Research Article  Open Access
Jian Wang, Junseok Kim, "Predicting Stock Price Trend Using MACD Optimized by Historical Volatility", Mathematical Problems in Engineering, vol. 2018, Article ID 9280590, 12 pages, 2018. https://doi.org/10.1155/2018/9280590
Predicting Stock Price Trend Using MACD Optimized by Historical Volatility
Abstract
With the rapid development of the financial market, many professional traders use technical indicators to analyze the stock market. As one of these technical indicators, moving average convergence divergence (MACD) is widely applied by many investors. MACD is a momentum indicator derived from the exponential moving average (EMA) or exponentially weighted moving average (EWMA), which reacts more significantly to recent price changes than the simple moving average (SMA). Traders find the analysis of 12 and 26day EMA very useful and insightful for determining buyandsell points. The purpose of this study is to develop an effective method for predicting the stock price trend. Typically, the traditional EMA is calculated using a fixed weight; however, in this study, we use a changing weight based on the historical volatility. We denote the historical volatility index as HVIX and the new MACD as MACDHVIX. We test the stability of MACDHVIX and compare it with that of MACD. Furthermore, the validity of the MACDHVIX index is tested by using the trend recognition accuracy. We compare the accuracy between a MACD histogram and a MACDHVIX histogram and find that the accuracy of using MACDHVIX histogram is 55.55% higher than that of the MACD histogram when we use the buyandsell strategy. When we use the buyandhold strategy for 5 and 10 days, the prediction accuracy of MACDHVIX is 33.33% and 12% higher than that of the traditional MACD strategy, respectively. We found that the new indicator is more stable. Therefore, the improved stock price forecasting model can predict the trend of stock prices and help investors augment their return in the stock market.
1. Introduction
Securities investment is a financial activity influenced by many factors such as politics, economy, and psychology of investors. Its process of change is nonlinear and multifractal [1]. The stock market has highrisk characteristics; i.e., if the stock price volatility is excessive or the stability is low, the risk is uncontrollable. Financial asset returns in the short term are persistent; however, those in the long term will be reversed [2].
Asness [3] reported that the stock, foreign exchange, and commodity markets have a trend. Hassan [4] noted that complex calculations are not particularly effective for predicting stock markets. Many trend analysis indicators and prediction methods for financial markets have been proposed. Pai [5] used Internet search trends and historical trading data to predict stock markets using the least squares support vector regression model. Lahmiri [6] accurately predicted the minuteahead stock price by using singular spectrum analysis and support vector regression. Researchers have also used other methods to forecast stock markets. Singh et al. [7] designed a forecasting model consisting of fuzzy theory and particle swarm optimization to predict stock markets using historical data from the State Bank of India. Lahmiri et al. [8] proposed an intelligent ensemble forecasting system for stock market fluctuations based on symmetric and asymmetric wavelet functions. Das et al. [9] proposed a hybridized machinelearning framework using a selfadaptive multipopulationbased Jaya algorithm for forecasting the currency exchange value. Laboissiere et al. [10] developed a model involving correlation analysis and artificial neural networks (NNs) to predict the stock prices of Brazilian electric companies. Lei [11] proposed a wavelet NN prediction method for the stock price trend based on rough set attribute reduction. Lahmiri [12] used variational mode decomposition to forecast the intraday stock price. Lahmiri [13] addressed the problem of technical analysis information fusion and reported that technical information fusion in an NN ensemble architecture improves the prediction accuracy. In [14], the authors argued that time series of stock prices are nonstationary and highly noisy. This led the authors to propose the use of a wavelet denoisingbased backpropagation (WDBP) NN for predicting the monthly closing price of the Shanghai composite index. Shynkevich et al. [15] investigated the impact of varying the input window length and the highest prediction performance was observed when the input window length was approximately equal to the forecast horizon. In [16], a prediction model based on the input/output data plan was developed by means of the adaptive neurofuzzy inference system method representing the fuzzy inference system. Zhou et al. [17] proposed a stock market prediction model based on highfrequency data using generative adversarial nets. Wang et al. [18] used a bimodal algorithm with a datadivider to predict the stock index. In [19], the author used multiresolution analysis techniques to predict the interest rate nextday variation. Using Kline patterns’ predictive power analysis, Tao et al. [20] found that their proposed approach can effectively improve prediction accuracy for stock price direction and reduce forecast error.
