The effect of market regulation on the profitability of technical analysis

(1)

The Effect of Market Regulation on

the Profitability of Technical Analysis

Ruben van den Eshof 10252584 31 June 2018

Abstract

In this paper, we find evidence that technical analysis outperforms buy-and-hold strategies on 'all-cap' indices. In addition, we find that higher government effectiveness has a significant negative effect on the profitability of technical analysis. For longer lag lengths of the moving average indicator, we find that domestic regulatory quality has a significant positive effect on the profitability of technical analysis.

Dhr. S. Kucinskas 2017/2018

Master Thesis Seminar Semester 2, period 3

(2)

2

Statement of Originality

This document is written by Ruben van den Eshof (10252584) who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are 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)

3

1 Introduction

One area of recurring disagreement between professionals and academics in the field of finance is the economic significance of technical analysis’s predictive power. Whereas traders have been using technical analysis to predict the profitability of investments for 30 years, finance academics question the merit of this technique because its basic assumptions are not in line with economic theory. Since the introduction of technical analysis to the financial literature, academics have been critical. Burton Malkiel, writer of the finance book A Random

Walk Down Wall Street, describes technical analysis as a way for brokers to profit at the

expense of their customers. In later writings, he refers to technical analysis as ‘voodoo

finance’ (Lo et al., 2000). The core of the disagreement between professionals and academics, is that technical analysis is not in line with the (weak-form) efficient market hypothesis.1 Whereas technical analysts believe in the predictability of excess returns, academics argue that asset returns take a random walk and are therefore unpredictable (Fama, 1965). The efficient market theory has long been disproved, and in the late 1990s, even Burton Maikiel had to re-evaluate his most conservative assumptions about the efficient market hypothesis (Lo et al., 2009).

Various studies have shown the relationship between past and future asset returns. For example, Rapach et al. (2013) have found that lagged US stock returns significantly predict stock returns in developed countries other than the US. In addition, a portfolio consisting of long positions in stock with a positive past six-month return and short positions in stock with a negative six-month return would result in a significant, abnormal portfolio return (Jegadeesh & Titman, 1993). Such portfolios yielded an excess yearly return of 12% in the six months after the initial investment was made between 1965 and 1989. These examples show that concepts such as price rationality and price efficiency are not accurate. Professional traders frequently use technical analysis to benefit from these price irregularities (Gehrig & Menkhoff, 2006). In fact, Gehrig and Menkhoff (2006) have discussed the results of two surveys they conducted over a span of 10 years, finding that technical analysis has overtaken fundamental analysis as the primary trade technique used by forex traders. Fund managers use technical analysis as their secondary tool to find profitable trading opportunities. These

groups of traders believe that market psychology impacts price in ways that can be predicted by technical analysis.

1

(5)

5

However, if asset markets are manipulated and the effect of market psychology no longer holds, technical analysis becomes useless. Technical analysis only predicts price movements if the underlying asset moves freely (Chou, 2011). Most developed economies have their own regulatory authorities to prevent market manipulation. For example, the Securities Exchange Act heavily fines and criminalizes information-based price manipulation in the US (Allen & Gale, 1992). The severity of such penalizations varies widely between different countries. Therefore, the question arises: to what extent does asset market regulation influence the profitability of technical analysis?

We structure the continuation of this paper in the following way: In Section 2, we expand on the existing literature on technical analysis and market manipulation. Section 3 explains the dataset and methodology. Section 4 discusses the results of the regressions and expands further on the subject by adding two case studies on manipulated stock. In the last section, we reach a conclusion.

2 Literature review

The first part of this literature review focuses on the underlying drivers of technical analysis and the proof of its profitability. The second section expounds upon market manipulation and how it influences the signals gleaned by technical analysis. In the final section, we focus on government policies that might prevent market manipulation.

2.1 The underlying drivers of technical analysis

One of the main drivers behind the predictability of technical analysis is the habit of traders to overreact or underreact to new information. Barberis et al. (1998) find that investors typically underreact to good news in the first 12 months after a positive announcement, due to the behavioural characteristics of conservatism bias.2 Investors overreact to good news in the three to five years after a pattern of good news. This can be attributed to representativeness heuristic.3 The 3-factor model of Fama and French (1993) can account for overreaction, but not for underreaction (Barberis et al., 1998). Technical indicators such as the relative strength index (RSI) and moving average strategies (MA) follow trends and are able to pick up on

2_{Conservatism bias is the behavioural process of prioritizing old beliefs over new information.}

3_{Representativeness heuristic bias is the tendency to overestimate the impact of minor events on the long term.}

For example, novice investors tend to interpreted the short term outperformance of one stock as a meaningful sign for future outperformance (Boussaidi, 2013).

(6)

6

such over- and underreactions in prices (Wilcox & Crittenden, 2009). Trading strategies that follow trends are even more profitable when publicly available information is limited (Zhang, 2006). When there is opaqueness about the credibility of good or bad news signals, feedback trading becomes more prominent. Feedback trading occurs when traders buy an asset as the price is rising (positive- feedback trading) and sell an asset when the price is declining

(negative-feedback trading). In other words, a limit to public information increases over- and underreaction (Jegadeesh & Titman, 1993). The majority of the literature on momentum focuses on the behavioural characteristics of the trader as the driver behind over- and

underreaction. However, Zhou and Zhu (2013) construct an equilibrium model in which they prove that both time-series momentum and moving-average price predictability can exist in a rational setting.

The most popular and widely used technical analysis rule is the moving average. The moving average rule signals a profitable trade when the price of an asset exceeds the moving average value. Conversely, it produces a ‘sell’ signal when the current price moves under the moving average value (Han et al., 2013). Empirical research on the Dow Jones Index from 1897–1986 shows highly significant excess returns for the moving average trading strategy (Brock et al., 1992). Following the moving average ‘buy’ signal resulted in an annual return rate of 12% over a time span of 90 years. The paper concludes that asset returns might be non-linear, and technical analysis picks up on this unknown pattern. Han et al. (2013) provide a comprehensive extension to the literature on the moving average excess return (MAER) anomaly. They divide all their stock observations across 10 different portfolios, based on the annual volatility of the daily closing prices.4 They implement the MA strategy on the 10 volatility portfolios and compare the returns with the buy-and-hold returns. Their results show that for all 10 portfolios, the MA strategy is significantly more profitable than the buy-and-hold strategy. Specifically, the highest volatility portfolio exhibited the highest excess return for the MA strategy, whereas the lowest volatility portfolio had the lowest excess return. These results are robust to the 3 factors of Fama and French (1993), distance to default, investor sentiment, forecast dispersion, and earnings volatility. The profitability of using trend-following technical indicators like the moving average has proven to exist in several different asset classes. In addition to the previous examples, the MA strategy has been shown to exert both direction predictions and profitable trading signals in stock (Wilcox &

Crittenden, 2009), government risk premiums (Goh et al., 2013), and foreign exchange

4

(7)

7 (LeBaron, 1999).

