Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Can Japanese Candlestick Patterns be Profitable on the Component Stocks of the SSE50 Index?

Can Japanese Candlestick Patterns be Profitable on the Component Stocks of the SSE50 Index? In this study, we investigate the profitability of 10 well-known Japanese candlestick charting patterns using daily-based data on the component stocks of the Chinese SSE50 index, which involves a lengthy sample period from January 2000 to December 2018. The main contribution of this paper is that we conduct the first predictive power examination of Japanese candlestick patterns on the Chinese SSE50 stocks while taking into account trend and overbought/oversold conditions, and their profitability over different holding periods. Experimental results indicate that several bullish candlestick patterns such as Long White and Bullish Gap can produce a significant positive average return over certain holding periods. In addition, empirical results show that none of the bearish candlestick patterns we examined offers predictive power. However, without considering trend and overbought/oversold conditions, we find that the bearish pattern Gravestone Doji over a 10-day holding period has superior profitability if it is applied as a contrary trading signal. The robustness of our results is confirmed based upon a bootstrap analysis and an out-of-sample test. The findings of this study are beneficial for the market traders engaged in transaction of the SSE50 component stocks. Keywords candlestick pattern, profitability test, trend condition, overbought/oversold condition, out-of-sample test practitioners but also researchers from many western coun- Introduction tries. Although there are many literature that have studied the Technical analysts use historical market trading data such as profitability of the Japanese candlestick patterns on stock price, volume, money flow, or turnover to predict move- markets (Goo et al., 2007; Lu, 2014; Lu et al., 2012), few of ments of future prices (Gehrig & Menkhoff, 2006; Murphy, them investigated the profitability of candlesticks on the 1999). Proponents of technical analysis believe that such his- Chinese SSE50 index in the Chinese market. Thus, in this torical trading data contains crucial information about direc- study, we examine the profitability of several well-known tion changes in future prices. However, according to Efficient Japanese candlestick patterns on the component stocks of the Market Hypothesis (EMH), in the weak form of an efficient SSE50 index, which is regarded as a blue chip and an essen- market, any information contained in past prices is therefore tial stock index of the Shanghai Stock Exchange. Since the reflected in current prices, such that historical price data sug- component stocks of the SSE50 index represent the top 50 gest no valuable information for forecasting future prices companies by capitalization, using those component stocks (Fama, 1970). Nevertheless, many researchers have proved of the SSE50 index for profitability examination will be that several technical analysis approaches could offer predic- highly representative of the Chinese stock market. tive power in financial markets (Brock et al., 1992; Chong & Ng, 2008; Deng et al., 2020; Fang et al., 2013; Fifield et al., China Three Gorges University, Yichang, China 2005; Metghalchi et al., 2012; Trivedi, 2022). Shandong University, Jinan, China Among the well-known technical analysis approaches, 3 Zhongnan University of Economics and Law, Wuhan, China the Japanese candlestick is the earliest one that was initially East China University of Science and Technology, Shanghai, China applied to the Japanese rice market in the 1700s. Since Nison Corresponding Author: (1991) first translated the knowledge of the Japanese candle- Shangkun Deng, College of Economics and Management, China Three stick in English and introduced it to the western world, it has Gorges University, Yichang 443002, China. Email: dsk8672@163.com drawn considerable attention from not only financial market Creative Commons CC BY: This article is distributed under the terms of the Creative Commons Attribution 4.0 License (https://creativecommons.org/licenses/by/4.0/) which permits any use, reproduction and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and Open Access pages (https://us.sagepub.com/en-us/nam/open-access-at-sage). 2 SAGE Open Other than applying just the Japanese candlestick pat- when applying the Japanese candlestick patterns, few prior terns, Goo et al. (2007) concluded that the profitability of researchers considered the overbought/oversold condition Japanese candlestick patterns could be improved by employ- for applying Japanese candlestick in the Chinese stock mar- ing other indicators simultaneously. Indeed, when market ket. Nevertheless, market or stock conditions are essential practitioners adopt the Japanese candlestick patterns for for traders when they make a transaction in actual trading. deciding if a buying or a short-selling signal is valid, they Hence, in this study, we test the profitability of Japanese would generally also take into account the underlying stock candlestick patterns with considering the uptrend/downtrend conditions, such as whether the stock price is in an uptrend or condition or overbought/oversold condition of stock prices. a downtrend, or whether it has already been overbought or To identify the trend condition, a well-known trend indicator oversold, rather than just using the Japanese candlestick pat- called moving average (MA), which is widely employed by terns. For instance, when a bullish candlestick pattern lots of market traders and researchers, is employed in this appears, professional market traders would generally also research. In addition, to confirm the overbought/oversold confirm the following two points before executing a long condition, though there are several momentum oscillators transaction: (1) The trading target is in an upward trend and that could be employed, one of the most common indicators (2) The trading target is currently not in an overbought area. called relative strength index (RSI) is employed by us in this Otherwise, it might imply an extremely high risk for trading. research. Subsequently, we investigate whether Japanese If either of those two conditions is not satisfied, it might not candlestick patterns could create values for market investors be appropriate to execute a long transaction at that time. with considering market conditions. Similarly, it would be more appropriate for market traders to Additionally, Park and Irwin (2007) pointed out that it is execute a short-selling transaction when a bearish candle- important to take data snooping, out-of-sample problems, and stick signal appears with condition confirmation that the transaction costs into consideration when examining the effec- stock price is in a downtrend and the stock is not in an over- tiveness of technical analysis approaches. Therefore, in this sold condition. study, we also adopt a bootstrap analysis methodology adopted by Marshall et al. (2006). Meanwhile, an out-of-sample test is conducted, and transaction cost is considered in the experi- Literature Review ments to examine whether Japanese candlestick patterns can In the last decade, the profitability of Japanese candlestick create values for market practitioners. Moreover, except for patterns on stock markets has been investigated by numerous the profitability test of Japanese candlestick patterns, we also researchers. Lu et al. (2012) and Lu and Shiu (2012) exam- conduct a simulation trading using the Japanese candlestick ined the profitability of Japanese candlestick patterns on the based trading strategies to examine their profitability. Taiwan stock market, and they found that several candlestick Compared with the profitability test of candlestick patterns, patterns have predictive power. Among the literature that simulation trading results would be closer to actual trading. investigated the effectiveness of Japanese candlestick pat- The main contributions and innovations of this study terns on U.S. stock markets, many researchers found that could be summarized as follows. First, to the best of our only a small fraction of the candlestick patterns could pro- knowledge, this paper is the first study that examines the vide correct direction predictions for market traders (Cohen, profitability of Japanese candlestick patterns in the Chinese 2021; Lu et al., 2015; Lu & Shiu, 2016; Qiu & Liu, 2019). stock market with considering the overbought/oversold con- Apart from the investigation of stock markets in developed ditions, and we compared the difference between the trading countries and areas (Naranjo and Santos, 2019; Xie et al., strategy results of the following four groups: candlestick pat- 2012), the profitability of candlestick patterns has also been tern only (unconditional), candlestick pattern appears in a examined in the stock markets of developing countries such trend, candlestick pattern appears with considering over- as the Vietnamese stock market (Thanh et al., 2018), Thailand bought/oversold condition, and candlestick pattern appears stock market (Tharavanij et al., 2017), Brazilian stock mar- with considering both the trend and overbought/oversold ket (Prado et al., 2013), and Chinese stock market (Chen conditions. Second, given the increased attention on the et al., 2016; Zhu et al., 2016). Chinese stock market from market traders and researchers all However, in most of the related literature that examined over the world, unlike most of the prior profitability test the profitability of Japanese candlestick patterns, researchers studies conducted on stock index of south Asian stock mar- generally did not consider the market conditions. Whereas, kets (Gunasekarage & Power, 2001) or European stock mar- identifying whether a stock price is in an overbought or an kets (Vasiliou et al., 2008), this study adopts the use of the oversold condition, or whether it is in an uptrend or a down- individual component stocks comprising the SSE50 index. trend condition would be an critical part of establishing the Since tests on index data are unable to be implemented and trading risk. In the current literature, only a limited amount biased as a result of nonsynchronous trading (Day & Wang, of them have considered the trend condition of stock prices 2002), investigations on the component stocks of a stock (Batten et al., 2018; Heinz et al., 2021; Lu et al., 2015) or index would be more practical for market traders. Third, overbought/oversold condition (Tharavanij et al., 2017) unlike previous studies that only researched the profitability Deng et al. 3 Table 1. A list of the five pairs of bullish and bearish Japanese candlestick patterns. No Candlestick Bullish pattern Bearish pattern 1 Long candle Long White Long Black 2 Doji Dragonfly Doji Gravestone Doji 3 Hammer Bullish Hammer Inverted Hammer 4 Harami Bullish Harami Bearish Harami 5 Gap Bullish Gap Bearish Gap tj + tj + test of candlestick patterns, we design the trading strategies (( P iS )/ − Pi () /) Lj >∈ 01 ,( ,....,T ),, (1) cc ∑ based on candlestick patterns to test whether those Japanese it =+ jS −+1 it =+ jL −+1 candlestick pattern based trading strategies could make prof- its for traders who participated in the Chinese stock market. which implies an uptrend condition identification at time The remainder of this paper is arranged as follows: a brief t + T. Similarly, a downward trend is identified if a short- introduction to the candlestick charting, trend identification, term moving average is below a long-term moving average, and overbought/oversold condition identification are and the condition lasts for a certain period: described in Section 3. In Section 4, we explain the research data, trading strategy, and research design adopted for this tj + tj + study. Empirical results of the profitability tests, bootstrap (2) (( P iS )/ − Pi () /) Lj <∈ 01 ,( ,....,T ),, cc ∑ analysis results, and out-of-sample test results are reported in it =+ jS −+1 it =+ jL −+1 Section 5. Conclusions and future work are provided in which implies a downtrend condition identification at time t + T. Section 6. A description of the examined Japanese candle- In the above equations (1) and (2), P (i) denotes the close stick patterns is provided in Appendix A. price at time i; T represents the minimum length of lasting period, which is set to be three in this research; L and S, Methodology respectively denote the parameters of a long-term and a short-term moving average, and the values of L and S are set Japanese Candlestick Pattern to be 20 and 10, respectively, as they have been widely used Candlestick charting was created by a Japanese rice trader by traders in financial markets. named Munehisa Homma. He observed that the emotions of investors reflected in the price candlesticks might exert sig- nificant influences on the rice markets (Nison, 1991). A sin- Overbought/Oversold Condition Identification gle Japanese candlestick line is created based upon four Relative Strength Index (RSI) is a momentum oscillator ini- prices (open, high, low, and close prices) of a certain period, tially developed by Wilder (1978). It is a measurement of stock and one single line or several lines may then comprise one price change momentum, which is displayed as an oscillator (a bullish/bearish pattern. There are many candlestick patterns line graph that moves between two extremes) that has a value to suggest a bullish signal, such as Long White, or suggest a range from 0 to 100. RSI is calculated based on prior sessions’ bearish signal, such as Long Black. Five pairs of well-known average gains versus losses. Traditional interpretation and use candlestick patterns examined in this study are displayed in of the RSI is as follows: an RSI value of 70 or above indicates Table 1, and the real examples of the candlestick patterns that a stock is overbought, and it might be primed for a trend rever- appeared in one component stock (Shanghai Pudong sal or corrective pullback in price. In contrast, an RSI value of Development Bank) of the SSE50 index are provided in 30 or below indicates an oversold or undervalued stock condi- Figure 1. Details of the candlestick patterns are described, tion. The calculation of RSI is as follows: with their shapes being illustrated in Appendix A. For each trading period t, an upward change U or down- ward change D is calculated as: Trend Identification In this study, the market trend is measured and identified by 00 () PP −≤  closet ,, closet−n U = , (3) a moving average (MA) rule (Kwok et al., 2009). MA is an P − P PP () −> P 0  closet , closet ,, −− nclose tclose,tn average of time series prices over several consecutive peri- ods. In this research, the MA is calculated using the close price of stocks. An uptrend is generally identified if a short- 00 () PP −≤  closet ,, -nclose t (4) term moving average is above a long-term moving average, D= . P − −− PP () P > 0  closet , -n closet ,, closet -nclose,t and the condition lasts for a certain period:  4 SAGE Open Figure1. (continued) Deng et al. 5 Figure 1. Real examples of the 10 candlestick patterns appeared in historical data of Shanghai Pudong Development Bank. The ratio of these averages is the relative strength (RS) or profitability analysis, with the whole sample period ranging relative strength factor: from January 2000 to December 2018. The trading price data of all component stocks are in four prices (open, high, low, and close) format. Among the 50 component stocks, 3 stocks EMAU (,n) (5) RS = , that without data before 31 December 2014 were excluded. EMAD (,n) Thus, the resulting sample included a total of 47 stocks. In addition, since the out-of-sample test is important for inves- (6) EMAP =+ αα () 1- EMA , tclose,tt−1 tors, the robustness of research results was also examined. Subsequently, the whole sample period was divided into two (7) α= , sub-periods, which were the in-sample (Sub-period 1) and n + 1 out-of-sample (Sub-period 2) periods with a length ratio of 3:1. The sample period before/after 1 January 2015 was where n represents the period parameter for calculating the employed as the in-sample/out-of-sample dataset. RSI, and EMA (Exponential Moving Average) is a type of Subsequently, the sub-period 1 ranges from 2000-Jan-4th to moving average that places a larger weight and significance on 2014-Dec-31st, while the date from 2015-Jan-5th to the most recent data points. The relative strength factor is then 2018-Dec-28th is employed for sub-period 2. converted to a relative strength index between 0 and 100: Experimental Design (8) RSI () t =− 100 . 1+ RS () t In practice, proponents of Japanese candlestick patterns also refer to the trend condition or overbought/oversold condi- In general, the greater/smaller the RSI value, the stronger tion. For example, if a Long White appears in an uptrend, and more protracted the bullish/bearish trend, whereas a more and the stock is not overbought simultaneously, it suggests a important usage is that it implies the trading target is at an over- stronger signal for making a long transaction. Similarly, if a bought/oversold level. Thus, it would be extremely risky for Bearish Gap pattern is formed in a downtrend, and the stock traders to make a long/short-selling transaction. In the experi- is not in an oversold condition simultaneously, it indicates an ments, the parameter n of RSI is set to be 12 due to its popular- agreement with the short signal generated from the bearish ity among traders in financial markets. Additionally, stocks are candlestick pattern. Taking into account both trend condition considered to be overbought when the RSI value is above 70 or and overbought/oversold condition might be beneficial for oversold when the indicator value is below 30. market traders to make a transaction in a relatively low-risk condition. In contrast, if the traders consider both of those Data and Experimental Design two conditions to be satisfied when applying the Japanese candlestick patterns, the number of tradable signals are Data fewer, and traders may miss several great chances of trading. The daily-based data on the 50 component stocks (see Therefore, it is necessary to examine whether or not the prof- Appendix B) of the SSE50 index were employed to conduct itability of candlestick patterns could be improved by taking 6 SAGE Open Table 2. Long and Short-Selling Trading Signal and Conditions for the Four Groups. Long signal Short-selling signal Group Bullish candlestick In an uptrend Not overbought Bearish candlestick In a downtrend Not oversold A 〇 × or 〇 × or 〇 〇 × or 〇 × or 〇 B 〇 〇 × or 〇 〇 〇 × or 〇 C 〇 × or 〇 〇 〇 × or 〇 〇 D 〇 〇 〇 〇 〇 〇 Note. “〇” means satisfied. “×” means not satisfied. into account the trend or overbought/oversold condition 5-day (1 week), and 10-day (about half a month) periods. since, indeed, many market investors tend to adopt additional Similarly, for short-selling transactions, the k-day return R is indicators when using candlestick patterns to achieve supe- expressed as: rior performance. Therefore, this research investigates the Pt () +− 1 Pt () + k sb profitability of Japanese candlestick patterns with consider- R = ×100%. (10) Pt () +1 ing trend condition and overbought/oversold condition. s Generally, Japanese candlestick patterns are used to cap- where P (t + k) is the close price for buying on day t + k; ture short-term price movements, and the most useful time P (t + 1) denotes the open price for short-selling on day t + 1; horizon is shorter than 10 days (Nison, 1991). Hence, in the k represents the length of days for the holding period, which experiments, we investigated the Japanese candlestick pat- is also comprised of 1-, 3-, 5-, and 10-day periods. terns with four different holding period lengths, which are 1-, 3-, 5-, and 10-day, for the profitability test. Note that the holding period is fixed when the period is decided, in other Experimental Results words, the opened position will be held until the expiration of the fixed period regardless of the daily candlestick recom- Summary Statistics mendation during the position holding period. However, the Stock price data used in the experiments were derived from profitability test considers all the subsequent candles during the Choice database (Database 1). Each candlestick line is the period after the pattern’s occurrence. constructed from four prices, which are open, high, low, and In this study, we design four groups of trading rules based close prices. The summary statistics on the selected 47 com- on candlestick pattern to identify their predictive power. The ponent stocks (the period before 1 January 2015) are dis- trading conditions of each trading group are shown in Table played in Table 3, with 102,184 observations of daily return. 2. Note that for the profitability test of each candlestick pat- Note that the natural logarithm of these daily returns was tern based trading rule, there will be no trade at the time point utilized for analysis. The distribution characteristics were where the conditions are not satisfied. examined using five statistics: mean, standard deviation, Regarding the profits generated by the candlesticks, the skewness, kurtosis, and JB statistics. All four series were returns are calculated using the difference between the sell- found to display positive mean and skewness, and the aver- ing and buying prices, expressed as a percentage of the buy- age returns of the time series have a near-zero mean. In addi- ing prices. It is assumed that the candlestick pattern is tion, it should be noted that skewness and kurtosis were formulated on day t, then the n-day return R is defined as: significantly different from 0 and 3, respectively, at the 1% significance level. Pt () +− kP () t + 1 sb (9) R = ×100% . Pt () + 1 Candlestick Pattern Returns Our analysis begins with a profitability examination of 10 where P (t + k) refers to the close price for selling on day candlestick patterns to examine their predictive power. The t + k, and P (t + 1) denotes the open price for buying on day profitability test of Japanese candlestick patterns was con- t + 1. Note that for return calculation, we assume that market ducted using a suitable test named the skewness adjusted traders cannot make the transaction on the same day that the two-tailed t-test (Mitra, 2011). On this test, the null hypoth- signal is generated since it is very difficult to determine the esis assumes that the average returns for the candlestick pat- trading signal until the close price is known. Therefore, a terns do not differ significantly from zero: more practical case is to enter at the open price of the trading day after the entry signal is generated. In addition, k repre- 1 1 sents the length of days for the holding period, which is com- 2 tn =+ () SS γγ ⋅+ , (11) sa prised of 1- and 3-day (about half a week for business days), 3 6n Deng et al. 7 Table 3. Descriptive Statistics of the Daily Returns. Open High Low Close −4 −4 −4 −4 M 5.32 × 10 5.33 × 10 5.35 × 10 5.37 × 10 SD 0.0272 0.0241 0.0241 0.0254 Skewness 0.86** 1.82** 1.21** 1.19** Kurtosis 29.24** 78.00** 48.53** 49.01** 6 7 7 7 JB statistics 3.65 × 10 ** 2.59 × 10 ** 1.01 × 10 ** 1.03 × 10 ** Note. There are 102,184 observations of daily return for the open, high, low, and close prices. The daily returns are measured as the natural logarithm difference of the open, high, low, and close prices. **Indicates statistical significance at the 1% level. with considering the uptrend condition. For Bullish Gap, it AR (12) S , was observed that its average return decreased from 2.7% to σ() AR 2.3% over a 5-day holding period, and it decreased from 3.8% to 2.6% over a 10-day holding period, thus indicating Bullish Gap is more appropriate to be applied without con- () AR − AR ∑ i sidering the uptrend condition, while Long White and Bullish (13) i=1 γ = . Harami patterns are suggested to be applied in the uptrend nA ⋅σ() R condition. Further, it also shows that for the profitable can- dlestick patterns Long White and Bullish Gap, when the where is the average return of the transactions, σ AR holding period increased, their average returns also improved, AR () is the sample standard deviation of the returns, and n denotes thereby indicating that it would be better to hold the trading the number of candlestick based trading signals. The returns position for a relatively long period. While for Bullish prior to transaction costs being taken into consideration for Harami with a 5-day holding period, it produced the best the 1-, 3-, 5-, and 10-day holding periods are presented in return, demonstrating that it is suitable to apply it with about Tables 4 and 5, which are the results for the five bullish and one week holding period. five bearish candlestick patterns, respectively. Next, we applied these bullish candlestick patterns based Returns of five bullish candlestick patterns without con- signals with considering an overbought condition rather than sidering the uptrend and overbought conditions are presented in an up-trend condition. As reported in panel C of Table 4, in panel A of Table 4, from which we find that at the 5% after considering the overbought condition, all Long White level, Long White yielded positive average returns over all (1-, 3-, 5-, and 10-day holding periods) remained to be profit- four holding periods, and Bullish Harami yielded positive able. While compared with the “Unconditional” trading sig- average returns over the 1-, 3-, and 5-day holding periods. nal group, all values of the average return decreased. In Bullish Gap produced positive average returns over the 5- addition, Bullish Harami (3- and 5-day holding periods) and 10-day holding periods. Following the approach adopted became not profitable after considering the overbought con- in Caginalp and Laurent (1998), we then considered total dition. Dragonfly Doji with the 5-day holding period became transaction costs, including execution, liquidity, and slippage profitable at the 5% level, while after taking into account the cost. Based upon our arbitrary assumption of a transaction transaction cost of 1%, it was unable to generate positive cost of 1% per round trip, Bullish Gap (5- and 10-day peri- average returns. The candlestick pattern Bullish Gap (10-day ods) and Long White (10-day period) were found to yield holding period) produced a positive average return signifi- positive returns and remain profitable even after deducting cantly, and the average return value was larger than the trans- the transaction cost. action cost 1%. However, compared with the “unconditional” Panel B of Table 4 shows the profitability test results for condition, the average return of Bullish Gap decreased from applying the five bullish candlestick patterns in an uptrend 3.8% to 3.6%, thus indicating that it was more appropriate to condition. Compared with the results shown in panel A, in apply the Bullish Gap candlestick pattern without consider- most cases for the candlestick patterns obtained positive ing the overbought condition. average returns, after applying them in an uptrend, they were Then, we applied those five bullish candlestick patterns found to produce superior returns. In addition, after consider- considering both the trend condition and overbought condi- ing the uptrend condition, Bullish Harami (10-day holding tion. From the results reported in panel D, it can be seen that period) yielded a positive average return at the 5% level, and the candlestick patterns yielded average returns over 1% (at Long White (5-day holding period) became profitable even the 5% level) were Bullish Gap (3-, 5-, and 10-day holding after deducting the 1% transaction cost. These results reveal periods) and Long White (10-day holding period), thus dem- that the returns of those candlestick patterns were improved onstrating that buying signals generated by those candlestick 8 SAGE Open Table 4. Profitability Test of the Five Bullish Candlestick Patterns Over the Period 2000 to 2015. Holding period length and profitability test result Japanese 1-day 3-day 5-day 10-day candlestick pattern N Return p-Value N Return p-Value N Return p-Value N Return p-Value Panel A (unconditional) Long White 27,218 0.002** <.01 27,218 0.004** <.01 27,218 0.006** <.01 27,218 0.012** <.01 Dragonfly Doji 588 0.001** <.01 588 0.011 .29 588 0.015 .21 588 0.021 .14 Bullish Hammer 1,976 −0.001 .01 1,976 −0.002 .06 1,976 −0.002 .06 1,976 −0.001 .79 Bullish Harami 2,332 0.003** <.01 2,332 0.002* .02 2,332 0.003* .03 2,332 0.004 .06 Bullish Gap 1,802 0.010 .92 1,802 0.020 .12 1,802 0.027* .03 1,802 0.038** <.01 Panel B (uptrend) Long White 12,940 0.002** <.01 12,940 0.006** <.01 12,940 0.011** <.01 12,940 0.018** <.01 Dragonfly Doji 237 0.002 .08 237 0.023 .41 237 0.031 .40 237 0.049 .16 Bullish Hammer 902 −0.001 .18 902 0.001 .84 902 0.002 .35 902 0.004 .18 Bullish Harami 999 0.004** <.01 999 0.004** <.01 999 0.006** <.01 999 0.005* .05 Bullish Gap 855 0.011 .86 855 0.019 .18 855 0.023** <.01 855 0.026** <.01 Panel C (not overbought) Long White 22,217 0.001** <.01 22,217 0.003** <.01 22,217 0.005** <.01 22,217 0.009** <.01 Dragonfly Doji 485 0.001* .02 485 0.004 .10 485 0.006* .02 485 −0.001 .91 Bullish Hammer 1,719 −0.002 <.01 1,719 −0.002 <.01 1,719 −0.003 .02 1,719 −0.002 .26 Bullish Harami 2,113 0.003** <.01 2,113 0.002 .06 2,113 0.002 .06 2,113 0.003 .11 Bullish Gap 1,258 0.008 .84 1,258 0.018 .16 1,258 0.027 .06 1,258 0.036* .02 Panel D (uptrend + not overbought) Long White 8,937 0.002** <.01 8,937 0.006** <.01 8,937 0.009** <.01 8,937 0.015** <.01 Dragonfly Doji 173 0.001 .07 173 0.003 .35 173 0.008 .10 173 0.009 .19 Bullish Hammer 761 −0.001 .13 761 −0.001 .70 761 0.001 .47 761 0.003 .24 Bullish Harami 942 0.004** <.01 942 0.005** <.01 942 0.006** <.01 942 0.006* .03 Bullish Gap 391 0.006** <.01 391 0.011** <.01 391 0.014** <.01 391 0.018** <.01 Note. “N” is the number of trading signals. “Return” refers to the average return from trading signals with its associated p-value. *and ** indicate generating positive average returns at the 5% and 1% level, respectively. patterns would on average produce positive returns. In addi- Dragonfly Doji, it yielded the best return of 0.6% when con- tion, we also find that by considering both the uptrend trend sidering the “not overbought” condition. However, none of and overbought conditions, Bullish Gap obtained a signifi- these candlestick patterns was able to generate an average cant average return greater than 1% over a 3-day holding return greater than the total transaction cost assumed per period. However, it produced a smaller average return over round turn in this study. Therefore, it seems that these pat- the 5- and 10-day holding periods than in the “unconditional” terns cannot create value for trading on the component stocks condition. Furthermore, we observe that all profitable can- of the SSE50 index. dlestick patterns yielded superior returns than in the “uncon- Similar to the bullish candlestick patterns, we further ditional” condition except Bullish Gap. examine the profitability of the five bearish candlestick pat- Finally, we focus on the best returns for each profitable terns, and their results are reported in Table 5. The average candlestick pattern that yielded an average return greater return results without considering the downtrend or oversold than 1%. For Bullish Gap, the best trading signal was the condition are presented in panel A, from which we observe “unconditional” condition over the 10-day holding period, in that under the “unconditional” condition over the 3-, 5-, and which it yielded an average return of about 3.8% before 10-day holding periods, none of their average returns was deducting the transaction cost. For Long White, it performed positive. Only Inverted Hammer, Bearish Harami, and better in an uptrend condition with a 10-day holding period, Bearish Gap with the 1-day holding period produced positive which yielded an average return of about 1.88% before con- returns. We then take into account the total transaction cost sidering the transaction cost. These two trading signals were of 1% per round trip. Obviously, none of those candlestick profitable on the transaction of the SSE50 component stocks. patterns was found to create values for investors. This indi- For Bullish Harami, it obtained the best average return of cates that by using only those bearish candlestick patterns, 0.6% (5- and 10-day holding periods) when considering both none of them showed predictive power. Additionally, we find the uptrend condition and overbought condition. For that over a 10-day holding period, for Gravestone Doji and Deng et al. 9 Table 5. Profitability Test Results of the Five Bearish Candlestick Patterns Over the Period 2000 to 2015. Holding period length and profitability test result 1-day 3-day 5-day 10-day Japanese candlestick pattern N Return p-Value N Return p-Value N Return p-Value N Return p-Value Panel A (unconditional) Long Black 26,187 −0.001 <.01 26,187 −0.003 <.01 26,187 −0.005 <.01 26,187 −0.008 <.01 Gravestone Doji 569 −0.008 .46 569 −0.005 .04 569 −0.006 .06 569 −0.011 .02 Inverted Hammer 1,982 0.001* .01 1,982 −0.006 <.01 1,982 −0.008 <.01 1,982 −0.015 <.01 Bearish Harami 3,067 0.002** <.01 3,067 −0.001 .40 3,067 −0.001 .36 3,067 −0.006 <.01 Bearish Gap 1,937 0.001 .16 1,937 −0.001 .86 1,937 −0.001 .75 1,937 −0.003 .30 Panel B (uptrend) Long Black 11,948 0.000 .93 11,948 −0.002 <.01 11,948 −0.002 <.01 11,948 −0.005 <.01 Gravestone Doji 235 −0.006 <.01 235 −0.007 .15 235 −0.009 .17 235 −0.013 .16 Inverted Hammer 855 0.001 .23 855 −0.004 <.01 855 −0.002 .16 855 −0.007 <.01 Bearish Harami 1,355 0.002** <.01 1,355 0.001 .66 1,355 0.002 .26 1,355 0.001 .73 Bearish Gap 1,116 0.001 .48 1,116 −0.001 .61 1,116 −0.002 .32 1,116 −0.007 .09 Panel C (not overbought) Long Black 22,292 −0.001 <.01 22,292 −0.002 <.01 22,292 −0.004 <.01 22,292 −0.006 <.01 Gravestone Doji 483 −0.005 <.01 483 −0.004 .16 483 −0.004 .19 483 −0.010 .07 Inverted Hammer 1,701 0.002** <.01 1,701 −0.005 <.01 1,701 −0.006 <.01 1,701 −0.011 <.01 Bearish Harami 2,707 0.002** <.01 2,707 0.001 .71 2,707 0.001 .66 2,707 −0.003 .04 Bearish Gap 1,426 −0.001 .25 1,426 0.001 .48 1,426 0.002 .39 1,426 −0.001 .96 Panel D (uptrend + not overbought) Long Black 8,938 0.000 .25 8,938 −0.001 .16 8,938 −0.001 .23 8,938 −0.002 <.01 Gravestone Doji 200 −0.006 <.01 200 −0.008 .17 200 −0.011 .17 200 −0.013 .20 Inverted Hammer 776 0.001 .16 776 −0.003 .02 776 −0.002 .33 776 −0.006 .03 Bearish Harami 1,280 0.002** <.01 1,280 0.001 .27 1,280 0.003* .03 1,280 0.002 .36 Bearish Gap 657 −0.002 .03 657 0.001 .59 657 0.001 .75 657 −0.003 .37 Note. “N” is the number of short-selling signals. “Return” refers to the average return from trading signals with its associated p-value. *and ** indicate generating positive average returns at the 5% and 1% level, respectively. Inverted Hammer which are generally regarded as the bear- return of more than 1%, thus indicating that even considering ish candlestick patterns, they suffered losses of more than the “not oversold” condition when applying those bearish −1% at the 5% level, thereby indicating that although they candlestick patterns. Still, none of those candlestick patterns are generally regarded as the bearish patterns to suggest offered predictive power, which is a finding similar to that of short-selling signals, they could be used as contrary signals the research conducted by Tharavanij et al. (2017). They to generate bullish signals for executing long transactions. found that even with filtering by stochastic indicators, can- Panel B of Table 5 reports the results for applying the dlestick patterns are still not useful. However, it is found that Japanese candlestick patterns in a downtrend condition. Inverted Hammer over a 10-day holding period suffered a Similar to the results of applying bearish candlesticks shown loss of more than −1% at the 5% level, therefore it reveals in panel A, none of these patterns was found to yield a sig- that although this candlestick pattern is generally regarded as nificantly positive average return of more than 1%, thus indi- a bearish signal, it could be used as a contrary signal for mak- cating that although these candlestick patterns were applied ing a long transaction. We find it could be used to generate a with considering a downtrend condition, all of them remained buying signal to make profits even after deducting the trans- no significant profitability. In addition, none of those candle- action cost in the in-sample period. stick patterns suffered an average loss of more than −1% at Furthermore, from the results presented in panel D of the 5% level, thereby demonstrating that there was no can- Table 5, we find that by considering both the “downtrend” dlestick pattern that could be applied as a contrary indicator. condition and “not oversold” condition, Bearish Harami pat- Next, panel C of Table 5 shows the profitability test results tern over a 5-day holding period produced a significantly of candlestick patterns considering the “not oversold” condi- positive average return. However, after deducting the 1% tion instead of applying them in a downtrend condition. Over transaction cost, it became not profitable. It demonstrates the 1-, 3-, 5-, and 10-day holding periods, none of these pat- that these bearish candlestick patterns could not make profits terns was found to produce a significantly positive average even though both the “not oversold” and “downtrend” 10 SAGE Open Figure 2. Average return results of the five bullish candlestick patterns over the period 2000 to 2015. Note. Each horizontal axis represents the length of holding days, and each vertical axis represents the value of average return. conditions were satisfied. Additionally, from the results yielded a significantly positive average return. However, we reported in panel D, none of these candlestick patterns suf- find evidence that Gravestone Doji and Inverted Hammer fered a loss of more than −1% at the 5% level. Therefore, over a 10-day holding period and under the “unconditional” none of these patterns could be used as a contrary signal for condition, as well as the Inverted Hammer over a 10-day market investors. holding period and under the “not oversold” condition, could Additionally, Figures 2 and 3 show how the length of the create values for investors with regarding those bearish can- hold period impacts the profitability of Japanese candlestick dlestick patterns as buying signals. patterns. In Figure 2, we observe that with the increase of hold- ing days, the return rises in most of the bullish candlestick pat- Bootstrap Analysis terns. Several signals based on Hammer, Dragonfly Doji, and To examine the robustness of profitability test results, we Harami are different from others, in which the largest or small- then employed the bootstrap method (Marshall et al., 2006) est return was produced by the middle-length holding period, to address the common features of stock return data, such as such as Bullish Harami in panel A and Dragonfly Doji in panel autocorrelation, skewness, and leptokurtosis. Subsequently, C. On the contrary, the return falls with the increase of holding a new dataset was generated while the characteristics of the days in most bearish candlestick patterns. For several candle- original dataset for the SSE50 index component stocks were stick patterns, the average return fluctuates with the increase of retained. We simulated 500 sets of open, high, low, and close holding days, such as Bearish Harami in panels B and D, and price series, comparing the profits accrued from the actual Inverted Hammer in panels B and D. Additionally, in some data and the random series to identify whether the profitabil- cases, the largest return is produced with the middle-length ity of those candlestick pattern is statistically significant. holding period, such as Gravestone Doji in panels A and C. The bootstrap p-value is the percentage of simulated average In summary, for those bearish candlestick patterns, aver- returns that are greater than the actual average returns. Note that age returns of all trading signals were insignificant with the we only report returns of the candlestick patterns yielded sig- exception of Bearish Harami with conditions in the “down- nificant average returns that over 1%. Among the candlestick trend” and “not oversold” over a 5-day holding period, which Deng et al. 11 Figure 3. Average return results of the five bearish candlestick patterns over the period 2000 to 2015. Note. Each horizontal axis represents the length of holding days, and each vertical axis represents the value of average return. patterns, 12 were found to generate significant average returns out-of-sample performance, a two-step procedure which is simi- greater than 1% on the component stocks of the SSE50 index. lar to the method proposed by Jensen (1967) is conducted, first The “Return” column in Table 6 denotes the average selecting the best-performing candlestick patterns and then using return rates generated by each candlestick pattern on the these on new (i.e., out-of-sample) data. Based on it, we conduct bootstrapped and original series. As shown in panels A, B, C, an out-of-sample test on the data by selecting the profitable can- and D, when transaction costs were taken into consideration, dlestick patterns and contrary candlestick patterns (with consider- those candlestick patterns were not found to be profitable on ing the transaction cost) in the in-sample period, and to investigate the random bootstrap series. In all cases, the average returns whether or not those successful candlestick patterns could remain on the original series were found to be greater than those on their predictive power in the following out-of-sample period. the bootstrap series, particularly for Bullish Gap under the The return results of the profitable and contrary candlestick pat- “unconditional” condition over the 5- and 10-day holding terns in the sub-period 1 and sub-period 2 are reported in Table 7, periods, Bullish Gap under a trend condition over the 5- and in which we find that both datasets appeared to be consistent in 10-day holding periods, as well as Bullish Gap considering most cases, thereby providing support for the conclusion that most the overbought condition over the 10-day holding period. of the profitable candlestick patterns have significant explanatory Our bootstrap results thus provide valuable evidence on the power for the component stocks of the SSE50 index. In addition, profitability of candlestick patterns. the bearish candlestick pattern Gravestone Doji (“unconditional” condition with the 10-day holding period) generated a negative average return lower than −1%, showing significant evidence that Robustness Test it could be used as a contrary indicator to generate a profitable buy- Despite the evidence provided on the profitability of technical ing signal. Inverted Hammer under the “unconditional” condition analysis, Park and Irwin (2007) pointed out the importance of with the 10-day holding period produced a significantly negative addressing the out-of-sample problem. Therefore, to examine the average return, whereas the profit by using it as a contrary 12 SAGE Open Table 6. The Bootstrapped Simulation Results for the Profitable Candlestick Patterns and Holding Periods. Original series Bootstrap series Candlestick pattern Number Return Number Return p-Value count Panel A (unconditional) Long White (10-day) 27,218 0.012 31,793 0.007 1 Bullish Gap (5-day) 1,802 0.027 1,582 0.003 1 Bullish Gap (10-day) 1,802 0.038 1,582 0.007 1 Panel B (trend condition) Long White (5-day) 12,940 0.010 14,944 0.004 1 Long White (10-day) 12,940 0.019 14,944 0.008 0 Bullish Gap (5-day) 855 0.022 753 0.004 1 Bullish Gap (10-day) 855 0.027 753 0.008 1 Panel C (overbought/oversold condition) Bullish Gap (10-day) 1,258 0.036 1,231 0.007 0 Panel D (trend + overbought/oversold condition) Long White (10-day) 8,937 0.015 11,945 0.008 1 Bullish Gap (3-day) 391 0.011 461 0.002 0 Bullish Gap (5-day) 391 0.014 461 0.004 0 Bullish Gap (10-day) 391 0.018 461 0.008 1 Note. The “Number” columns refer to the number of patterns in the bootstrapped and original series. “p-value count” columns refer to the number of stocks for which the average return is larger on the original series than the 500 bootstrapped series; “Return” columns are the average return rates generated by each candlestick pattern on the bootstrapped series and original series. Table 7. The In-Sample (2000-Jan-4th to 2014-Dec-31st) and Out-of-Sample (2015-Jan-5th to 2018-Dec-28th) Profitability Results of the Best-Performing Japanese Candlestick Patterns. Sub-period 1 (In-sample) Sub-period 2 (Out-of-sample) Candlestick pattern N Return p-Value N Return p-Value Panel A (unconditional) Long White (10-day) 27,218 0.012** <.01 12,366 0.006** <.01 Bullish Gap (5-day) 1,802 0.027* .03 907 0.028** <.01 Bullish Gap (10-day) 1,802 0.038** <.01 907 0.050** <.01 Gravestone Doji (10-day) 569 −0.011* .02 247 −0.022* .01 Inverted Hammer (10-day) 1,982 −0.015** <.01 849 −0.006* .02 Panel B (trend condition) Long White (5-day) 12,940 0.010** <.01 5,952 0.003** <.01 Long White (10-day) 12,940 0.018** <.01 5,952 0.008** <.01 Bullish Gap (5-day) 855 0.022* .01 372 0.027* .01 Bullish Gap (10-day) 855 0.027** <.01 372 0.050* .01 Panel C (overbought/oversold condition) Bullish Gap (10-day) 1,258 0.036* .02 670 0.037* .01 Inverted Hammer (10-day) 1,701 −0.011** <.01 721 −0.003* .16 Panel D (trend + overbought/oversold condition) Long White (10-day) 8,937 0.015** <.01 4,089 0.006** <.01 Bullish Gap (3-day) 391 0.011** <.01 179 0.002 .53 Bullish Gap (5-day) 391 0.014** <.01 179 0.002 .69 Bullish Gap (10-day) 391 0.018** <.01 179 0.012 .10 Note. “N” is the number of trading signal for each candlestick pattern. “Return” refers to the average return from trading signals with its associated p-value. *and ** indicate generating positive average returns at the 5% and 1% level, respectively. indicator was still unable to cover the transaction cost 1%. For not-overbought condition, we can observe that none of the trading Bullish Gap with considering the trend condition and signals created a significant average return over 1%. Deng et al. 13 Table 8. Trading Simulation Results of the Five Bullish Candlestick Pattern Based Trading Strategies Over the period 2000 to 2018 Using the Component Stocks of the SSE50 Index. Holding period length and profitability test results Candlestick 1-day 3-day 5-day 10-day pattern based trading strategy SR AR SD SR AR SD SR AR SD SR AR SD Panel A (unconditional) Long White −0.899 −0.393 0.47 −0.073 −0.005 0.43 0.126 0.080 0.43 0.326 0.169 0.44 Dragonfly Doji −0.537 −0.348 0.70 −0.180 −0.055 0.45 0.034 0.041 0.45 0.231 0.138 0.48 Hammer −1.459 −0.329 0.24 −0.539 −0.090 0.21 −0.396 −0.041 0.17 −0.035 0.020 0.18 Bullish Harami −0.555 −0.086 0.20 −0.154 0.000 0.17 0.118 0.044 0.15 0.183 0.055 0.16 Bullish Gap 0.594 0.465 0.74 0.611 0.429 0.66 0.544 0.356 0.61 0.482 0.340 0.65 Panel B (uptrend) Long White −0.623 −0.228 0.41 0.168 0.092 0.39 0.313 0.154 0.41 0.409 0.206 0.44 Dragonfly Doji −0.373 −0.175 0.54 −0.040 0.010 0.41 0.140 0.078 0.37 0.256 0.127 0.40 Hammer −1.470 −0.206 0.16 −0.440 −0.039 0.15 −0.188 0.003 0.12 −0.017 0.024 0.10 Bullish Harami −0.711 −0.099 0.18 −0.348 −0.025 0.15 −0.132 0.009 0.13 −0.040 0.022 0.11 Bullish Gap 0.522 0.363 0.65 0.505 0.282 0.51 0.472 0.231 0.43 0.376 0.169 0.38 Panel C (not overbought) Long White −1.002 −0.436 0.46 −0.171 −0.047 0.43 0.032 0.040 0.44 0.218 0.121 0.44 Dragonfly Doji −0.907 −0.415 0.49 −0.322 −0.099 0.39 −0.022 0.017 0.42 0.139 0.078 0.37 Hammer −1.358 −0.307 0.25 −0.593 −0.092 0.20 −0.455 −0.043 0.15 −0.177 0.002 0.13 Bullish Harami −0.408 −0.065 0.22 −0.197 −0.010 0.18 0.062 0.037 0.17 0.154 0.050 0.16 Bullish Gap 0.445 0.159 0.30 0.489 0.185 0.33 0.385 0.158 0.34 0.400 0.183 0.39 Panel D (uptrend + not overbought) Long White −0.722 −0.256 0.39 0.057 0.048 0.39 0.178 0.096 0.39 0.257 0.126 0.39 Dragonfly Doji −0.760 −0.242 0.35 −0.170 −0.028 0.32 0.075 0.050 0.32 0.154 0.070 0.28 Hammer −1.318 −0.167 0.15 −0.452 −0.029 0.12 −0.227 0.003 0.10 −0.096 0.018 0.08 Bullish Harami −0.574 −0.079 0.18 −0.337 −0.022 0.14 −0.113 0.012 0.13 −0.038 0.022 0.11 Bullish Gap 0.244 0.056 0.12 0.286 0.058 0.11 0.218 0.048 0.10 0.109 0.034 0.07 Note. “SR” refers to the Sharpe ratio of each candlestick pattern based trading strategy. “AR” refers to the average annual return. “SD” refers to the standard deviation of the annual returns. the simulation trading return of each year is calculated for Simulation Trading Results each candlestick pattern based trading strategy. Using the Except for the profitability test of candlestick patterns, we annual returns, it is possible to calculate the Sharpe ratio of also designed the candlestick pattern based trading strategies each trading strategy, and the 1-year Yield of China’s trea- in the experiments, and simulation trading was adopted to sury bond is adopted as the risk-free rate. Additionally, the evaluate their trading performance. We introduced trading buy-and-hold strategy and sell-and-hold strategy, which are position management and transaction cost in our model. By two famous benchmarks, are used for comparing with the designing the candlestick pattern based trading strategy, it is bullish and bearish candlestick based trading strategies. possible to calculate the profit per year of each candlestick The trading simulation results on the component stocks of pattern, and its trading performance can be evaluated and the SSE50 index are shown in Tables 8 and 9. Compared with compared with other benchmarks. In the experiments, we the previous results, all the trading strategies of Bullish Gap adopted a “separation strategy” in position design, which and Long White with 5- or 10-day holding periods could earn means that we distribute the full position for each holding positive profits for investors. Since the Sharp ratio for sell-and- days averagely, and the daily position for each stock that hold was negative, the trading results were compared with the appears the trading signal is also distributed averagely. By buy-and-hold strategy. Considering the buy-and-hold strategy’s using this portfolio management, we can ensure that every Sharpe ratio was 0.291, for Bullish Gap, most trading strategies satisfied stock could be traded by the designed candlestick based on it performed better than the benchmark except panel pattern based trading strategy. For instance, if we adopt the D. Possible reason could be that the frequency of transaction 3-day holding period and there is a bullish candlestick signal was too low for the condition of panel D, thus it might miss the appears in three stocks at the same date D, we will distribute trading chances to produce considerable profits. For Long 1/9 long position to each stock at date D + 1. Subsequently, White based trading strategy, only the 10-day holding length in 14 SAGE Open Table 9. Trading Simulation Results of the Five Bearish Candlestick Pattern Based Trading Strategies Over the Period 2000 to 2018 Using the Component Stocks of the SSE50 Index. Holding period length and profitability test results Candlestick pattern 1-day 3-day 5-day 10-day based trading strategy SR AR SD SR AR SD SR AR SD SR AR SD Panel A (unconditional) Long Black −1.300 −0.585 0.47 −1.003 −0.386 0.41 −0.844 −0.353 0.45 −0.646 −0.251 0.43 Gravestone Doji −0.539 −0.275 0.56 −0.575 −0.102 0.22 −0.694 −0.071 0.14 −0.436 −0.078 0.24 Inverted Hammer −0.795 −0.162 0.24 −0.468 −0.074 0.21 −0.492 −0.058 0.17 −0.518 −0.069 0.18 Bearish Harami −0.485 −0.150 0.36 −0.581 −0.139 0.28 −0.526 −0.104 0.25 −0.558 −0.115 0.25 Bearish Gap −0.237 −0.036 0.26 0.024 0.031 0.20 −0.261 −0.015 0.16 −0.397 −0.027 0.13 Panel B (uptrend) Long Black −1.039 −0.315 0.33 −0.785 −0.202 0.29 −0.825 −0.176 0.24 −0.691 −0.133 0.23 Gravestone Doji −1.110 −0.066 0.08 −0.694 −0.011 0.05 −0.711 −0.010 0.05 −0.950 −0.004 0.03 Inverted Hammer −0.886 −0.092 0.13 −0.399 −0.016 0.11 −0.411 −0.009 0.09 −0.399 −0.003 0.07 Bearish Harami −0.559 −0.081 0.19 −0.492 −0.059 0.17 −0.487 −0.035 0.13 −0.590 −0.043 0.12 Bearish Gap −0.028 0.022 0.14 0.022 0.030 0.15 −0.143 0.011 0.11 −0.367 −0.004 0.08 Panel C (not overbought) Long Black −1.399 −0.568 0.42 −1.019 −0.353 0.37 −0.855 −0.297 0.38 −0.585 −0.206 0.40 Gravestone Doji −1.083 −0.139 0.15 −0.698 −0.049 0.11 −1.041 −0.040 0.06 −0.981 −0.017 0.04 Inverted Hammer −0.666 −0.137 0.25 −0.420 −0.056 0.20 −0.467 −0.046 0.15 −0.540 −0.047 0.14 Bearish Harami −0.498 −0.146 0.35 −0.508 −0.115 0.28 −0.470 −0.079 0.22 −0.560 −0.088 0.20 Bearish Gap −0.424 −0.078 0.25 −0.054 0.014 0.22 −0.242 −0.013 0.16 −0.385 −0.018 0.12 Panel D (uptrend + not overbought) Long Black −1.088 −0.274 0.28 −0.617 −0.162 0.31 −0.560 −0.128 0.27 −0.497 −0.102 0.26 Gravestone Doji −1.166 −0.061 0.07 −0.639 −0.001 0.04 −0.703 −0.007 0.05 −0.991 −0.005 0.03 Inverted Hammer −0.734 −0.064 0.12 −0.250 −0.003 0.11 −0.326 −0.002 0.09 −0.395 −0.003 0.07 Bearish Harami −0.444 −0.060 0.19 −0.427 −0.050 0.18 −0.420 −0.027 0.13 −0.599 −0.044 0.12 Bearish Gap −0.211 −0.006 0.15 −0.019 0.023 0.15 −0.113 0.014 0.11 −0.399 −0.003 0.07 Note. “SR” refers to the Sharpe ratio of each candlestick pattern based trading strategy. “AR” refers to the average annual return. “SD” refers to the standard deviation of the annual returns. panels A and B, as well as the 5-day holding length in panel B, the benchmark. Among the bearish candlestick patterns, performed better than the buy-and-hold. Some trading strate- Bearish Gap based trading strategies performed better than the gies produced positive annual returns, but they did not perform buy-and-hold strategy in terms of Sharpe ratio. better than the buy-and-hold in terms of annual average return or Sharpe ratio. Additionally, for the bearish candlestick pattern Conclusion based trading strategies, none of them could perform better than the buy-and-hold in terms of both average annual return Traditional research has done little to examine the effectiveness and Sharpe ratio. of Japanese candlestick patterns to the component stocks of the Our further research used all the listed company data in the Chinese SSE 50 index when taking into consideration the over- Chinese stock market. The results are shown in Tables 10 and bought/oversold condition. In this study, we aimed to fill this 11, from which it is found that all Bullish Gap based trading gap, and comprehensive analysis has been conducted on the strategies generated positive returns, and most of them per- component stocks of the SSE 50 index to investigate the pre- formed better than buy-and-hold, which produced an average dictive power of 10 well-known Japanese candlestick patterns. return of 11.22% and a Sharpe ratio value of 0.183. Unlike the For the experiments, daily-based data on the component stocks previous results, Long White based trading strategy failed to of the SSE50 index that covers the period from 2000 to 2018 perform better than the benchmark in terms of Sharpe ratio. was employed. In addition, to avoid data mining, we divided Similar results as the trading simulation test on the SSE50 data- the whole dataset period into two sub-periods, in which the out- sets, several Bullish Harami based trading strategies could of-sample period was further examined to confirm the robust- yield positive returns, while they could not perform better than ness of the results that whether or not the successful candlestick Deng et al. 15 Table 10. Trading Simulation Results of the Five Bullish Candlestick Patterns Based Trading Strategies Over the Period 2000 to 2018 Using All Stocks Data. Holding period length and profitability test results Candlestick 1-day 3-day 5-day 10-day pattern based trading strategy SR AR SD SR AR SD SR AR SD SR AR SD Panel A (unconditional) Long White −0.897 −0.519 0.61 −0.411 −0.198 0.55 −0.271 −0.118 0.53 0.006 0.029 0.54 Dragonfly Doji −0.878 −0.515 0.62 −0.290 −0.141 0.58 −0.131 −0.045 0.55 0.058 0.057 0.54 Hammer −0.749 −0.521 0.73 −0.437 −0.228 0.58 −0.223 −0.105 0.59 −0.051 −0.002 0.55 Bullish Harami −0.769 −0.482 0.66 −0.339 −0.140 0.49 −0.050 0.002 0.47 −0.013 0.020 0.45 Bullish Gap 0.756 2.343 3.06 0.689 1.960 2.81 0.604 1.583 2.58 0.587 1.243 2.07 Panel B (uptrend) Long White −1.048 −0.662 0.66 −0.491 −0.268 0.60 −0.350 −0.181 0.59 −0.119 −0.043 0.58 Dragonfly Doji −1.095 −0.653 0.62 −0.425 −0.226 0.59 −0.233 −0.105 0.56 −0.067 −0.011 0.56 Hammer −0.689 −0.487 0.74 −0.407 −0.220 0.61 −0.243 −0.117 0.59 −0.103 −0.029 0.53 Bullish Harami −0.774 −0.479 0.65 −0.234 −0.108 0.57 −0.076 −0.009 0.46 −0.068 −0.004 0.44 Bullish Gap 0.981 1.739 1.75 0.784 1.129 1.41 0.624 0.793 1.23 0.489 0.505 0.98 Panel C (not overbought) Long White −0.870 −0.489 0.59 −0.397 −0.185 0.53 −0.249 −0.106 0.53 −0.005 0.023 0.53 Dragonfly Doji −0.824 −0.485 0.62 −0.299 −0.144 0.57 −0.144 −0.052 0.54 0.036 0.045 0.53 Hammer −0.689 −0.487 0.74 −0.407 −0.220 0.61 −0.243 −0.117 0.59 −0.103 −0.029 0.53 Bullish Harami −0.394 −0.344 0.94 −0.228 −0.101 0.56 −0.013 0.020 0.51 −0.015 0.019 0.47 Bullish Gap 0.587 2.144 3.61 0.594 2.052 3.41 0.540 1.738 3.17 0.535 1.458 2.68 Panel D (uptrend + not overbought) Long White −1.002 −0.616 0.64 −0.451 −0.242 0.59 −0.298 −0.150 0.59 −0.125 −0.046 0.57 Dragonfly Doji −1.045 −0.616 0.62 −0.432 −0.226 0.58 −0.249 −0.107 0.54 −0.104 −0.029 0.53 Hammer −0.827 −0.501 0.64 −0.593 −0.250 0.46 −0.368 −0.150 0.48 −0.213 −0.075 0.48 Bullish Harami −0.382 −0.311 0.88 −0.185 −0.074 0.54 −0.023 0.015 0.49 −0.022 0.016 0.45 Bullish Gap 0.477 0.523 1.04 0.499 0.436 0.82 0.320 0.276 0.78 0.229 0.182 0.68 Note. “SR” refers to the Sharpe ratio of each candlestick based trading strategy. “AR” refers to the average annual return. “SD” refers to the standard deviation of the annual returns. patterns in the in-sample period could remain their predictabil- White based trading strategies, not only generated positive ity in the out-of-sample period. average annual return but also performed better than the bench- Our primary analysis results from the in-sample period mark in terms of Sharpe ratio. revealed that certain bullish patterns were found to produce Overall, our results suggest that several candlestick patterns positive and significant average returns on the component and candlestick pattern based trading strategies, particularly stocks of the SSE50 index. The out-of-sample experimental Bullish Gap and Long White patterns with 5- or 10-day holding results of the robustness test demonstrated that signals gener- periods, can create values for investors trading on the compo- ated by Bullish Gap over a 10-day holding period under all four nent stocks of the SSE50 index. Therefore, experimental results conditions were proved to be profitable for investors. demonstrated that several candlestick pattern based trading Additionally, experimental results indicated that the typical use strategies had significant predictive power, whereas investors of the bearish patterns investigated in this research was not need to consider different market conditions when employing profitable on the component stocks of the SSE50 index over those Japanese candlestick patterns. Future studies could test the studied period. Further, based upon a bootstrap analysis, the their profitability on the medium-market-value or small-market- robustness of the experiment results was subsequently con- value stocks in the Chinese stock market, or examine their prof- firmed. Finally, we designed the candlestick pattern based trad- itability in other financial markets such as FX or commodity ing strategies and examined their profitability on the component markets. Moreover, this research studied the profitability of stocks of the SSE50 index, as well as all stocks in the Chinese candlestick patterns by using daily data, while other researchers stock market. Experimental results showed that most Bullish/ could investigate the high-frequency financial time series data Bearish Gap based trading strategies, as well as several Long such as 2-hour, 1-hour, or 30-minutes in the future. 16 SAGE Open Table 11. Trading Simulation Results of the Five Bearish Candlestick Pattern Based Trading Strategies Over the Period 2000 to 2018 Using All Stocks Data. Holding period length and profitability test results Candlestick pattern 1-day 3-day 5-day 10-day based trading strategy SR AR SD SR AR SD SR AR SD SR AR SD Panel A (unconditional) Long Black −1.192 −0.611 0.53 −0.609 −0.298 0.53 −0.505 −0.245 0.54 −0.406 −0.188 0.53 Gravestone Doji −1.123 −1.818 1.64 −0.636 −0.612 1.00 −0.505 −0.404 0.85 −0.389 −0.263 0.74 Inverted Hammer −1.310 −0.926 0.73 −0.483 −0.254 0.58 −0.358 −0.184 0.59 −0.307 −0.143 0.55 Bearish Harami −0.557 −0.268 0.53 −0.117 −0.040 0.56 −0.048 −0.003 0.61 −0.127 −0.043 0.55 Bearish Gap 1.033 1.408 1.34 0.968 1.075 1.08 0.753 0.725 0.93 0.343 0.335 0.90 Panel B (uptrend) Long Black −1.016 −0.468 0.49 −0.469 −0.242 0.57 −0.395 −0.219 0.62 −0.325 −0.270 0.91 Gravestone Doji −0.836 −1.491 1.81 −0.462 −0.473 1.08 −0.368 −0.328 0.96 −0.333 −0.156 0.55 Inverted Hammer −1.124 −0.859 0.79 −0.442 −0.318 0.78 −0.383 −0.204 0.60 −0.317 −0.136 0.51 Bearish Harami −0.584 −0.211 0.41 −0.133 −0.089 0.87 −0.165 −0.186 1.29 −0.195 −0.112 0.71 Bearish Gap 0.890 1.176 1.29 0.900 0.919 0.99 0.651 0.607 0.89 0.091 0.140 1.25 Panel C (not overbought) Long Black −1.184 −0.602 0.53 −0.594 −0.286 0.53 −0.491 −0.229 0.52 −0.372 −0.164 0.51 Gravestone Doji −0.998 −1.608 1.64 −0.588 −0.509 0.91 −0.448 −0.373 0.89 −0.370 −0.212 0.65 Inverted Hammer −1.339 −0.954 0.79 −0.501 −0.275 0.60 −0.342 −0.178 0.60 −0.272 −0.119 0.53 Bearish Harami −0.683 −0.312 0.50 −0.171 −0.071 0.57 −0.078 −0.021 0.60 −0.130 −0.044 0.54 Bearish Gap 0.808 1.062 1.28 0.829 0.868 1.02 0.648 0.569 0.84 0.415 0.314 0.69 Panel D (uptrend + not overbought) Long Black −0.914 −0.437 0.51 −0.413 −0.221 0.60 −0.366 −0.170 0.54 −0.316 −0.195 0.70 Gravestone Doji −0.812 −1.508 1.89 −0.460 −0.454 1.04 −0.356 −0.310 0.95 −0.325 −0.151 0.54 Inverted Hammer −1.104 −0.897 0.84 −0.409 −0.292 0.78 −0.304 −0.161 0.62 −0.263 −0.094 0.46 Bearish Harami −0.560 −0.197 0.40 −0.139 −0.094 0.86 −0.160 −0.161 1.17 −0.205 −0.107 0.65 Bearish Gap 0.798 0.661 0.80 0.791 0.547 0.66 0.628 0.357 0.53 0.263 0.153 0.48 Note. “SR” refers to the Sharpe ratio of each candlestick pattern based trading strategy. “AR” refers to the average annual return. “SD” refers to the standard deviation of the annual returns. 17 Appendix A. Japanese Candlestick Patterns. Japanese Pattern shape candlestick pattern example Description Dragonfly Doji Dragonfly Doji consists of one candlestick line. It is identified when the length of the candlestick body is smaller or equal to 1/10 of the total length of the candlestick. Long White For Long White, the open price of the candlestick line is less than the close price. The candlestick body length must be greater or equal to the median of the candlestick body length of 20 preceding candlesticks. Bullish Hammer Bullish Hammer is a 1-day candlestick. It must satisfy three requirements: (1) The length of the lower shadow line must be greater than two-thirds of the length of the entire candlestick length; (2) The upper shadow line must be less than or equal to one-tenth of the length of the entire candlestick line; (3) The length of the candlestick body must be greater than 1/10 of the length of the entire candlestick line. Bullish Harami In Bullish Harami, a black candlestick is followed by a white candlestick. The white candlestick body is located within the vertical range of the black candlestick body. The Bullish Harami is a black candlestick engulfing a white candlestick. Bullish Gap Bullish Gap consists of two candlesticks; the low price of the second candlestick should be greater than the high price of the first candlestick line. Gravestone Doji Gravestone Doji consists of one candlestick. The open and the close prices of the candlestick are identical or very close to each other. The body of the candlestick is smaller or equal to 1/10 of its overall length. It has a long upper shadow but no lower shadow. Long Black For Long Black, the open price of the candlestick line is greater than the close price. The candlestick body length must be greater or equal to the median of the body length of 20 preceding candlesticks. Inverted Hammer Inverted Hammer consists of one candlestick. It must satisfy three requirements: (1) The length of the upper shadow line must be greater than two-thirds of the length of the entire candlestick. (2) The lower shadow line must be less than or equal to one-tenth of the length of the entire candlestick. (3) The length of the candlestick body must be greater than one-tenth of the length of the entire candlestick. Bearish Harami In Bearish Harami, a white candlestick is followed by a black candlestick. The black candlestick body is located within the vertical range of the white candlestick body. The Bearish Harami is a white candlestick engulfing a black candlestick. Bearish Gap Bearish Gap consists of two candlesticks; the high price of the second candlestick line is lower than the low price of the first candlestick line. 18 SAGE Open Appendix B. A List of the Component Stocks for the SSE 50 Index. No Company name No Company name 1 Shanghai Pudong Development Bank 2 China Minsheng Banking Corporation 3 Baoshan Iron & Steel 4 PetroChina Company 5 China Southern Airlines Company 6 Citic Securities Company 7 China Merchants Bank 8 Poly Developments and Holdings Group 9 China United Network Communications 10 SAIC Motor Corporation 11 Shanghai Fosun Pharmaceutical (Group) 12 Jiangsu Hengrui Medicine 13 Wanhua Chemical Group 14 China Fortune Land Development 15 Kweichow Moutai 16 Shandong Gold Mining 17 Anhui Conch Cement Company 18 Greenland Holdings Corporation 19 Haier Smart Home 20 Sanan Optoelectronics 21 Inner Mongolia Yili Industrial Group 22 Daqin Railway 23 Industrial Bank 24 Ping an Insurance (Group) 25 China Life Insurance Company 26 China Shenhua Energy Company 27 Bank of Beijing 28 China Railway Construction Corporation 29 Guotai Junan Securities 30 Agricultural Bank of China Limited 31 Bank of Communications 32 New China Life Insurance Company 33 360 Security Technology Inc. 34 China Railway Group 35 Industrial and Commercial Bank of China 36 China Pacific Insurance (Group) 37 China State Construction Engineering Corporation 38 Huatai Securities 39 CRRC Corporation Limited 40 China Communications Construction 41 China Everbright Bank Company 42 PetroChina Company 43 China Tourism Group Duty Free Corporation 44 China Construction Bank 45 Bank of China 46 China Shipbuilding Industry Company 47 China Molybdenum 48 Foxconn Industrial Internet 49 Bank of Shanghai 50 WuXi Application Technology Acknowledgments ORCID iD We thank the editors and anonymous referees for their insightful Shangkun Deng https://orcid.org/0000-0002-6041-485X comments. All errors and omissions are our responsibility. Additionally, the scientific calculations in this paper have been References done on the HPC Cloud Platform of Shandong University. Batten, J. A., Lucey, B. M., Mcgroarty, F., Peat, M., & Urquhart, A. (2018). Does intraday technical trading have predictive power Data Availability Statement in precious metal markets? Journal of International Financial The data that support the findings of this study are openly available Markets, Institutions and Money, 52, 102–113. in the Choice database at http://data.eastmoney.com. Brock, W., Lakonishok, J., & LeBaron, B. (1992). Simple technical trading strategies and the stochastic properties of stock returns. Declaration of Conflicting Interests The Journal of Finance, 47(5), 1731–1764. Caginalp, G., & Laurent, H. (1998). The predictive power of price The author(s) declared no potential conflicts of interest with respect patterns. Applied Mathematical Finance, 5, 181–205. to the research, authorship, and/or publication of this article. Chen, S., Bao, S., & Zhou, Y. (2016). The predictive power of Japanese candlestick charting in Chinese stock market. Physica Funding A: Statistical Mechanics and its Applications, 457, 148–165. The author(s) disclosed receipt of the following financial support Chong, T. L., & Ng, W. K. (2008). Technical analysis and the lon- for the research, authorship, and/or publication of this article: This don stock exchange: Testing the macd and rsi strategies using work was supported by the Natural Science Foundation of Hubei the ft30. Applied Economics Lettersm, 15(14), 1111–1114. Province [Grant Number 2021CFB175]; National Social Science Cohen, G. (2021). Optimizing candlesticks patterns for Bitcoin’s Foundation of China [Grant Number 19BGL131]; and Innovation trading systems. Review of Quantitative Finance and and Entrepreneurship Programs of China Three Gorges University Accounting, 57, 1155–1167. [Grant Numbers S202111075045 and S202111075066X]. In addi- Day, T. E., & Wang, P. (2002). Dividends, nonsynchronous prices, tion, we are grateful to the editors and anonymous reviewers for and the returns from trading the dow jones industrial average. their comments and discussions. Journal of Empirical Finance, 9(4), 431–454. Deng et al. 19 Deng, S., Yu, H., Wei, C., Yang, T., & Tatsuro, S. (2020). The Metghalchi, M., Marcucci, J., & Chang, Y. H. (2012). Are moving profitability of Ichimoku Kinkohyo based trading rules in stock average trading strategies profitable? Evidence from the European markets and FX markets. International Journal of Finance and stock markets. Applied Economics, 44(12), 1539–1559. Economics, 26, 5321–5326. https://doi.org/10.1002/ijfe.2067 Mitra, S. K. (2011). How rewarding is technical analysis in the Fama, E. F. (1970). Efficient capital markets: A review of theory Indian stock market? Quantitative Finance, 11, 287–297. and empirical work. The Journal of Finance, 25(2), 383–417. Murphy, J. J. (1999). Technical analysis of the financial markets: A Fang, J., Jacobsen, B., & Qin, Y. (2013). Predictability of the sim- comprehensive guide to trading methods and applications. In ple technical trading strategies: An out-of-sample test. Review Technical analysis of the financial markets. New York Institute of Financial Economics, 23(1), 30–45. of Finance. Fifield, S. G., Power, D. M., & Sinclair, C. D. (2005). An analy- Naranjo, R., & Santos, M. (2019). A fuzzy decision system for sis of trading strategies in eleven European stock markets. The money investment in stock markets based on fuzzy candle- European Journal of Finance, 11(6), 531–548. sticks pattern recognition. Expert Systems with Applications, Gehrig, T., & Menkhoff, L. (2006). Extended evidence on the use of 133, 34–48. technical analysis in foreign exchange. International Journal Nison, S. (1991). Japanese candlestick charting techniques. New of Finance and Economics, 11, 327–338. York Institute of Finance. Goo, Y., Chen, D., & Chang, Y. (2007). The application of Park, C. H., & Irwin, S. H. (2007). What do we know about the Japanese candlestick trading strategies in Taiwan. Investment profitability of technical analysis? Journal of Economic Management and Financial Innovations, 4, 49–71. Surveys, 21, 786–826. Gunasekarage, A., & Power, D. M. (2001). The profitability of Prado, H. A. D., Ferneda, E., Morais, L. C. R., & Matsura, E. moving average trading rules in south Asian stock markets. (2013). On the effectiveness of candlestick chart analysis for Emerging Markets Review, 2(1), 1–33. the Brazilian stock market. Procedia Computer Science, 22, Heinz, A., Jamaloodeen, M., Saxena, A., & Pollacia, L. (2021). 1136–1145. Bullish and Bearish Engulfing Japanese candlestick patterns: Qiu, H., & Liu, F. (2019, March 15–18). Candlestick analysis in A statistical analysis on the S&P 500 Index. The Quarterly forecasting U.S. stock market: Are they informative and effec- Review of Economics and Finance, 79, 221–244. tive [Conference session]. IEEE International Conference on Jensen, M. C. (1967). Random walks: Reality or mythcomment. Big Data Analytics, Suzhou, China. Financial Analysts Journal, 23, 77–85. Thanh, L. V., Anh, B. N., Ban, T. Q., Hoa, D. X., & Son, N. T. Kwok, N. M., Fang, G., & Ha, Q. P. (2009). Moving average-based (2018). An analyze on effectiveness of candlestick reversal pat- stock trading rules from particle swarm optimization [Conference terns for vietnamese stock market [Conference session]. 2018 session]. 2009 International Conference on Artificial Intelligence International Conference. and Computational Intelligence (Vol. 1, pp. 149–153). November Tharavanij, P., Siraprapasiri, V., & Rajchamaha, K. (2017). 7–8, 2009. Washington, DC, United States. Profitability of candlestick charting patterns in the stock Lu, T. H. (2014). The profitability of candlestick charting in the exchange of Thailand. SAGE Open, 8(4), 1–18. Taiwan stock market. Pacific-Basin Finance Journal, 26, 65– Trivedi, S. R. (2022). Technical analysis using Heiken Ashi 78. Stochastic: To catch a trend, use a HASTOC. International Lu, T. H., Chen, Y. C., & Hsu, Y. C. (2015). Trend definition or Journal of Finance and Economics, 27, 1836–1847. https://doi. holding strategy: What determines the profitability of candle- org/10.1002/ijfe.2245. stick charting? Journal of Banking & Finance, 61, 172–183. Vasiliou, D., Eriotis, N., & Papathanasiou, S. (2008). Technical Lu, T. H., & Shiu, Y. M. (2012). Tests for two-day candlestick trading profitability in Greek stock market. The Empirical patterns in the emerging equity market of Taiwan. Emerging Economics Letters, 7(7), 1681–8997. Markets Finance & Trade, 48, 41–57. Wilder, J. W. (1978). New concepts in technical trading systems. Lu, T. H., & Shiu, Y. M. (2016). Can 1-day candlestick patterns be Trend Research. profitable on the 30 component stocks of the DJIA? Applied Xie, H., Zhao, X., & Wang, S. (2012). A comprehensive look at Economics, 48(34–36), 3345–3354. the predictive information in Japanese candlestick. Procedia Lu, T. H., Shiu, Y. M., & Liu, T. C. (2012). Profitable candle- Computer Science, 9, 1219–1227. stick trading strategies: The evidence from a new perspective. Zhu, M., Atri, S., & Yegen, E. (2016). Are candlestick trading Review of Financial Economics, 21, 63–68. strategies effective in certain stocks with distinct features? Marshall, B. R., Young, M. R., & Rose, L. C. (2006). Candlestick Pacific-Basin Finance Journal, 37(4), 116–127. Database 1. technical trading strategies: Can they create value for inves- Choice database. Retrieved May 1, 2022, from http://choice. tors? Journal of Banking and Finance, 30, 2303–2323. eastmoney.com/ http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png SAGE Open SAGE

Can Japanese Candlestick Patterns be Profitable on the Component Stocks of the SSE50 Index?

Deng, Shangkun; Su, Zhihao; Ren, Yanmei; Yu, Haoran; Zhu, Yingke; Wei, Chenyang
SAGE Open , Volume OnlineFirst: 1 – Aug 10, 2022
Download PDF
19 pages

Loading next page...