We will introduce the concept of moving average convergence divergence (MACD) and help the readers understand its principle and application in Section 2. Although the MACD oscillator is one of the most popular technical indicators, it is a lagging indicator. In Section 3, we propose an improved model called MACDHVIX to deal with the lag factor. In Section 4, data for empirical research are described. Finally, in Section 5, we develop a trading strategy using MACDHVIX and employ actual market data to verify its validity and reliability. We also compare the prediction accuracy and cumulative return of the MACDHVIX histogram with those of the MACD histogram. The performance of MACDHVIX exceeds that of MACD. Therefore, the trading strategy based on the MACDHVIX index is useful for trading. Section 6 presents our conclusion.
2. MACD and Its Strategy
MACD evolved from the exponential moving average (EMA), which was proposed by Gerald Appel in the 1970s. It is a common indicator in stock analysis. The standard MACD is the 12day EMA subtracted by the 26day EMA, which is also called the DIF. The MACD histogram, which was developed by T. Aspray in 1986, measures the signed distance between the MACD and its signal line calculated using the 9day EMA of the MACD, which is called the DEA. Similar to the MACD, the MACD histogram is an oscillator that fluctuates above and below the zero line. The construction formula is as follows: where , , and . The weight number is a fixed value equal to . The number of the MACD histogram is usually called the MACD bar or OSC. The analysis process of the cross and deviation strategy of DIF and DEA includes the following three steps.
(i) Calculate the values of DIF and DEA.
(ii) When DIF and DEA are positive, the MACD line cuts the signal line in the uptrend, and the divergence is positive, there is a buy signal confirmation.
(iii) When DIF and DEA are negative, the signal line cuts the MACD line in the downtrend, and the divergence is negative, there is a sell signal confirmation.
3. MACDHVIX Weighted by Historical Volatility and Its Strategy
The essence of a good technical indicator is a smooth trading strategy; i.e., the constructed index must be a stationary process. We present an empirical study in Section 5. The validity and sensitivity of MACD have a strong relationship with the choice of parameters. Different investors choose different parameters to achieve the best return for different stocks. In this study, the weight is based on the historical volatility. It is expected that the accuracy and stability of MACD can be improved. The construction formula is as follows:Here, the weight changes over time; HVIX is the change index of the historical volatility of a stock. The HVIX in this paper is the change index of the volatility in the past days. It is similar to the market volatility index VIX used by the Chicago options exchange. It reflects the panic of the market to a certain extent; thus, it is also called the panic index. The above process is expressed by the code shown in Algorithm 1.

The analysis process of the cross and deviation strategy of DIFHVIX and DEAHVIX includes the following three steps.(i)Calculate the values of DIFHVIX and DEAHVIX.(ii)When DIFHVIX and DEAHVIX are positive, the MACDHVIX line cuts the signal line of HVIX in the uptrend, and the divergence is positive, there is a buy signal confirmation.(iii)When DIFHVIX and DEAHVIX are negative, the signal line of HVIX cuts the MACDHVIX line in the downtrend, and the divergence is negative, there is a sell signal confirmation.
4. Data Description
We first perform an empirical study on the buyandsell strategy, which involves buying today and selling tomorrow. We use the historical data for the stock “zgrs” from November 2, 2015, to September 21, 2017, from the Shanghai stock market. First, we develop the strategy for the new index and calculate the prediction accuracy and cumulative return of the stock with two different indicators. Then, we compare the accuracy rate and cumulative return. The accuracy here is calculated according to whether the stock price rises on the second day. Furthermore, we test a buyandhold strategy for the proposed model. The buyandhold strategy is a trading strategy in which the traders hold the stock for a while instead of selling it on the next trading day. We use the historical data for the stock “dggf” from July 27, 2009, to November 3, 2017, from the Shanghai stock market to test a 5 d buyandhold strategy and use the historical data for the stock “payh” from June 22, 1993, to May 10, 2010, from the Shanghai stock market to test a 10 d buyandhold strategy. The detailed trading strategy is similar to the buyandsell strategy. Here, we use , , and
5. Empirical Results
5.1. Empirical Results of BuyandSell Strategy
From the “zgrs” stock data chosen in Section 4, we calculate the HVIX of the past m days and the past n days. A higher stock index means that investors feel anxiety regarding the stock market, and a lower stock index means that the rate of change of the stock price will decrease. Figure 1 shows the HVIX index.
