Hybrid approach to the Japanese candlestick method for financial forecasting Takenori Kamo * , Cihan Dagli 1 Smart Engineering Systems Lab, University of Missouri – Rolla, Rolla, MO 65401, USA article info Keywords: Financial forecasting Committee machine Gating networks Neural networks abstract This paper discusses an experimental study of the Japanese candlestick method as used in hybrid stock market forecasting models. Two models are presented in this paper. Model 1 is a committee machine with simple generalized regression neural networks (GRNN) experts. This model also has a simple gating network. Model 2 has a similar committee machine along with a hybrid type gating network that con- tains fuzzy logic. Model 1 was developed to introduce the candlestick method and examine whether using the candle- stick method improves performance. Model 2 is developed to determine whether the application of fuzzy logic could improve the former model. This model uses standard IF-THEN rules based fuzzy logic. In the experiment, a few simple Japanese candlestick patterns are integrated into the models. Both models use the same simple candlestick patterns to provide a basis for comparison. The Japanese candle- stick method is implemented in the gating network. Model 1 uses features of candlestick patterns in the gating network. Model 2 uses candlestick patterns for recognizing the strength of market conditions. To investigate the performance of these models, the daily stock quotes of Hewlett-Packard, Bank of America, Ford, DuPont, and Yahoo are used as input data sets. The performance of the models was satis- factory based on the mean squared error. Ó 2008 Elsevier Ltd. All rights reserved. 1. Introduction Many financial forecasting models have been developed to pre- dict stock market price. Some of them are neural network based models, many of which have limitations because several factors af- fect market movement and it is difficult to handle large data sets in one neural network. However, a neural network has the ability to learn from the past data set in order to calculate the data relation and can generate a new data set as an output for predicting an un- known case (Ruggiero, 2006). For this reason, neural networks have been widely used to develop financial forecasting models. A large data set can also be divided into smaller pieces for han- dling by several neural networks. A committee machine is the best candidate for this purpose because it can handle many factors or tasks affecting problem solving. Disorntetiwat and Dagli’s (2000) study showed that it is best to let one GRNN forecast one category or one factor to generate an output. To mediate the outputs from each neural network, the gating network is used to generate one output. Forecasting stock prices is very complex and many factors are involved in stock price movement. In this environment, the com- mittee machines’ experts best control their specialized area. Each expert is assigned to manage only one task, then the addition of gating networks adjusts the results from the experts and combines them to produce the final output. One key feature of the models is the introduction of the Japa- nese candlestick technique. The Japanese candlestick technique is believed to be a useful indicator for forecasting stock prices (Nis- son, 2001). This model uses some basic Japanese candlestick pat- terns to observe whether the candlestick method improves the performance of the models. In reality, the Japanese candlestick method is a perception- based technique in which precision is not necessary. Therefore, a way of handling the imprecise knowledge of candlestick patterns is required. To solve this problem, a fuzzy logic model is intro- duced. Fuzzy logic provides a wide variety of concepts and tech- niques for representing knowledge that is uncertain or imprecise (Zadeh, 1992) and can be used to analyze the Japanese candlestick patterns. Two models are presented in this paper. Model 1 is a committee machine with simple generalized regression neural networks (GRNN) experts and a gating network. Model 2 has a similar com- mittee machine with a hybrid type gating network that is con- trolled by fuzzy logic. Model 1 uses Japanese candlestick patterns’ features in the gating network as a key element of medi- ation. Model 2 uses fuzzy logic for recognizing the candlestick pat- terns and stock market conditions. 0957-4174/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2008.06.050 * Corresponding author. Tel.: +1 573 341 6556; fax: +1 573 341 6567. E-mail addresses: takenori@umr.edu (T. Kamo), dagli@umr.edu (C. Dagli). 1 Tel.: +1 573 341 7211; fax: +1 573 341 7238. Expert Systems with Applications 36 (2009) 5023–5030 Contents lists available at ScienceDirect Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa