Technical analysis via candlestick patterns : an empirical study on European stock markets

(1)

Master Thesis | Anh Nguyen - 10839186

TECHNICAL ANALYSIS VIA CANDLESTICK

PATTERNS

A

N

E

MPIRICAL

S

TUDY ON

E

UROPEAN

S

TOCK

M

ARKETS

MSc Business Economics - Finance

Thesis Supervisor: Dr. Philippe Versijp

(2)

STATEMENT OF ORIGINALITY

This document is written by Student Anh Nguyen who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

ABSTRACT

This paper examines whether technical analysis works by testing the predictive capability of the popular Japanese candlestick patterns through various statistical tests. Data of individual component stocks for 7 major European stock markets for a 30-year sample period are utilized. Overall, the results provide strong evidence that short-term market inefficiency is exhibited during this period for most of the European markets under study. However, although some candlestick patterns appear to have predictive power through the standard t-tests, the regression results suggest that this predictive ability is mainly attributed to the direction of the short-term preceding trend and changes in stock return volatility prior to these patterns. An interesting finding, notably, is that for candlestick patterns that do display predictive power in certain markets, there is evidence that this power weakens with increasing stock return volatility but no evidence that this power strengthens with the confirming short-term prior trend. In addition, the candlestick patterns, with or without the prior confirming short-term trend, are shown to outperform the popular Western short-term momentum strategy for specific periods and markets. Furthermore, the significant results of the short-term preceding trend prompted me to perform a supplementary test on the predictive ability of the current market trend, which provides strong evidence supporting the contrarian trading strategy.

(4)

ACKNOWLEDGEMENTS

First and foremost, I would like to extend my sincere gratitude to my professors at Amsterdam Business School, who have given us immense support and equipped us with the necessary knowledge and tools to complete our thesis throughout the past six months. I am especially thankful to my thesis supervisor – Dr. Philippe Versijp – whose tremendous guidance and valuable insights have been essential throughout the course of this research.

Many thanks are extended to my fellow friends in the Msc Business Economics course and my friends in Amsterdam and Eindhoven for their help and encouragement during this period.

Last but not least, I am deeply grateful to my family who has always been there for me every step of the way and is the reason for me to try and to strive for the past 12 years away from home.

(5)

I. INTRODUCTION

For centuries, financial market practitioners have come up with ways to analyze and predict future directions of stock prices in an attempt to make a profit from the market. Two clearly segregated paradigms were developed as a result: fundamental analysis and technical analysis. Fundamental analysts focus on the true value of a financial asset by assessing the fundamentals such as economic variables and growth potential. Technical analysts instead purely observe the market and use historical prices to make predictions on future price movements based on a variety of trading techniques, ranging from moving average, trading range break to the Japanese candlestick patterns. But can technical analysis truly be used to predict future prices such that market participants are able to take advantage of it? This requires an understanding of whether or not these different trading techniques actually have predictive power with regards to future prices. This paper conducts such a study on technical analysis with the focus on the Japanese candlestick patterns.

Despite being widely used in practice (Taylor and Allen 1992), technical analysis has received heavy criticisms mostly from the mainstream academics who center their arguments around the theory of Efficient Market Hypothesis (Fama 1970). Yet many critics and studies have shown evidences of market anomalies against the notion of market efficiency (Jensen 1978) and suggested that although market sometimes makes mistakes, such mispricing will disappear quickly (Malkiel 2003). In addition, many disregard technical analysis as being self-fulfilling in the short run because as many people use the common signals, prices may be pushed in the predicted directions (Taylor and Allen 1992). Another important aspect is the emerging field of behavioral finance whose principles lie in the fact that markets’ movements are based on people’s expectations which are driven not only by rational facts but also by irrational human emotions and behaviors (Hirshleifer 2001, Shleifer and Lawrence 1990). These views provide support to short-term technical analysis that the market can be inefficient in the short run and profits can possibly be achieved by exploiting certain trading signals to predict short-term price movements.

However, the facts that market can be inefficient in the short run and short-term technical analysis can work do not necessarily mean that short-term technical analysis always works. As the findings of Terence (1997) show, trading rules seem to work during 1935-1974 in the London stock exchange when the FT30 market index was growing moderately in value but do not perform as well from 1980s onwards when the market trend started to dominate. This may suggest that volatility has a part

(7)

to play in driving stock returns and influencing the effectiveness of technical trading rules. In addition, short-term momentum, or short-term trend, prior to a trading signal, also plays a role in the construction of the trading signal as important as the signals themselves, especially in the case of candlestick patterns (Marshall et al. 2006). Hence, short-term momentum may possibly affect the returns and effectiveness of these signals as well. Yet, these two aspects, stock return volatility and short-term preceding trend, have not been examined anywhere in the existing literature alongside the technical trading rules. In this study, I will look at the effect of stock return volatility and short-term preceding trend in conjunction with the trading signals to investigate the effectiveness of short-term technical analysis. On top of that, short-term momentum, a rather Western approach, has also been widely used by practitioners as a strategy on its own, which offers an interesting question as to whether combining the candlestick patterns and short-term momentum creates a strategy that indeed outperforms the short-term momentum strategy itself.

Numerous studies have been conducted throughout the past century to assess the profitability and predictive power of various Western methods of technical analysis. However, little has been documented on Japanese candlestick analysis, a short-term market-timing technique which has been used for many years to predict short-term future price movements through visual displays of the opening, close, high and low prices across trading days. This seems puzzling as Japanese candlestick charts have been shown to provide advantages over the Western technical analysis techniques in many ways (Nison 1991, Caginalp and Laurent 1998). Despite being tested and used for many years in Japan, the topic only started to attract more attention in the 1990s after Steve Nison introduced it to the Western world, which explains the dearth of literature on this ancient trading method prior to the 1990s. Besides, as the focus of this research is on term technical analysis, the use of short-term trading indicators such as the Japanese candlestick charts is thus very relevant.

The majority of existing studies examine the Asian markets such as Japan, China, Taiwan, Hong Kong and the US market with mostly consistent results supporting the candlestick method in Asia yet conflicting results towards the value of candlestick charts in the US (Goo et al. 2007, Lu and Shiu 2012, Marshall et al. 2006, Marshall et al. 2008, Fock et al. 2005). One may question whether these findings are because Japanese candlestick analysis originates from the East and is therefore more effective in the Eastern markets compared to the Western markets. Perhaps cultural differences in investor behaviors, investors’ knowledge and market environment are the culprits of different levels of short-term market inefficiency in different markets, resulting in varying results towards the predictive power of certain trading techniques. The answer to this question possibly requires writing

(8)

another paper on its own and is therefore not within the scope of this paper. However, with the aim to further provide evidence in order to examine the above proposition and expand the existing literature in my study, I would like to examine the application of this Eastern trading technique in the context of the European stock markets. The emphasis on European market is relevant as this region consists of important players in the world market, which have had significant influence on the world economy as we have seen in the past decade with the European Debt Crisis. Besides, regulations, market environment and investor behavior in Europe might be different from those of other regions and therefore may yield different results from studies on other regions.

