Jens, 2014). As is well known that predicting the financial crisis refers to an important task in manager's daily work. If the financial crisis cannot be solved in time, the manager will lose the capability to tackle it and then go into the bankruptcy (Berger T, Missong M, Financial Crisis, 2014; De Sousa Gabriel Vtor Manuel, 2014). Hence, forecasting the corporate financial crisis has important theoretical and practical significance.

In the capital market, company can collect the low cost fund from capital market and boost enterprise development. The investor should fully utilize capital market operation to achieve higher reward. Modern company is faced with the gradually dangerous marketplace environment (Metescu Ana-Maria. et al, 2013; Halbleib Roxana, Pohlmeier Winfried, 2012). Financial crisis not only affects enterprises' subsistence and development, but also influences the benefits of investors. With rapid development of capital market and the reform of market economy system, the complexity and uncertainty in economic field have been increasingly evident(Svetlova Ekaterina, 2012). Furthermore, financial crisis happens frequently, meanwhile, bankruptcy occurs in enterprises as well (Zaleskiewicz Tomasz, 2011). In this paper, we focus on the problem of financial crisis short term forecasting by a hybrid PSO-SVM model.

For the problem of financial crisis forecasting, existing methods are listed as follows. Chen et al. used Adaptive Markov chain Monte Carlo method to forecast financial risk via a computational Bayesian framework (Chen Cathy W S. 2012). Chen et al. utilized Z-Score value to measure multinomial financial crisis index for financial crisis forecasting, and exploited Grey Markov forecasting for performance valuation (Chen Li-Hui, Guo Tsuei-Yang, 2011). Yu et al. proposed a multiscale neural network learning paradigm to forecast financial crisis events for short term purposes (Yu Lean. et al, 2010). Cipollini et al. exploited principal components analysis to achieve vulnerability indicators to forecast financial crisis (Cipollini A, Kapetanios G, 2009). Hajek et al. proposed a novel framework using Latent Semantic Indexing to forecast financial crisis, and this work assume that equity markets can forecast even sharp changes in monetary policy during a quarter ahead of such a change (Hajek Petr. et al, 2009). Oh et al. establish an alarm zone in the daily financial condition indicator, which is utilized to forecast a potential financial crisis with high accuracy. Particularly, this paper is able to provide an early warning signal via neural networks and nonlinear programming (Oh KJ. et al, 2005).

Different from the above research work, we proposed a novel financial crisis short term forecasting method based on a hybrid particle swarm optimization and support vector machine model. In particularly, in our work, to improve the performance of particle swarm optimization we provide a high inertia weight to build up a new searching space, and inertia weight decreases with paths varying for different particle number.

2. Framework of the Financial Crisis Short Term Forecasting System

In this section, we will discuss the design of financial crisis short term forecasting system. Framework of the financial crisis short term forecasting system is illustrated in Fig. 1. The proposed financial crisis short term forecasting system extracts high dimension financial data from the financial database. Afterwards, to reduce the time cost of the financial crisis short term forecasting task, we proposed an index system to map high dimension financial data to a low dimension space. Afterwards training dataset and testing dataset are built up based on the low dimension experiment data (Oliveira, J. A., Ferreira, J., Figueiredo, M., Dias, L., & Pereira, G., 2014). Next, we utilize Particle Swarm Optimization to optimize parameters of SVM, and then training dataset is exploited to train a SVM classifier. In the end, financial crisis forecasting results can be gained for a given testing sample.

[FIGURE 1 OMITTED]

To carry out the financial crisis prediction, we should build a reasonable index system in advance (shown in Fig. 2). Moreover, high quality index system is able to help us to find the reasons of financial crisis. If the financial crisis can be precisely forecasted, management method of companies can be significantly improved. Based on the above analysis, we should make clear the principle of enterprise financial crisis evaluation index system construction, and then choose appropriate financial evaluation index system to cover all influences of enterprise financial crisis.

