• No results found

The volatility effect after the financial crisis : a historic perspective

N/A
N/A
Protected

Academic year: 2021

Share "The volatility effect after the financial crisis : a historic perspective"

Copied!
30
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

1

The volatility effect after the financial crisis;

a historic perspective

June 2016

Author: Roger van Buuren

Student number: 10557954

Economics and Business

Finance and Organization track

Supervised by: Esther Eiling

(2)

1

Table of contents

Verklaring eigen werk ... 1

Abstract ... 2 Introduction ... 2 Literature ... 5 Methodology ... 9 Data ... 11 Results ... 12 Robustness ... 16 Conclusion ... 17 Reference list ... 19

Appendix A: Stata commands for creating decile portfolio’s ... 20

Appendix B: CRSP Index returns ... 22

Appendix C: Low and High portfolio returns ... 24

Appendix D: Subsample regressions of Low and High portfolio ... 24

Appendix E: Penny stocks ... 25

Appendix F: Large companies ... 25

Appendix G1: Portfolio characteristics, robust ... 26

Appendix G2: Low and High portfolio returns, robust ... 26

Appendix G3: Portfolio regressions, robust ... 27

Appendix G4: High-Low regressions per period, robust ... 27

Appendix G5: High and Low portfolio regressions per period, robust ... 27

Appendix G6: Regressions per subperiod, robust ... 28

Appendix H1: CAPM Portfolio regressions ... 28

Appendix H2: High and Low portfolio regressions per period, CAPM robust ... 28

Appendix H3: Regressions per subperiod, CAPM robust ... 29

Verklaring eigen werk

Hierbij verklaar ik, Roger van Buuren, dat ik deze scriptie zelf geschreven heb en dat ik de volledige verantwoordelijkheid op me neem voor de inhoud ervan.

Ik bevestig dat de tekst en het werk dat in deze scriptie gepresenteerd wordt origineel is en dat ik geen gebruik heb gemaakt van andere bronnen dan die welke in de tekst en in de referenties worden

genoemd.

De Faculteit Economie en Bedrijfskunde is alleen verantwoordelijk voor de begeleiding tot het inleveren van de scriptie, niet voor de inhoud.

(3)

2

Abstract

This thesis considers the volatility effect otherwise known as the beta puzzle. It tries to answer the question if the recent financial crisis has impacted the size of the volatility effect, the theory that a portfolio of low volatile stocks outperform the market. Through regressing subsamples before and after recessions a broad perspective of the effect is established. A negative volatility effect is

discovered, contradicting most existing literature. This is more in line with traditional financial theory and suggests the effect is an anomaly rather than a undocumented value factor. The negative effect is found to be higher before a recession hits the economy than after one, but not significantly.

Introduction

Risk versus reward is a key element of asset management. According to the traditional paradigm, the efficient portfolio is the market portfolio. This should maximize the Sharpe ratio through

diversification and give the investor the most return for the risk he’s willing to run (the amount of risk can be altered by borrowing or lending additional funds). However, recent papers have shown that portfolios with low volatility stocks outperform the market significantly. This is known as the volatility effect. Authors have tried to explain the effect by theoretical deduction. Among the

possible causes were a couple that have been influenced by the recent financial crisis (Causes named by Haugen and Baker (2012) and Blitz and Van Vliet (2007)). This thesis tries to answer the question if the financial crisis has affected the size of the volatility effect.

One of these possible explanations (Haugen & Baker, 2012, pp. 10-11) can be found in the origin of institutional asset management, the nature of the portfolio manager compensation. The use of a High Water Mark as a benchmark for portfolio compensation causes the managers to act as a risk-seeking agent when reaching the benchmark seems unlikely using conservative trading strategies. This directly contradicts the theory that financial agents are risk averse and should therefore be compensated for taking on risks. As volatility is a proxy for risk, this means that demand for high-volatility might be higher than predicted, causing prices to rise and returns to fall.

Since the financial crisis in 2008, management bonuses have become high-profile. The management performance fees have been heavily criticized and described online as unfair, rarely risk-adjusted and frequently complex (Bacon, 2010). Journals investigated the structure like Van Doesburg and De Kleer (2011) and academics felt the responsibility to further research the pay policy’s as well (for instance, Guasoni & Obłój, 2016). This media exposure would suggest the volatility effect might have changed, as it is fueled by the the manager compensation factor to a lesser extent. The shifting negative attitude to this kind of bonus structure might have caused

(4)

3 institutions to update their policy and reduce the impact of the compensation structure on risky stock returns.