The ultimate goal of this research is therefore to examine whether technical analysis works, or in other words whether the Efficient Market Hypothesis is violated, during the more recent periods through the use of candlestick chart patterns with the focus on the European stock markets. Specifically, I am interested to see if technical analysis works in the short term and whether the predictive power of technical analysis, through the use of candlestick reversal patterns in particular, weakens with increasing stock returns volatility and strengthens with the appropriate short-term prior momentum, especially since other studies have suggested the possible influence of these two factors on the effectiveness of trading signals. In addition, it is also of interest to examine if a short-term strategy based on a combination of the candlestick patterns and short-term prior momentum actually outperforms the short-term momentum strategy itself.

The remainder of this paper will be structured as follows: section II reviews some of the relevant past research on the criticisms and support towards technical analysis including the Efficient Market Hypothesis and the Random Walk Theory, as well as existing studies on the Japanese candlestick patterns and short-term momentum strategy; section III describes the objectives of this research as well as the data and methodologies used to test the hypotheses in question; section IV reports the various empirical test results and their implications; and section V finally concludes and provides recommendations for future research.

(9)

II. LITERATURE REVIEW

1. Criticism on Technical Analysis

Criticisms towards technical analysis are established mainly on the grounds of Efficient Market Hypothesis (EMH) and Random Walk Theory (RWT). As academics are interested in whether prices can indeed be forecasted rather than how they can be forecasted, the credibility of these theories, together with the lack of academic substantiation for technical analysis, have made technical analysis an easy target to pick on (Brock 1992).

In an efficient market according to the Efficient Market Hypothesis, all information, either public or private, is known to every market participant and has been accounted for in the market price. There are three main forms of EMH, namely the weak form, semi-strong form and strong form pertaining to the different nature of the information involved, but for the purpose of this study, we are only interested in the weak form of EMH, which is associated with historical price information. According to this weak-form EMH, prices in the market at any particular point in time trade at fair values, meaning that they have fully incorporated all past price information and hence, it is impossible to employ any form of technical analysis to predict future price movements and achieve abnormal returns (Fama 1970). Many studies in the past have tested this proposition and found consistent results in favor of it in a wide range of markets and products. Yielding the same conclusion with respect to the value of technical analysis, the Random Walk Theory suggests that stock prices take a random path and any identified trends in historical prices are purely spurious. Many early empirical works found support for this theory, showing no evidence of linear dependence among successive price changes (Cootner 1962, Fama 1965).

Yet, there has been increasing evidence that has shown inconsistency with these two theories and in turn provides support for technical analysis. Some of these studies are discussed in the next section.

2. Support for Technical Analysis

Despite criticisms from the mainstream academics, technical analysis has nowadays become widely known and used in practice. Taylor and Allen (1992), through a questionnaire survey to British chief foreign exchange dealers, found that 90% of the survey respondents base their trading decisions on

(10)

technical analysis and that reliance on technical analysis as compared to fundamental analysis increases as the horizon length becomes shorter. Oberlechner (2001) performed a questionnaire and an interview survey on the perceived importance of technical and fundamental analysis in the European foreign exchange market and found consistent results with Taylor and Allen (1992) that technical analysis is perceived as more important for shorter forecasting horizons while fundamental analysis is more important for longer forecasting horizons. However, most of these survey studies focus on the futures and foreign exchange markets and no such survey is found for stock markets (Park and Irwin 2004).

Academically, there are also many studies that defend the use of technical analysis. Treynor and Ferguson (1985) found that past prices are useful only when combined with other non-price information available as “it is the non-price information that creates the opportunity” that allows effective exploitation of past prices. Other studies suggest the predictability of historical prices. Lo et al (1988) showed evidence that rejects the random walk theory for weekly stock market returns while French and Roll (1986) found negative serial correlation among weekly and daily returns for individual stocks. In addition, many sources provide evidences of market anomalies that are not in line with efficient market hypothesis and random walk theory such as the January effect, the Weekend Effect, the Holiday Effect. Malkiel (2003) explained, with reference to these anomalies, that although the market may not be perfectly efficient, it would quickly correct itself in the long run, which then allows for short-term opportunities to be realized. Grossman and Stigliz (1980) also argued that due to the costly nature of information, price cannot perfectly reflect all available information, meaning that the market cannot be perfectly efficient because otherwise, there would be no incentives for those who spend resources to obtain such information. The above papers support the view that the market can be inefficient in the short run and it may be possible to achieve short-term profits with technical analysis after all. This sets up the focus on short-short-term technical analysis in my study.

The emerging field of behavioral finance also lends support to technical analysis. While conventional finance idealizes a world where market participants are all rational “wealth maximizers”, behavioral finance recognizes and seeks to gain insights into why people sometimes make irrational decisions. Prices, essentially, are determined by forces of demand and supply, both of which are very much affected by human emotions. Most academic pricing models, however, suffer from not taking into account this aspect of investor psychology (namely human emotions and irrationality), as Hirshleifer (2001) suggests, and hence, the market itself remains the only place that

(11)

provides its participants with the most accurate information. On a related note, Shleifer and Lawrence (1990), look at an alternative to the traditional efficient market approach by introducing the concept of “limits of arbitrage” which suggests that in an environment where both rational and irrational agents exist and interact, arbitrage is risky and hence limited. Arbitrage is an act of exploiting mispricing situations of financial instruments, made by rational investors. Although the existence of arbitrage is a result of market inefficiency, arbitrage helps to minimize any substantial price deviation from its fair value for too long and quickly regain the equilibrium state of market efficiency. However, due to two types of risk, namely fundamental risk and uncertainty of future resale prices, arbitrage is, as mentioned above, risky and limited. This failure of perfect arbitrage, according to the paper, allows changes in investor sentiment to be a significant determinant of prices, which in turn is the reason why market inefficiency and crises occur.

3. Japanese Candlestick Analysis as a form of Technical Analysis

Technical analysis, as mentioned above, is a paradigm that purely makes use of information on historical prices to predict future price movements. Many studies have been performed to assess the predictive power of Western trading techniques of different kinds such as Moving Average, Relative Strength Index, MACD, Trading Range Break-out, etc. However, as explained earlier, our focus of short-term technical analysis would be on the Japanese candlestick analysis. The oldest known form of technical analysis, candlesticks were first invented in Japan in the 18th century, long before the idea of technical analysis was introduced in the Western world by Charles Dow (Marshall et al. 2006) in the late 19th century. However, it was only until the introduction of candlestick charting to the Western world by Steve Nison in 1991 that sparked research interest in the field with mixed results from various studies. Caginalp and Laurent (1998) examined the predictive capability of some candlestick patterns on S&P 500 stocks and found strong evidence supporting the method. Subsequently, Goo et al. (2007) and Lu and Shiu (2012) showed findings in support of this method in the context of Taiwan market. In contrast, results from Fock et al. (2005) and Marshall et al. (2006, 2008) suggest that there is little value in candlestick charting. In particular, Marshall et al. (2006) employed a bootstrapping method on the DIJA index stocks during 1992-2002 periods which overlap with those of Caginalp and Laurent (1998) but concluded that none of the candlestick patterns used provides signals that are better than what would be expected by chance. The differences in results, however, could possibly arise from differences in methods used, markets or timeframes examined.