[FIGURE 2 OMITTED]

As is shown in Fig. 2, the factors that affect the financial position of the enterprise are classified into four aspects: 1) Profitability, 2) Solvency, 3) Operating capacity, and 4) Composition of capital. Furthermore, the proposed index system also contains special characteristics of the global economic environments.

3. The Proposed Financial Crisis Short Term Forecasting Method

The main idea of this paper is that we introduce an artificial intelligence based classification algorithm (that is, support vector machine) to forecast short term financial crisis. As is well known that support vector machine based on statistical learning theory has been widely utilized in many applications of machine learning, and SVM is able to solve the classification problem with high accuracy and good generalization capability (Garcia Nieto P J. et al, 2016; Garcia Nieto P J. et al, 2015; Garcia-Gonzalo E. et al, 2015; Hsieh Ching-Tang, Hu Chia-Shing, 2014). In order to an optimal SVM prediction model, it is very crucial to select a kernel function and parameters to gain a soft margin. Parameters of SVM classifier greatly affect performance of classification accuracy (Sudheer Ch. et al, 2014; Selakov A. et al, 2014; Liu Baoling. et al, 2013).

In this section, we utilize particle swarm optimization to SVM parameters. Particle swarm optimization is designed to be initialized with a population of random solutions, which are named as particles (Wang Wen-chuan. et al, 2013). In particular, each particle goes in the search space at a velocity which is dynamically updated according to its own and neighbors' historical behavior. Different from the genetic algorithm, particle swarm optimization is more effective (Subasi Abdulhamit, 2013). Therefore, this paper adopts particle swarm optimization to optimize the SVM parameters.

SVM is designed to nonlinearly map the training data to a high dimensional feature space, and utilize a linear regression model in the feature space. In support vector machine, the key task is to optimize the following equation.

min J (w, [xi]) = [1/2] [w.sup.t]w + [[gamma]/2] [l.summation over (i=1)] [[xi].sup.2.sub.i] (1)

[y.sub.i] = [w.sup.T] [phi]([x.sub.i]) + b + [[xi].sub.i], i [member of] {1,2, ..., l} (2)

where [phi]([x.sub.i]) denotes a high dimensional feature space and parameter [xi] is error vector. Moreover, parameter w and b are used to seek the classification boundary, [[xi].sub.i] refers to a slack variable, and parameter [gamma] is value of penalty for mis-classification. Then, the level of financial crisis short term is gained by solving the following equation.

f(x) = wx + b = [Ns.summation over (i=1)] [y.sub.i] [[alpha].sub.i] [phi] ([x.sub.i]) [phi] (x)] + b (3)

However, classification accuracy of SVM is significantly affected by parameter estimation. Hence, in this paper, we introduce particle swarm optimization to optimize parameters of SVM. In particle swarm optimization, each solution is defined as a particle which is used to obtain the best position. In particular, for each iteration, PSO aims to search for two best solutions: 1) pbest and 2) [DELTA]y [not equal to] c. pbest is defined as the fitness in the current iteration, and gbest means the global best solution.

[v.sup.d.sub.i] (N + 1) = [v.sup.d.sub.i] (N) + [c.sub.1] x [rand1.sup.d.sub.i] x ([pbest.sup.d.sub.i] (N) - [x.sup.d.sub.i] (N)) + [c.sub.2] * [rand2.sup.d.sub.i] x ([gbest.sup.d] (N) - [x.sup.d.sub.i] (N)) (4)

[x.sup.d.sub.i] (N + 1) = [x.sup.d.sub.i] (N) + [v.sup.d.sub.i] (N +1) (5)

where [v.sup.d.sub.i] (N) is the current velocity, d means the dth particle, i refers to the ith variable, and [rand2.sup.d.sub.i] means iteration number. [rand1.sup.d.sub.i] and [rand2.sup.d.sub.i] are random numbers in the range [0,1]. Furthermore, [c.sub.1] and [c.sub.1] denote acceleration degrees. In order to avoid premature convergence, weighting factors are defined as follows.