Next, using the calculated volatility index, we calculate the weight of the EMA formula in Section 3 and obtain the values of MACDHVIX, DEAHVIX, and OSC.
Figure 2 shows the candlestick chart and MACD histogram. In the candlestick chart, the blue line represents the 12d EMA, and the red line represents the 26d EMA. Candlesticks are usually composed of a red and green body, as well as an upper wick and a lower wick. The area between the opening and the closing prices is called the body, and price excursions above or below the real body are called the wick. The body indicates the opening and closing prices, and the wick indicates the highest and lowest traded prices of a stock during the time interval represented. For a red body, the opening price is at the bottom, and the closing price is at the top. For a green body, the opening price is at the top, and the closing price is at the bottom.
In the MACD histogram, the solid line represents the DIF, the dotted line represents the DEA, and the histogram represents the MACD bar. According to the strategy described in Section 3, we buy the stock when the DIF and DEA are positive, the DIF cuts the DEA in an uptrend, and the divergence is positive. We sell the stock when the DEA cuts the DIF in a downtrend, and the divergence is negative. As shown in Figure 2, we sell the stock on days 155 and 355 and buy the stock on days 212, 290, 310, 381, and 393. The buyandsell signals in the candlestick chart and the MACD histogram are shown in Figure 3.
Figure 4 shows the candlestick chart and MACD histogram of HVIX. In the candlestick chart, the blue line represents the 12d EMAHVIX, and the red line represents the 26d EMAHVIX. In the MACDHVIX histogram, the solid line represents the DIFHVIX, the dotted line represents the DEAHVIX, and the histogram represents the MACDHVIX bar. According to the strategy described in Section 3, we buy the stock when the DIFHVIX and DEAHVIX are positive, the DIFHVIX cuts the DEAHVIX in an uptrend, and the divergence is positive, and we sell the stock when the DEAHVIX cuts the DIFHVIX in a downtrend, and the divergence is negative. As shown in Figure 4, we sell the stock on days 118 and 187 and buy the stock on days 222, 231, 241, 243, 292, 415, and 447. The buyandsell signals in the candlestick chart and the MACD histogram are shown in Figure 5.
To verify the stability of the new indicator, we compare the MACD and MACDHVIX in Figure 6. The MACD and MACDHVIX have basically the same trend and the stability of the MACDHVIX is better than that of the MACD.
Table 1 shows a comparison of the specific values of the buyingselling points for the MACD index and MACDHVIX index, as well as a comparison of the predicted and actual trends. Here, we see that the prediction accuracy of MACDHVIX is 0.667 and that of MACD is 0.4286. By using the proposed indicator, we can improve the prediction accuracy by 55.55% compared with the traditional MACD. The “Price” in the table represents the closing price of stock. Next, we compare the cumulative returns for the two indicators. According to the trading points shown in the table, we perform a simulation test. We assume that the initial fund is 1 million. The cumulative returns under the two indexes are 1.1136 million and 1.3365 million, for MACD and MACDHVIX indices, respectively.

5.2. Empirical Results of BuyandHold Strategy
Using the “dggf” stock data chosen in Section 4, we first investigate the buy or sell points for both the indicators with the buyandhold strategy applied for 5 d. Then, we compare the prediction accuracy between the two indicators. The MACD histogram shown in Figure 7 indicates the buyandsell points; we should buy the stock at a buy point on days 391, 1,071, 1,181, 1,326, and 1,481, and sell the stock at a sell point on days 791, 881, and 911. The prediction situation is shown in Table 2.

The MACDHVIX histogram in Figure 8 indicates the buyandsell points. We should buy the stock at a buy point on days 371, 1,201, 1,331, and 1,561 and sell the stock at a sell point on days 751 and 771. The prediction situation is shown in Table 2. A comparison between MACD and MACDHVIX is shown in Figure 9.
Table 2 shows a comparison of the specific values of the buyingselling points for the MACD and MACDHVIX indices with the buyandhold strategy for 5 d, as well as a comparison of the predicted and actual trends. Here, we observe that the prediction accuracy of MACDHVIX is 0.8333 and that of MACD is 0.6250. By using the proposed indicator, we can improve the prediction accuracy by 33.33% compared with the traditional MACD. The “Price” in the table represents the closing price of the stock.