Another named factor that plays a role in the volatility puzzle is the borrowing availability (Blitz & Van Vliet, 2007, pp. 10-11). Profiting from the optimal Sharpe ratio requires that asset managers can adjust their risk-level by altering their debt. Applying leverage to a portfolio allows investors to fully utilise the potential market return. This means investors should be able to borrow funds for the risk-free rate or buy bonds yielding the risk-risk-free rate. A possible explanation for the volatility effect is that investors are not all able to do so. When an investor cannot access leverage in the portfolio, it will resort to trading more volatile stocks to meet the return requirements.

Low interest rates have made it easier to borrow funds, providing quicker access to leverage to investors that require it as can be seen in Figure 1. This volatility effect might have weakened by the recent decline in interest rates, making it easier to borrow funds, hence making it less attractive to purchase riskier stocks.

Figure 1 Movements of the monthly risk-free rate (1-month Treasury bill) during the sample period. Interest rates have declined since the recent financial crisis. These low interest rates have made it easier to borrow funds the last couple of

years.

Finally the structure of asset management firms can have had an effect that influenced the volatility effect (Blitz & Van Vliet, 2007, p. 11). Many investment firms have separate departments with different objectives in respect to returns. This insinuates that risk-seeking departments exist within firms, looking for additional risk to maximize returns. These risk-loving departments increase demand for high volatile stocks, driving up their prices and reducing the highly volatile stock returns.

(5)

4 The crisis has changed market sentiment and might have caused firms to become better aware of all risks concerning investments, making them more conservative, hence closing risk-seeking

departments or lowering the return objectives. Expected market volatility has risen during the crisis and saw some steep spikes shortly afterwards as can be seen in Figure 2.

Figure 2 VIX-index during the credit crunch. This reflects the implied volatility of the S&P 500 stock options, giving information about expected volatility by investors. More expected volatility means higher uncertainty about future returns and can be used as an indication of market sentiment where low implied volatility means less market risk and more room for

opportunism (Baker & Wurgler, 2007, p. 137). Data found via WRDS.

For an even closer look on investor attitude we can take a look at the Market Sentiment Index introduced by Baker & Wurgler (2007) in Figure 3. We can tell that the market sentiment took a hit when the crisis unfolded and by the end of 2013 the sentiment had not fully recovered yet. If institutions reorganise accordingly it would mean that risk-seeking behaviour would not be encouraged as much as before, thereby creating a smaller premium on risk and a decrease in the volatility effect.

(6)

5

Figure 3 Investor Sentiment Index during the credit crunch. The index is a combination of 6 proxies for measuring the market sentiment. Higher sentiment means more bullish attitude towards the market and expecting rising stock prices. Lower

sentiment means a bearish attitude and more carefulness about the stock prices for the next period. Data found via http://people.stern.nyu.edu/jwurgler/.

Changes in compensation, interest rate and attitude described above suggest that the volatility effect has decreased during the financial crisis. However, in recent years more papers on this subject have been published causing more awareness and more institutions to invest accordingly, thereby reducing the low volatility premium and possibly contradicting the hypothesis. The effect could be a market inefficiency that can be traded away, but it is also a possible pricing factor that imposes a permanent imbalance on the market due to limits to arbitrage as Haugen and Baker argued in one of their early papers (1991).

Literature

The volatility effect has been discussed in several papers throughout the years. Both practitioners and academics find the subject to be appealing and interesting enough for further research.

In 1991 Haugen and Baker first acknowledged the volatility effect. Their paper consists of a theoretical and an empirical part. In both sections they provide evidence that even with an

informationally efficient market the market cap-weighted portfolios are inefficient. Their explanation for the effect was mainly aimed at the investors differing assumptions and preferences. Investors might have different opinions on riskiness and future returns and not all investors have the same

(7)

6 investment permissions, such as short-selling (Haugen & Baker, 1991, pp. 36-38). Some investors have to face different taxation than others and some might have access to different investment opportunities than others (outside of the index). Also foreign investors are active in the market, who build their portfolios in a way to be efficient relative to their own domestic market.