[v.sup.d.sub.i] (N +1) = [omega] x [v.sup.d.sub.i] (N) + [c.sub.1] x [rand1.sup.d.sub.i] x ([pbest.sup.d.sub.i] (N) - [x.sup.d.sub.i] (N)) + [c.sub.2] x [rand2.sup.d.sub.i] x ([gbest.sup.d] (N) - [x.sup.d.sub.i] (N)) (6)

where parameter w refers to a weighting factor.

Afterwards, the proposed financial crisis short term forecasting algorithm based on a hybrid particle swarm optimization and support vector machine model is illustrated as follows.

Algorithm: Financial crisis short term forecasting algorithm.

Input: Parameter of financial crisis short term forecasting algorithm.

Output: Financial crisis short term forecasting results.

Step 1: Designing a function to estimate initial dynamic inertia weight:

[bar.[gamma]] = [([i.sub.max] - [i.sub.cur]/[i.sub.max]).sup.n] x ([[gamma].sub.ini] - [[gamma].sub.fin]) + [[gamma].sub.fin] (7)

Step 2: Calculating inertia weight:

[[gamma].sup.*] = [(d).sup.r] x [[gamma].sub.ini], d [member of] [0.1,1] (8)

Step 3: If f([P.sub.gd-new]) is lower than f ([P.sub.gd-old])

Step 4: Set r = r - 1

Step 5: Else set r = r + 1

Step 6: Computing stochastically initialization population, velocity, and fitness value for each particle.

Step 7: Initializing position of gbest and pbest.

Step 8: Initializing position of Ibest using the best particle in the initialized population.

Step 9: End If.

Step 10: While max endgen for a generation is not satisfied then.

Step 11: k = k + 1

Step 12: Producing a swarm for the next generation and then evaluate it.

Step 13: End while.

Step 14: Using the above improved PSO to optimize parameters of SVM.

Step 15: Output the level of financial crisis by the SVM classifier.

4. Experiment

As is well known that Quoted company is defined as the company that has a wide influence on the business in modern economic society. From the point of view of enterprise management, quoted company with its accounting information is able to enhance management quality and then promote enterprises' social responsibility. In general, financial crisis short term forecasting is crucial for modern society management. To make performance comparison, standard support vector regression (SVR) and standard SVM classifier are used to compare with our method.

At first, weight of each index is calculated by the Analytic Hierarchy Process (AHP) (shown in Fig. 3).

[FIGURE 3 OMITTED]

Next, we collect financial data from several quoted companies to build up a dataset, and compare our proposed with support vector regression and support vector machine. Error rates of financial crisis short term forecasting with different algorithms are listed in Table. 1 as follows.

Table. 1 demonstrates that the average error rate of our algorithm is 2.92%, and our algorithm obviously performs better than SVR and SVM. Next, we take quoted company 1 and 2 as examples to show forecasting values and error rates for different algorithm. In particularly, financial data in 100 days are extracted from quoted company 1 and 2 are used as testing dataset.

[FIGURE 4 OMITTED]

[FIGURE 5 OMITTED]

[FIGURE 6 OMITTED]

[FIGURE 7 OMITTED]

Combining the experimental results in Fig. 4 to Fig. 7, we can see that the forecasting values of our proposed is much closer to real values than SVR and SVM, and forecasting error rates of our proposed method are much lower than other algorithm. The reasons lie in the following aspects: 1) We convert the financial crisis short term prediction problem to an artificial intelligence based classification problem, 2) In order to promote the performance of particle swarm optimization we design a high inertia weight to construct a new searching space, 3) For various values of particle number, inertia weight decreases with paths varying, and 4) final inertia weight is gained if at the max number of iterations is reached.

5. Conclusion

This paper aims to tackle the financial crisis short term forecasting problem. An index system is provided to map high dimension financial data to a low dimension space, and four factors are included: a) Profitability, b) Solvency, c) Operating capacity, and d) Composition of capital. The main innovation of this paper is that short term financial crisis is predicted by integrating particle swarm optimization and support vector machine together. Finally, experimental results prove the effectiveness of our proposed algorithm.

