INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 8, ISSUE 12, DECEMBER 2019 ISSN 2277-8616 Adoptive Trend Following Strategy In Financial Time Series With Multi-Objective Function Ajit Kumar Rout, Satish Muppidi Abstract: Given the complex nature of the stock market, determining the stock market timing - when to buy or sell is a challenge for investors. There are two basic methodologies used for prediction in financial time series; fundamental and technical analysis. The fundamental analysis depends on external factors such as economic environment, industry and company performance. Technical analysis utilizes financial time series data such as stock price and trading volume. Trend following (TF) is an investment strategy based on the technical analysis of market prices. Trend followers do not aim to forecast nor do they predict specific price levels. They simply jump on the uptrend and ride on it until the end of this uptrend. Most of the trend followers determine the establishment and termination of uptrend based on their own rules. In this paper, we propose a TF algorithm that employs multiple pairs of thresholds to determine the stock market timing. The experimental result on 7 stock market indexes shows that the proposed multi-threshold TF algorithm with multi-objective function is superior when it is compared to static, dynamic, and float encoding genetic algorithm based TF. Index Terms: financial time series, technical analysis, trend following Algorithm. —————————— —————————— 1. INTRODUCTION identify the establishment of an uptrend and determine It is believed that the future behavior can be predicted by whether the uptrend could persist for a certain period of time studying the past behaviors. This theory is the foundation for based on their own strategies. The uptrend which could persist many application domains. For instance, meteorologists for a certain period of time is identified as a strong uptrend. forecast weather based on the analysis of past climate data. Trend followers enter in market after a strong uptrend is This theory can also be applied in financial trading. Financial established and exit at the end of this strong uptrend. markets are complex dynamic systems with a high number of Therefore, how to identify the establishment and termination of active agents (investors, traders and hedgers), influenced by strong uptrend is the key issue in TF. In recent years, several each other and by external information (news, economic data, TF algorithms have been developed. Static TF algorithm [5] events). This produces a behavior with high randomness and employs a pair of thresholds P and Q to generate buy and sell noise which is very difficult to predict [13]. Thus, various tools signals. P and Q are constants and their values can be and methods have been developed to help investors study determined from the historical data. Adaptive TF algorithm [6] these behaviors in financial markets. These studies are very also employs a pair of thresholds Pt and Qt to determine the useful for predicting future movements. In financial trading, market timing. In an adaptive TF algorithm, the values of Pt there are two basic types of analysis fundamental and and Qt are adjusted dynamically to adapt to the market trend. technical. Fundamental analysis is a methodology of Adaptive TF algorithm can be considered as a statistical evaluating a security by attempting to measure its intrinsic method since thresholds Pt and Qt are not related to value by examining related economic, financial and other historical data. For instance, in [13], Machado et al. develop an qualitative and quantitative factors [2]. Fundamental analysis adaptive TF algorithm with multi-time frame trading rules using attempts to study all of related factors that can affect the Genetic Algorithms. security’s value. Therefore, it is an arduous task to find as In this paper, we propose a multi-threshold TF algorithm to many related factors as possible and to evaluate each factor’s determine the right moment to buy and sell shares. Each influence on the price movement. In contrast, technical threshold is designed to respond specific cases of price analysis evaluates a security only based on financial time movement in its own working region. To reduce human bias, series data such as daily price and trading volume. Technical the search spaces of threshold s and parameter s are analysts locate interesting patterns in the time series and use determined from the price data rather than input by the users. these patterns to predict the future movements. Technical This paper makes following contributions to the trend following analysis is based on the theory that, at any given point in time, literature. A