2.2 Market manipulation

As mentioned in the introduction section, technical analysis works best if prices can move freely. Once assets are manipulated, the signals exerted by technical analysis become less reliable (Chou, 2011). One of the best examples to illustrate how technical trading signals are diluted by manipulation is ‘pump and dump’ schemes. In a ‘pump and dump’ scheme,

colluders artificially inflate the price of an asset. In this process, positive-feedback traders are attracted to the asset because they believe the colluders have inside information in favour of the asset. When enough positive-feedback traders have bought into the asset, the colluders dump the asset to unknowing market participants (Asim and Atif, 2003). In such schemes, trend-following technical indicators would signal a ‘buy’ opportunity, while in reality market participants would buy into a losing trade. Such schemes are more common in countries with weak market regulations (Asim and Atif, 2003). A more common form of market

manipulation is insider trading. Insider trading is when company insiders sell or buy stocks of their own firms based on non-public information.5 One can imagine that buying stock in one’s own firm on the basis of insider knowledge is detrimental to other market participants.

Although insiders are allowed to trade their own stock, they have a mandatory duty to disclose those trades. Insider trades that are not disclosed are punishable by law (Aggarwal, 2003). Another way to manipulate the market is the ‘short squeeze’. During a short squeeze, a group of colluders acquires a large portion of the supply of a specific asset. Due to the

combination of constant demand and lack of free-moving supply, the price of this asset surge to irrational high prices. At the peak asset price, the colluders unload the shares on other market participants, which leads to a price crash. A primary example of a ‘short squeeze’ is the case of Salomon Brothers in 1991 (Jordan and Jordan, 1996). The US Treasury issued 12.3 billion two-year T-bills. By using escrow accounts and over-the-counter markets,

Salomon Brothers cornered the market and gained control over 86% of the total supply.6 Due to the lack of supply owned by other market participants, the price surged from $0.16 to $0.25 in a matter of days. In this instance, the Securities and Exchange Commission suspected a short squeeze and started an investigation. By US law, a single entity is allowed to own only

5_{The term ‘insider’ is defined by domestic regulatory authorities, and differs slightly per country. The American}

SEC defines ‘insider’ as anyone who is in possession of non-public information (SEC, 2011).

6

‘Cornering the market’ refers to acquiring a controlling fraction of the supply. Controlling this fraction allows the owner to influence the price.

(8)

8

35% of a total T-bill auction. Salomon Brothers finally settled with the SEC for $290 million dollar in fines and reparations (Jordan and Jordan, 1996).

2.3 Government policy

Whereas the SEC intervened in this very instance, market manipulation practices occur on a regular basis. The only cases of market manipulation known to the public occur when regulatory authorities actually come to action. In emerging markets, much more market manipulation takes place due to weaker government regulation (Aggarwal, 2003). In addition, the ease of spreading false information to manipulate prices has increased in frequency due to the popularity of the Internet. Market manipulation bears additional costs to unknowing market participants by increasing price volatility due to sudden price surges and crashes (Aggarwal, 2003). Even though the profitability of technical analysis increases with price volatility (Han et al., 2013), trend-following trading strategies like the moving average become less reliable when the underlying asset is manipulated. Therefore, the question arises: does the degree of market regulation influence the profitability of technical analysis? We hypothesize that the profitability of technical analysis is higher in a market without manipulation.

3 Data and methodology

In order to test our research question, we select 24 countries that constitute both emerging and developed economies. Our sample includes 10 European countries, 8 Asian countries, 2 Latin-American countries, 2 African countries, Canada, and the United States of America. A full list of the selected countries and their descriptive statistics can be found in Appendix 1. For these countries we have collected data on the ‘all-cap’ indices from 2004 to 2016. This is the value-weighted index per country that includes all publicly listed stocks. These index calculations are provided by the FTSE and are extracted from Datastream. Then, we conduct technical analysis on every calendar year, for every ‘all-cap’ index, and compare the returns of technical analysis to the returns of a buy-and-hold strategy.

(9)

9 3.1 Trading strategy

As a tool for the measurement of the profitability of technical analysis, we use the moving average timing strategy. Moving average strategies are widely used in technical analysis worldwide (Brock et al., 1991) and provide buy and sell signals that have proven to be more profitable than buy-and-hold strategies (Han et al., 2013). Therefore, we use their method for the calculation of the 10-day MA and their associated trading strategy:

(1) 𝑀𝐴_{𝑥,𝑡,𝐿}

=

𝑃𝑥,𝑡+𝑃𝑥,(𝑡−1)+⋯+𝑃𝑥,(𝑡−(𝐿−1))+𝑃𝑥,(𝑡−𝐿)

𝐿

In equation (1), 𝑀𝐴_{𝑥,𝑡,𝐿} is the moving average of index x, on day t of lag L. Further, 𝑃_𝑥,𝑡 is the price of the index of country x on day t. As default settings for the moving average timing strategy, we use the 10-day moving average (L = 10). The 10-day MA strategy gives the strongest trading signals. A comparison between the strengths of the trading signals for L = 10, L = 20, and L = 50 can be found in Appendix 2. We then use the following trading

strategy: if the closing price is larger than the 10-day moving average on the previous day, we will invest in the index. If the closing price of the previous day is smaller than the 10-day moving average, we will invest in T-bills. The T-bills earn a daily interest of 0.0067%. This is the average daily T-bill return over our sample from 2004 till 2016 (Treasury.gov, 2018). The cumulative yearly return of the 10-day MA strategy is then compared with the cumulative yearly return of the buy-and-hold strategy per year, per country. This is the formula for the MAER:

(2) 𝑀𝐴𝐸𝑅𝑥,𝑇,𝐿 = 𝑅(𝑀𝐴)𝑥,𝑇,𝐿− 𝑅(𝐵𝐻)𝑥,𝑇,𝐿

In equation (2), 𝑀𝐴𝐸𝑅_{𝑥,𝑇,𝐿} is the MAER of country x in year T of lag L, while 𝑅(𝑀𝐴)_{𝑥,𝑇,𝐿} is the cumulative yearly return of the MA strategy of country x in year T of lag L. Finally,

𝑅(𝐵𝐻)_{𝑥,𝑇,𝐿} is the cumulative yearly return of the buy-and-hold strategy of country x in year T of lag L. Appendix 3 offers a visualization of our trading strategy.

Since the moving average trading rules trades fairly frequent, we have included Table 1 to dilute any concerns about transactions costs. The MA(10) rule has a turnover of roughly

(10)

10

39 trades per year, and has an average yearly return of 15.17% (measured over 13 years over 24 countries). This means that if the average transaction cost equal 38.62 basis points per buy or sell order, the trading rule breaks even. In this methodology we adopt the assumptions of Han (2006) & Lynch and Balduzzi (2000), who assume that trading T-bills does not lead to trading expenses. In addition, Lynch and Balduzzi (2000) use transaction costs of 25 basis points. In the present time 25 basis point may be a bit conservative. We can assume that due to the advance in technology those transactions costs have decreased. This would imply that our trading strategy is still profitable after deducting transaction costs. Since the MA(20) and MA(50) rule are slower moving averages, they adapt less quickly to trends. Therefore they make fewer trades and can handle higher transactions costs to break even.

Table 1. This table demonstrates transactions costs that would make the moving average strategy

break even. The average return is the yearly average of all yearly country observations. The transactions costs are shown in basis points.