Recebido/Submission: 12-09-2015

Aceitacao/Acceptance: 09-11-2015

References

Audrino Francesco. (2014). Forecasting Correlations During the Late-2000s Financial Crisis: the Short-Run Component, the Long-Run Component, and Structural Breaks. Computational Statistics & Data Analysis, 76, 43-60.

Berger T, Missong M, Financial Crisis. (2014). Value-At-Risk Forecasts And The Puzzle Of Dependency Modeling. International Review of Financial Analysis, 33, 33-38.

Cao Wei, Cao Longbing. (2015). Financial Crisis Forecasting via Coupled Market State Analysis. IEEE Intelligent Systems, 30(2), 18-25.

Chen Cathy W S, Gerlach Richard, Lin Edward M H, Lee W C W. (2012). Bayesian Forecasting for Financial Risk Management, Pre and Post the Global Financial Crisis. Journal of Forecasting, 31(8), 661-687.

Chen Li-Hui, Guo Tsuei-Yang. (2011). Forecasting Financial Crises for an Enterprise by Using the Grey Markov Forecasting Model. Quality & Quantity, 45(4), 911-922.

Cipollini A, Kapetanios G. (2009). Forecasting Financial Crises and Contagion in Asia Using Dynamic Factor Analysis. Journal of Empirical Finance, 16(2), 188-200.

De Sousa Gabriel Vtor Manuel. (2014). Multivariate Models to Forecast Portfolio Value at Risk: from the Dot-Com crisis to the global financial crisis. Rbgn-Revista Brasileira De Gestao De Negocios, 16(51), 299-318.

Garcia Nieto P J, Alonso Fernandez J R, Gonzalez Suarez V M, Diaz Muniz C, Garcia-Gonzalo E, Mayo Bayon R. (2015). A Hybrid PSO Optimized SVM-based Method for Predicting of the Cyanotoxin Content from Experimental Cyanobacteria Concentrations in the Trasona Reservoir: A Case Study in Northern Spain. Applied Mathematics and Computation, 260, 170-187.

Garcia Nieto P J, Garcia-Gonzalo E, Alonso Fernandez J R, Diaz Muniz C. (2016). A Hybrid PSO Optimized SVM-based Model for Predicting a Successful Growth Cycle of the Spirulina Platensis from Raceway Experiments Data. Journal of Computational and Applied Mathematics, 291, 293-303.

Garcia Nieto P J, Garcia-Gonzalo E, Sanchez Lasheras F, de Cos Juez F J. (2015). Hybrid PSO-SVM-based Method for Forecasting of the Remaining Useful Life for Aircraft Engines and Evaluation of its Reliability. Reliability Engineering & System Safety, 138, 219-231.

Hajek Petr, Strizik Michal, Praks Pavel, Kaderabek Petr. (2009). Possibilities of Financial Crises Forecasting with Latent Semantic Indexing. Politicka Ekonomie, 57(6), 754-768.

Halbleib Roxana, Pohlmeier Winfried. (2012). Improving the Value at Risk Forecasts: Theory and Evidence from the Financial Crisis. Journal of Economic Dynamics & Control, 36(8), 1212-1228.

Hsieh Ching-Tang, Hu Chia-Shing. (2014). Fingerprint Recognition by Multi-objective Optimization PSO Hybrid with SVM. Journal of Applied Research and Technology, 12(6), 1014-1024.

Kunze Frederik, Kramer Jens. (2014). Rudschuck Norman, Interest Rate Forecasts in Times of Financial Crisis: What Might be Interesting to Know. European Journal of Political Economy, 34, S45-S52.

Liu Baoling, Hou Dibo, Huang Pingjie, Liu Banteng, Tang Huayi, Zhang Wubo, Chen Peihua, Zhang Guangxin. (2013). An Improved PSO-SVM Model for Online Recognition Defects in Eddy Current Testing. Nondestructive Testing and Evaluation, 28(4), 367-385.