In the empirical section (Haugen & Baker, 1991, pp. 38-40) they composed low volatility portfolio’s out of the 1000 largest companies in the Wilshere 5000 index in such a way that total volatility is minimised. To control for a single stock or industry to dilute the returns they constrained the portfolio to have no more than 1.5% of the portfolio invested in a single stock, no more than 15% of the portfolio invested in a single industry. To re-enact short selling restrictions, minimally 0% is invested in each firm. Volatility was measured in the past 24 months and the portfolio would be held for one quarter (further on in the thesis it would be referred to as a 24/0/3 selection). After the quarter the portfolios are redesigned and (transaction costs of 2% are assumed) and this ran for the period between 1972 and 1989.

This portfolio is compared to a set of randomly selected cap-weighted portfolios and randomly selected portfolio with the same weighing structure as the main portfolio. The low volatility portfolio dominates the Wilshere 5000 index and the other random portfolios in terms of risk, but also returns are greater or equal than the market.

Later in 2012 Haugen and Baker updated their old study with more and recent data. This time their research covered 21 developed markets and 12 emerging markets between 1990 and 2011 in which they created decile portfolios ranked on realised volatility, again based on 24 months of returns. Instead of quarterly turnover they choose to compose monthly portfolios. The decile with the highest volatility is compared with the least volatile decile in respect to risk, return and Sharpe ratio. They show that the volatility effect is observable all over the world (Haugen & Baker, 2012, pp. 4-8).

Haugen and Baker criticise (2012, p. 8) the Fama-French model approach because alpha is according to them essentially the difference in return between value stocks with high book-to-market ratio’s and growth stocks (low book-to-book-to-market ratio’s) if the book-to-market does not outperform the risk-free rate, small firms do not outperform large firms and value stocks do not outperform growth stocks. In such a market environment, it makes sense that alpha is zero. Other measures including more sophisticated modelling is called statistical maneuvering (Haugen & Baker, 2012, p. 9) and only prevents researchers to observe the volatility effect and come up with explanations. Possible causes for the effect are a preferred third moment of returns by investors or a flatter security market line due to limits to arbitrage.

Haugen and Baker (2012, p. 12) add that the manager compensation for professional investment managers could be an important factor. They argue that the option-like bonus structure

(8)

7 imposes an incentive on the portfolio managers to hold or recommend to volatile portfolio instead of a portfolio with low volatility. This way their total expected pay-off increases. In addition they

describe the process of compiling a model portfolio of a large investment firm. The analyst meetings are set up in such a way that employees are stimulated to impress their supervisors with highly volatile stocks. Both of these theories result in an overrepresentation of volatile stocks in

institutional investor portfolios, resulting in turn in overpricing and lower returns. To back-up this hypothesis they find that institutional ownership is in fact higher in volatile stocks (Haugen & Baker, 2012, p. 14).

They claim to see a paradigm shift now that practitioners start to understand that risk has a negative reward, undermining the fundamentals of traditional finance (Haugen & Baker, 2012, p. 16). Because of this change they plead for different financial schooling with more focus on practise than theory.

Researchers Clarke, De Silva and Thorley (2006), Blitz and Van Vliet (2007) and Dutt and Humphery-Jenner (2013) expanded on the work by Haugen and Baker. Clarke et al discover that minimum variance portfolios have a bias for value and small-size factors (2006, p. 21), where Dutt & Humphery-Jenner try to explain the effect by a fundamental analysis and find a positive relation between low volatility and high operating performance (2013, p. 1015). Blitz & Van Vliet find

additional evidence on the volatility effect using weekly data for the volatility proxy with even higher results (2007, p. 12).

Ang, Hodrick, Xing and Zhang discuss in their paper (2006) the volatility effect in a Fama French setting. In the second section of the essay they examine the relationship between idiosyncratic risk and expected returns, where idiosyncratic risk is defined relative to the Fama French model. They argue that if aggregate volatility can (partially) explain returns, it should show up in the residuals of the regressions. In the introduction they already assure us that the effect is observable in all kinds of market circumstances (Ang et al, 2006, p. 262). This means that both in bull and bear markets, during both economic expansions and recessions and in both volatile and stable periods the risky portfolios underperform the low volatility portfolios.