Trades/Year Av. Return Max. bp

MA(10) 39.29 15.17% 38.62

MA(20) 26.03 14.21% 54.58

MA(50) 15.01 13.05% 86.96

Because we use ‘all-cap’ indices of entire countries, it is difficult to calculate daily bid-ask spreads. For example, the ‘all-cap’ index of India contains 5144 different stock. However, previous papers have shown that (the lack of) liquidity is not the underlying factor that results in the excess return of the MA trading rule. Han et al. (2013) use the liquidity factor described by Pastor and Stambaugh (2003). They include this liquidity factor in a model with the 3 factors of Fama and French (1993). They find that the liquidity factor provides no significant statistical explanation for the excess return of the MA trading rule. To more clearly depict the profitability of the MA strategy, we look at the

determinants of the MAER. We divide the yearly MAER observations into four groups, based on the yearly volatility of the daily closing prices. Group 1 contains the observations with the lowest volatility, and Group 4 contains the observations with the highest volatility. Previous papers found that the profitability of the MA strategy increases as the volatility of an asset increases (Han et al., 2013). However, we conduct this experiment on value-weighted ‘all-cap’ indices of entire countries. As such, volatility is much lower, since the ‘all-‘all-cap’ index is highly diversified (Rumelt, 1982). Therefore, we expect a lower degree of profitability of the MA strategy, and small differences in profitability between the four volatility portfolio’s.

(11)

11

To adjust for market risk and momentum, we use the 3-factor model and the

‘momentum’ factor (Fama and French, 1993, 2012). The regular approach is to only use the 3-factor model, since the ‘momentum factor’ is insignificant in most models. However, technical analysis indicators rely partly on positive-feedback trading (Hong and Stein, 1999; Edmans et al., 2012), so therefore the ‘momentum factor’ should be a useful addition. Since we are dealing with an international sample, we choose to include the global version of these 4 factors. In this 4-factor model, ‘MKT’ represents the adjustment for global market risk. ‘MOM’ is the term for momentum; this corrects for the pattern that rising stocks rise more, and falling stocks to fall more. The factor ‘SMB’ corrects for the outperformance of small firms against big firms, in terms of stock return. Finally, ‘HML’ corrects for the

outperformance of firms with a high market ratio against firms with small book-to-market ratios (Fama and French, 1993). We derive the following expression to investigate the drivers of the MAER, and it yielded the following results:

(3) 𝑀𝐴𝐸𝑅𝑥,𝑇,10= 𝑎𝑥+ 𝛽𝑥,𝑀𝐾𝑇𝑟𝑀𝐾𝑇,𝑇+ 𝛽𝑥,𝑀𝑂𝑀𝑟𝑀𝑂𝑀,𝑇+ 𝛽𝑥,𝑆𝑀𝐵𝑟𝑆𝑀𝐵,𝑇+ 𝛽𝑥,𝐻𝑀𝐿𝑟𝐻𝑀𝐿,𝑇+ 𝜀𝑥,𝑇

Table 1. Results of equation (3). These are the descriptive statistics by volatility quartile that expound

the excess return of the 10-day moving average strategy (MA). In this table ‘Group 1’ contains the observations with the lowest yearly volatility in daily closing prices. ‘Group 4’ holds the observations with the highest volatility in daily closing prices. Global market return is represented by MKT, MOM represents momentum, SMB is small-minus-big, and HML is high-minus-low.

Group 1 Group 2 Group 3 Group 4 Sample

Alpha 4.18 (2.41) * −0.09 (0.30) −1.82 (4.52) 13.52 (2.91) *** 5.78 (1.27) *** MKT −0.45 (0.20) ** −0.03 (0.23) −0.07 (0.32) −0.49 (0.11) *** −0.59 (0.07) *** MOM −0.03 (0.15) −0.02 (0.16) 0.21 (0.15) 0.38 (0.14) *** 0.12 (0.06) ** SMB 0.48 (0.22) ** −0.59 (0.30)* 0.08 (0.49) 0.26 (0.66) 0.11 (0.19) HML −0.43 (0.24) −0.55 (0.26) ** −0.65 (0.43) −0.63 (0.48) −0.29 (0.16) * Adj. R^2 0.21 0.10 0.03 0.46 0.30 * = Significant at p = .10 ** = Significant at p = .05 *** = Significant at p = .01

(12)

12

Although the results for Groups 2 and 3 are insignificant, a large difference appears between the alphas for Group 1, Group 4, and the entire sample. In this regression, the alpha represents the excess return of the 10-day MA strategy over the buy-and-hold strategy. The alpha of Group 1 is twice as small as the alpha of Group 4. This is line with the theory that technical analysis is more profitable for assets with higher volatility (Han et al., 2013). In addition, the 10-day MA strategy fares significantly better in a bear market. If we look at the entire sample, for every 1% decrease in the market index, the 10-day moving average return increases by 0.58%. Again, this finding confirms theoretical frameworks on technical analysis (Neely et al., 2013). Finally, positive momentum contributes to the excess return of the 10-day moving average. This contribution can be dedicated to the presence of positive- feedback traders (Hong and Stein, 1999). All in all, the magnitude of the alpha is much larger than the coefficient of MKT and MOM. This difference in size gives us reason to conclude that applying the moving average trading strategies to volatility portfolios results in a good hedge against a market portfolio.

3.2 Quantifying regulatory strength

Now that we have established the presence of a significant alpha, and therefore a significant excess return for the 10-day MA strategy, we can develop the next step of our research. In order to test the correlation between the degree of market regulation and the profitability of technical analysis, we must find the appropriate proxy variables for market regulation. These variables can be found at the World Bank. They determine the World Governance Indicators (WGIs). The World Bank describes these indicators as follows:

‘The Worldwide Governance Indicators (WGI) are a research dataset summarizing the views on the quality of governance provided by a large number of enterprise, citizen and expert survey respondents in industrial and developing countries. These data are gathered from a number of survey institutes, think tanks, non-governmental organizations, international organizations, and private sector firms.’

The WGIs are available for 214 different countries and are independently constructed by a non-profit company. These indicators are proven to represent an accurate representation of the current governance situation in a country, as captured by empirical research (World Bank, 2017). The first indicator we choose is ‘Government Effectiveness’, which reflects the perception of the quality of public services and the degree of its independence from political

(13)

13

pressure, the quality of policy formulation and implementation, and the credibility of the government’s commitment to such policies (World Bank, 2017). The second indicator we choose is ‘Regulatory Quality’, which reflects the perception of the ability of the government to formulate and implement sound policies and regulations that permit and promote private sector development (World Bank. 2017). Both indicators range between -2.5 and 2.5, where 2.5 indicates the highest quality of regulatory facilities and government effectiveness. It is important to note that these indicators mirror the perception of the inhabitants of the quality of the rated country. This means that if the perception of the regulatory quality or governmental effectiveness is high, one is less inclined to manipulate the market, since the likelihood of this manipulation being discovered and punished is perceived to be higher.