(12)

Most recently, Lu and Chen (2013), in one of the very few papers that examined the profitability of candlesticks pattern in the European markets, found that after accounting for transaction costs, three different patterns are profitable in the three different markets studied (UK, Germany and France). The reason why these three particular markets were chosen is unclear, however, their results suggest that candlestick charting method does work in the European markets and motivates me to conduct a more extensive research to assess the predictive power of candlestick analysis across all major European stock markets. However, while Lu and Chen categorize all two-day patterns by adopting a 1x4 vector approach, this study uses the 3-day reversal patterns as introduced by Nison (1991) and Morris (1995). These patterns have been developed and tested over centuries and are more well-known among investors who would rely on signals generated by these patterns to make trading decisions. Moreover, this paper also examined the impact of the global financial crisis and found that the crisis weakened the predictive power of the candlestick trading method in all three markets examined. As crisis periods are usually associated with high volatility, this supports the notion that volatility may have a negative impact on returns and thereby influence the predictive power of technical trading techniques. Hence, taking into account volatility in assessing the effectiveness of candlestick method adds a unique aspect to this research.

In addition, as mentioned earlier, Japanese candlestick charts provide advantages in many ways over the Western techniques. Unlike the traditional focus on daily closing prices of Western trading techniques, the emphasis on open-close-high-low price relationship of Japanese candlesticks provides more significant insights into the strategies and sentiments of other players in the market and thereby the direction the market might be heading (Caginalp and Laurent 1998). In addition, these charts are able to display changes in volatility and momentum without the use of Western techniques like Oscillators or Moving Average. Although they are fairly similar to the Western bar charts, they provide signals through well-defined chart patterns, which are not available from bar charts (Nison 1991). Based on these advantages, although no such study has been conducted before, it may be interesting to hypothesize that the Japanese candlestick analysis outperforms the Western trading techniques. Short-term momentum is a simple Western trading technique that has been in use by financial market practitioners around the world. The general idea behind this strategy is that traders will only go long or short on a stock for a short period of time when it is clear that momentum is in a particular direction. This strategy signals rather the continuation of the short-term prevailing trend while the reversal candlestick patterns strategy signals instead a reversal of that trend. Therefore, it would be interesting to perform a simple test to compare the performance of this short-term momentum strategy with that of the Japanese candlestick analysis.

(13)

4. Short-term Momentum Strategies

The idea of momentum trading has been a topic of interest among academicians in the past few decades. Notably, Jegadeesh and Titman (1993) examined the profitability of momentum-exploiting strategies that buy well-performing stocks and sell poor-performing stocks over the previous horizon of 2 to 12 months for the 1965 to 1989 period in the US market. They found that these momentum strategies generate significant returns of about one percent per month for the following year. Although these findings were generally well-accepted, the authors subsequently published another paper (Jegadeesh and Titman 2001) with evidence of continued profitability of these momentum strategies in the 1990s, refuting the notion that their original findings were the outcome of data snooping process. Rouwenhorst (1998, 1999) also addressed the data snooping problem by examining momentum strategies in an international context. Rouwenhorst (1998) found that momentum strategies yield significant returns of approximately one percent per month across all 12 European markets in the sample from 1978 through to 1995, consistent with findings by Jegadeesh and Titman (1993). In another paper, Rouwenhorst (1999) also found that momentum is exhibited in 20 emerging stock markets in his sample. Many other papers that ensue offer a variety of theoretical, behavioral and risk-based models to explain this phenomenon (Daniel et al. 1998, Hong and Stein 1999, Sagi and Seasholes 2007). These findings all challenge the view of Efficient Market Hypothesis and support technical analysis and particularly the momentum strategies.

Momentum strategies are generally viewed as a short-term trading technique. Khoroshilov (2012) found that investors with shorter-term horizon tend to rely more on momentum strategies than those with longer-term horizon. However, the short-term notion can range from a few hours up to a year depending on traders’ preferences. Despite strong findings in support of momentum strategies, few existing studies shed light on how the chosen time horizon over which momentum is observed may affect the profitability of the strategies. In a recent paper, Novy-Marx (2012) found that momentum is mainly driven by past returns over the intermediate horizon of 12 to 7 months, rather than past returns over the recent 6 to 2 months, suggesting that strategies based on the tradition view of momentum – the recent past performance – are less profitable. However, Gong et al. (2014) argued that Novy-Marx’s findings are due to an estimation bias and concluded that “momentum is really short-term momentum”. In this paper, since we are interested in comparing the short-term momentum strategy with the short-term candlestick patterns, it is thus more appropriate to use the short-term momentum strategy based on the most recent 3-day horizon.

(14)

III. OBJECTIVES AND METHODOLOGY

1. Objectives

Before testing the predictive power of the various chart patterns, it is ideal to test for market efficiency during the period examined, which will give an idea on whether market is indeed inefficient to ensure validity of the subsequent tests. As the literature review suggests that market can make mistakes and cannot be perfectly efficient in the short run, we are only interested in testing the weak-form short-term market inefficiency in the sense that we can make use of the immediate preceding days’ prices to predict the short-term price movements that follow. Hence, the first hypothesis goes as follows:

Hypothesis 1: The European market as a whole exhibits short-term market inefficiency during the period examined. The same behavior can be observed for each of the individual European markets under study.

As the main objective of the study is to examine the predictive power of the candlestick reversal chart patterns, given that the market cannot be perfectly efficient in the short run following my argument earlier, it might be possible to achieve short-term profits through candlestick patterns. In addition, as the findings by Lu and Chen (2013) and Terence (1997) suggest, increasing volatility in stock returns may possibly weaken the predictive power of these candlestick patterns. Also, short-term trend prior to the appearance of the signals also plays a very important role in the construction of trading signals (Marshall et al. 2006). In particular, an uptrend prior to a sell signal and a downtrend prior to a buy signal help to strengthen these signals and thus, increase the predictive power of these reversal patterns. Thus, my second hypothesis is as follows:

Hypothesis 2: Candlestick reversal patterns have predictive power.

2a. The predictive ability of candlestick reversal patterns weakens with increasing stock returns volatility.

2b. The predictive ability of candlestick reversal patterns strengthens with the confirming short-term preceding momentum.

(15)

Finally, based on the advantages that candlestick method has over the Western trading technique, I would like to investigate if the Japanese candlestick analysis outperforms the Western trading techniques. For the purpose and convenience of this study, I will compare the candlestick method, which incorporates the preceding momentum prior to the candlestick signals, with the simplest form of Western trading method - the short-term momentum strategy, to examine this relationship on the same dataset. My last hypothesis is as follows:

Hypothesis 3: The combination of candlestick reversal patterns and the short-term preceding trend outperforms the short-term momentum strategy itself.