The researchers construct monthly value-weighted quintile portfolios ranked on realised volatility, based on daily 1-month data (Ang et al, 2006, p. 283-285). These portfolios are held for a month and then rearranged for the period between July 1963 and December 2000. A gap in market capitalisation is found between quintile 1 and quintile 5 as the low volatility portfolio holds 41.7% of the market and the high volatility portfolio only 2,4%.

(9)

8 The paper observes a significant negative alpha for a High volatility minus Low volatility trading strategy and check for robustness by controlling for different effects such as size, particular exchanges, value, leverage, liquidity, analyst forecasts and momentum (Ang et al, 2006, pp. 285-296). On top of that they report to see the effect as well for different definitions of realised volatility (12 months daily data for instance) and with a lagging period between the sorting of the portfolios and the acquisition of one month. The holding period is altered as well.

Ang et al are also the first to investigate the volatility effect in relation to market

circumstances. They observe the effect in subsamples of the sample period (Ang et al, 2006, p. 292-296). They focused on subperiods indicated by the NBER as economic expansion, recession and also stable and volatile periods and conclude that the volatility effect is robust to all these factors.

As the researchers cannot explain the effect or the underlying drivers they conclude that the anomaly remains a puzzle.

Bali and Cakici (2008) pick up where Ang et al stopped. They continued to search for explanations for the overpricing of risky portfolios using similar methodology as Ang et al. In their paper they

introduced screens for size, price and liquidity to an extended sample period of July 1963 until December 2004. Bali and Cakici show that the portfolio composition is key to observing the anomaly (2008, pp. 31-54). When the robustness screen is included the difference between the high volatility quintile and the low volatility quintile is negligible, which implies that the results of Ang et al are mainly driven by small illiquid stocks. Also the effect disappears when portfolios are

equally-weighted instead of value-equally-weighted or when portfolio market capitalisation is restricted to be equal or capped by choosing a different exchange.

As an alternative to using the traditional weighing schemes, Bali and Cakici weight stocks in portfolios on inverse volatility as well (2008, p.48). This way the size effects should be completely eliminated and a pure relation between risk and return can be measured. However these portfolios do not result in significant return differences.

Moreover they show that volatility proxy’s based on daily data have a negative significant effect on cross-sectional expected returns where monthly based proxy’s show a flat relation (Bali & Cakici, 2008, p. 34). They argue that monthly based measures of volatility are better predictors of future volatility, which leads them to believe there is no relation between idiosyncratic volatility and expected return.

Where the practitioners papers use a practical approach and look at ways to exploit the effect and the underlying drivers without worrying too much about methodological factors, the academics remain more reserved about the observation of the volatility effect. Researchers like Haugen, Baker,

(10)

9 Blitz and Van Vliet have discovered and investigated a wide range of practical volatility effect aspects that promise overperforming risk-adjusted returns as a constant risk factor to be accounted for with practical causes.

The academic research papers by Dutt & Humphery-Jenner (2013), Ang et al (2006) and Bali & Cakici (2008) seem to treat the volatility effect more as an anomaly and attempt to find which specific part of the construction of the volatile portfolios can explain the abnormal behaviour with statistical certainty. Especially Bali and Cakici came a long regarding the reasons for the discovered conclusions in the data, but the volatility puzzle has not yet been solved completely.

Methodology

The focus of interest is the American equity market. As the financial crisis has started in America it is the market that should be influenced the most. Because it is also the largest market in capitalization, it has been researched before in previous literature the most. Both the volatility effect and the financial crisis are known to the market. The CRSP database is the universe we will examine. This way the American market as a whole is covered.

Monthly data is used from January 1960 until December 2015. This way we can research a large timeframe, giving insight in the historical perspective of the overpricing of risky portfolio’s. Each month the market is split up in decile portfolios, meaning a tenth of the companies is assigned to portfolio 1, a tenth to portfolio 2 and so on. The companies are assigned to a portfolio based on their 1-year realised volatility. As a result the 10% stocks with the lowest standard deviation regarding the returns, including dividends, of the past 36 months will end up in portfolio 1, from now on often referred to as the top portfolio. We build value-weighted portfolios to optimize market index comparisons. All portfolios are held for a month and then the returns including dividends 𝑅𝑖 can be computed. After the end of the month the portfolios are rebalanced and stocks are resorted.