Next, using the “payh” stock data chosen in Section 4, we investigate the buy or sell points for both the indicators with the buyandhold strategy applied for 10d. Then, we compare the prediction accuracy between the two indicators. The MACD histogram in Figure 10 indicates the buyandsell points. We should buy the stock at a buy point on day 901, and sell the stock at a sell point on days 621, 1,971, 2,071, 2,291, 2,431, and 2,661. The prediction situation is shown in Table 3.

The MACDHVIX histogram in Figure 11 indicates the buyandsell points. We should buy the stock at a buy point on day 901 and sell the stock at a sell point on days 2,071, 2,421, 2,661, and 2,741. The prediction situation is shown in Table 3. A comparison of MACD and MACDHVIX is presented in Figure 12.
Table 3 shows the comparison of the specific values of the buyingselling points for the MACD and MACDHVIX indices with the buyandhold strategy applied for 5 d, as well as a comparison of the predicted and actual trends. Here, we observe that the prediction accuracy of MACDHVIX is 0.8 and that of MACD is 0.7143. By using the proposed indicator, we can improve the prediction accuracy by 12% compared with the traditional MACD. The “Price” in the table represents the closing price of the stock.
5.3. Computational Complexity
The computational complexity of the MACD and MACDHVIX for a stock which has a length of n are and , respectively. In terms of trend prediction processing time, the average time required to process a buyandsell strategy, a buyandhold strategy for 5 days, and a buyandhold strategy for 10 days with the MACD approach (MACDHVIX) are, respectively, 1.25 (1.51), 1.12 (1.35), and 1.41 seconds (1.58) using Matlab R2017b on an Intel(R) Core(TM) i56200 CPU @ 2.30GHz processor.
6. Conclusion
As indicated by Tables 1, 2, and 3, we buyandsell stock based on improved MACD; then we found all the accuracy is higher than that before the improvement. Therefore, the improved model has higher maneuverability in securities investment and allows investors to capture every buyandsell points in the market. For both the buyandsell strategy and the buyandhold strategy, the empirical results indicated that the proposed model can make more precise predictions than the traditional model. The proposed model was tested with three different stocks and it generated the high prediction accuracy for all the cases. In addition, while a smoothing index is used to construct the MACD index and the impact of the past price declines exponentially, the MACDHVIX does not have this property. Although the MACDHVIX index is improved compared with the MACD index, the stationarity of the MACDHVIX index is difficult prove theoretically. Test shows that it is stable; however, in the everchanging market, an abnormal situation can cause incalculable losses to investors. In future research, we will investigate other factors for the model by constantly updating the data and the training model to obtain a better prediction effect.
Data Availability
The data used to support the findings of this study are available from the corresponding author upon request.
Conflicts of Interest
The authors declare that there are no conflicts of interest regarding the publication of this paper.
Acknowledgments
The first author (Jian Wang) was supported by the China Scholarship Council (201808260026). The corresponding author (J.S. Kim) expresses thanks for the support from the BK21 PLUS program.