 
/lp/sage/can-japanese-candlestick-patterns-be-profitable-on-the-component-TkSI7yZku9

References (56)

Publisher
SAGE
Copyright
© The Author(s) 2022
ISSN
2158-2440
eISSN
2158-2440
DOI
10.1177/21582440221117803
Publisher site
See Article on Publisher Site

Abstract

In this study, we investigate the profitability of 10 well-known Japanese candlestick charting patterns using daily-based data on the component stocks of the Chinese SSE50 index, which involves a lengthy sample period from January 2000 to December 2018. The main contribution of this paper is that we conduct the first predictive power examination of Japanese candlestick patterns on the Chinese SSE50 stocks while taking into account trend and overbought/oversold conditions, and their profitability over different holding periods. Experimental results indicate that several bullish candlestick patterns such as Long White and Bullish Gap can produce a significant positive average return over certain holding periods. In addition, empirical results show that none of the bearish candlestick patterns we examined offers predictive power. However, without considering trend and overbought/oversold conditions, we find that the bearish pattern Gravestone Doji over a 10-day holding period has superior profitability if it is applied as a contrary trading signal. The robustness of our results is confirmed based upon a bootstrap analysis and an out-of-sample test. The findings of this study are beneficial for the market traders engaged in transaction of the SSE50 component stocks. Keywords candlestick pattern, profitability test, trend condition, overbought/oversold condition, out-of-sample test practitioners but also researchers from many western coun- Introduction tries. Although there are many literature that have studied the Technical analysts use historical market trading data such as profitability of the Japanese candlestick patterns on stock price, volume, money flow, or turnover to predict move- markets (Goo et al., 2007; Lu, 2014; Lu et al., 2012), few of ments of future prices (Gehrig & Menkhoff, 2006; Murphy, them investigated the profitability of candlesticks on the 1999). Proponents of technical analysis believe that such his- Chinese SSE50 index in the Chinese market. Thus, in this torical trading data contains crucial information about direc- study, we examine the profitability of several well-known tion changes in future prices. However, according to Efficient Japanese candlestick patterns on the component stocks of the Market Hypothesis (EMH), in the weak form of an efficient SSE50 index, which is regarded as a blue chip and an essen- market, any information contained in past prices is therefore tial stock index of the Shanghai Stock Exchange. Since the reflected in current prices, such that historical price data sug- component stocks of the SSE50 index represent the top 50 gest no valuable information for forecasting future prices companies by capitalization, using those component stocks (Fama, 1970). Nevertheless, many researchers have proved of the SSE50 index for profitability examination will be that several technical analysis approaches could offer predic- highly representative of the Chinese stock market. tive power in financial markets (Brock et al., 1992; Chong & Ng, 2008; Deng et al., 2020; Fang et al., 2013; Fifield et al., China Three Gorges University, Yichang, China 2005; Metghalchi et al., 2012; Trivedi, 2022). Shandong University, Jinan, China Among the well-known technical analysis approaches, 3 Zhongnan University of Economics and Law, Wuhan, China the Japanese candlestick is the earliest one that was initially East China University of Science and Technology, Shanghai, China applied to the Japanese rice market in the 1700s. Since Nison Corresponding Author: (1991) first translated the knowledge of the Japanese candle- Shangkun Deng, College of Economics and Management, China Three stick in English and introduced it to the western world, it has Gorges University, Yichang 443002, China. Email: dsk8672@163.com drawn considerable attention from not only financial market Creative Commons CC BY: This article is distributed under the terms of the Creative Commons Attribution 4.0 License (https://creativecommons.org/licenses/by/4.0/) which permits any use, reproduction and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and Open Access pages (https://us.sagepub.com/en-us/nam/open-access-at-sage). 2 SAGE Open Other than applying just the Japanese candlestick pat- when applying the Japanese candlestick patterns, few prior terns, Goo et al. (2007) concluded that the profitability of researchers considered the overbought/oversold condition Japanese candlestick patterns could be improved by employ- for applying Japanese candlestick in the Chinese stock mar- ing other indicators simultaneously. Indeed, when market ket. Nevertheless, market or stock conditions are essential practitioners adopt the Japanese candlestick patterns for for traders when they make a transaction in actual trading. deciding if a buying or a short-selling signal is valid, they Hence, in this study, we test the profitability of Japanese would generally also take into account the underlying stock candlestick patterns with considering the uptrend/downtrend conditions, such as whether the stock price is in an uptrend or condition or overbought/oversold condition of stock prices. a downtrend, or whether it has already been overbought or To identify the trend condition, a well-known trend indicator oversold, rather than just using the Japanese candlestick pat- called moving average (MA), which is widely employed by terns. For instance, when a bullish candlestick pattern lots of market traders and researchers, is employed in this appears, professional market traders would generally also research. In addition, to confirm the overbought/oversold confirm the following two points before executing a long condition, though there are several momentum oscillators transaction: (1) The trading target is in an upward trend and that could be employed, one of the most common indicators (2) The trading target is currently not in an overbought area. called relative strength index (RSI) is employed by us in this Otherwise, it might imply an extremely high risk for trading. research. Subsequently, we investigate whether Japanese If either of those two conditions is not satisfied, it might not candlestick patterns could create values for market investors be appropriate to execute a long transaction at that time. with considering market conditions. Similarly, it would be more appropriate for market traders to Additionally, Park and Irwin (2007) pointed out that it is execute a short-selling transaction when a bearish candle- important to take data snooping, out-of-sample problems, and stick signal appears with condition confirmation that the transaction costs into consideration when examining the effec- stock price is in a downtrend and the stock is not in an over- tiveness of technical analysis approaches. Therefore, in this sold condition. study, we also adopt a bootstrap analysis methodology adopted by Marshall et al. (2006). Meanwhile, an out-of-sample test is conducted, and transaction cost is considered in the experi- Literature Review ments to examine whether Japanese candlestick patterns can In the last decade, the profitability of Japanese candlestick create values for market practitioners. Moreover, except for patterns on stock markets has been investigated by numerous the profitability test of Japanese candlestick patterns, we also researchers. Lu et al. (2012) and Lu and Shiu (2012) exam- conduct a simulation trading using the Japanese candlestick ined the profitability of Japanese candlestick patterns on the based trading strategies to examine their profitability. Taiwan stock market, and they found that several candlestick Compared with the profitability test of candlestick patterns, patterns have predictive power. Among the literature that simulation trading results would be closer to actual trading. investigated the effectiveness of Japanese candlestick pat- The main contributions and innovations of this study terns on U.S. stock markets, many researchers found that could be summarized as follows. First, to the best of our only a small fraction of the candlestick patterns could pro- knowledge, this paper is the first study that examines the vide correct direction predictions for market traders (Cohen, profitability of Japanese candlestick patterns in the Chinese 2021; Lu et al., 2015; Lu & Shiu, 2016; Qiu & Liu, 2019). stock market with considering the overbought/oversold con- Apart from the investigation of stock markets in developed ditions, and we compared the difference between the trading countries and areas (Naranjo and Santos, 2019; Xie et al., strategy results of the following four groups: candlestick pat- 2012), the profitability of candlestick patterns has also been tern only (unconditional), candlestick pattern appears in a examined in the stock markets of developing countries such trend, candlestick pattern appears with considering over- as the Vietnamese stock market (Thanh et al., 2018), Thailand bought/oversold condition, and candlestick pattern appears stock market (Tharavanij et al., 2017), Brazilian stock mar- with considering both the trend and overbought/oversold ket (Prado et al., 2013), and Chinese stock market (Chen conditions. Second, given the increased attention on the et al., 2016; Zhu et al., 2016). Chinese stock market from market traders and researchers all However, in most of the related literature that examined over the world, unlike most of the prior profitability test the profitability of Japanese candlestick patterns, researchers studies conducted on stock index of south Asian stock mar- generally did not consider the market conditions. Whereas, kets (Gunasekarage & Power, 2001) or European stock mar- identifying whether a stock price is in an overbought or an kets (Vasiliou et al., 2008), this study adopts the use of the oversold condition, or whether it is in an uptrend or a down- individual component stocks comprising the SSE50 index. trend condition would be an critical part of establishing the Since tests on index data are unable to be implemented and trading risk. In the current literature, only a limited amount biased as a result of nonsynchronous trading (Day & Wang, of them have considered the trend condition of stock prices 2002), investigations on the component stocks of a stock (Batten et al., 2018; Heinz et al., 2021; Lu et al., 2015) or index would be more practical for market traders. Third, overbought/oversold condition (Tharavanij et al., 2017) unlike previous studies that only researched the profitability Deng et al. 3 Table 1. A list of the five pairs of bullish and bearish Japanese candlestick patterns. No Candlestick Bullish pattern Bearish pattern 1 Long candle Long White Long Black 2 Doji Dragonfly Doji Gravestone Doji 3 Hammer Bullish Hammer Inverted Hammer 4 Harami Bullish Harami Bearish Harami 5 Gap Bullish Gap Bearish Gap tj + tj + test of candlestick patterns, we design the trading strategies (( P iS )/ − Pi () /) Lj >∈ 01 ,( ,....,T ),, (1) cc ∑ based on candlestick patterns to test whether those Japanese it =+ jS −+1 it =+ jL −+1 candlestick pattern based trading strategies could make prof- its for traders who participated in the Chinese stock market. which implies an uptrend condition identification at time The remainder of this paper is arranged as follows: a brief t + T. Similarly, a downward trend is identified if a short- introduction to the candlestick charting, trend identification, term moving average is below a long-term moving average, and overbought/oversold condition identification are and the condition lasts for a certain period: described in Section 3. In Section 4, we explain the research data, trading strategy, and research design adopted for this tj + tj + study. Empirical results of the profitability tests, bootstrap (2) (( P iS )/ − Pi () /) Lj <∈ 01 ,( ,....,T ),, cc ∑ analysis results, and out-of-sample test results are reported in it =+ jS −+1 it =+ jL −+1 Section 5. Conclusions and future work are provided in which implies a downtrend condition identification at time t + T. Section 6. A description of the examined Japanese candle- In the above equations (1) and (2), P (i) denotes the close stick patterns is provided in Appendix A. price at time i; T represents the minimum length of lasting period, which is set to be three in this research; L and S, Methodology respectively denote the parameters of a long-term and a short-term moving average, and the values of L and S are set Japanese Candlestick Pattern to be 20 and 10, respectively, as they have been widely used Candlestick charting was created by a Japanese rice trader by traders in financial markets. named Munehisa Homma. He observed that the emotions of investors reflected in the price candlesticks might exert sig- nificant influences on the rice markets (Nison, 1991). A sin- Overbought/Oversold Condition Identification gle Japanese candlestick line is created based upon four Relative Strength Index (RSI) is a momentum oscillator ini- prices (open, high, low, and close prices) of a certain period, tially developed by Wilder (1978). It is a measurement of stock and one single line or several lines may then comprise one price change momentum, which is displayed as an oscillator (a bullish/bearish pattern. There are many candlestick patterns line graph that moves between two extremes) that has a value to suggest a bullish signal, such as Long White, or suggest a range from 0 to 100. RSI is calculated based on prior sessions’ bearish signal, such as Long Black. Five pairs of well-known average gains versus losses. Traditional interpretation and use candlestick patterns examined in this study are displayed in of the RSI is as follows: an RSI value of 70 or above indicates Table 1, and the real examples of the candlestick patterns that a stock is overbought, and it might be primed for a trend rever- appeared in one component stock (Shanghai Pudong sal or corrective pullback in price. In contrast, an RSI value of Development Bank) of the SSE50 index are provided in 30 or below indicates an oversold or undervalued stock condi- Figure 1. Details of the candlestick patterns are described, tion. The calculation of RSI is as follows: with their shapes being illustrated in Appendix A. For each trading period t, an upward change U or down- ward change D is calculated as: Trend Identification In this study, the market trend is measured and identified by 00 () PP −≤  closet ,, closet−n U = , (3) a moving average (MA) rule (Kwok et al., 2009). MA is an P − P PP () −> P 0  closet , closet ,, −− nclose tclose,tn average of time series prices over several consecutive peri- ods. In this research, the MA is calculated using the close price of stocks. An uptrend is generally identified if a short- 00 () PP −≤  closet ,, -nclose t (4) term moving average is above a long-term moving average, D= . P − −− PP () P > 0  closet , -n closet ,, closet -nclose,t and the condition lasts for a certain period:  4 SAGE Open Figure1. (continued) Deng et al. 5 Figure 1. Real examples of the 10 candlestick patterns appeared in historical data of Shanghai Pudong Development Bank. The ratio of these averages is the relative strength (RS) or profitability analysis, with the whole sample period ranging relative strength factor: from January 2000 to December 2018. The trading price data of all component stocks are in four prices (open, high, low, and close) format. Among the 50 component stocks, 3 stocks EMAU (,n) (5) RS = , that without data before 31 December 2014 were excluded. EMAD (,n) Thus, the resulting sample included a total of 47 stocks. In addition, since the out-of-sample test is important for inves- (6) EMAP =+ αα () 1- EMA , tclose,tt−1 tors, the robustness of research results was also examined. Subsequently, the whole sample period was divided into two (7) α= , sub-periods, which were the in-sample (Sub-period 1) and n + 1 out-of-sample (Sub-period 2) periods with a length ratio of 3:1. The sample period before/after 1 January 2015 was where n represents the period parameter for calculating the employed as the in-sample/out-of-sample dataset. RSI, and EMA (Exponential Moving Average) is a type of Subsequently, the sub-period 1 ranges from 2000-Jan-4th to moving average that places a larger weight and significance on 2014-Dec-31st, while the date from 2015-Jan-5th to the most recent data points. The relative strength factor is then 2018-Dec-28th is employed for sub-period 2. converted to a relative strength index between 0 and 100: Experimental Design (8) RSI () t =− 100 . 