Metescu Ana-Maria, Borisov Daniela, Banica Ramona Mihaela. (2013). Forecasting Rates of Return Using Geometric Brownian Motion During the Global Financial Crisis Period. Metalurgia International, 18(4), 134-137.

Oh KJ, Kim TY, Lee HY, Lee H, Zhang S, Jarvis R. (2005). Using Neural Networks to Support Early Warning System for Financial Crisis Forecasting. Advances in Artificial Intelligence, 3809, 284-296.

Oliveira, J. A., Ferreira, J., Figueiredo, M., Dias, L., & Pereira, G. (2014). Sistema de Apoio a Decisao para o Transporte Nao Urgente de Doentes em Veiculo Partilhado. RISTI--Revista Iberica de Sistemas e Tecnologias de Informacao, 2014(13), 17-33.

Selakov A, Cvijetinovic D, Milovic L, Mellon S, Bekut D. (2014). Hybrid PSO-SVM Method for Short-Term Load Forecasting During Periods with Significant Temperature Variations in City of Burbank. Applied Soft Computing, 16, 80-88.

Subasi Abdulhamit. (2013). Classification of EMG Signals Using PSO Optimized SVM for Diagnosis of Neuromuscular Disorders. Computers in Biology and Medicine, 43(5), 576-586.

Sudheer Ch, Maheswaran R, Panigrahi B K, Mathur Shashi. (2014). A Hybrid SVM-PSO Model for Forecasting Monthly Streamflow. Neural Computing & Applications, 24(6), 1381-1389.

Svetlova Ekaterina. (2012). Talking about the Crisis: Performance of Forecasting in Financial Markets. Culture And Organization, 18(2), 155-169.

Wang Wen-chuan, Xu Dong-mei, Chau Kwok-wing, Chen Shouyu. (2013). Improved Annual Rainfall-Runoff Forecasting Using Pso-Svm Model Based on EEMD. Journal of Hydroinformatics, 15(4), 1377-1390.

Yu Lean, Wang Shouyang, Lai Kin Keung, Wen Fenghua. (2010). A Multiscale Neural Network Learning Paradigm for Financial Crisis Forecasting. Neurocomputing, 73(4-6), 716-725.

Zaleskiewicz Tomasz. (2011). Financial Forecasts during the Crisis: Were Experts More Accurate Than Laypeople. Journal of Economic Psychology, 32(3), 384-390.

Guo Zhiqiang, Wang Huaiqing, Liu Quan. (2013). Financial Time Series Forecasting Using Lpp and Svm Optimized by PSO. Soft Computing, 17(5), 805-818.

Smola Alex J. (2004). Scholkopf Bernhard, A tutorial on support vector regression. Statistics and computing, 14(3), 199-222.

Jihong Huang (1)

huang_j_h@yahoo.com.cn

(1) Anhui Audit College, 230601, Heifei, China.

DOI: 10.17013/risti.17A.125-135

Table 1--Error rates of Financial Crisis Short Term Forecasting with Different Algorithms. No. of SVR (%) SVM (%) Our quoted algorithm (%) company 1 12.12 11.47 2.49 2 7.32 6.56 4.43 3 8.20 8.08 3.25 4 5-79 4.78 2.06 5 11.35 10.80 2.29 6 10.74 9.78 3.04 7 6.12 5.18 4.96 8 6.98 5.85 1.51 9 9.93 9.29 1.06 10 0.47 0.42 4.13 Average 7.9 7.22 2.92

Printer friendly Cite/link Email Feedback | |

Author: | Huang, Jihong |
---|---|

Publication: | RISTI (Revista Iberica de Sistemas e Tecnologias de Informacao) |

Date: | Mar 15, 2016 |

Words: | 3069 |

Previous Article: | A clustering method for high-dimensional data analysis in stock market. |

Next Article: | Three dimensional weld pool surface reconstruction based on shape from shading algorithm. |

Topics: |