References
 F. Schmitt, D. Schertzer, and S. Lovejoy, “Multifractal Fluctuations in Finance,” International Journal of Theoretical and Applied Finance, vol. 03, no. 03, pp. 361–364, 2000. View at: Publisher Site  Google Scholar
 T. J. Moskowitz, Y. H. Ooi, and L. H. Pedersen, “Time series momentum,” Journal of Financial Economics, vol. 104, no. 2, pp. 228–250, 2012. View at: Publisher Site  Google Scholar
 C. S. Asness, T. J. Moskowitz, and L. H. Pedersen, “Value and Momentum Everywhere,” Journal of Finance, vol. 68, no. 3, pp. 929–985, 2013. View at: Publisher Site  Google Scholar
 M. R. Hassan, “A combination of hidden Markov model and fuzzy model for stock market forecasting,” Neurocomputing, vol. 72, no. 1618, pp. 3439–3446, 2009. View at: Publisher Site  Google Scholar
 P. Pai, L. Hong, and K. Lin, “Using Internet Search Trends and Historical Trading Data for Predicting Stock Markets by the Least Squares Support Vector Regression Model,” Computational Intelligence and Neuroscience, vol. 2018, Article ID 6305246, 15 pages, 2018. View at: Publisher Site  Google Scholar
 S. Lahmiri, “Minuteahead stock price forecasting based on singular spectrum analysis and support vector regression,” Applied Mathematics and Computation, vol. 320, pp. 444–451, 2018. View at: Publisher Site  Google Scholar  MathSciNet
 P. Singh and B. Borah, “Forecasting stock index price based on Mfactors fuzzy time series and particle swarm optimization,” International Journal of Approximate Reasoning, vol. 55, no. 3, pp. 812–833, 2014. View at: Publisher Site  Google Scholar  MathSciNet
 S. Lahmiri and M. Boukadoum, “Intelligent Ensemble Forecasting System of Stock Market Fluctuations Based on Symetric and Asymetric Wavelet Functions,” Fluctuation and Noise Letters, vol. 14, no. 4, 2015. View at: Google Scholar
 S. R. Das, D. Mishra, and M. Rout, “A hybridized ELM using selfadaptive multipopulationbased Jaya algorithm for currency exchange prediction: an empirical assessment,” Neural Computing and Applications, pp. 1–24, 2018. View at: Publisher Site  Google Scholar
 L. A. Laboissiere, R. A. S. Fernandes, and G. G. Lage, “Maximum and minimum stock price forecasting of Brazilian power distribution companies based on artificial neural networks,” Applied Soft Computing, vol. 35, pp. 66–74, 2015. View at: Publisher Site  Google Scholar
 L. Lei, “Wavelet Neural Network Prediction Method of Stock Price Trend Based on Rough Set Attribute Reduction,” Applied Soft Computing, vol. 62, pp. 923–932, 2018. View at: Publisher Site  Google Scholar
 S. Lahmiri, “Intraday stock price forecasting based on variational mode decomposition,” Journal of Computational Science, vol. 12, pp. 23–27, 2016. View at: Publisher Site  Google Scholar
 S. Lahmiri, “A Technical Analysis Information Fusion Approach for Stock Price Analysis and Modeling,” Fluctuation and Noise Letters, vol. 17, no. 01, p. 1850007, 2018. View at: Publisher Site  Google Scholar
 J.Z. Wang, J.J. Wang, Z.G. Zhang, and S.P. Guo, “Forecasting stock indices with back propagation neural network,” Expert Systems with Applications, vol. 38, no. 11, pp. 14346–14355, 2011. View at: Publisher Site  Google Scholar
 Y. Shynkevich, T. M. McGinnity, S. A. Coleman, A. Belatreche, and Y. Li, “Forecasting price movements using technical indicators: Investigating the impact of varying input window length,” Neurocomputing, vol. 264, pp. 71–88, 2017. View at: Publisher Site  Google Scholar
 I. Svalina, V. Galzina, R. Lujić, and G. Šimunović, “An adaptive networkbased fuzzy inference system (ANFIS) for the forecasting: the case of close price indices,” Expert Systems with Applications, vol. 40, no. 15, pp. 6055–6063, 2013. View at: Publisher Site  Google Scholar
 X. Zhou, Z. Pan, G. Hu, S. Tang, and C. Zhao, “Stock Market Prediction on HighFrequency Data Using Generative Adversarial Nets,” Mathematical Problems in Engineering, vol. 2018, Article ID 4907423, 11 pages, 2018. View at: Publisher Site  Google Scholar
 Z. Wang, J. Hu, and Y. Wu, “A Bimodel Algorithm with DataDivider to Predict Stock Index,” Mathematical Problems in Engineering, vol. 2018, Article ID 3967525, 14 pages, 2018. View at: Publisher Site  Google Scholar
 S. Lahmiri, “Interest rate nextday variation prediction based on hybrid feedforward neural network, particle swarm optimization, and multiresolution techniques,” Physica A: Statistical Mechanics and its Applications, vol. 444, pp. 388–396, 2016. View at: Publisher Site  Google Scholar
 L. Tao, Y. Hao, H. Yijie, and S. Chunfeng, “KLine Patterns' Predictive Power Analysis Using the Methods of Similarity Match and Clustering,” Mathematical Problems in Engineering, vol. 2017, Article ID 3096917, 11 pages, 2017. View at: Publisher Site  Google Scholar
Copyright
Copyright © 2018 Jian Wang and Junseok Kim. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.