Since we want to isolate the effect of ‘Government Effectiveness’ and ‘Regulatory Quality’ on the profitability of technical analysis, we add country fixed effects and

adjustments for risk (Stock and Watson, 2015). We include country fixed effects, the global 3-factor model, and the ‘momentum’ 3-factors from (Fama and French, 1993, 2012). Country fixed effects are included to prevent omitted variable bias in panel data when omitted variables are different in each country (Stock and Watson, 2015). The first country fixed effect we use is the domestic real interest rate. The domestic real interest rate is the interest rate corrected for domestic inflation. The second country fixed effect we use is the ‘market capitalization of all domestic listed companies, as a percentage of the GDP’. We assume that countries with a higher percentage of public companies in relation to their GDP will have better and stronger market regulations. This effect should therefore be isolated. Finally, we include the ‘number of listed firms’ as one of the country fixed effects. The ‘number of listed firms’ differs from ‘market capitalization of listed firms as percentage of GDP’ since it reflects different types of domestic economies. Some countries contain a large number of small companies with a combined market capitalization that is similar to countries in which only a few large companies dominate. An example of the latter is Finland, where Nokia dominates the total market capitalization of all listed firms (OMX, 2017). Prior research has shown that smaller companies have larger stock price volatility (Ang et al., 2009) and hence represent greater profitability for MA trading strategies (Han et al., 2013). For the

implementation of country fixed effects, we use the ‘entity demeaned’ OLS algorithm in Stata, better known as the ‘within estimator’. This algorithm subtracts the entity-specific average from each variable. It subsequently runs the regression on the ‘entity-demeaned’ variables (Stock and Watson, 2015). For risk-adjustment, we again use the global 3 factors and the ‘momentum factor’ described by Fama and French (1993, 2012). These 4 factors are

(14)

14

MKT (i.e., adjustment for global market risk), MOM (i.e., momentum, which represents the tendency for rising stocks to rise more and falling stocks to fall more), SMB (which corrects for the outperformance of small firms against big firms, in terms of stock return), and HML (which corrects for the outperformance of firms with a high book-to-market ratio against firms with small book-to-market ratio) (Fama and French, 1993, 2012). A correlation matrix of all the variables in the equation can be found in the Appendix 4. Our final regression equations looks as follows:

(4) 𝑀𝐴𝐸𝑅𝑥,𝑇,10= 𝑎𝑥+ (𝛽1𝑥𝐺𝑜𝑣𝐸𝑓𝑓) + (𝛽2𝑥𝑅𝑒𝑔𝑄) + 𝑅𝑒𝑎𝑙𝐼𝑛𝑡𝑒𝑟𝑒𝑠𝑡𝑥+ 𝑀𝐶𝐿𝑖𝑠𝑡𝑒𝑑𝑥 +

𝐿𝑖𝑠𝑡𝑒𝑑𝐹𝑖𝑟𝑚𝑠𝑥+ (𝛽𝑥,𝑀𝐾𝑇𝑟𝑀𝐾𝑇,𝑇) + (𝛽𝑥,𝑀𝑂𝑀𝑟𝑀𝑂𝑀,𝑇) + (𝛽𝑥,𝑆𝑀𝐵𝑟𝑆𝑀𝐵,𝑇) + (𝛽𝑥,𝐻𝑀𝐿𝑟𝐻𝑀𝐿,𝑇) +

𝜀𝑥,𝑇

In equation (4), 𝛽1𝑥 is the coefficient for country x of Government Effectiveness, and 𝛽2𝑥 is

the coefficient for country x of Regulatory Quality. 𝑅𝑒𝑎𝑙𝐼𝑛𝑡𝑒𝑟𝑒𝑠𝑡_𝑥 is the intercept for country

x for the real interest rate. 𝑀𝐶𝐿𝑖𝑠𝑡𝑒𝑑𝑥 is the intercept for country x for the market

capitalization of all listed firms, as a percentage of the GDP. 𝐿𝑖𝑠𝑡𝑒𝑑𝐹𝑖𝑟𝑚𝑠_𝑥 is the intercept for country x for the number of publicly listed firms. For the year effects, we use 𝛽_{𝑥,𝑀𝐾𝑇}, 𝛽_{𝑥,𝑀𝑂𝑀}, 𝛽_{𝑥,𝑆𝑀𝐵}, 𝛽_{𝑥,𝐻𝑀𝐿} as the coefficients for the 3-factor model and the ‘momentum factor’, as described in Fama and French (1993, 2012).

4 Results and discussion

In this section, we analyse the results of the regressions. To make this analysis as clear as possible, we conduct the regression in steps. First, we do a simple regression of the 10-day MAER (henceforth, “MAER(10)”) on Government Effectiveness and Regulatory Quality in regression [1]. We then regress the MAER(10) on the WGI variables and the country fixed effects in regression [2]. In regression [3], we regress the MAER(10) on the global 3 factors and on the momentum factor from Fama and French (1993, 2012). In the final regression [4], we regress MAER(10) on the country fixed effects, the global 3-factor model, and the

momentum factor. In the paragraphs that follow these results, we will do a thorough robustness check. Appendix 4 presents a correlation matrix of all the variables used in this model. Although Government Effectiveness and Regulatory Quality are highly correlated with a magnitude of 0.95, their demeaned values have a lower correlation (0.46). The lower

(15)

15

correlation of the demeaned variables allows us to include both WGI variables in the same regression equation.

4.1 Regression results

In regression [1], Government Effectiveness has a significant negative coefficient, and Regulatory Quality has a significant positive coefficient. Note that the adjusted R^2 of regression [1] is very low (0.03), and should therefore be interpreted with caution. The addition of the country fixed effects in regression [2] has some impact on the adjusted R^ 2. The magnitude of the adjusted R^2 increases only to 0.10. The significance of the coefficients of the country fixed effects tell a consistent story. All the included country fixed effects contribute significantly, but their magnitude is relatively small compared to the WGI variables.

In regression [3], MAER(10) is regressed on Government Effectiveness, Regulatory Quality, global 3-factor model, and the momentum factor. Adding these factors notably improves predictive power of the model. The adjusted R^2 jumps from 0.10 in the previous model to 0.35. Government Effectiveness has a significant negative impact on the MAER(10), but Regulatory Quality loses its significant positive influence. In addition, MKT negatively impacts MAER(10). A small positive impact on MAER(10) was found for MOM.

Table 2. This table shows the regressions of the 10-day moving average abnormal return on perceived

market regulation strength. In regression [1] we regress MAER(10) on ‘Government Effectiveness’ and ‘Regulatory Quality’. In regression [2] we regress MAER(10) on the World Governance Indicator (WGI) variables and country fixed effects. In regression [3] we regress MAER(10) on the WGI

variables, the global 3-factors, and the momentum factor. In the final regression [4], we find the results of equation (4). We regress MAER(10) on the WGI variables, the country fixed effects, the global 3-factors, and the momentum factor.

(16)

16 [1] [2] [3] [4] Alpha 6.96 (9.07) 3.98 (10.71) 11.48 (7.54) 8.74 (8.89) GovEff −28.42 (11.46) ** −31.42 (11.49) *** −21.09 (9.57) ** −23.17 (9.79) ** RegQ 21.23 (11.40) * 23.02 (11.07) ** 13.53 (9.44) 14.92 (9.34) RealInterest −1.72 (0.55) *** −1.26 (0.47) *** MCListed −0.17 (0.06) *** −0.03 (0.02) ListedFirms 0.01 (0.01) ** 0.01 (0.01) * MKT −0.59 (0.07) *** −0.56 (0.07) *** MOM 0.12 (0.06) ** 0.11 (0.06) ** SMB 0.18 (0.19) 0.12 (0.19) HML −0.24 (0.15) −0.28 (0.15) * Sample size 265 265 265 265 Adj. R² 0.03 0.10 0.35 0.38 * = Significant at p = .10 ** = Significant at p = .05 *** = Significant at p = .01

In the final regression [4], country fixed effects, the global 3 factors, and the momentum factor are added to the equation with the WGI variables. In this regression, the alpha is as insignificant as in all the previous regressions. We can conclude that the significant excess return of the 10-day MA strategy fades due to the addition of the WGIs and the country fixed effects. Three out of the four factors from Fama and French (1993, 2012) show varying degrees of significance. The most notable is MKT. For every positive 1% yearly increase in the domestic ‘allcap’ exchange, the profitability of the 10day MA strategy decreases by -0.56%. This supports the finding in various papers that technical analysis is more profitable during economic downturns (Neely et al., 2013). In addition, HML is also negative and significant. Value stocks are generally stocks with higher volatility than growth stocks. Controlling for this fact reduces the profitability of the 10-day MA strategy, since higher volatility stocks are more profitable candidates for technical analysis. For the last factor, MOM, although its coefficient is significant and positive, its magnitude is very small and therefore negligible.