1+ RS () t In practice, proponents of Japanese candlestick patterns also refer to the trend condition or overbought/oversold condi- In general, the greater/smaller the RSI value, the stronger tion. For example, if a Long White appears in an uptrend, and more protracted the bullish/bearish trend, whereas a more and the stock is not overbought simultaneously, it suggests a important usage is that it implies the trading target is at an over- stronger signal for making a long transaction. Similarly, if a bought/oversold level. Thus, it would be extremely risky for Bearish Gap pattern is formed in a downtrend, and the stock traders to make a long/short-selling transaction. In the experi- is not in an oversold condition simultaneously, it indicates an ments, the parameter n of RSI is set to be 12 due to its popular- agreement with the short signal generated from the bearish ity among traders in financial markets. Additionally, stocks are candlestick pattern. Taking into account both trend condition considered to be overbought when the RSI value is above 70 or and overbought/oversold condition might be beneficial for oversold when the indicator value is below 30. market traders to make a transaction in a relatively low-risk condition. In contrast, if the traders consider both of those Data and Experimental Design two conditions to be satisfied when applying the Japanese candlestick patterns, the number of tradable signals are Data fewer, and traders may miss several great chances of trading. The daily-based data on the 50 component stocks (see Therefore, it is necessary to examine whether or not the prof- Appendix B) of the SSE50 index were employed to conduct itability of candlestick patterns could be improved by taking 6 SAGE Open Table 2. Long and Short-Selling Trading Signal and Conditions for the Four Groups. Long signal Short-selling signal Group Bullish candlestick In an uptrend Not overbought Bearish candlestick In a downtrend Not oversold A 〇 × or 〇 × or 〇 〇 × or 〇 × or 〇 B 〇 〇 × or 〇 〇 〇 × or 〇 C 〇 × or 〇 〇 〇 × or 〇 〇 D 〇 〇 〇 〇 〇 〇 Note. “〇” means satisfied. “×” means not satisfied. into account the trend or overbought/oversold condition 5-day (1 week), and 10-day (about half a month) periods. since, indeed, many market investors tend to adopt additional Similarly, for short-selling transactions, the k-day return R is indicators when using candlestick patterns to achieve supe- expressed as: rior performance. Therefore, this research investigates the Pt () +− 1 Pt () + k sb profitability of Japanese candlestick patterns with consider- R = ×100%. (10) Pt () +1 ing trend condition and overbought/oversold condition. s Generally, Japanese candlestick patterns are used to cap- where P (t + k) is the close price for buying on day t + k; ture short-term price movements, and the most useful time P (t + 1) denotes the open price for short-selling on day t + 1; horizon is shorter than 10 days (Nison, 1991). Hence, in the k represents the length of days for the holding period, which experiments, we investigated the Japanese candlestick pat- is also comprised of 1-, 3-, 5-, and 10-day periods. terns with four different holding period lengths, which are 1-, 3-, 5-, and 10-day, for the profitability test. Note that the holding period is fixed when the period is decided, in other Experimental Results words, the opened position will be held until the expiration of the fixed period regardless of the daily candlestick recom- Summary Statistics mendation during the position holding period. However, the Stock price data used in the experiments were derived from profitability test considers all the subsequent candles during the Choice database (Database 1). Each candlestick line is the period after the pattern’s occurrence. constructed from four prices, which are open, high, low, and In this study, we design four groups of trading rules based close prices. The summary statistics on the selected 47 com- on candlestick pattern to identify their predictive power. The ponent stocks (the period before 1 January 2015) are dis- trading conditions of each trading group are shown in Table played in Table 3, with 102,184 observations of daily return. 2. Note that for the profitability test of each candlestick pat- Note that the natural logarithm of these daily returns was tern based trading rule, there will be no trade at the time point utilized for analysis. The distribution characteristics were where the conditions are not satisfied. examined using five statistics: mean, standard deviation, Regarding the profits generated by the candlesticks, the skewness, kurtosis, and JB statistics. All four series were returns are calculated using the difference between the sell- found to display positive mean and skewness, and the aver- ing and buying prices, expressed as a percentage of the buy- age returns of the time series have a near-zero mean. In addi- ing prices. It is assumed that the candlestick pattern is tion, it should be noted that skewness and kurtosis were formulated on day t, then the n-day return R is defined as: significantly different from 0 and 3, respectively, at the 1% significance level. Pt () +− kP () t + 1 sb (9) R = ×100% . Pt () + 1 Candlestick Pattern Returns Our analysis begins with a profitability examination of 10 where P (t + k) refers to the close price for selling on day candlestick patterns to examine their predictive power. The t + k, and P (t + 1) denotes the open price for buying on day profitability test of Japanese candlestick patterns was con- t + 1. Note that for return calculation, we assume that market ducted using a suitable test named the skewness adjusted traders cannot make the transaction on the same day that the two-tailed t-test (Mitra, 2011). On this test, the null hypoth- signal is generated since it is very difficult to determine the esis assumes that the average returns for the candlestick pat- trading signal until the close price is known. Therefore, a terns do not differ significantly from zero: more practical case is to enter at the open price of the trading day after the entry signal is generated. In addition, k repre- 1 1 sents the length of days for the holding period, which is com- 2 tn =+ () SS γγ ⋅+ , (11) sa prised of 1- and 3-day (about half a week for business days), 3 6n Deng et al. 7 Table 3. Descriptive Statistics of the Daily Returns. Open High Low Close −4 −4 −4 −4 M 5.32 × 10 5.33 × 10 5.35 × 10 5.37 × 10 SD 0.0272 0.0241 0.0241 0.0254 Skewness 0.86** 1.82** 1.21** 1.19** Kurtosis 29.24** 78.00** 48.53** 49.01** 6 7 7 7 JB statistics 3.65 × 10 ** 2.59 × 10 ** 1.01 × 10 ** 1.03 × 10 ** Note. There are 102,184 observations of daily return for the open, high, low, and close prices. The daily returns are measured as the natural logarithm difference of the open, high, low, and close prices. **Indicates statistical significance at the 1% level. with considering the uptrend condition. For Bullish Gap, it AR (12) S , was observed that its average return decreased from 2.7% to σ() AR 2.3% over a 5-day holding period, and it decreased from 3.8% to 2.6% over a 10-day holding period, thus indicating Bullish Gap is more appropriate to be applied without con- () AR − AR ∑ i sidering the uptrend condition, while Long White and Bullish (13) i=1 γ = . Harami patterns are suggested to be applied in the uptrend nA ⋅σ() R condition. Further, it also shows that for the profitable can- dlestick patterns Long White and Bullish Gap, when the where is the average return of the transactions, σ AR holding period increased, their average returns also improved, AR () is the sample standard deviation of the returns, and n denotes thereby indicating that it would be better to hold the trading the number of candlestick based trading signals. The returns position for a relatively long period. While for Bullish prior to transaction costs being taken into consideration for Harami with a 5-day holding period, it produced the best the 1-, 3-, 5-, and 10-day holding periods are presented in return, demonstrating that it is suitable to apply it with about Tables 4 and 5, which are the results for the five bullish and one week holding period. five bearish candlestick patterns, respectively. Next, we applied these bullish candlestick patterns based Returns of five bullish candlestick patterns without con- signals with considering an overbought condition rather than sidering the uptrend and overbought conditions are presented in an up-trend condition. As reported in panel C of Table 4, in panel A of Table 4, from which we find that at the 5% after considering the overbought condition, all Long White level, Long White yielded positive average returns over all (1-, 3-, 5-, and 10-day holding periods) remained to be profit- four holding periods, and Bullish Harami yielded positive able. While compared with the “Unconditional” trading sig- average returns over the 1-, 3-, and 5-day holding periods. nal group, all values of the average return decreased. In Bullish Gap produced positive average returns over the 5- addition, Bullish Harami (3- and 5-day holding periods) and 10-day holding periods. Following the approach adopted became not profitable after considering the overbought con- in Caginalp and Laurent (1998), we then considered total dition. Dragonfly Doji with the 5-day holding period became transaction costs, including execution, liquidity, and slippage profitable at the 5% level, while after taking into account the cost. Based upon our arbitrary assumption of a transaction transaction cost of 1%, it was unable to generate positive cost of 1% per round trip, Bullish Gap (5- and 10-day peri- average returns. The candlestick pattern Bullish Gap (10-day ods) and Long White (10-day period) were found to yield holding period) produced a positive average return signifi- positive returns and remain profitable even after deducting cantly, and the average return value was larger than the trans- the transaction cost. action cost 1%. However, compared with the “unconditional” Panel B of Table 4 shows the profitability test results for condition, the average return of Bullish Gap decreased from applying the five bullish candlestick patterns in an uptrend 3.8% to 3.6%, thus indicating that it was more appropriate to condition. Compared with the results shown in panel A, in apply the Bullish Gap candlestick pattern without consider- most cases for the candlestick patterns obtained positive ing the overbought condition. average returns, after applying them in an uptrend, they were Then, we applied those five bullish candlestick patterns found to produce superior returns. In addition, after consider- considering both the trend condition and overbought condi- ing the uptrend condition, Bullish Harami (10-day holding tion. From the results reported in panel D, it can be seen that period) yielded a positive average return at the 5% level, and the candlestick patterns yielded average returns over 1% (at Long White (5-day holding period) became profitable even the 5% level) were Bullish Gap (3-, 5-, and 10-day holding after deducting the 1% transaction cost. These results reveal periods) and Long White (10-day holding period), thus dem- that the returns of those candlestick patterns were improved onstrating that buying signals generated by those candlestick 8 SAGE Open Table 4. Profitability Test of the Five Bullish Candlestick Patterns Over the Period 2000 to 2015. Holding period length and profitability test result Japanese 1-day 3-day 5-day 10-day candlestick pattern N Return p-Value N Return p-Value N Return p-Value N Return p-Value Panel A (unconditional) Long White 27,218 0.002** <.01 27,218 0.004** <.01 27,218 0.006** <.01 27,218 0.012** <.01 Dragonfly Doji 588 0.001** <.01 588 0.011 .29 588 0.015 .21 588 0.021 .14 Bullish Hammer 1,976 −0.001 .01 1,976 −0.002 .06 1,976 −0.002 .06 1,976 −0.001 .79 Bullish Harami 2,332 0.003** <.01 2,332 0.002* .02 2,332 0.003* .03 2,332 0.004 .06 Bullish Gap 1,802 0.010 .92 1,802 0.020 .12 1,802 0.027* .03 1,802 0.038** <.01 Panel B (uptrend) Long White 12,940 0.002** <.01 12,940 0.006** <.01 12,940 0.011** <.01 12,940 0.018** <.01 Dragonfly Doji 237 0.002 .08 237 0.023 .41 237 0.031 .40 237 0.049 .16 Bullish Hammer 902 −0.001 .18 902 0.001 .84 902 0.002 .35 902 0.004 .18 Bullish Harami 999 0.004** <.01 999 0.004** <.01 999 0.006** <.01 999 0.005* .05 Bullish Gap 855 0.011 .86 855 0.019 .18 855 0.023** <.01 855 0.026** <.01 Panel C (not overbought) Long White 22,217 0.001** <.01 22,217 0.003** <.01 22,217 0.005** <.01 22,217 0.009** <.01 Dragonfly Doji 485 0.001* .02 485 0.004 .10 485 0.006* .02 485 −0.001 .91 Bullish Hammer 1,719 −0.002 <.01 1,719 −0.002 <.01 1,719 −0.003 .02 1,719 −0.002 .26 Bullish Harami 2,113 0.003** <.01 2,113 0.002 .06 2,113 0.002 .06 2,113 0.003 .11 Bullish Gap 1,258 0.008 .84 1,258 0.018 .16 1,258 0.027 .06 1,258 0.036* .02 Panel D (uptrend + not overbought) Long White 8,937 0.002** <.01 8,937 0.006** <.01 8,937 0.009** <.01 8,937 0.015** <.01 Dragonfly Doji 173 0.001 .07 173 0.003 .35 173 0.008 .10 173 0.009 .19 Bullish Hammer 761 −0.001 .13 761 −0.001 .70 761 0.001 .47 761 0.003 .24 Bullish Harami 942 0.004** <.01 942 0.005** <.01 942 0.006** <.01 942 0.006* .03 Bullish Gap 391 0.006** <.01 391 0.011** <.01 391 0.014** <.01 391 0.018** <.01 Note. “N” is the number of trading signals. “Return” refers to the average return from trading signals with its associated p-value. *and ** indicate generating positive average returns at the 5% and 1% level, respectively. patterns would on average produce positive returns. In addi- Dragonfly Doji, it yielded the best return of 0.6% when con- tion, we also find that by considering both the uptrend trend sidering the “not overbought” condition. However, none of and overbought conditions, Bullish Gap obtained a signifi- these candlestick patterns was able to generate an average cant average return greater than 1% over a 3-day holding return greater than the total transaction cost assumed per period. However, it produced a smaller average return over round turn in this study. Therefore, it seems that these pat- the 5- and 10-day holding periods than in the “unconditional” terns cannot create value for trading on the component stocks condition. Furthermore, we observe that all profitable can- of the SSE50 index. dlestick patterns yielded superior returns than in the “uncon- Similar to the bullish candlestick patterns, we further ditional” condition except Bullish Gap. examine the profitability of the five bearish candlestick pat- Finally, we focus on the best returns for each profitable terns, and their results are reported in Table 5. The average candlestick pattern that yielded an average return greater return results without considering the downtrend or oversold than 1%. For Bullish Gap, the best trading signal was the condition are presented in panel A, from which we observe “unconditional” condition over the 10-day holding period, in that under the “unconditional” condition over the 3-, 5-, and which it yielded an average return of about 3.8% before 10-day holding periods, none of their average returns was deducting the transaction cost. For Long White, it performed positive. Only Inverted Hammer, Bearish Harami, and better in an uptrend condition with a 10-day holding period, Bearish Gap with the 1-day holding period produced positive which yielded an average return of about 1.88% before con- returns. We then take into account the total transaction cost sidering the transaction cost. These two trading signals were of 1% per round trip. Obviously, none of those candlestick profitable on the transaction of the SSE50 component stocks. patterns was found to create values for investors. This indi- For Bullish Harami, it obtained the best average return of cates that by using only those bearish candlestick patterns, 0.6% (5- and 10-day holding periods) when considering both none of them showed predictive power. Additionally, we find the uptrend condition and overbought condition. For that over a 10-day holding period, for Gravestone Doji and Deng et al. 9 Table 5. Profitability Test Results of the Five Bearish Candlestick Patterns Over the Period 2000 to 2015. Holding period length and profitability test result 1-day 3-day 5-day 10-day Japanese candlestick pattern N Return p-Value N Return p-Value N Return p-Value N Return p-Value Panel A (unconditional) Long Black 26,187 −0.001 <.01 26,187 −0.003 <.01 26,187 −0.005 <.01 26,187 −0.008 <.01 Gravestone Doji 569 −0.008 .46 569 −0.005 .04 569 −0.006 .06 569 −0.011 .02 Inverted Hammer 1,982 0.001* .01 1,982 −0.006 <.01 1,982 −0.008 <.01 1,982 −0.015 <.01 Bearish Harami 3,067 0.002** <.01 3,067 −0.001 .40 3,067 −0.001 .36 3,067 −0.006 <.01 Bearish Gap 1,937 0.001 .16 1,937 −0.001 .86 1,937 −0.001 .75 1,937 −0.003 .30 Panel B (uptrend) Long Black 11,948 0.000 .93 11,948 −0.002 <.01 11,948 −0.002 <.01 11,948 −0.005 <.01 Gravestone Doji 235 −0.006 <.01 235 −0.007 .15 235 −0.009 .17 235 −0.013 .16 Inverted Hammer 855 0.001 .23 855 −0.004 <.01 855 −0.002 .16 855 −0.007 <.01 Bearish Harami 1,355 0.002** <.01 1,355 0.001 .66 1,355 0.002 .26 1,355 0.001 .73 Bearish Gap 1,116 0.001 .48 1,116 −0.001 .61 1,116 −0.002 .32 1,116 −0.007 .09 Panel C (not overbought) Long Black 22,292 −0.001 <.01 22,292 −0.002 <.01 22,292 −0.004 <.01 22,292 −0.006 <.01 Gravestone Doji 483 −0.005 <.01 483 −0.004 .16 483 −0.004 .19 483 −0.010 .07 Inverted Hammer 1,701 0.002** <.01 1,701 −0.005 <.01 1,701 −0.006 <.01 1,701 −0.011 <.01 Bearish Harami 2,707 0.002** <.01 2,707 0.001 .71 2,707 0.001 .66 2,707 −0.003 .04 Bearish Gap 1,426 −0.001 .25 1,426 0.001 .48 1,426 0.002 .39 1,426 −0.001 .96 Panel D (uptrend + not overbought) Long Black 8,938 0.000 .25 8,938 −0.001 .16 8,938 −0.001 .23 8,938 −0.002 <.01 Gravestone Doji 200 −0.006 <.01 200 −0.008 .17 200 −0.011 .17 200 −0.013 .20 Inverted Hammer 776 0.001 .16 776 −0.003 .02 776 −0.002 .33 776 −0.006 .03 Bearish Harami 1,280 0.002** <.01 1,280 0.001 .27 1,280 0.003* .03 1,280 0.002 .36 Bearish Gap 657 −0.002 .03 657 0.001 .59 657 0.001 .75 657 −0.003 .37 Note. “N” is the number of short-selling signals. “Return” refers to the average return from trading signals with its associated p-value. *and ** indicate generating positive average returns at the 5% and 1% level, respectively. Inverted Hammer which are generally regarded as the bear- return of more than 1%, thus indicating that even considering ish candlestick patterns, they suffered losses of more than the “not oversold” condition when applying those bearish −1% at the 5% level, thereby indicating that although they candlestick patterns. Still, none of those candlestick patterns are generally regarded as the bearish patterns to suggest offered predictive power, which is a finding similar to that of short-selling signals, they could be used as contrary signals the research conducted by Tharavanij et al. (2017). They to generate bullish signals for executing long transactions. found that even with filtering by stochastic indicators, can- Panel B of Table 5 reports the results for applying the dlestick patterns are still not useful. However, it is found that Japanese candlestick patterns in a downtrend condition. Inverted Hammer over a 10-day holding period suffered a Similar to the results of applying bearish candlesticks shown loss of more than −1% at the 5% level, therefore it reveals in panel A, none of these patterns was found to yield a sig- that although this candlestick pattern is generally regarded as nificantly positive average return of more than 1%, thus indi- a bearish signal, it could be used as a contrary signal for mak- cating that although these candlestick patterns were applied ing a long transaction. We find it could be used to generate a with considering a downtrend condition, all of them remained buying signal to make profits even after deducting the trans- no significant profitability. In addition, none of those candle- action cost in the in-sample period. stick patterns suffered an average loss of more than −1% at Furthermore, from the results presented in panel D of the 5% level, thereby demonstrating that there was no can- Table 5, we find that by considering both the “downtrend” dlestick pattern that could be applied as a contrary indicator. condition and “not oversold” condition, Bearish Harami pat- Next, panel C of Table 5 shows the profitability test results tern over a 5-day holding period produced a significantly of candlestick patterns considering the “not oversold” condi- positive average return. However, after deducting the 1% tion instead of applying them in a downtrend condition. Over transaction cost, it became not profitable. It demonstrates the 1-, 3-, 5-, and 10-day holding periods, none of these pat- that these bearish candlestick