2. Data & Methodology

2.1. Datasets

This study requires data on the daily opening, high, low and closing unadjusted prices as well as the total return index (which takes into account effects of dividends and stock splits and will be used to calculate the holding period returns measure in this study) of individual component stocks for the 7 major European stock markets for the sample period from 1984 to 2014. These markets include UK, France, Germany, Netherlands, Spain, Sweden and Switzerland. These daily stock data are extracted from Datastream and are converted to the common currency, Euro, for the comparisons across regional markets to be valid from a European investor’s perspective. Although the Euro only came into existence on January 1, 1999 as the official currency of the Eurozone, Datastream created the synthetic euro rates for the periods 1976-1998 “based on the weights that each component currency has in the real Euro” (Bris et al. 2009). For periods after January 1, 1999, daily exchange rates from Datastream are used for automatic conversion to Euro.

The use of 30 years sample period allows for robustness tests by splitting them into 3 smaller sub-periods of 10 years each (1984-1993, 1994-2003, 2004-2014). As argued by Caginalp and Laurent (1998) and Marshall et al. (2006), since the data used to develop candlestick pattern analysis by Japanese rice traders were from a different time period and subsequently used for a different market, our approach in this study is clearly out-of-sample.

(16)

2.2. Candlestick Terminology

Japanese candlestick method involves analyzing candlestick charts that are made up of individual candlesticks, each representing a specific trading range – one minute, one hour, one day, one month or one year, etc. For the ease of data collection and analysis, this study only looks at one-trading-day candlesticks. An example of the black and white candlesticks are shown below:

Figure 1: Example of a white candlestick and a black candlestick

Below are some of the basic terminologies associated with candlestick charting method (Caginalp and Laurent 1998):

- Each candlestick is made up of four main component prices of a trading session, namely opening, high, low and closing prices.

- The body of a candlestick denotes the difference between the opening and closing prices. The body can be either black or white. The white candlestick opens below the closing price and represents a bullish session while the black candlestick opens above the closing price and represents a bearish session. In a special case when the opening price is equal to the closing price, the body becomes a single horizontal line and is called a ‘Doji’.

- The small vertical lines above and below the body of each candlestick are called upper and lower shadows and represent the trading range of a particular trading day.

(17)

2.3. Candlestick Patterns

As noted in Caginalp and Laurent (1998), the Japanese candlestick charting system consists of patterns with different time scales (from one to three days) and patterns with three-day period are said to be more reliable because they display the changing balance of demand and supply in the market more accurately. Thus, the focus of this study would be on three-day candlestick patterns.

The Japanese candlestick method comprises two main types of charts – continuation and reversal chart patterns. Although they are both important and being widely used in practice, one drawback of strategies using continuation patterns is that once a trend is identified, the trader might be getting in too late and therefore may not be able to capture the most profit from the prevailing trend. However, a reversal pattern is able to signal more critical turning points arising from any new event in the market and allows the trader to identify and exploit the new trend right from the beginning. Thus, I find it more meaningful to examine only reversal patterns in my study.

The following 5 pairs of three-day reversal patterns have been selected for the purpose of this study. They are among the most commonly used patterns by practitioners and researchers in this field and most of these patterns were ranked among the top-performing 3-day reversal patterns in terms of predictive power in Bulkowski’s book titled “Encyclopedia of Candlestick Charts” (2008):

1. Three White Solders // Three Black Crows 2. Three Inside Up // Three Inside Down 3. Three Outside Up // Three Outside Down 4. Morning Star // Evening Star

5. Morning Doji Star // Evening Doji Star

2.4. Pattern Definitions and Conditions

A set of definitions is outlined below to define each candlestick pattern used in this study and coding will be subsequently performed in STATA based on these definitions to identify the corresponding signals from the data set collected. These definitions have been consolidated mainly based on Nison (1991) and Caginalp and Laurent (1998):

(18)

Candlestick

Reversal Pattern Pattern Description

Prior Trend Signal Type (Bullish/Bearish) Illustration Three White Soldiers

A group of three consecutive white candlesticks, each with a higher closing price than the previous session and each closing at or near its high. In addition, it should show a gradual rise with each day opening within the previous session’s real body range.

Down Bullish

Three Black Crows

A group of three consecutive black candlesticks, each with a lower closing price than the previous session and each closing at or near its low. In addition, it should show a gradual fall with each day opening within the previous session’s real body range.

Up Bearish

Three Inside Up

This pattern is formed with the first session’s long black candlestick whose real body engulfs the second session’s white candlestick (both its high and low). The last candlestick is white and closes higher than the previous 2 days.

Down Bullish

Three Inside Down

This pattern is formed with the first session’s long white candlestick whose real body contains the second session’s black candlestick (both its high and low). The last candlestick is black and closes lower than the previous 2 days.

Up Bearish

Three Outside Up

This pattern is formed with the second session’s long white candlestick whose real body engulfs the first session’s small black candlestick (both its high and low). The third candlestick is white and closes higher than the previous 2 days.

Down Bullish

Three Outside Down

This pattern is formed with the second session’s long black candlestick whose real body engulfs the first session’s small white candlestick (both its high and low). The third candlestick is black and closes lower than the previous 2 days.

(19)

Morning Star

This pattern comprises a long black candlestick followed by a small second day’s candlestick, with either black or white real body gapping below the first day’s closing price. The third day is a white candlestick that closes above the midpoint of the first day’s black body.

Down Bullish

Evening Star

This pattern comprises a long white candlestick followed by a small second day’s candlestick, with either black or white real body gapping above the first day’s closing price. The third day is a black candlestick that closes below the midpoint of the first day’s white body.

Up Bearish

Morning Doji Star

This pattern comprises a long black candlestick followed by a doji with equal opening and closing prices on the second day gapping below the first day’s closing price. The third day is a white candlestick that closes above the midpoint of the first day’s black body.

Down Bullish

Evening Doji Star

This pattern comprises a long white candlestick followed by a doji with equal opening and closing prices on the second day gapping above the first day’s closing price. The third day is a black candlestick that closes below the midpoint of the first day’s white body.

Up Bearish

2.5. Holding Period Returns and Short-term Trend

2.5.1. Holding Period Returns

In this study, the predictive power of candlestick patterns will be tested and compared with the simple buy-and-hold strategy and the short-term momentum strategy on the basis of holding period returns. Due to the short-term nature of candlestick analysis, we calculate these raw returns on each day for holding periods (HPR) of 2, 5 and 10 days. Following Brock et al. (1992), Terence (1997) and Lu and Chen (2013), we calculate the HPR from a buy-and-hold perspective, with positive returns and negative return denoting profitable buying and selling strategies respectively.

(20)

Specifically, the n-day holding period returns for a particular stock are measured as log differences of the total return indexes between dates t and t + n as shown below, with n = 2, 5 and 10 and RIt denoting the total return index on day t:

𝐻𝑃𝑅_𝑡,𝑛 = ln⁡(𝑅𝐼𝑡+𝑛 𝑅𝐼_𝑡 )

The holding period returns apply to hypothesis 2 and 3 mostly. For hypothesis 1, however, only daily returns are required and can be calculated in the same way with n equal to 1.