(17)

17

After controlling for the year and country characteristics in regression [4], ‘Government Effectiveness’ still has a significant negative influence on the excess return of the 10-day MA strategy. This means that higher national trust in the capabilities of the domestic government, leads to reduced profitability for technical analysis. Several explanations are possible for this finding. It is possible that higher trust in the domestic government leads to less overreaction to economic shocks. A higher degree of confidence that the government is able to formulate and implement quality policies, for example, leads to less panic during recessions and economic downturn (Bachman and Sims, 2012). When the effects of a negative stock market shock are smaller in countries with higher Government Effectiveness, this relative stability would imply that the volatility of the daily closing prices of these domestic stock indices is lower. In Section 3, we showed that lower volatility in daily closing prices leads to a lower excess return of the 10-day MA strategy (Han et al., 2013).

Another explanation that is supported by prior research is associated with the disclosure of private information. The profitability of technical analysis, especially the profitability of the MA strategy, is produced by the presence of positive and

negative-feedback traders. These traders simply sell whenever they observe signals that fellow traders are selling, and they buy whenever they receive information that fellow traders are buying. Hong and Stein (1999) show in their paper that the overshooting of the fundamental value of an asset by feedback traders is much more severe when little private information is available about an asset. In countries with better government policies, the disclosure of private

information about a firm’s fundamentals is more in favour of minority shareholders. A good measure for this is the dealing index, described by Djankov (2008). The anti-self-dealing index is a measure of minority shareholder protection, and it uses disclosure laws regarding private information as one of the major variables. Although the anti-self-dealing index is only available for a limited number of countries and years, Ng et al. (2016) show that it is positively correlated with Government Effectiveness. To put this into perspective,

countries with a higher level of Government Effectiveness, have higher levels of minority shareholder protection and disclose more private information. More disclosure of private information leads to less overshooting by feedback traders, which in turn leads to lower excess return in the 10-day MA strategy.

However, there might be another possible explanation of the significant coefficient of Government Effectiveness. It is possible that we have overlooked the existence of an

unknown common response variable. This variable would simultaneously affect MAER(10) and Government Effectiveness. This would lead to a spurious regression where causality

(18)

18

seems to apply, but is non-existent. Previous papers have encountered similar situations in their analyses of moving average profitability. LeBaron (1999) found a significant correlation between periods of Federal Bank intervention and profitability of technical analysis. After ruling out volatility and time patterns as confounding factor, they concluded that the single common factor that influenced both variables was difficult to find.

4.3 Robustness check

Past research has suggested that the profitability of technical analysis relies heavily on investor sentiment. Baker and Wurgler (2006), for example, found that both asset price and asset volatility depend investor sentiment. To test whether our results are robust to investor sentiment, we use the yearly consumer confidence index of every country as a proxy. These consumer confidence indices are collected from Datastream. Consumer confidence is inherently linked to (stock) investment (Lemmon and Portniaguina, 2006; Delong et al., 1990). To test the effect of ‘Sentiment’ on the profitability of the MA strategy, we use the following regression equations:

(5) 𝑀𝐴𝐸𝑅𝑥,𝑇,10= 𝑎𝑥+ (𝛽1𝑥𝐺𝑜𝑣𝐸𝑓𝑓) + (𝛽2𝑥𝑅𝑒𝑔𝑄) + (𝛽3𝑥𝑆𝑒𝑛𝑡𝑖𝑚𝑒𝑛𝑡) + 𝑅𝑒𝑎𝑙𝐼𝑛𝑡𝑒𝑟𝑒𝑠𝑡𝑥+

𝑀𝐶𝐿𝑖𝑠𝑡𝑒𝑑𝑥 + 𝐿𝑖𝑠𝑡𝑒𝑑𝐹𝑖𝑟𝑚𝑠𝑥+ (𝛽𝑇,𝑀𝐾𝑇𝑟𝑀𝐾𝑇,𝑇) + (𝛽𝑇,𝑀𝑂𝑀𝑟𝑀𝑂𝑀,𝑇) +

(𝛽𝑇,𝑆𝑀𝐵𝑟𝑆𝑀𝐵,𝑇) + (𝛽𝑇,𝐻𝑀𝐿𝑟𝐻𝑀𝐿,𝑇) + 𝜀𝑥,𝑇

If we look at the results of equation (5) in Table 3, we find that Sentiment has no significant effect on the excess return of the 10-day moving average. This finding is in line with the findings of Han et al. (2013), who also identified no correlation between the two concepts. Government Effectiveness still has a significant positive impact on MAER(10), although the relationship becomes weaker. This can be explained by our proxy for Sentiment, namely consumer confidence. It is plausible that consumer confidence relies in part on Government Effectiveness. The WGI variable measures the perception of domestic inhabitants about the strength of the government. It is likely that consumer confidence is higher in countries where the government is perceived to be stronger. Hence, by adding consumer confidence to the equation, the significance of Government Effectiveness declines. Note that the sample of this regression is slightly smaller than the sample in the main regression (251 observations versus 265 observations). This divergence arises due to the lack of availability of consumer

(19)

19

confidence data for some African and South American countries.

Table 3. In regression [5] we regress MAER(10) on the World Governance Indicator (WGI) variables,

the global 3-factors, the momentum factor, and ‘consumer confidence’. In this regression ‘consumer confidence’ is a proxy for investor sentiment.

[5] Alpha 9.70 (11.77) GovEff −17.87 (9.67) * RegQ 11.23 (8.93) RealInterest −1.09 (0.48) ** MCListed −0.03 (0.02) * ListedFirms 0.01 (0.01) ** MKT −0.51 (0.07) *** MOM 0.15 (0.06) *** SMB 0.26 (0.18) HML −0.17 (0.15) ConC −0.10 (0.10) Sample size 251 Adj. R² 0.39 * = Significant at p = .10 ** = Significant at p = .05 *** = Significant at p = .01

For the second robustness check, we take a look at different lag lengths for the MA strategy. Whereas in the previous regression we looked exclusively at the 10-day moving average, we will now incorporate the 20-day and 50-day moving averages. We use the similar set up as our main regression, but now the dependent variable will change to MAER(20) (i.e., with the 20-day average) and MAER(50) (i.e., with the 50-day average).