patterns could not make profits terns was found to produce a significantly positive average even though both the “not oversold” and “downtrend” 10 SAGE Open Figure 2. Average return results of the five bullish candlestick patterns over the period 2000 to 2015. Note. Each horizontal axis represents the length of holding days, and each vertical axis represents the value of average return. conditions were satisfied. Additionally, from the results yielded a significantly positive average return. However, we reported in panel D, none of these candlestick patterns suf- find evidence that Gravestone Doji and Inverted Hammer fered a loss of more than −1% at the 5% level. Therefore, over a 10-day holding period and under the “unconditional” none of these patterns could be used as a contrary signal for condition, as well as the Inverted Hammer over a 10-day market investors. holding period and under the “not oversold” condition, could Additionally, Figures 2 and 3 show how the length of the create values for investors with regarding those bearish can- hold period impacts the profitability of Japanese candlestick dlestick patterns as buying signals. patterns. In Figure 2, we observe that with the increase of hold- ing days, the return rises in most of the bullish candlestick pat- Bootstrap Analysis terns. Several signals based on Hammer, Dragonfly Doji, and To examine the robustness of profitability test results, we Harami are different from others, in which the largest or small- then employed the bootstrap method (Marshall et al., 2006) est return was produced by the middle-length holding period, to address the common features of stock return data, such as such as Bullish Harami in panel A and Dragonfly Doji in panel autocorrelation, skewness, and leptokurtosis. Subsequently, C. On the contrary, the return falls with the increase of holding a new dataset was generated while the characteristics of the days in most bearish candlestick patterns. For several candle- original dataset for the SSE50 index component stocks were stick patterns, the average return fluctuates with the increase of retained. We simulated 500 sets of open, high, low, and close holding days, such as Bearish Harami in panels B and D, and price series, comparing the profits accrued from the actual Inverted Hammer in panels B and D. Additionally, in some data and the random series to identify whether the profitabil- cases, the largest return is produced with the middle-length ity of those candlestick pattern is statistically significant. holding period, such as Gravestone Doji in panels A and C. The bootstrap p-value is the percentage of simulated average In summary, for those bearish candlestick patterns, aver- returns that are greater than the actual average returns. Note that age returns of all trading signals were insignificant with the we only report returns of the candlestick patterns yielded sig- exception of Bearish Harami with conditions in the “down- nificant average returns that over 1%. Among the candlestick trend” and “not oversold” over a 5-day holding period, which Deng et al. 11 Figure 3. Average return results of the five bearish candlestick patterns over the period 2000 to 2015. Note. Each horizontal axis represents the length of holding days, and each vertical axis represents the value of average return. patterns, 12 were found to generate significant average returns out-of-sample performance, a two-step procedure which is simi- greater than 1% on the component stocks of the SSE50 index. lar to the method proposed by Jensen (1967) is conducted, first The “Return” column in Table 6 denotes the average selecting the best-performing candlestick patterns and then using return rates generated by each candlestick pattern on the these on new (i.e., out-of-sample) data. Based on it, we conduct bootstrapped and original series. As shown in panels A, B, C, an out-of-sample test on the data by selecting the profitable can- and D, when transaction costs were taken into consideration, dlestick patterns and contrary candlestick patterns (with consider- those candlestick patterns were not found to be profitable on ing the transaction cost) in the in-sample period, and to investigate the random bootstrap series. In all cases, the average returns whether or not those successful candlestick patterns could remain on the original series were found to be greater than those on their predictive power in the following out-of-sample period. the bootstrap series, particularly for Bullish Gap under the The return results of the profitable and contrary candlestick pat- “unconditional” condition over the 5- and 10-day holding terns in the sub-period 1 and sub-period 2 are reported in Table 7, periods, Bullish Gap under a trend condition over the 5- and in which we find that both datasets appeared to be consistent in 10-day holding periods, as well as Bullish Gap considering most cases, thereby providing support for the conclusion that most the overbought condition over the 10-day holding period. of the profitable candlestick patterns have significant explanatory Our bootstrap results thus provide valuable evidence on the power for the component stocks of the SSE50 index. In addition, profitability of candlestick patterns. the bearish candlestick pattern Gravestone Doji (“unconditional” condition with the 10-day holding period) generated a negative average return lower than −1%, showing significant evidence that Robustness Test it could be used as a contrary indicator to generate a profitable buy- Despite the evidence provided on the profitability of technical ing signal. Inverted Hammer under the “unconditional” condition analysis, Park and Irwin (2007) pointed out the importance of with the 10-day holding period produced a significantly negative addressing the out-of-sample problem. Therefore, to examine the average return, whereas the profit by using it as a contrary 12 SAGE Open Table 6. The Bootstrapped Simulation Results for the Profitable Candlestick Patterns and Holding Periods. Original series Bootstrap series Candlestick pattern Number Return Number Return p-Value count Panel A (unconditional) Long White (10-day) 27,218 0.012 31,793 0.007 1 Bullish Gap (5-day) 1,802 0.027 1,582 0.003 1 Bullish Gap (10-day) 1,802 0.038 1,582 0.007 1 Panel B (trend condition) Long White (5-day) 12,940 0.010 14,944 0.004 1 Long White (10-day) 12,940 0.019 14,944 0.008 0 Bullish Gap (5-day) 855 0.022 753 0.004 1 Bullish Gap (10-day) 855 0.027 753 0.008 1 Panel C (overbought/oversold condition) Bullish Gap (10-day) 1,258 0.036 1,231 0.007 0 Panel D (trend + overbought/oversold condition) Long White (10-day) 8,937 0.015 11,945 0.008 1 Bullish Gap (3-day) 391 0.011 461 0.002 0 Bullish Gap (5-day) 391 0.014 461 0.004 0 Bullish Gap (10-day) 391 0.018 461 0.008 1 Note. The “Number” columns refer to the number of patterns in the bootstrapped and original series. “p-value count” columns refer to the number of stocks for which the average return is larger on the original series than the 500 bootstrapped series; “Return” columns are the average return rates generated by each candlestick pattern on the bootstrapped series and original series. Table 7. The In-Sample (2000-Jan-4th to 2014-Dec-31st) and Out-of-Sample (2015-Jan-5th to 2018-Dec-28th) Profitability Results of the Best-Performing Japanese Candlestick Patterns. Sub-period 1 (In-sample) Sub-period 2 (Out-of-sample) Candlestick pattern N Return p-Value N Return p-Value Panel A (unconditional) Long White (10-day) 27,218 0.012** <.01 12,366 0.006** <.01 Bullish Gap (5-day) 1,802 0.027* .03 907 0.028** <.01 Bullish Gap (10-day) 1,802 0.038** <.01 907 0.050** <.01 Gravestone Doji (10-day) 569 −0.011* .02 247 −0.022* .01 Inverted Hammer (10-day) 1,982 −0.015** <.01 849 −0.006* .02 Panel B (trend condition) Long White (5-day) 12,940 0.010** <.01 5,952 0.003** <.01 Long White (10-day) 12,940 0.018** <.01 5,952 0.008** <.01 Bullish Gap (5-day) 855 0.022* .01 372 0.027* .01 Bullish Gap (10-day) 855 0.027** <.01 372 0.050* .01 Panel C (overbought/oversold condition) Bullish Gap (10-day) 1,258 0.036* .02 670 0.037* .01 Inverted Hammer (10-day) 1,701 −0.011** <.01 721 −0.003* .16 Panel D (trend + overbought/oversold condition) Long White (10-day) 8,937 0.015** <.01 4,089 0.006** <.01 Bullish Gap (3-day) 391 0.011** <.01 179 0.002 .53 Bullish Gap (5-day) 391 0.014** <.01 179 0.002 .69 Bullish Gap (10-day) 391 0.018** <.01 179 0.012 .10 Note. “N” is the number of trading signal for each candlestick pattern. “Return” refers to the average return from trading signals with its associated p-value. *and ** indicate generating positive average returns at the 5% and 1% level, respectively. indicator was still unable to cover the transaction cost 1%. For not-overbought condition, we can observe that none of the trading Bullish Gap with considering the trend condition and signals created a significant average return over 1%. Deng et al. 13 Table 8. Trading Simulation Results of the Five Bullish Candlestick Pattern Based Trading Strategies Over the period 2000 to 2018 Using the Component Stocks of the SSE50 Index. Holding period length and profitability test results Candlestick 1-day 3-day 5-day 10-day pattern based trading strategy SR AR SD SR AR SD SR AR SD SR AR SD Panel A (unconditional) Long White −0.899 −0.393 0.47 −0.073 −0.005 0.43 0.126 0.080 0.43 0.326 0.169 0.44 Dragonfly Doji −0.537 −0.348 0.70 −0.180 −0.055 0.45 0.034 0.041 0.45 0.231 0.138 0.48 Hammer −1.459 −0.329 0.24 −0.539 −0.090 0.21 −0.396 −0.041 0.17 −0.035 0.020 0.18 Bullish Harami −0.555 −0.086 0.20 −0.154 0.000 0.17 0.118 0.044 0.15 0.183 0.055 0.16 Bullish Gap 0.594 0.465 0.74 0.611 0.429 0.66 0.544 0.356 0.61 0.482 0.340 0.65 Panel B (uptrend) Long White −0.623 −0.228 0.41 0.168 0.092 0.39 0.313 0.154 0.41 0.409 0.206 0.44 Dragonfly Doji −0.373 −0.175 0.54 −0.040 0.010 0.41 0.140 0.078 0.37 0.256 0.127 0.40 Hammer −1.470 −0.206 0.16 −0.440 −0.039 0.15 −0.188 0.003 0.12 −0.017 0.024 0.10 Bullish Harami −0.711 −0.099 0.18 −0.348 −0.025 0.15 −0.132 0.009 0.13 −0.040 0.022 0.11 Bullish Gap 0.522 0.363 0.65 0.505 0.282 0.51 0.472 0.231 0.43 0.376 0.169 0.38 Panel C (not overbought) Long White −1.002 −0.436 0.46 −0.171 −0.047 0.43 0.032 0.040 0.44 0.218 0.121 0.44 Dragonfly Doji −0.907 −0.415 0.49 −0.322 −0.099 0.39 −0.022 0.017 0.42 0.139 0.078 0.37 Hammer −1.358 −0.307 0.25 −0.593 −0.092 0.20 −0.455 −0.043 0.15 −0.177 0.002 0.13 Bullish Harami −0.408 −0.065 0.22 −0.197 −0.010 0.18 0.062 0.037 0.17 0.154 0.050 0.16 Bullish Gap 0.445 0.159 0.30 0.489 0.185 0.33 0.385 0.158 0.34 0.400 0.183 0.39 Panel D (uptrend + not overbought) Long White −0.722 −0.256 0.39 0.057 0.048 0.39 0.178 0.096 0.39 0.257 0.126 0.39 Dragonfly Doji −0.760 −0.242 0.35 −0.170 −0.028 0.32 0.075 0.050 0.32 0.154 0.070 0.28 Hammer −1.318 −0.167 0.15 −0.452 −0.029 0.12 −0.227 0.003 0.10 −0.096 0.018 0.08 Bullish Harami −0.574 −0.079 0.18 −0.337 −0.022 0.14 −0.113 0.012 0.13 −0.038 0.022 0.11 Bullish Gap 0.244 0.056 0.12 0.286 0.058 0.11 0.218 0.048 0.10 0.109 0.034 0.07 Note. “SR” refers to the Sharpe ratio of each candlestick pattern based trading strategy. “AR” refers to the average annual return. “SD” refers to the standard deviation of the annual returns. the simulation trading return of each year is calculated for Simulation Trading Results each candlestick pattern based trading strategy. Using the Except for the profitability test of candlestick patterns, we annual returns, it is possible to calculate the Sharpe ratio of also designed the candlestick pattern based trading strategies each trading strategy, and the 1-year Yield of China’s trea- in the experiments, and simulation trading was adopted to sury bond is adopted as the risk-free rate. Additionally, the evaluate their trading performance. We introduced trading buy-and-hold strategy and sell-and-hold strategy, which are position management and transaction cost in our model. By two famous benchmarks, are used for comparing with the designing the candlestick pattern based trading strategy, it is bullish and bearish candlestick based trading strategies. possible to calculate the profit per year of each candlestick The trading simulation results on the component stocks of pattern, and its trading performance can be evaluated and the SSE50 index are shown in Tables 8 and 9. Compared with compared with other benchmarks. In the experiments, we the previous results, all the trading strategies of Bullish Gap adopted a “separation strategy” in position design, which and Long White with 5- or 10-day holding periods could earn means that we distribute the full position for each holding positive profits for investors. Since the Sharp ratio for sell-and- days averagely, and the daily position for each stock that hold was negative, the trading results were compared with the appears the trading signal is also distributed averagely. By buy-and-hold strategy. Considering the buy-and-hold strategy’s using this portfolio management, we can ensure that every Sharpe ratio was 0.291, for Bullish Gap, most trading strategies satisfied stock could be traded by the designed candlestick based on it performed better than the benchmark except panel pattern based trading strategy. For instance, if we adopt the D. Possible reason could be that the frequency of transaction 3-day holding period and there is a bullish candlestick signal was too low for the condition of panel D, thus it might miss the appears in three stocks at the same date D, we will distribute trading chances to produce considerable profits. For Long 1/9 long position to each stock at date D + 1. Subsequently, White based trading strategy, only the 10-day holding length in 14 SAGE Open Table 9. Trading Simulation Results of the Five Bearish Candlestick Pattern Based Trading Strategies Over the Period 2000 to 2018 Using the Component Stocks of the SSE50 Index. Holding period length and profitability test results Candlestick pattern 1-day 3-day 5-day 10-day based trading strategy SR AR SD SR AR SD SR AR SD SR AR SD Panel A (unconditional) Long Black −1.300 −0.585 0.47 −1.003 −0.386 0.41 −0.844 −0.353 0.45 −0.646 −0.251 0.43 Gravestone Doji −0.539 −0.275 0.56 −0.575 −0.102 0.22 −0.694 −0.071 0.14 −0.436 −0.078 0.24 Inverted Hammer −0.795 −0.162 0.24 −0.468 −0.074 0.21 −0.492 −0.058 0.17 −0.518 −0.069 0.18 Bearish Harami −0.485 −0.150 0.36 −0.581 −0.139 0.28 −0.526 −0.104 0.25 −0.558 −0.115 0.25 Bearish Gap −0.237 −0.036 0.26 0.024 0.031 0.20 −0.261 −0.015 0.16 −0.397 −0.027 0.13 Panel B (uptrend) Long Black −1.039 −0.315 0.33 −0.785 −0.202 0.29 −0.825 −0.176 0.24 −0.691 −0.133 0.23 Gravestone Doji −1.110 −0.066 0.08 −0.694 −0.011 0.05 −0.711 −0.010 0.05 −0.950 −0.004 0.03 Inverted Hammer −0.886 −0.092 0.13 −0.399 −0.016 0.11 −0.411 −0.009 0.09 −0.399 −0.003 0.07 Bearish Harami −0.559 −0.081 0.19 −0.492 −0.059 0.17 −0.487 −0.035 0.13 −0.590 −0.043 0.12 Bearish Gap −0.028 0.022 0.14 0.022 0.030 0.15 −0.143 0.011 0.11 −0.367 −0.004 0.08 Panel C (not overbought) Long Black −1.399 −0.568 0.42 −1.019 −0.353 0.37 −0.855 −0.297 0.38 −0.585 −0.206 0.40 Gravestone Doji −1.083 −0.139 0.15 −0.698 −0.049 0.11 −1.041 −0.040 0.06 −0.981 −0.017 0.04 Inverted Hammer −0.666 −0.137 0.25 −0.420 −0.056 0.20 −0.467 −0.046 0.15 −0.540 −0.047 0.14 Bearish Harami −0.498 −0.146 0.35 −0.508 −0.115 0.28 −0.470 −0.079 0.22 −0.560 −0.088 0.20 Bearish Gap −0.424 −0.078 0.25 −0.054 0.014 0.22 −0.242 −0.013 0.16 −0.385 −0.018 0.12 Panel D (uptrend + not overbought) Long Black −1.088 −0.274 0.28 −0.617 −0.162 0.31 −0.560 −0.128 0.27 −0.497 −0.102 0.26 Gravestone Doji −1.166 −0.061 0.07 −0.639 −0.001 0.04 −0.703 −0.007 0.05 −0.991 −0.005 0.03 Inverted Hammer −0.734 −0.064 0.12 −0.250 −0.003 0.11 −0.326 −0.002 0.09 −0.395 −0.003 0.07 Bearish Harami −0.444 −0.060 0.19 −0.427 −0.050 0.18 −0.420 −0.027 0.13 −0.599 −0.044 0.12 Bearish Gap −0.211 −0.006 0.15 −0.019 0.023 0.15 −0.113 0.014 0.11 −0.399 −0.003 0.07 Note. “SR” refers to the Sharpe ratio of each candlestick pattern based trading strategy. “AR” refers to the average annual return. “SD” refers to the standard deviation of the annual returns. panels A and B, as well as the 5-day holding length in panel B, the benchmark. Among the bearish candlestick patterns, performed better than the buy-and-hold. Some trading strate- Bearish Gap based trading strategies performed better than the gies produced positive annual returns, but they did not perform buy-and-hold strategy in terms of Sharpe ratio. better than the buy-and-hold in terms of annual average return or Sharpe ratio. Additionally, for the bearish candlestick pattern Conclusion based trading strategies, none of them could perform better than the buy-and-hold in terms of both average annual return Traditional research has done little to examine the effectiveness and Sharpe ratio. of Japanese candlestick patterns to the component stocks of the Our further research used all the listed company data in the Chinese SSE 50 index when taking into consideration the over- Chinese stock market. The results are shown in Tables 10 and bought/oversold condition. In this study, we aimed to fill this 11, from which it is found that all Bullish Gap based trading gap, and comprehensive analysis has been conducted on the strategies generated positive returns, and most of them per- component stocks of the SSE 50 index to investigate the pre- formed better than buy-and-hold, which produced an average dictive power of 10 well-known Japanese candlestick patterns. return of 11.22% and a Sharpe ratio value of 0.183. Unlike the For the experiments, daily-based data on the component stocks previous results, Long White based trading strategy failed to of the SSE50 index that covers the period from 2000 to 2018 perform better than the benchmark in terms of Sharpe ratio. was employed. In addition, to avoid data mining, we divided Similar results as the trading simulation test on the SSE50 data- the whole dataset period into two sub-periods, in which the out- sets, several Bullish Harami based trading strategies could of-sample period was further examined to confirm the robust- yield positive returns, while they could not perform better than ness of the results that whether or not the successful candlestick Deng et al. 15 Table 10. Trading Simulation Results of the Five Bullish Candlestick Patterns Based Trading Strategies Over the Period 2000 to 2018 Using All Stocks Data. Holding period length and profitability test results Candlestick 1-day 3-day 5-day 10-day pattern based trading strategy SR AR SD SR AR SD SR AR SD SR AR SD Panel A (unconditional) Long White −0.897 −0.519 0.61 −0.411 −0.198 0.55 −0.271 −0.118 0.53 0.006 0.029 0.54 Dragonfly Doji −0.878 −0.515 0.62 −0.290 −0.141 0.58 −0.131 −0.045 0.55 0.058 0.057 0.54 Hammer −0.749 −0.521 0.73 −0.437 −0.228 0.58 −0.223 −0.105 0.59 −0.051 −0.002 0.55 Bullish Harami −0.769 −0.482 0.66 −0.339 −0.140 0.49 −0.050 0.002 0.47 −0.013 0.020 0.45 Bullish Gap 0.756 2.343 3.06 0.689 1.960 2.81 0.604 1.583 2.58 0.587 1.243 2.07 Panel B (uptrend) Long White −1.048 −0.662 0.66 −0.491 −0.268 0.60 −0.350 −0.181 0.59 −0.119 −0.043 0.58 Dragonfly Doji −1.095 −0.653 0.62 −0.425 −0.226 0.59 −0.233 −0.105 0.56 −0.067 −0.011 0.56 Hammer −0.689 −0.487 0.74 −0.407 −0.220 0.61 −0.243 −0.117 0.59 −0.103 −0.029 0.53 Bullish Harami −0.774 −0.479 0.65 −0.234 −0.108 0.57 −0.076 −0.009 0.46 −0.068 −0.004 0.44 Bullish Gap 0.981 1.739 1.75 0.784 1.129 1.41 0.624 0.793 1.23 0.489 0.505 0.98 Panel C (not overbought) Long White −0.870 −0.489 0.59 −0.397 −0.185 0.53 −0.249 −0.106 0.53 −0.005 0.023 0.53 Dragonfly Doji −0.824 −0.485 0.62 −0.299 −0.144 0.57 −0.144 −0.052 0.54 0.036 0.045 0.53 Hammer −0.689 −0.487 0.74 −0.407 −0.220 0.61 −0.243 −0.117 0.59 −0.103 −0.029 0.53 Bullish Harami −0.394 −0.344 0.94 −0.228 −0.101 0.56 −0.013 0.020 0.51 −0.015 0.019 0.47 Bullish Gap 0.587 2.144 3.61 0.594 2.052 3.41 0.540 1.738 3.17 0.535 1.458 2.68 Panel D (uptrend + not overbought) Long White −1.002 −0.616 0.64 −0.451 −0.242 0.59 −0.298 −0.150 0.59 −0.125 −0.046 0.57 Dragonfly Doji −1.045 −0.616 0.62 −0.432 −0.226 0.58 −0.249 −0.107 0.54 −0.104 −0.029 0.53 Hammer −0.827 −0.501 0.64 −0.593 −0.250 0.46 −0.368 −0.150 0.48 −0.213 −0.075 0.48 Bullish Harami −0.382 −0.311 0.88 −0.185 −0.074 0.54 −0.023 0.015 0.49 −0.022 0.016 0.45 Bullish Gap 0.477 0.523 1.04 0.499 0.436 0.82 0.320 0.276 0.78 0.229 0.182 0.68 Note. “SR” refers to the Sharpe ratio of each candlestick based trading strategy. “AR” refers to the average annual return. “SD” refers to the standard deviation of the annual returns. patterns in the in-sample period could remain their predictabil- White based trading strategies, not only generated positive ity in the out-of-sample period. average annual return but also performed better than the bench- Our primary analysis results from the in-sample period mark in terms of Sharpe ratio. revealed that certain bullish patterns were found to produce Overall, our results suggest that several candlestick patterns positive and significant average returns on the component and candlestick pattern based trading strategies, particularly stocks of the SSE50 index. The out-of-sample experimental Bullish Gap and Long White patterns with 5- or 10-day holding results of the robustness test demonstrated that signals gener- periods, can create values for investors trading on the compo- ated by Bullish Gap over a 10-day holding period under all four nent stocks of the SSE50 index. Therefore, experimental results conditions were proved to be profitable for investors. demonstrated that several candlestick pattern based trading Additionally, experimental results indicated that the typical use strategies had significant predictive power, whereas investors of the bearish patterns investigated in this research was not need to consider different market conditions when employing profitable on the component stocks of the SSE50 index over those Japanese candlestick patterns. Future studies could test the studied period. Further, based upon a bootstrap analysis, the their profitability on the medium-market-value or small-market- robustness of the experiment results was subsequently con- value stocks in the Chinese stock market, or examine their prof- firmed. Finally, we designed the candlestick pattern based trad- itability in other financial markets such as FX or commodity ing strategies and examined their profitability on the component markets. Moreover, this research studied the profitability of stocks of the SSE50 index, as well as all stocks in the Chinese candlestick patterns by using daily data, while other researchers stock market. Experimental results showed that most Bullish/ could investigate the high-frequency financial time series data Bearish Gap based trading strategies, as well as several Long such as 2-hour, 1-hour, or 30-minutes in the future. 16 SAGE Open Table 11. Trading Simulation Results of the Five Bearish Candlestick Pattern Based Trading Strategies Over the Period 2000 to 2018 Using All Stocks Data. Holding period length and profitability test results Candlestick pattern 1-day 3-day 5-day 10-day based trading strategy SR AR SD SR AR SD SR AR SD SR AR SD Panel A (unconditional) Long Black −1.192 −0.611 0.53 −0.609 −0.298 0.53 −0.505 −0.245 0.54 −0.406 −0.188 0.53 Gravestone Doji −1.123 −1.818 1.64 −0.636 −0.612 1.00 −0.505 −0.404 0.85 −0.389 −0.263 0.74 Inverted Hammer −1.310 −0.926 0.73 −0.483 −0.254 0.58 −0.358 −0.184 0.59 −0.307 −0.143 0.55 Bearish Harami −0.557 −0.268 0.53 −0.117 −0.040 0.56 −0.048 −0.003 0.61 −0.127 −0.043 0.55 Bearish Gap 1.033 1.408 1.34 0.968 1.075 1.08 0.753 0.725 0.93 0.343 0.335 0.90 Panel B (uptrend) Long Black −1.016 −0.468 0.49 −0.469 −0.242 0.57 −0.395 −0.219 0.62 −0.325 −0.270 0.91 Gravestone Doji −0.836 −1.491 1.81 −0.462 −0.473 1.08 −0.368 −0.328 0.96 −0.333 −0.156 0.55 Inverted Hammer −1.124 −0.859 0.79 −0.442 −0.318 0.78 −0.383 −0.204 0.60 −0.317 −0.136 0.51 Bearish Harami −0.584 −0.211 0.41 −0.133 −0.089 0.87 −0.165 −0.186 1.29 −0.195 −0.112 0.71 Bearish Gap 0.890 1.176 1.29 0.900 0.919 0.99 0.651 0.607 0.89 0.091 0.140 1.25 Panel C (not overbought) Long Black −1.184 −0.602 0.53 −0.594 −0.286 0.53 −0.491 −0.229 0.52 −0.372 −0.164 0.51 Gravestone Doji −0.998 −1.608 1.64 −0.588 −0.509 0.91 −0.448 −0.373 0.89 −0.370 −0.212 0.65 Inverted Hammer −1.339 −0.954 0.79 −0.501 −0.275 0.60 −0.342 −0.178 0.60 −0.272 −0.119 0.53 Bearish Harami −0.683 −0.312 0.50 −0.171 −0.071 0.57 −0.078 −0.021 0.60 −0.130 −0.044 0.54 Bearish Gap 0.808 1.062 1.28 0.829 0.868 1.02 0.648 0.569 0.84 0.415 0.314 0.69 Panel D (uptrend + not overbought) Long Black −0.914 −0.437 0.51 −0.413 −0.221 0.60 −0.366 −0.170 0.54 −0.316 −0.195 0.70 Gravestone Doji −0.812 −1.508 1.89 −0.460 −0.454 1.04 −0.356 −0.310 0.95 −0.325 −0.151 0.54 Inverted Hammer −1.104 −0.897 0.84 −0.409 −0.292 0.78 −0.304 −0.161 0.62 −0.263 −0.094 0.46 Bearish Harami −0.560 −0.197 0.40 −0.139 −0.094 0.86 −0.160 −0.161 1.17 −0.205 −0.107 0.65 Bearish Gap 0.798 0.661 0.80 0.791 0.547 0.66 0.628 0.357 0.53 0.263 0.153 0.48 Note. “SR” refers to the Sharpe ratio of each candlestick pattern based trading strategy. “AR” refers to the average annual return. “SD” refers to the standard deviation of the annual returns. 17 Appendix A. Japanese Candlestick Patterns. Japanese Pattern shape candlestick pattern example Description Dragonfly Doji Dragonfly Doji consists of one candlestick line. It is identified when the length of the candlestick body is smaller or equal to 1/10 of the total length of the candlestick. Long White For Long White, the open price of the candlestick line is less than the close price. The candlestick body length must be greater or equal to the median of the candlestick body length of 20 preceding candlesticks. Bullish Hammer Bullish Hammer is a 1-day candlestick. It must satisfy three requirements: (1) The length of the lower shadow line must be greater than two-thirds of the length of the entire candlestick length; (2) The upper shadow line must be less than or equal to one-tenth of the length of the entire candlestick line; (3) The length of the candlestick body must be greater than 1/10 of the length of the entire candlestick line. Bullish Harami In Bullish Harami, a black candlestick is followed by a white candlestick. The white candlestick body is located within the vertical range of the black candlestick body. The Bullish Harami is a black candlestick engulfing a white candlestick. Bullish Gap Bullish Gap consists of two candlesticks; the low price of the second candlestick should be greater than the high price of the first candlestick line. Gravestone Doji Gravestone Doji consists of one candlestick. The open and the close prices of the candlestick are identical or very close to each other. The body of the candlestick is smaller or equal to 1/10 of its overall length. It has a long upper shadow but no lower shadow. Long Black For Long Black, the open price of the candlestick line is greater than the close price. The candlestick body length must be greater or equal to the median of the body length of 20 preceding candlesticks. Inverted Hammer Inverted Hammer consists of one candlestick. It must satisfy three requirements: (1) The length of the upper shadow line must be greater than two-thirds of the length of the entire candlestick. (2) The lower shadow line must be less than or equal to one-tenth of the length of the entire candlestick. (3) The length of the candlestick body must be greater than one-tenth of the length of the entire candlestick. Bearish Harami In Bearish Harami, a white candlestick is followed by a black candlestick. The black candlestick body is located within the vertical range of the white candlestick body. The Bearish Harami is a white candlestick engulfing a black candlestick. Bearish Gap Bearish Gap consists of two candlesticks; the high price of the second candlestick line is lower than the low price of the first candlestick line. 18 SAGE Open Appendix B. A List of the Component Stocks for the SSE 50 Index. No Company name No Company name 1 Shanghai Pudong Development Bank 2 China Minsheng Banking Corporation 3 Baoshan Iron & Steel 4 PetroChina Company 5 China Southern Airlines Company 6 Citic Securities Company 7 China Merchants Bank 8 Poly Developments and Holdings Group 9 China United Network Communications 10 SAIC Motor Corporation 11 Shanghai Fosun Pharmaceutical (Group) 12 Jiangsu Hengrui Medicine 13 Wanhua Chemical Group 14 China Fortune Land Development 15 Kweichow Moutai 16 Shandong Gold Mining 17 Anhui Conch Cement Company 18 Greenland Holdings Corporation 19 Haier Smart Home 20 Sanan Optoelectronics 21 Inner Mongolia Yili Industrial Group 22 Daqin Railway 23 Industrial Bank 24 Ping an Insurance (Group) 25 China Life Insurance Company 26 China Shenhua Energy Company 27 Bank of Beijing 28 China Railway Construction Corporation 29 Guotai Junan Securities 30 Agricultural Bank of China Limited 31 Bank of Communications 32 New China Life Insurance Company 33 360 Security Technology Inc. 34 China Railway Group 35 Industrial and Commercial Bank of China 36 China Pacific Insurance (Group) 37 China State Construction Engineering Corporation 38 Huatai Securities 39 CRRC Corporation Limited 40 China Communications Construction 41 China Everbright Bank Company 42 PetroChina Company 43 China Tourism Group Duty Free Corporation 44 China Construction Bank 45 Bank of China 46 China Shipbuilding Industry Company 47 China Molybdenum 48 Foxconn Industrial Internet 49 Bank of Shanghai 50 WuXi Application Technology Acknowledgments ORCID iD We thank the editors and anonymous referees for their insightful Shangkun Deng https://orcid.org/0000-0002-6041-485X comments. All errors and omissions are our responsibility. Additionally, the scientific calculations in this paper have been References done on the HPC Cloud Platform of Shandong University. Batten, J. A., Lucey, B. M., Mcgroarty, F., Peat, M., & Urquhart, A. (2018). Does intraday technical trading have predictive power Data Availability Statement in precious metal markets? Journal of International Financial The data that support the findings of this study are openly available Markets, Institutions and Money, 52, 102–113. in the Choice database at http://data.eastmoney.com. Brock, W., Lakonishok, J., & LeBaron, B. (1992). Simple technical trading strategies and the stochastic properties of stock returns. Declaration of Conflicting Interests The Journal of Finance, 47(5), 1731–1764. Caginalp, G., & Laurent, H. (1998). The predictive power of price The author(s) declared no potential conflicts of interest with respect patterns. Applied Mathematical Finance, 5, 181–205. to the research, authorship, and/or publication of this article. Chen, S., Bao, S., & Zhou, Y. (2016). The predictive power of Japanese candlestick charting in Chinese stock market. Physica Funding A: Statistical Mechanics and its Applications, 457, 148–165. The author(s) disclosed receipt of the following financial support Chong, T. L., & Ng, W. K. (2008). Technical analysis and the lon- for the research, authorship, and/or publication of this article: This don stock exchange: Testing the macd and rsi strategies using work was supported by the Natural Science Foundation of Hubei the ft30. Applied Economics Lettersm, 15(14), 1111–1114. Province [Grant Number 2021CFB175]; National Social Science Cohen, G. (2021). Optimizing candlesticks patterns for Bitcoin’s Foundation of China [Grant Number 19BGL131]; and Innovation trading systems. Review of Quantitative Finance and and Entrepreneurship Programs of China Three Gorges University Accounting, 57, 1155–1167. [Grant Numbers S202111075045 and S202111075066X]. In addi- Day, T. E., & Wang, P. (2002). Dividends, nonsynchronous prices, tion, we are grateful to the editors and anonymous reviewers for and the returns from trading the dow jones industrial average. their comments and discussions. Journal of Empirical Finance, 9(4), 431–454. Deng et al. 19 Deng, S., Yu, H., Wei, C., Yang, T., & Tatsuro, S. (2020). The Metghalchi, M., Marcucci, J., & Chang, Y. H. (2012). Are moving profitability of Ichimoku Kinkohyo based trading rules in stock average trading strategies profitable? Evidence from the European markets and FX markets. International Journal of Finance and stock markets. Applied Economics, 44(12), 1539–1559. Economics, 26, 5321–5326. https://doi.org/10.1002/ijfe.2067 Mitra, S. K. (2011). How rewarding is technical analysis in the Fama, E. F. (1970). Efficient capital markets: A review of theory Indian stock market? Quantitative Finance, 11, 287–297. and empirical work. The Journal of Finance, 25(2), 383–417. Murphy, J. J. (1999). Technical analysis of the financial markets: A Fang, J., Jacobsen, B., & Qin, Y. (2013). Predictability of the sim- comprehensive guide to trading methods and applications. In ple technical trading strategies: An out-of-sample test. Review Technical analysis of the financial markets. New York Institute of Financial Economics, 23(1), 30–45. of Finance. Fifield, S. G., Power, D. M., & Sinclair, C. D. (2005). An analy- Naranjo, R., & Santos, M. (2019). A fuzzy decision system for sis of trading strategies in eleven European stock markets. The money investment in stock markets based on fuzzy candle- European Journal of Finance, 11(6), 531–548. sticks pattern recognition. Expert Systems with Applications, Gehrig, T., & Menkhoff, L. (2006). Extended evidence on the use of 133, 34–48. technical analysis in foreign exchange. International Journal Nison, S. (1991). Japanese candlestick charting techniques. New of Finance and Economics, 11, 327–338. York Institute of Finance. Goo, Y., Chen, D., & Chang, Y. (2007). The application of Park, C. H., & Irwin, S. H. (2007). What do we know about the Japanese candlestick trading strategies in Taiwan. Investment profitability of technical analysis? Journal of Economic Management and Financial Innovations, 4, 49–71. Surveys, 21, 786–826. Gunasekarage, A., & Power, D. M. (2001). The profitability of Prado, H. A. D., Ferneda, E., Morais, L. C. R., & Matsura, E. moving average trading rules in south Asian stock markets. (2013). On the effectiveness of candlestick chart analysis for Emerging Markets Review, 2(1), 1–33. the Brazilian stock market. Procedia Computer Science, 22, Heinz, A., Jamaloodeen, M., Saxena, A., & Pollacia, L. (2021). 1136–1145. Bullish and Bearish Engulfing Japanese candlestick patterns: Qiu, H., & Liu, F. (2019, March 15–18). Candlestick analysis in A statistical analysis on the S&P 500 Index. The Quarterly forecasting U.S. stock market: Are they informative and effec- Review of Economics and Finance, 79, 221–244. tive [Conference session]. IEEE International Conference on Jensen, M. C. (1967). Random walks: Reality or mythcomment. Big Data Analytics, Suzhou, China. Financial Analysts Journal, 23, 77–85. Thanh, L. V., Anh, B. N., Ban, T. Q., Hoa, D. X., & Son, N. T. Kwok, N. M., Fang, G., & Ha, Q. P. (2009). Moving average-based (2018). An analyze on effectiveness of candlestick reversal pat- stock trading rules from particle swarm optimization [Conference terns for vietnamese stock market [Conference session]. 2018 session]. 2009 International Conference on Artificial Intelligence International Conference. and Computational Intelligence (Vol. 1, pp. 149–153). November Tharavanij, P., Siraprapasiri, V., & Rajchamaha, K. (2017). 7–8, 2009. Washington, DC, United States. Profitability of candlestick charting patterns in the stock Lu, T. H. (2014). The profitability of candlestick charting in the exchange of Thailand. SAGE Open, 8(4), 1–18. Taiwan stock market. Pacific-Basin Finance Journal, 26, 65– Trivedi, S. R. (2022). Technical analysis using Heiken Ashi 78. Stochastic: To catch a trend, use a HASTOC. International Lu, T. H., Chen, Y. C., & Hsu, Y. C. (2015). Trend definition or Journal of Finance and Economics, 27, 1836–1847. https://doi. holding strategy: What determines the profitability of candle- org/10.1002/ijfe.2245. stick charting? Journal of Banking & Finance, 61, 172–183. Vasiliou, D., Eriotis, N., & Papathanasiou, S. (2008). Technical Lu, T. H., & Shiu, Y. M. (2012). Tests for two-day candlestick trading profitability in Greek stock market. The Empirical patterns in the emerging equity market of Taiwan. Emerging Economics Letters, 7(7), 1681–8997. Markets Finance & Trade, 48, 41–57. Wilder, J. W. (1978). New concepts in technical trading systems. Lu, T. H., & Shiu, Y. M. (2016). Can 1-day candlestick patterns be Trend Research. profitable on the 30 component stocks of the DJIA? Applied Xie, H., Zhao, X., & Wang, S. (2012). A comprehensive look at Economics, 48(34–36), 3345–3354. the predictive information in Japanese candlestick. Procedia Lu, T. H., Shiu, Y. M., & Liu, T. C. (2012). Profitable candle- Computer Science, 9, 1219–1227. stick trading strategies: The evidence from a new perspective. Zhu, M., Atri, S., & Yegen, E. (2016). Are candlestick trading Review of Financial Economics, 21, 63–68. strategies effective in certain stocks with distinct features? Marshall, B. R., Young, M. R., & Rose, L. C. (2006). Candlestick Pacific-Basin Finance Journal, 37(4), 116–127. Database 1. technical trading strategies: Can they create value for inves- Choice database. Retrieved May 1, 2022, from http://choice. tors? Journal of Banking and Finance, 30, 2303–2323. eastmoney.com/

Journal

SAGE Open – SAGE

Published: Aug 10, 2022

Keywords: candlestick pattern; profitability test; trend condition; overbought/oversold condition; out-of-sample test

There are no references for this article.