Throughout the analysis penny stocks are excluded because they dilute the data with high volatility and low returns. Their performance is not comparable with average stocks because the tick size change results in a relatively high performance change. These companies are often on the brink of bankruptcy and make huge percentile jumps. These stocks attract a different type of investor as it is often referred to as gambling rather than investing. The robustness section provides additional information on penny stock behaviour in respect to our portfolios.

Our strategy is called a 36/0/1 strategy by Jegadeesh and Titman (1993) and Ang et al (2006, p. 283), because stock volatility is measured in 36 months, after which there is no waiting period and the portfolios are kept in the portfolios for 1 month. No waiting period and a one month holding period

(11)

10 are used in all previous research described in the literature section, however the selection period is up for debate. Using 36 months is a compromise between the one month of Ang et al (2006), 24 months of Baker & Haugen (1991, 2012), 36 months of Blitz & Van Vliet (2007) and 60 months of Clarke et al (2006) and will get us enough observations to fairly estimate the monthly volatility of each monthly return without a noisy standard deviation.

Because it takes 36 observations before a portfolio can be computed, the portfolios are compiled from January 1963 until December 2015. This means a total of 6360 portfolios are

constructed. Stocks with missing returns are wiped out of the analysis. For detailed Stata commands on building the portfolios, see Appendix A.

An additional fictional portfolio is created named High-Low. The returns of this portfolio are the ones of the bottom portfolio (high volatility, portfolio number 10) minus the top portfolio returns (low volatility, portfolio number 1).

The portfolio returns will serve as an input for the regression that will be core in this thesis. After subtracting the monthly risk-free rate 𝑅𝑓 given by the 1-month Treasury Bill rate, the excess return will be the dependant variable in the regression. The Fama French 3 factor model will be used as the basis of the regression as is described in the paper by Fama and French in 1992. This is for the reason that it is the dominant asset pricing model and it is used in previous literature as well, which makes it easier to bridge the gap between the thesis and established science. For this reason additional factors such as momentum and liquidity are omitted. According to the Fama French 3 factor model, the market excess return, Small minus Big (market capitalization) and High minus Low (book-to-market ratio) should be used as control variables, leaving alpha the unexplained return or ‘free lunch’. Robust standard errors are assumed throughout the thesis.

𝑅𝑖− 𝑅𝑓= 𝛼𝑖 + 𝛽𝑖∙ (𝑅𝑚− 𝑅𝑓) + 𝑠𝑖∙ 𝑆𝑀𝐵 + ℎ𝑖∙ 𝐻𝑀𝐿 + 𝜀𝑖

𝑅𝑖− 𝑅𝑓 Portfolio excess return 𝑅𝑚 Fama French market return

𝑅𝑓 One month Treasury Bill rate 𝑆𝑀𝐵 Small minus Big

𝐻𝑀𝐿 High minus Low

The high-low portfolio alpha will capture the volatility effect. The lower the negative alpha, the higher the volatility effect.

(12)

11 To examine the changes of the effect, the dataset is split up in 8 subsample periods. These periods are separated by US recessions officially registered by the Business Cycle Dating Committee of the National Bureau of Economic Research (2010). Every first month of economic contraction is the start of a new subgroup in the database, because that is when optimism turns into negativism, creating hypothetically new incentives to alter the trading strategy.

Furthermore each period is divided in two subperiods. The first half of the period is labelled as ‘Post-recession’, because it is just after the start of the recession thereby holding on to the negativity concerned with it. The second half is labelled ‘Pre-recession’ for the reason that it is the period with optimism building up towards the inevitable next recession. Observations in months between post and pre subperiods (when the period has an odd number of months) are excluded for the remaining analysis to avoid biases.

Comparing the different high-low returns for post subperiods and pre subperiods will lead to a conclusion about the impact of crises on the volatility effect. Period 1 is excluded in this section, because it starts between two recessions (April 1960 and December 1969) and would therefore dilute the database.

Data

The Centre for Research in Security Prices database includes all stocks listed on American exchanges. The CRSP index (excess) returns differ from the Fama French market returns. Differences in the rates can be seen in the table in Appendix B, where the results are graphically displayed in a frequency distribution of the differences as well. However to avoid an imbalance in the Fama French factor weights the Fama French market proxy is still preferred.