(6) 𝑀𝐴𝐸𝑅𝑥,𝑇,50= 𝑎𝑥+ (𝛽1𝑥𝐺𝑜𝑣𝐸𝑓𝑓) + (𝛽2𝑥𝑅𝑒𝑔𝑄) + 𝑅𝑒𝑎𝑙𝐼𝑛𝑡𝑒𝑟𝑒𝑠𝑡𝑥+ 𝑀𝐶𝐿𝑖𝑠𝑡𝑒𝑑𝑥

(20)

20

(7) 𝑀𝐴𝐸𝑅𝑥,𝑇,50= 𝑎𝑥+ (𝛽1𝑥𝐺𝑜𝑣𝐸𝑓𝑓) + (𝛽2𝑥𝑅𝑒𝑔𝑄) + 𝑅𝑒𝑎𝑙𝐼𝑛𝑡𝑒𝑟𝑒𝑠𝑡𝑥+ 𝑀𝐶𝐿𝑖𝑠𝑡𝑒𝑑𝑥

𝐿𝑖𝑠𝑡𝑒𝑑𝐹𝑖𝑟𝑚𝑠𝑥+ (𝛽𝑥,𝑀𝐾𝑇𝑟𝑀𝐾𝑇,𝑇) + (𝛽𝑥,𝑀𝑂𝑀𝑟𝑀𝑂𝑀,𝑇) + (𝛽𝑥,𝑆𝑀𝐵𝑟𝑆𝑀𝐵,𝑇) + (𝛽𝑥,𝐻𝑀𝐿𝑟𝐻𝑀𝐿,𝑇) + 𝜀𝑥,𝑇

Table 4. This table shows the regressions results of equation (6): the 20-day moving average abnormal

return on perceived market regulation strength. In regression [6] we regress MAER(20) on

‘Government Effectiveness’ and ‘Regulatory Quality’. In regression [7] we regress MAER(20) on the World Governance Indicator (WGI) variables and country fixed effects. In regression [8] we regress MAER(20) on the WGI variables, the global 3-factors, and the momentum factor. In the final regression [9] we regress MAER(20) on the WGI variables, the country fixed effects, the global 3-factors, and the momentum factor.

[6] [7] [8] [9] Alpha −0.07 (8.73) −1.73 (10.40) 4.44 (7.32) 2.92 (8.82) GovEff −26.93 (11.03) ** −28.78 (11.16) ** −20.49 (9.30) ** −21.44 (9.61) ** RegQ 27.77 (10.97) ** 29.47 (10.75) ** 20.37 (9.17) ** 21.65 (9.16) ** RealInterest −1.15 (0.54) ** −0.69 (0.46) MCListed −0.07 (0.02) *** −0.03 (0.02) * ListedFirms 0.01 (0.01) * 0.01 (0.01) MKT −0.56 (0.06) *** −0.52 (0.07) *** MOM 0.12 (0.06) ** 0.12 (0.06) ** SMB 0.13 (0.18) 0.08 (0.18) HML −0.12 (0.15) −0.15 (0.15) Sample size 265 265 265 265 Adj. R² 0.03 0.09 0.35 0.36 * = Significant at p = .10 ** = Significant at p = .05 *** = Significant at p = .01

Table 5. This table shows the regressions results of equation (7): the 50-day moving average abnormal

return on perceived market regulation strength. In regression [10] we regress MAER(50) on ‘Government Effectiveness’ and ‘Regulatory Quality’. In regression [11] we regress MAER(50) on the World Governance Indicator (WGI) variables and country fixed effects. In regression [12] we regress MAER(50) on the WGI variables, the global 3-factors, and the momentum factor. In the final regression [13] we regress MAER(50) on the WGI variables, the country fixed effects, the global 3-factors, and the momentum factor.

(21)

21 [10] [11] [12] [13] Alpha −11.17 (9.15) −12.82 (10.83) −4.32 (6.93) −6.21 (8.34) GovEff −21.94 (11.56) * −23.58 (11.62) ** −14.85 (8.80) * −15.62 (9.09) * RegQ 35.59 (11.50) *** 37.67 (11.19) *** 25.59 (8.68) *** 26.96 (8.67) *** RealInterest −0.93 (0.56) * −0.33 (0.43) MCListed −0.09 (0.02) *** −0.04 (0.02) ** ListedFirms 0.01 (0.01) * 0.01 (0.004) MKT −0.73 (0.06) *** −0.69 (0.06) *** MOM 0.11 (0.05) ** 0.11 (0.05) ** SMB 0.17 (0.17) 0.12 (0.18) HML 0.10 (0.14) 0.06 (0.14) Sample size 265 265 265 265 Adj. R² 0.04 0.10 0.47 0.48 * = Significant at p = .10 ** = Significant at p = .05 *** = Significant at p = .01

The results are fairly similar to the main regression where L = 10. The positive significance of Government Effectiveness is robust to changes in lag length. In all eight regressions, the positive effect persists. In the main regression with L = 10, ‘Regulatory Quality’ lost its significance after the introduction of the 3 global factors and momentum. However, the positive significance of the coefficient of Regulatory Quality is consistent in the four regressions where L = 20 and the four regression where L = 50. One can imagine that lower trust in domestic regulatory authorities will lead to more attempts to manipulate the asset markets. A higher degree of market regulation results in higher trade volume and more efficient markets (Aggarwal, 2003). When asset markets are manipulated, technical indicators by themselves become less reliable. Most technical indicators are ‘lagged indicators’ (among them the moving average). They use information from the past to make assumptions about the future. However, when markets are manipulated, the past prices are not representative of the future movements of the assets. Manipulation schemes like ‘pump and dumps’ might cause sudden price crashes, unforeseen by technical analysis. Therefore, a higher coefficient for Regulatory Quality might have a positive effect on the abnormal return of the 10-day MA strategy. Another possibility is the presence of a common response variable that effects both MAER(20) and MAER(50), on the one hand, and Regulatory Quality, on the other. Although we tested a wide variety of macroeconomic and risk-related variables in our search for

(22)

22

appropriate country fixed effects, we did not find this missing factor.

4.4 Case studies

In the previous sections we have established a weak statistical relationship between higher domestic trust in regulatory policies and the profitability of the 10-day moving average indicator. In the robustness check, we found that this weak relationship does not hold for alternative lag lengths (i.e., for 20- and 50-day moving average). However, we also argued that the missing correlation might be due to the nature of the observations. The ‘all-cap’ indices on country level exert very low volatility, hence lower profitability for technical analysis. Apart from the low volatility of the ‘all-cap’ indices, one can imagine that it is very difficult to significantly manipulate an entire index. In his famous book The Dow Theory, Robert Rhea wrote the following of this matter:

“Manipulation is possible for a limited number of stock in the day to day movement of the averages and may give an entirely false view of the situation. It is impossible, however, to manipulate the whole list so that the average price of 20 active stocks will show changes sufficiently important to draw market deductions from them. For instance, on September 1, 1929, the total market value of all stocks listed on the New York Stock Exchange was reported to have amounted to more than $89,000,000,000. Imagine the money which would have been involved in depressing such a mass of values even 10 per cent!”

The weak significant relationship between our regulation variable and the moving average profitability partly confirms Robert Rhea’s theory. In order to test the relationship between manipulated individual stock and technical analysis, we will conduct two anecdotal case studies of famous instances of stock fraud.