2.5.2. Short-term Trend

The identification of short-term trend is required in the testing of candlestick patterns with short-term preceding trend as well as the short-term momentum strategy. This applies to both hypothesis 2 and 3. This paper adopts the same method employed by Caginalp and Laurent (1998) and Fock et al. (2005) to identify the short-term trend, which is to use the three-day moving averages of the most recent 6 days. The three-day moving average at time t of a stock can be defined as follows, with MAt,3 and RIt denoting the 3-day moving average and the total return index on day t respectively:

𝑀𝐴𝑡,3=

1

3(𝑅𝐼𝑡−2+ 𝑅𝐼𝑡−1+ 𝑅𝐼𝑡)

An uptrend on day t is then defined by:

𝑀𝐴𝑡−6,3< 𝑀𝐴𝑡−5,3 < 𝑀𝐴𝑡−4,3< 𝑀𝐴𝑡−3,3 < 𝑀𝐴𝑡−2,3< 𝑀𝐴𝑡−1,3 < 𝑀𝐴𝑡,3

Analogously, a downtrend on day t is defined by:

𝑀𝐴_𝑡−6,3> 𝑀𝐴_𝑡−5,3 > 𝑀𝐴_𝑡−4,3> 𝑀𝐴_𝑡−3,3 > 𝑀𝐴_𝑡−2,3> 𝑀𝐴_𝑡−1,3 > 𝑀𝐴_𝑡,3

(21)

2.6. Empirical Methods

Hypothesis 1: The European market as a whole exhibits short-term market inefficiency during the period examined. The same behavior can be observed for each of the individual European markets under study.

The first hypothesis regarding market inefficiency can be tested by running a linear autoregressive model on the daily stock returns and examining the magnitude and significance of the coefficients, with 2, 5 and 10 lags respectively for the overall European market as well as each of the individual markets under study separately:

Returnt = α + β1 Returnt-1 +β2 Returnt-2 + εt

Returnt = α + β1 Returnt-1 +β2 Returnt-2 + … + β5 Returnt-5 + εt

Returnt = α + β1 Returnt-1 +β2 Returnt-2 +… + β10 Returnt-10 + εt Where: Returnt = ln(Pt/Pt-1)

In addition to the lagged returns, the length of the candlestick measured by the high and low prices in the most recent preceding days, which we will denote as HML (high minus low), may also be a good source of information in influencing the returns at time t, as suggested by Nison (1991) and Morris (1992) that lower and upper shadows are believed by technical analysts to be predictive and are hence frequently used to predict future prices. Thus, multiple lags of this additional predictor will be included in another so-called autoregressive distributed lag (ADL) model below:

Returnt = α + β1 Returnt-1 +β2 Returnt-2 + δ1 HMLt-1 + δ2 HMLt-2 + εt

Returnt = α + β1 Returnt-1 +β2 Returnt-2 +… + β5 Returnt-5 +

δ1 HMLt-1 + δ2 HMLt-2 + … + δ5 HMLt-5 + εt

Returnt = α + β1 Returnt-1 +β2 Returnt-2 +… + β10 Returnt-10 +

δ1 HMLt-1 + δ2 HMLt-2 + … + δ10 HMLt-10 + εt

Expected Results: If it can be shown that any of the βn and δn values are significantly different from zero, we can then reject the null hypothesis that the market exhibits short-term efficiency where the short-term period is represented by the respective lag.

(22)

Hypothesis 2: Candlestick reversal patterns have predictive power. However,

a. The predictive ability of candlestick reversal patterns weakens with increasing stock returns volatility.

b. The predictive ability of candlestick reversal patterns strengthens with the confirming short-term preceding momentum.

The predictive power of each of the 3-day candlestick patterns in the overall European market as well as in individual markets can be tested using dummy variable OLS regressions with and without some control variables and their respective interaction terms as follows:

HPR = α + β1 Pattern + ε

HPR = α + β1 Pattern + β2 Volatility + β3 Down Pre Trend + β4 Up Pre Trend +

Β5 Pattern x Volatility + β6 Pattern x Down Pre Trend + β7 Pattern x Up Pre Trend + ε

Where Pattern denotes the dummy variable for each pattern, Up Pre Trend and Down Pre Trend denote dummy variables for the preceding trends, and Volatility denotes the time-varying stock return volatility over the most recent 30-day period.

The logic of the above regression is as follows. Suppose that a short-term trend (Up, Down, No Trend) is observed from time t-6 to time t before the start of the candlestick pattern. The candlestick pattern is then observed over time t+1, t+2 and t+3. By the end of day t+3, a trading signal (Buy, Sell, No Signal) is generated, a trading position is opened at the start of day t+4 and held until day

t+6, t+9 and t+14 when the position is closed for a holding period of 2, 5 and 10 days respectively.

The returns of each of these holding periods are then measured and are indeed the dependent variables in the above regression.

Expected Results:

There are three sub-parts to the second hypothesis as follows (together with the respective coefficients of interest in the above regression equation):

1. Candlestick reversal patterns have predictive power. (β1)

2. This predictive ability weakens with increasing stock returns volatility. (β5)

(23)

The direction and significance of these coefficients of interest would allow us to form a conclusion regarding the validity of the second hypothesis:

1. For the first sub-part, due to a buy-and-hold perspective as explained earlier, we would expect β1 to be positive for profitable buy signals and negative for profitable sell signals in order to conclude that a particular candlestick pattern is likely to have predictive power in a particular market. This is because each type of candlestick reversal patterns we are looking at is only indicative of a specific short-term direction the stock price is heading, i.e. either a buy signal or a sell signal.

2. For the second sub-part, β5 is expected to be negative for buy signals and positive for sell signals, which means that the higher the stock returns volatility is, the lower the returns generated by the candlestick patterns’ signals are and hence the weaker the predictive ability of these patterns will be.

3. For the final sub-part, we would expect β6 to be positive and β7 to be negative. This means that pattern signals with the confirming short-term preceding trend, either a buy signal with a prior down trend or a sell signal with a prior up trend, have even greater predictive power than the signals on their own. In other words, the short-term trend preceding the formation of each candlestick pattern plays an important part in the formation, and hence, the predictive ability of the candlestick reversal patterns.

Hypothesis 3: The combination of candlestick reversal patterns and the short-term preceding trend outperforms the short-term momentum strategy itself.

To test the last hypothesis, for simplicity and consistency, we will apply traditional tests for significance (standard t-statistics) similar to those used in Brock et al. (1992) and Terrence (1997), except that instead of the general tests of differences in the mean returns between buy and sell signals, we will compare the mean buy returns and mean sell returns between the two mentioned strategies – namely the candlestick reversal patterns with short-term preceding trend strategy and the short-term momentum strategy. This method will be applied to each of the candlestick patterns in the overall European market as well as in each of the individual markets. At the same time, the t-statistics for the mean buy or sell returns for each strategy with respect to the unconditional mean will also be provided, which would give a general idea of how each strategy performs as compared to the simple buy-and-hold strategy.