novel approach is proposed to identify time series data already reflects all known factors affecting unprofitable weak up trends in trend following. By using supply and demand for that particular market [3]. Trend working regions, we propose an approach for responding to Following (TF) is a popular investment strategy based on stock soar and slump in trend following. By using RSI based technical analysis. In a stock market, a trend can be defined thresholds, we show how to determine the strength of as the general direction of the market movement. Uptrend is uptrend’s and downtrends in trend following. We propose a the upward movement and downtrend is the downward new multi-threshold trend following algorithm with multi- movement of the stock price or the market as measured by an objective function. The algorithm was tested on seven stock average or index over a period of time [2]. Trend followers indices. Experimental results show that the proposed algorithm usually do not predict specific price levels. Rather, they simply achieves best result in maximizing SR and AROI when it is compared with static TF, dynamic TF, and TF with Float ———————————————— Encoding Genetic algorithms. The remainder of the paper is Ajit Kumar Rout is currently working as Professor and HOD in organized into four sections. A review the existing TF Department of IT, GMRIT, Rajam, Andhra Pradesh, India, E-mail: algorithms are given in Section 2. The proposed multi- [email protected] threshold TF algorithm is detailed in Section 3. The Satish Muppidi is currently working as an Assistant Professor in Department of CSE. GMRIT, Rajam, Andhra Pradesh, India E-mail: experimental results are given in 4. We conclude the paper in [email protected] Section 5. 3746 IJSTR©2019 www.ijstr.org INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 8, ISSUE 12, DECEMBER 2019 ISSN 2277-8616 2 RELATED WORK repeat TF can be seen as a simple trading idea since it generates Compute EMA(t); buy and sell signals only based on the observation of trend If EMA(t)>EMA(t−1)&&EMA(t−2)>EMA(t−1) then changes. TF has many manifestations so far. Many breakout upt=EMA(t−1); systems, moving average systems and volatility systems can Current trend is an uptrend be considered to be TF in nature. The essential operation elseif EMA(t)<EMA(t−1)&&EMA(t−2)<EMA(t−1) then phases of TF can be summarized into open, hold and, close dtp=EMA(t−1); positions. A position is a commitment to buy or sell a given Current trend is a downtrend amount of shares. ―Open a position‖ is the action of buying in end a security. Likewise, ―close a position‖ is the action of selling if no position is opened && current trend is uptrend then out the held security. In technical analysis, technical indicators if EMA(t) – utp ≥ P then are very useful tools in identifying the direction of market trend Open aposition and analyzing the market circumstances. However, exclusively end using technical indicators to implement TF often fails to else if the position is opened && current trend is downtrend produce a good result. In recent years, several TF algorithms then have been developed such as static, dynamic and fuzzy TF if dtp−EMA(t)≥Q then algorithms. Close the position end 2.1 Static TF Algorithm end Static TF algorithm was proposed by Fong et al. [5]. The main if at the end of market then idea of static TF algorithm is to smooth the price data by Close the opened position using Exponential Moving Average (EMA) and determine the end trading timing based on a pair of thresholds P and Q. until at the end of market; Threshold P (Q) is the vertical length of an upward (downward) Algorithm-2: Adaptive TF algorithm EMA. Threshold P is used for opening a position and threshold repeat Q is used for closing the opened position. Threshold P and Q Compute RSI(t) and RSI(EMA(t)); are static constants and their optimal values can be manually If RSI(t) >50then set by the traders or obtained from the study of historical price current trend is a uptrend data. The concept of thresholds P and Q is depicted in Figur-1 else if RSI(t) <50 then and the pseudo-code of static TF algorithm is given in current trend is a down trend Algorithm-1. For the example in Figure-1, a position will be end if RSI(t) >RSI(EM A(t)) &&40 <RSI(EM A(t)) >60 && opened at time point b and closed at time point d. current trend is a uptrend&& no position is opened then 2.2 Adaptive TF Algorithm Open a position, Pt In adaptive TF algorithm, the thresholds P t and Qt are else if RSI(t) < RSI (EM A(t)) &&40 <RSI(EM A(t)) >60 && variables instead of static constants. The values of thresholds current P t and Qt are dynamically changed according to the market trend is a down trend && the position is opened then trend.