In the introduction of this paper, we stressed the sceptical perception of technical analysis by most finance academics. Part of this scepticism arises from the fact that a wide array of technical indicators exists. The availability of hundreds of technical tools opens the door for datamining, especially when technical analysis is conducted after the facts. To discourage scepticism about potential datamining, we constructed a method to limit those concerns. We used the search words ‘Top technical analysis indicators’ and ‘Top technical analysis books’ on Google to decide which technical analysis tools we should use. This resulted in three technical analysis methods: the RSI, the moving average convergence

(23)

23

divergence (MACD) and the on-balance volume (OBV). Unfortunately, the OBV indicator was not usable because the daily trading volume data is not available for stock traded 15 years ago. Subsequently, we use the definition of the RSI and MACD methods as described in the top Google hits for the search words ‘Top technical analysis books’. Finally, we conduct our case studies on Bre-X Minerals and WorldCom. These instances of stock fraud were high on the all the lists of Google search results for the search words ‘most famous cases of stock fraud’. More importantly, they had public daily closing price data for the fraudulent period. The exact output of the search engine methodology can be found in Appendix 5.

4.4.1 The relative strength index (RSI)

The first indicator we will use for the analysis of manipulated stock, is the RSI. The RSI is the most widely used indicator in Western technical analysis (Deng and Sarkurai, 2013). Over large samples, following only RSI strategies obtains profitable trading results. The standard deviation of RSI returns is much lower than the returns of buy-and-hold strategies (Marek and Sediva, 2017).

The RSI is a momentum oscillator. It measures momentum as the balance between the recent gains and losses of an asset (Pring, 2014). The RSI was developed by J. Welles Wilder in 1978. It uses 14 periods for its calculations (14 days on the daily timeframe, 14 weeks on the weekly timeframe). Some traders use shorter periods of time, but the RSI becomes more volatile and less reliable in such cases (Murphy, 1999). The RSI graph should be read as follows: when the RSI reaches 70, the asset is ‘overbought’, and a downward correction is likely. The term ‘overbought’ simply means that statistically the recent buy pressure has been exhausted by the excess demand in the past days. When the RSI reaches 30, the asset is ‘oversold’, and an upward correction is probable. The term ‘oversold’ has the inverse interpretation of the ‘overbought’ principle (Pring, 2014).

The RSI is calculated as follows:

(8) 𝑅𝑆𝐼 = 100 − 100

(1+𝑅𝑆)

(9) 𝑅𝑆 = 𝐴𝑣𝑒𝑟𝑎𝑔𝑒 14 𝑑𝑎𝑦 𝑐𝑙𝑜𝑠𝑒 𝑈𝑃

(24)

24

The slope of the RSI is highly correlated with the speed of trading. If the price decreases very fast, the RSI will do so as well (Pring, 2014). The RSI trend usually precedes future

movements of the price. The top (RSI > 70) and the bottom (RSI < 30) limits of a price movement can usually be predicted by the RSI trend. In addition, technical analysis

formations, like pennants, flags, and support and resistance levels, often appear on the RSI chart before they appear on the price chart. The strongest signals that are sent by the RSI are the so called ‘divergence signals’. Divergence between the price and the RSI indicates a strong case for market reversal. When the price decreases as the RSI increases, the downtrend of the price will likely be reversed. This is called ‘bullish divergence’. Conversely, when the price increases and the RSI decreases, a move downwards is more likely. This is called ‘bearish divergence’ (Pring, 2014; Murphy, 1999).

4.4.2 Moving average convergence divergence (MACD)

The MACD is another momentum oscillator. This and the RSI are the most commonly used technical analysis tools to assess momentum (Murphy, 1999). For the calculation of the MACD, we need to construct four different lines:

1. Exponential moving average (EMA) – 12-day period 2. Exponential moving average (EMA) – 26-day period

3. MACD line = 12-day EMA minus 26-day EMA (this is the blue line) 4. Signal line = 9-day EMA of the MACD line (this is the red line)

(25)

25

On the line graph of the MACD indicator, only the MACD line and the signal line are

displayed. The MACD gives off a buy signal or a sell signal once the MACD line crosses the signal line. If the MACD line crosses over the signal line, this should be interpreted as a buy signal. If the MACD line crosses under the signal line, this should be seen as sell signal (Murphy, 1999). The MACD also incorporates certain aspects of the RSI. When both the MACD line and the signal line are under the ‘zero’ border, this is an ‘oversold’ condition. A price movement out of the ‘oversold’ area should be considered as a bullish signal.

Conversely, if both lines are in the area above the zero line, this is an ‘overbought’ condition. When the price moves out of the ‘overbought’ area, this is an argument for bearish reversal. One of the limitations of the MACD indicator is the low reliability of its signals during periods of very low volatility. In such times, the MACD does not give signals with long-term interpretability (Pring, 2014).

4.4.3 Bre-X Minerals manipulation

Bre-X Minerals was a Canadian gold mining firm, founded in 1989 by David Walsh. Until 1993, it traded on insignificant exchanges as a penny stock. In March 1993, the company bought a large piece of jungle in Borneo, Indonesia. One year after acquiring the site in Borneo, Bre-X disclosed the discovery of large amounts of gold in the jungle. In 1995 they reported the mining potential of 840.000 kilograms of gold; in 1997 this estimate increased to 2 million kilograms of gold (Schneider, 1997). During these years, major stock analysts like

(26)

26

J.P. Morgan tipped Bre-X as a great long-term investment (Kirby, 2011). The market capitalization of Bre-X increased to $4.4 billion in 1997, and the stock was listed on the NASDAQ exchange. David Walsh, the seated CEO and founder, had large amounts of stock options. He cashed out $100 million of his own stock in 1996 when Bre-X was in its best years.

In January 1997, the Indonesian government intervened with Bre-X, claiming a piece of the potential profits. The company and the Indonesian government agreed to a deal, where Bre-X had a claim to 45% of the mining gains (Schneider, 1997). The government hired an independent American firm for the exploitation of the gold mine in Borneo. On the March 26, 1997, after a week of due diligence on the site, the independent exploitation firm claimed there was no gold at all in the jungle ground. The Indonesian government pulled out of the deal, and Bre-X stock plummeted. In May 1997, a third-party mining company proved that the initial gold samples from Bre-X had been falsified with gold that Bre-X had bought on the local market in Borneo. Days after this news, the value of Bre-X went to zero, and the stock was delisted from the NASDAQ (NY Times, 1997).

4.4.4 Bre-X Minerals technical analysis

In Image 1, we can see the development of the RSI, and in Image 2, we can see the price development in the same timeframe. As mentioned in the section on the RSI, divergences between the RSI and the price are the strongest indication of trend reversals given by the RSI. Divergences are marked by the red lines on the RSI graph. We can see that the corresponding red lines on the price graph move in the opposite direction. The first significant RSI

divergence occurs in February 1996. We can see an increase in the price, while the RSI is steeply declining. This bearish divergence would imply a downturn in the price. However, we can see that this expected trend reversal does not completely materialize. The price responds with a brief pause in the uptrend, and moves sideways for some weeks. The next divergence can be seen between the end of September 1996 and October 17, 1996. The price and the RSI move steeply in opposite directions. The bullish divergence in this timeframe notes the end of the downtrend in price. Divergences are stronger indicators of future price movements when the span of the divergence is longer (Murphy, 1999). The last instance of bearish divergence takes place at the end of 1996. While the price is increasing, the RSI is decreasing on the daily timeframe in the same time interval (between 17-12-1992 and 12-02-1993). Relative strength index theory dictates that this kind of divergence over such a long time period is the strongest

(27)

27 ‘sell’ signal it can possibly exert.