(24)

Following Brock’s definitions, the t-statistics for the mean returns of each strategy with respect to the unconditional mean are computed as follows:

µ_𝑆− µ (𝜎2_{/𝑁 + 𝜎}2_/𝑁

𝑆)1/2

where µS and NS are the mean return and number of signals for the strategy, and µ and N are the

unconditional mean and total number of observations for the sample. 𝝈𝟐 is the estimated variance for

that entire sample. For the mean returns between the candlestick patterns strategy and the short-term momentum strategy, the t-statistics are computed as follows:

µ𝐶− µ𝑀

(𝜎2_/𝑁

𝐶+ 𝜎2/𝑁𝑀)1/2

where µC and NC are the mean return and number of signals for the candlestick patterns strategy, and

µM and NM are the mean return and number of signals for the short-term momentum strategy.

Expected Results: The standard t-statistics test is expected to show that the differences in the mean

buy returns and mean sell returns of the two strategies mentioned above to be significantly different from zero, with the former displaying higher mean returns (more positive for buy signals and more negative for sell signals) than the latter strategy. In other words, this would show that the former strategy outperforms the latter strategy in predicting stock price movements. In addition, the t-statistics comparing the mean returns of each strategy with respect to the unconditional mean would also ideally be expected to be positive for buy signals and negative for sell signals, which would indicate that these strategies outperform the simple buy-and-hold strategy.

(25)

IV. EMPIRICAL RESULTS

1. Descriptive Statistics

Appendix A reports the general descriptive statistics of the main data. Table I displays summary statistics for 2-day, 5-day and 10-day holding period returns for the full sample as well as the three sub-periods for the overall European market. The holding period returns (HPRs) are measured as log differences of the individual stock prices, represented in this study by the total return index, over that specific time period (2, 5 or 10 days). Overall, we could see that the HPRs show some signs of negative skewness and are strongly leptokurtic in all periods examined and for all three holding periods of returns, although as the holding horizon increases, kurtosis tends to decrease significantly for the first 2 sub-periods (1984-1993 and 1994-2003) and remains largely the same for the 2004-2014 sub-period. Volatility, measured by standard deviation in this case, is generally large but seems to be significantly larger in the 1994-2003 sub-period than the other sub-periods for all horizon lengths. This coincides with the period of dot-com crisis when many tech firms went burst and generated much uncertainty in the world market. After this sub-period, volatility appears to have declined in the 2004-2014 sub-period, which includes the period of the 2008-2009 financial crisis, but is still slightly higher than the first sub-period. In addition, volatility increases with increasing horizon length, with 10-day HPRs having volatility that doubles that of 2-day HPRs, which suggests that the mean returns for shorter holding periods are more certain than those for longer holding periods.

Similar results can be observed with all the individual European markets under study, with the only exception of the Spanish market. As shown in table II which displays similar summary statistics for IBEX35 market, volatility appears to be lowest for the 1994-2003 sub-period despite the dot-com bubble, as opposed to results observed for all other markets during the same period. Moreover, kurtosis during the 2004-2014 sub-period for this market is exceptionally large and increases substantially when holding period increases from 2 to 5 days.

Table III contains a breakdown of the frequency, in number and relative percentage, of each of the 10 candlestick patterns observed in each of the 7 European markets for the full sample. Generally, over the 30-year sample period, the two patterns that are consistently observed with the highest frequency (0.7% - 1%) are The White Soldiers and The Black Crows, followed by Morning Star and

(26)

Evening Star patterns (0.5% of the time). The other 6 patterns are observed only roughly 0.1% of the time. The low frequencies with which these patterns are observed indicate that opportunities arising from these patterns are not frequent and traders should not expect to take advantage of this type of trading technique on a daily basis. However, they do not necessarily rule out the hypothesis that trading based on these patterns is profitable.

2. Market Efficiency Test Results

Tables 1 and 2 below and Appendix B display the regression results of the market efficiency tests for the overall market as well as for all individual markets. Table 1 below reports the results for both the autoregressive (AR) and autoregressive distributed lag (ADL) models that test market efficiency in the overall European market only for the 2004-2014 sub-period. This table shows strongly significant coefficients for most of the lagged values of returns and candlestick lengths of the ADL(2,2) and ADL(5,5) models and suggests that daily returns are dependent on up to the most recent 5 days’ returns and candlestick lengths. This implication, however, is only true for the 2004-2014 sub-period in that the coefficients of the lagged values of both the returns and candlestick lengths are significantly different from zero. The candlestick lengths for all the other sub-periods have coefficients that are insignificant. Upon further examining the results for all the individual markets, the significant results for the overall market seem to come mainly from the UK market which displays similar results for the same sub-period. For simplicity, only the results of the overall European market are included. The results suggest that both the daily returns and the intra-day length of the candlesticks, measured by the difference between the high price and the low price during a trading session, have short-term predictive power in the UK market during the 2004-2014 period.

As the ADL model does not provide any additional interesting insights with regards to other markets except the UK during the 2004-2014 period, detailed regression results for only the AR model with 2, 5 and 10 lags of returns for the full sample and three sub-periods for all markets are included in table I through to table VII under Appendix B. Table 2 below gives an overall view on the results of this test for the overall European market. It shows that in general, the coefficients of the lagged returns are strongly and consistently significant at 1% significance level up to the sixth lag and across most sub-periods for the overall European market. The seventh to tenth lags appear to be rather insignificant and thus, are not predictive of future prices. The behavior of the seventh to tenth lags seems consistent across all individual markets.

(27)

Table 1 – Testing Market Efficiency in Overall European Market – AR & ADL Models

Below table reports the AR & ADL regression results for the market efficiency test for the Overall European market. All regressions are performed with robust standard errors.

2004 - 2014

AR(10) ADL(10,10) AR(5) ADL(5,5) AR(2) ADL(2,2) 1st lag Return 0.0126*** 0.00993*** 0.0130*** 0.0128*** 0.0130*** 0.0124*** (4.62) (3.33) (4.78) (4.51) (4.78) (4.51) 2nd lag Return -0.0217*** -0.0208*** -0.0218*** -0.0239*** -0.0215*** -0.0225*** (-7.65) (-7.01) (-7.70) (-8.40) (-7.62) (-8.22) 3rd lag Return -0.0272*** -0.0351*** -0.0268*** -0.0318*** (-9.65) (-11.47) (-9.54) (-11.07) 4th lag Return 0.0106*** 0.0151*** 0.0111*** 0.0139*** (4.59) (5.88) (4.79) (5.77) 5th lag Return -0.0288*** -0.0300*** -0.0288*** -0.0299*** (-11.56) (-11.08) (-11.57) (-11.53) 6th lag Return -0.0128*** -0.0120*** (-5.15) (-4.36) 7th lag Return 0.000314 0.00334 (0.13) (1.22) 8th lag Return -0.000876 0.00221 (-0.41) (0.95) 9th lag Return -0.00313 -0.00626* (-1.37) (-2.48) 10th lag Return -0.00472* -0.00668** (-2.04) (-2.69) 1st lag HighMinusLow -0.0000103** -0.000013*** -0.000014*** (-2.72) (-3.64) (-5.06) 2nd lag HighMinusLow 0.00000859* 0.00000686* 0.00000786** (2.32) (2.09) (2.84) 3rd lag HighMinusLow 0.000013*** 0.000012*** (3.34) (3.39) 4th lag HighMinusLow -0.00000458 -0.00000697* (-1.41) (-2.36) 5th lag HighMinusLow -0.00000173 -0.00000678* (-0.51) (-2.18) 6th lag HighMinusLow -0.000022*** (-6.03) 7th lag HighMinusLow -0.00000223 (-0.63) 8th lag HighMinusLow 0.000011*** (3.37) 9th lag HighMinusLow 0.00000230 (0.60) 10th lag HighMinusLow -0.00000267 (-0.80) Constant 0.00037*** 0.00037*** 0.00037*** 0.00033*** 0.00035*** 0.000333*** (14.69) (11.65) (14.69) (10.90) (14.23) (11.41) N 749976 619936 750176 675693 750296 715339 Adj. R-sq 0.002 0.003 0.002 0.003 0.001 0.001

