Financial forecasting using ANFIS networks with Quantum-behaved Particle Swarm Optimization

Abstract

To be successful in financial market trading it is necessary to correctly predict future market trends. Most professional traders use technical analysis to forecast future market prices. In this paper, we present a new hybrid intelligent method to forecast financial time series, especially for the Foreign Exchange Market (FX). To emulate the way real traders make predictions, this method uses both historical market data and chart patterns to forecast market trends. First, wavelet full decomposition of time series analysis was used as an Adaptive Network-based Fuzzy Inference System (ANFIS) input data for forecasting future market prices. Also, Quantum-behaved Particle Swarm Optimization (QPSO) for tuning the ANFIS membership functions has been used. The second part of this paper proposes a novel hybrid Dynamic Time Warping (DTW)-Wavelet Transform (WT) method for automatic pattern extraction. The results indicate that the presented hybrid method is a very useful and effective one for financial price forecasting and financial pattern extraction.

Introduction

The FX is the largest global market today. According to the September 2013 report of the Bank for International Settlements (2013), Global FX turnover reached $5.3 trillion a day in 2013. There are two main groups that trade on the FX market. The first group is companies and governments that use the FX market to convert domestic currency into a foreign currency for international business transactions. The second group consists of investors that trade in order to make a profit on the Forex market. Speculators on the FX market range from large banks to home-based operators (Archer, 2010).

As with other financial markets, the most important factor for being successful in FX trading is the ability to correctly predict future market fluctuations. If a speculator can “buy low and sell high”, then he or she will make a profit. There are wild variations in exchange rates on the FX market, and it is difficult for traders to make the right decision to buy or sell. Forecasting future FX exchange rates is an intriguing subject for many speculators. They use artificial intelligent models to forecast future market values and look for complex chart patterns. The objective of this paper is to propose a hybrid artificial intelligence model as a trading advisory system. An ANFIS-QPSO hybrid system is used as a one-step-ahead forecasting method. Wavelet coefficients of time series are used as the ANFIS input parameters. The paper also presents a hybrid Dynamic Time Warping (DTW)-Wavelet Transform (WT) method for automatic pattern extraction from a financial time series. This study attempts to make correct trading signals based on forecasted market values and identified chart patterns. The proposed model can help traders to reduce trading risks and to increase their profit.

According to the efficient market theory, it is nearly impossible to accurately make long-term predictions based on historical market data. But in the short-term, there are some hidden repeating patterns that, if we can identify them, could help us make a profit from the market (Liu & Kwong, 2007). Professional traders use two major types of analysis to make accurate decisions in financial markets: fundamental and technical. Fundamental analysis is based on the overall state of the economy, the state of the industry and a company’s overall financial situation. Technical analysis, on the other hand, relies on charts and historical data, and is based on the idea that history will repeat itself. Therefore, by analyzing past data, we can forecast future market trends. The two main methods used for technical analysis are statistical-based indicators and chart patterns. Statistical methods include the use of moving averages to find a mathematical relationship between past data that can be used for evaluating future market values. Finally, if a specific pattern appears in a chart, chart patterns analysis can be used to predict the future trends.

There are many papers dealing with financial market forecasting. Many of them employ soft computing techniques, such as genetic algorithms, neural networks, and neuro-fuzzy systems. Most previous works have presented methods to try to accurately predict the market. Obviously, based on the current human knowledge, it is impossible to correctly predict the exact market values. Therefore, researchers have used complicated methods to minimize forecasting error. To be successful in real market trading, professional traders place great value on predicting future market trends. Instead of trying to forecast an exact market value, the model proposed in this paper provides a trading advisory signal generated from predicted market trends.

The innovations of the presented method are as follows:

• The majority of earlier papers used just one of the technical indicators or chart patterns as a forecasting method. The proposed method uses wavelet decomposition of time series and extracted chart patterns as the system inputs.

• An ANFIS-QPSO hybrid method has rarely been used for financial forecasting. A novel method for tuning ANFIS membership functions by QPSO has been proposed. Experimental results have shown that this model is very accurate and highly efficient for forecasting in financial markets, especially in the Forex market.

• DTW has rarely been used for finding patterns in financial time series. In this paper a state-of-the-art hybrid algorithm, combining Dynamic Time Warping and Wavelet Transform, is presented to extract the shaped patterns in financial time series.

• The majority of earlier papers employed a single shape template or time series as a target pattern. These patterns are generally independent from the input time series. The presented method creates an adaptive pattern based on the main features of pattern and input time series to accurately predict the current data.

Section 2 of this paper provides the literature review, Section 3 reviews basic methods. The proposed method is described in Section 4. Section 5 describes the experiments and makes comparisons. Finally, Section 6 examines the conclusions of the study.