It is important in technical analysis to combine different technical indicators to form a judgement of future price movement. As explained in the introduction, we look at the MACD as well. Image 3 shows the MACD for Bre-X Minerals. The black line is the ‘signal line’, and the MACD line changes between green and red. When the MACD line crosses above the signal line, it is a bullish signal (hence the green colour). When the MACD line crosses under the signal line, it is a bearish signal (hence the red colour). From Image 3, we can conclude that the MACD gives off signals on much shorter timeframes. However, some signals are stronger than others. When the MACD line and the signal line move from the ‘oversold’ area (under zero) to the ‘overbought’ area, while simultaneously the MACD line crosses over the signal line, this is a strong bullish sign. This scenario occurs at the beginning of 1997. The MACD line and the signal line move upwards from −1.5 on the vertical axis, while the MACD line moves over the signal line. However, only a few weeks later, this uptrend reverses. In the second week of February 1997, the MACD dives under the signal line into ‘oversold’ territory. This dive coincides with the bearish divergence on the RSI on February 12, 1997. The combination of these two signals produces very strong sell advice. Only five weeks later, the stock price imploded, after the fraud scheme of Bre-X Minerals was revealed.

Image 1. The relative strength index (RSI) of Bre-X Minerals. The RSI is decreases between the

(28)

28

Image 2. The price development of Bre-X Minerals from 1995 to 1997. The price increases between

the December 17, 1992, and February 12, 1993.

Image 3. The MACD graph of Bre-X minerals. The red dots on the horizontal axis highlight the buy

(29)

29 4.4.5 WorldCom market manipulation

WorldCom was founded in 1995 after a merger between the companies WilTel and LDDS. Only three years after this merger, WorldCom closed the largest merger deal in history by acquiring a company called MCI for $40 billion. In the years that followed, the stock price thrived, and WorldCom was one of the largest firms in the USA. However, Bernard Ebbers, the CEO of WorldCom, had an alternative agenda. The increasing stock price of WorldCom made him very rich. In fact, during his time as CEO, he had leveraged long positions on WorldCom stock at various brokers (SEC, 2008). When the WorldCom stock price began to decrease at the end of 2000, he borrowed $400 million from the company to cover the margin calls on his leveraged WorldCom long positions (SEC, 2008). The board agreed with this loan, because they were frightened that Bernard Ebbers would otherwise be forced to sell off a large chunk of his WorldCom stock, which would result in a decreasing stock price. As the strategy failed, Bernard Ebbers and his directors started to artificially pump the stock price. They used accounting schemes to inflate the revenue and decrease the expenditures. During this time, the net worth of the company increased by $11 billion in ‘fake’ value (SEC, 2008). The first signs of the company’s downturn emerged on the March 11, 2002, when the SEC filed a request for information on the company’s accounting methods and loans to workers. Only a few weeks later, Bernard Ebbers was fired for taking $400 million worth in personal loans from the company (NY Times, 2005). The stock price of WorldCom

completely collapsed. On June 26, the SEC charged the WorldCom board with accounting fraud. One month later, WorldCom filed for bankruptcy, the largest bankruptcy in the history of Wall Street (NY Times, 2005). During the fall of the stock in the first quarter of 2001, major American stock analysts like Salomon Brothers never exerted ‘sell’ advice for WorldCom. Only after the stock had fallen by 94% in April 2002 they changed their advice from ‘buy’ to ‘neutral’ (PBS, 2003).

4.4.6 WorldCom technical analysis

As mentioned in the previous paragraph, the catalyst opinion on WorldCom remained positive, and ‘hold’ advice was upheld during the downfall. However, when we look at the technical indicators for the historical price graph of WorldCom, a different picture arises. If we analyse the RSI in Image 4, we can spot the first divergence signal during the end of October 2001. Over a timespan of three months, the price and RSI moved in opposing

(30)

30

directions. This bullish divergence results in a price increase to $15. The second instance of significant divergence occurs between 18-10-2001 and 28-11-2001. Whereas the RSI declines in this period, the price of WorldCom ascends. This is another example of ‘bearish

divergence’. Bearish divergence is the clearest ‘sell’ signal possible from the RSI. This ‘sell’ signal was given four months prior to the SEC accusations. After this ‘sell’ signal, the RSI did not again signal a buying opportunity.

Whereas the fundamental analysis of the analysts failed to inform their customers of the negative spiral of WorldCom prior to the SEC news, a basic understanding of technical analysis would have done the opposite. Although the combination of the RSI and the MACD (Image 6) would have signalled a buying opportunity at the beginning of September 2001, the inverse would happen at the end of November 2001. On the December 20, 2001, the MACD line dives under the signal line, and stays consistently below it until March 2002. The MACD gives off a ‘sell’ signal after the price has started the nose dive downwards, but this is still prior to the public disclosure of accounting fraud.

Image 4. The relative strength index (RSI) of WorldCom between 1-12-2000 and 1-6-2002. The RSI

decreases between 18-10-2001 and 28-11-2001, marked by the red line. This is a few weeks prior to the exposure of the accounting fraud.

(31)

31

Image 5. The price of WorldCom between 1-12-2000 and 1-6-2002. The price increases between

18-10-2001 and 28-11-2001, marked by the red line.

Image 6. The moving average convergence divergence (MACD0 graph of WorldCom. The red dots

(32)

32 4.4.7 Limitations

In these two case studies, we have showed that technical analysis can be helpful when fundamental analysis fails to deliver. For both Bre-X Minerals and WorldCom, technical analysis gave ‘sell’ signals before the fall of the stock prices. In these two cases, the

momentum oscillators (RSI and MACD) picked up signals that the price was losing upward pressure. Note, however, that concerning other price-manipulation schemes, these momentum oscillators would likely give incorrect signals. A good example of this is a ‘pump and dump’ scheme, which usually happens in illiquid penny stock markets, where public data is

unfortunately scarce. In a ‘pump and dump’ scheme, the MACD is likely to exert ‘buy’ signals due to the high buy pressure. These ‘buy’ signals will likely persist until the price crash.

We realize that the sample of only two stocks is extremely small. In addition, even though we limited the possibility of datamining as much as possible, we did conduct the research 15 years after the facts. This lessens the strength of our case in favour of technical analysis. For this reason, the case studies should be seen as illustrational and should be interpreted with caution.

5 Conclusion

In this thesis, we tried to conclude whether the degree of domestic market regulation has an effect on the profitability of the moving average trading rule. Since price manipulation of an asset can cause unexpected price surges and crashes, we hypothesized that stricter market regulation will enhance the reliability of technical indicators. We conducted technical analysis on the ‘all-cap’ indices of 24 countries in the timespan 2004-2016. We find a significant excess return for the 10-day moving average strategy compared to a buy-and-hold strategy. This result is robust to the 3-factor model and the momentum factor of Fama and French (1993, 2012) and transaction costs.

As a proxy for market regulation we used the World Governance Indicators for regulatory quality and government effectiveness. In addition to these WGI variables, we added country fixed effects and the four factors described above. We find that government effectiveness has a weak significant effect on the profitability of the 10-day MA strategy. Regulatory quality is insignificant for the 10-day moving average, but has a significant positive effect on the profitability of the 20-day and 50-day MA strategy. Although there are

The effect of market regulation on the profitability of technical analysis