(28)

Table 2 – Testing Market Efficiency in Overall European Market – AR Model

Below table reports the AR regression results for the market efficiency test for the Overall European market. All regressions are performed with robust standard errors.

t statistics in parentheses * p<0.05, ** p<0.01, *** p<0.001

All Periods 1984 - 1993 1994 – 2003 2004 - 2014

AR(10) AR(5) AR(2) AR(10) AR(5) AR(2) AR(10) AR(5) AR(2) AR(10) AR(5) AR(2) 1st lag Return 0.0290*** 0.0293*** 0.0295*** 0.0655*** 0.0658*** 0.0659*** 0.0306*** 0.0311*** 0.0316*** 0.0126*** 0.0130*** 0.0130*** (14.29) (14.48) (14.47) (13.41) (13.50) (13.52) (8.35) (8.49) (8.50) (4.62) (4.78) (4.78) 2nd lag Return -0.0148*** -0.0149*** -0.0152*** -0.00352 -0.00355 -0.00435 -0.0135** -0.0135** -0.0138** -0.0217*** -0.0218*** -0.0215*** (-6.24) (-6.29) (-6.43) (-1.00) (-1.02) (-1.25) (-2.70) (-2.71) (-2.78) (-7.65) (-7.70) (-7.62) 3rd lag Return -0.0236*** -0.0234*** -0.0114*** -0.0114*** -0.0261*** -0.0258*** -0.0272*** -0.0268*** (-12.32) (-12.21) (-3.37) (-3.38) (-7.51) (-7.41) (-9.65) (-9.54) 4th lag Return 0.00624*** 0.00647*** 0.0128*** 0.0124*** -0.00319 -0.00269 0.0106*** 0.0111*** (4.24) (4.39) (4.01) (3.90) (-1.37) (-1.15) (4.59) (4.79) 5th lag Return -0.0199*** -0.0203*** 0.00300 0.00229 -0.0205*** -0.0211*** -0.0288*** -0.0288*** (-13.24) (-13.56) (0.97) (0.74) (-9.15) (-9.46) (-11.56) (-11.57) 6th lag Return -0.0144*** -0.0165*** -0.0168*** -0.0128*** (-9.82) (-5.61) (-7.78) (-5.15) 7th lag Return 0.0000804 0.00539 -0.00343 0.000314 (0.05) (1.81) (-1.53) (0.13) 8th lag Return 0.00443** 0.00505 0.00910*** -0.000876 (3.25) (1.86) (4.06) (-0.41) 9th lag Return 0.00284 0.0184*** 0.00196 -0.00313 (1.86) (6.38) (0.73) (-1.37) 10th lag Return -0.00153 0.0174*** -0.00777*** -0.00472* (-1.08) (5.12) (-3.83) (-2.04) Constant 0.00041*** 0.00041*** 0.00040*** 0.00052*** 0.00054*** 0.00055*** 0.00038*** 0.00037*** 0.00035*** 0.00037*** 0.00037*** 0.00035*** (23.98) (24.21) (23.69) (15.80) (16.77) (17.34) (11.61) (11.46) (10.99) (14.69) (14.69) (14.23) N 1661548 1662948 1663788 369991 370846 371359 541581 541926 542133 749976 750176 750296 Adj. R-sq 0.002 0.002 0.001 0.006 0.005 0.004 0.003 0.002 0.001 0.002 0.002 0.001

(29)

Tables I to table VII report the breakdown of the same regression results across all individual markets. We can see that the coefficients are strongly significant up to the fifth lag, albeit with some lags at 10% significance level and at most one insignificant lag, for all markets except for Spanish (IBEX35) market with only 2 significant lags. The sixth lagged returns are only significant for the UK market, which contributes to the result observed for the overall market. In terms of sub-periods, it is found that the extent of significance of the above results appears to increase with time. This means that most markets, with the exception of the German market which shows the other way around, exhibit greater market inefficiency in the later sub-period (2004-2014) than in the earlier sub-period (1984-1993). This goes in line with findings by Terence (2007) who found that the trading rules that he tested worked for the London Stock Exchange FT30 index at least up to the early 1980s and became less effective ever since (up to 1994 as per his study). Clearly, if the market became more efficient, or less predictive, during this period (early 1980s-1994), trading techniques would not have been able to generate returns greater than a simple buy-and-hold strategy.

Therefore, based on the above results, it can be concluded that the European market as a whole exhibits short-term (up to 5 lag days) market inefficiency during all periods and sub-periods examined. For the individual markets, however, all markets except for Spain display varying degree of short-term market inefficiency for the full sample period. For Spain in particular, only the first lag return is strongly significant in all periods, which suggests that the market is only inefficient in an extremely short horizon and it would be difficult to exploit any of the 3-day candlestick reversal patterns to generate abnormal returns. In addition, most of the markets seem to exhibit greater degree of market inefficiency in the later sub-periods than in the earlier sub-periods, except Germany which appears to have greater market inefficiency in the earlier sub-periods. This would mean that if the candlestick patterns work, they are expected to have more predictive power in general during the later sub-periods as compared to the earlier sub-periods. In addition, since it has been shown that the market displays much greater level of inefficiency during the shorter holding periods, Hypothesis 2 and 3’s results will be focused only on the 2-day and 5-day holding periods.

(30)

3. Candlestick Patterns’ Predictive Power Results

3.1. General t-tests Results

Appendix C provides the results for all the t-test and regressions used to test the predictive power of the 10 candlestick patterns in the European market. Tables I to IV display the standard two-sided t-test results for the mean comparison of the candlestick patterns, with and without preceding trend, and the short-term momentum strategies as well as the comparison of their means with regards to the unconditional mean for the overall European market.

