Reliability analysis of technical trading rules in security market based on conditional random field model

Abstract

Conditional random field model is a powerful model for dealing with time series with hidden states and is becoming more and more widely used in various fields. This model generalizes the hidden Markov chain model by releasing the independence assumptions between observations and obtains more accurate estimation results. Although conditional random field model has been successfully applied in both academics and industry, it has rarely been used for analyzing financial time series. In financial industry, there are mainly two types of skills for investment analysis: fundamental analysis, and technical analysis. Although these two kinds of analysis are equally important to security market participants, technical analysis does not receive enough attention from academics. Technical analysis is deemed to be a pseudoscience such as astrology and alchemy by the researchers of mainstream finance. One of the most important reasons is that technical analysis is contradictory with efficient market hypothesis, which is the theoretical foundation of modern finance. Also, the subjective nature of technical analysis makes it unfalsifiable, which constricts its applications. There is positive evidence of the profitability of technical analysis resulting from continued empirical studies testing the effectiveness of technical analysis in different security markets. The rapid development of computer science makes it possible to process large-scale calculations allowing increased interest in technical analysis. Some researchers have used behavioral economics to explain extra profits of technical analysis and several have used statistics and machine learning techniques to investigate the profitability of technical analysis. However, a uniform framework for technical analysis is still not available. Even the definitions of technical analysis in different studies are not consistent. In this thesis, we concentrate on technical analysis. Given trading signal sequence indicated by technical trading rules, we apply conditional random field to analyze this time series. Considering the issues mentioned above technical trading signals may be unreliable. The reliability analysis is important for the application of technical analysis. The research objective of this thesis is to apply conditional random field model to improve the reliability of technical analysis. Technical analysis depends on trading signals generated by technical trading rules, and these signals form a time series which is the target for our research. To achieve this objective the following problems should be solved. 1. How to scientifically measure the reliability of technical analysis? 2. What are the quantitative characteristics of trading signals for conditional random model? 3. How to use conditional random model to improve the reliability? For solving these research questions, we firstly redefine technical analysis scientifically, and then develop a framework based on conditional random field model for reliability of technical analysis. The technical analysis is defined quantitatively. The framework for reliability analysis is based on conditional random field model and also employs statistical methods such as two tailed t-test. Heng Seng Index stocks in Hong Kong stock market were taken as examples for experiments, and the famous simple moving average rule and Bollinger Bands rule were tested. Compared with the trading strategy without technical trading rules, i.e., buy-and-hold strategy, it has been found that most simple moving average rules cannot bring significantly larger returns while a Bollinger Bands rule is effective to indicate good trading chances.