Financial market monitoring by case-based reasoning

https://doi.org/10.1016/j.eswa.2006.01.044Get rights and content

Abstract

This paper shows that case-based reasoning (CBR), an artificial intelligence technique, is a quite efficient tool in monitoring financial market against its possible collapse. For this purpose, daily financial condition indicator (DFCI) monitoring financial market is built on CBR and its performance is compared to DFCI on neural network. This study is empirically done for the Korean financial market.

Introduction

During the last decade, every financial market in the world more often than not had bump into sudden collapses that cause a severe blow to the market (Eichengreen et al., 1995, Frankel and Rose, 1996, Kaminsky and Reinhart, 1999). To avoid or prepare for such sudden collapse, it is necessary to have a proper tool that monitors the market efficiently. Recently Kim et al., 2004, Kim et al., 2004 found that neural network (NN) may monitor financial market effectively thanks to its over-fitting tendency. Problem with NN as a monitoring tool, however, is its difficulty in updating since it usually requires a large amount of training data for its proper functioning (Refenes, 1995). This is not desirable for monitoring financial market since modern financial market tends to undergo change of its mechanism over a short period of time and hence needs to be updated regularly with relatively small amount of data. To resolve such updating difficulty, we propose case-based reasoning (CBR) for financial market monitoring and examine its efficiency. It will be shown that CBR is more efficient than NN. Recall that CBR is known as a very useful artificial intelligence technique that could be efficiently trained on relatively small amount of data.

For our discussion, daily financial condition indicator (DFCI) is built with CBR and then compared to its counter part (DFCI with NN). Specifically DFCI’s built with CBR and NN respectively are constructed for the Korean financial market and compared. This paper consists of as follows. Following Sections 1 Introduction, 2 DFCI construction describes DFCI construction procedure and reviews CBR briefly. Section 3 establishes and compares DFCI’s for the Korean financial market. Concluding remarks are given in Section 4.

Section snippets

DFCI construction

The core of CBR is the case base which stores a collection of cases or memories from the past. If a new event or problem occurs, CBR recognizes it as a target case and retrieves multiple exemplars of the target case from the case base (i.e., source case is found). Then CBR is to produce a response or output corresponding to the new event. Inside such procedure, two key parameters, the locale k (or number of neighbors for the source case) and the weights of input variables are to be specified (

An empirical study

Korea had experienced the economic crisis during 1997–1998 which had initially started as financial market collapses in late 1997. Since then much attention has been paid to efficient monitoring of financial market. Kim et al., 2004, Kim et al., 2004 proposed using NN for stock market monitoring but found that the trained neural network is hard to update. In this section, DFCI is reconstructed by employing CBR instead of NN. Throughout this section, the variable names given in Table 1 are used.

Concluding remarks

Modern financial market tends to be easily unstable since external or internal shocks spread easily through fast and massive electronic communication or transaction system. Recent economic or financial crises in the 1990s seem to provide typical examples of these, which emphasizes the need to have a proper monitoring tool of financial markets against its possible crash. Recently Kim et al., 2004, Kim et al., 2004 found that NN may monitor financial market effectively thanks to its over-fitting

Acknowledgement

This research was supported by the Program for the Training of Graduate Students in Regional Innovation which was conducted by the Ministry of Commerce Industry and Energy of the Korean Government.

References (14)

There are more references available in the full text version of this article.

Cited by (0)

1

Tel.: +82 53 580 5533.

View full text