3.1.1. Mean comparison with Unconditional Mean

Tables I to IV of Appendix C contain significantly sufficient evidence indicating that the short-term momentum (STM) strategy consistently underperforms the simple buy-and-hold strategy in the overall European market. This point is illustrated in column STM – μ, computed by taking the difference between STM mean and unconditional mean. The summary of this column’s results is shown in table 3 below. Generally, the results report negative difference for buy signals and positive difference for sell signals, except for sub-period 1984-2003 when the short-term momentum strategy outperforms the buy-and-hold strategy by roughly 0.06 percentage points for 2-day holding periods and 0.15 percentage points for 5-day holding periods for both buy and sell signals.

Table 3 – Mean Difference between Short-term Momentum Strategy and Unconditional Mean

Below table reports the mean difference between the Short-term Momentum Strategy and the unconditional mean for the Overall European market. The stars display the significance level of the standard two-sided t-tests.

2-day HPR 5-day HPR

Periods STM (Buy) – μ STM (Sell) – μ STM (Buy) – μ STM (Sell) – μ Full Sample -0.0006*** 0.00096*** -0.00116*** 0.00161*** (-11.04375) (16.41335) (-13.56755) (17.52885) 1984-1993 0.00063*** -0.00051*** 0.00108*** -0.00171*** (6.09529) (-4.55846) (6.4858) (-9.58826) 1994-2003 -0.00096*** 0.00159*** -0.00219*** 0.00276*** (-9.04201) (14.39591) (-13.19512) (15.98921) 2004-2014 -0.00094*** 0.00124*** -0.00152*** 0.00244*** (-12.04869) (14.2376) (-12.54725) (18.06446)

(31)

In addition, tables I to IV of Appendix C also suggest, through columns Pattern with Trend – μ and

Pattern Only – μ, that while candlestick patterns with confirming preceding trend do have some,

albeit few, evidence of superior performance over the simple buy-and-hold strategy, the candlestick patterns alone regardless of the prior trend display significant signs of outperformance over the same strategy by up to 0.4 percentage points, a proof that candlestick patterns do have predictive power. Results for the 2-day holding periods and full sample are summarized in Table 4 below. For patterns with confirming trends, the Three Inside Up and Three Outside Up patterns significantly outperform the buy-and-hold strategy by 0.35 percentage points while the Three Black Crows and Evening Star patterns outperform the buy-and-hold strategy by 0.12 and 0.4 percentage points. For patterns only, all patterns except for the Three White Soldiers, Three Black Crows and Three Inside Down display significant outperformance over the buy-and-hold strategy by up to 0.4 percentage points.

Table 4 – Mean Difference between Patterns (with & without confirming trend) and Unconditional Mean

Below table reports the mean difference in the 2-day HPRs between the candlestick patterns and the unconditional mean for the Overall European market and the full sample. The stars display the significance level of the two-sided t-tests.

Patterns Pattern w/ trend - μ Pattern Only - μ Patterns Pattern w/ trend - μ Pattern Only - μ TWS 0.00053 -0.00072** TBC -0.00121** -0.0003 (0.78318) (-2.43502) (-2.34428) (-1.21021) TIU 0.00341** 0.00197*** TID 0.00282* -0.00011 (2.47092) (2.95261) (1.89724) (-0.15079) TOU 0.00363*** 0.00126** TOD 0.00102 -0.00198*** (3.12019) (2.37765) (0.99622) (-3.97809) MS 0.00052 0.00148*** ES -0.00395*** -0.00395*** (0.61677) (3.86805) (-4.7561) (-9.75244) MDS 0.003* 0.00291*** EDS -0.0012 -0.00354*** (1.84922) (4.01875) (-0.74124) (-4.53168)

This predictive power, however, decreases with longer holding periods, as evident in the lower significance of the t-tests in panel B and D of table 5 and 6 below. These tables, together with table V and VI of appendix C, display a summary of these mean differences across all markets and all sub-periods. Noted exceptions are the Three White Soldiers and Three Black Crows which show substantial evidence of underperformance over the buy-and-hold strategy across most markets and sub-periods. The candlestick patterns also seem to have better predictive power in the UK market than in most other markets while the Evening Star pattern especially appears to be effective across almost all markets with the outperformance estimated at around 0.4 percentage points on average.

(32)

Table 5 – Mean Difference between Pattern with Confirming Trend and Unconditional Mean (FULL SAMPLE) Panel A – Buy Signals & 2-day HPRs

Patterns France CAC40 German DAX30 Netherlands AEX Spain IBEX35 Sweden OMXS30 Swiss Market Index UK FTSE100 Overall European TWS 0.00019 -0.00041 0.0003 -0.00137 -0.00619 0.0034 0.00095 0.00053 (0.09951) (-0.19101) (0.11122) (-0.55428) (-1.63223) (1.58111) (1.00542) (0.78318) TIU 0.00391 -0.00037 -0.00943** 0.01073** 0.00363 0.00281 0.00544*** 0.00341** (1.01183) (-0.1025) (-1.99193) (2.14844) (0.60739) (0.58514) (2.60667) (2.47092) TOU 0.00702** 0.00264 0.00547 0.00062 0.01099* 0.00094 0.00297* 0.00363*** (2.0934) (0.81904) (1.1693) (0.13071) (1.87037) (0.23751) (1.82312) (3.12019) MS -0.00086 0.00018 0.00061 -0.0037 0.00029 0.00336 0.00197 0.00052 (-0.37505) (0.08111) (0.21552) (-1.25751) (0.09467) (1.32615) (1.42119) (0.61677) MDS 0.00951 0.00142 0.0011 0.00714 0.00197 0.00638 0.00207 0.003* (1.56869) (0.31497) (0.19627) (1.21702) (0.35997) (1.30703) (0.85843) (1.84922) Panel B – Buy Signals & 5-day HPRs

Patterns France CAC40 German DAX30 Netherlands AEX Spain IBEX35 Sweden OMXS30 Swiss Market Index UK FTSE100 Overall European TWS -0.00416 0.0013 -0.0026 -0.00039 -0.01123* 0.0023 -0.00106 -0.00126 (-1.41955) (0.38935) (-0.60968) (-0.09861) (-1.89847) (0.68917) (-0.71534) (-1.18817) TIU -0.00209 0.00869 -0.02876*** -0.00488 -0.01071 -0.00324 0.00463 -0.00083 (-0.34776) (1.55357) (-3.8903) (-0.61412) (-1.14931) (-0.43463) (1.41132) (-0.38165) TOU 0.00563 0.00458 0.01177 0.00224 0.00793 0.00345 0.00233 0.00377** (1.07887) (0.90809) (1.61153) (0.2995) (0.86481) (0.56408) (0.90941) (2.06307) MS 0.00486 -0.00233 -0.00739* -0.00784* 0.00676 0.00192 -0.00091 -0.00051 (1.37006) (-0.67623) (-1.65993) (-1.67415) (1.39639) (0.48835) (-0.41794) (-0.38318) MDS 0.00451 0.00027 0.00312 0.00722 0.00513 0.01018 0.00229 0.00366 (0.47804) (0.03797) (0.35499) (0.77438) (0.60175) (1.34389) (0.60343) (1.43972)

Technical analysis via candlestick patterns : an empirical study on European stock markets

Master Thesis | Anh Nguyen - 10839186