The computed portfolios have particular characteristics, which are shown below. Highly volatile portfolios perform better than low volatile portfolios in terms of return. This is in accordance with traditional financial theory as higher systematic risk means higher expected return, however recent papers found this effect to be diminished by the high risk attached to the portfolios. Volatility puzzle theory predicts that the low volatile portfolios have the highest Sharpe ratios because there is compensation for being risk averse. Our data does not reflect that thought as highly volatile

(13)

12

Table 1 Characteristics of the decile portfolios. The mean returns are annualized means of the monthly returns of the portfolios, ranging from 9.1% to 39.6% for the most volatile portfolio. Standard deviations are also annualized by multiplying with the square root of 12, increasing as the volatility increases, which means that realised volatility can predict future volatility in a way. Each portfolio has 636 months of data. Sharpe ratio’s range between 0.79 and 1.27, which is more than the Blitz and Van Vliet global portfolios (0.05 – 0.72) (2007, p. 15) and the large cap market portfolio Sharpe ratio of 0.36 of Clarke et al (2006, p. 14). Market Capitalization is more or less equally distributed over the ten portfolios. Correlation between returns of portfolio 1 and 10 is 0,4939.

Portfolio member

Mean Excess Return (Annualized)

Std. Dev.

(Annualized) n

Sharpe

Ratio Sharpe # Market Cap %

1 9.1% 11.5% 636 0.79 9 9.8% 2 11.1% 14.2% 636 0.79 10 9.7% 3 12.9% 15.6% 636 0.82 8 10.4% 4 15.8% 17.7% 636 0.89 7 10.0% 5 18.8% 20.0% 636 0.94 6 10.3% 6 21.2% 21.7% 636 0.98 5 10.3% 7 24.2% 24.6% 636 0.98 4 10.1% 8 30.0% 26.8% 636 1.12 3 10.1% 9 35.4% 30.0% 636 1.18 2 9.8% 10 39.6% 31.2% 636 1.27 1 9.7% Total 21.8% 6360 0.98 100%

Appendix C shows the historical performance of the Low volatility and the High volatility portfolios. The Low portfolio remains stable just above the 0% return in the graph while the High portfolio goes through a rough couple of spikes during the years, especially in the period beginning in the late 90’s and ending early 2000’s.

Results

Regressing the portfolio returns against the Fama French 3 factor model results in the same unexpected patterns as for table 1. High minus Low are the returns if the High portfolio returns are subtracted by the Low portfolio returns. This gives us an estimate of the volatility effect size. A positive volatility effect would provide us with negative alpha’s for the portfolio of interest. However a negative effect has been established by the data, contradicting volatility effect theory.

(14)

13

Table 2 Portfolio regressions, *indicates significant alpha with α ≤ 10% (** = α ≤ 5%, *** = α ≤ 1%) total values are simple averages. All portfolios have highly significant positive monthly alpha’s, ranging from 0.35% to 2.09%. Alpha’s generally become more significant when the volatility increases. The volatility effect is the size of the High minus Low alpha, which is 1.74% monthly, being statistically significant at the highest level.

Portfolio member Alpha Std. Err. T-value P>|t|

Low 0.0035*** 0.0006 5.79 0.000 2 0.0042*** 0.0006 7.18 0.000 3 0.0047*** 0.0006 8.09 0.000 4 0.0065*** 0.0006 10.91 0.000 5 0.0080*** 0.0007 11.52 0.000 6 0.0096*** 0.0008 11.28 0.000 7 0.0117*** 0.0011 10.87 0.000 8 0.0150*** 0.0012 12.51 0.000 9 0.0182*** 0.0015 12.43 0.000 High 0.0209*** 0.0017 12.52 0.000

High minus Low 0.0174*** 0.0020 8.87 0.000

Total 0.0102 0.0009

In order to check if this negative volatility effect persists in all periods, all subsamples of periods are regressed separately. Results of the regressions split up in the Low and High portfolio per period are available in Appendix D.

Table 3 Regressions per period. Indicated at the left are the dates of separation, the dates starting new recessions. April 1960 was the start of the last regression before December 1969, but our dataset starts on January 1963, so for that reason those months are indicated as well. N is the number of months before the next recession strikes and the total values are simple averages. Alpha’s are significant in all periods except for period 4 (a flawed period due to the low number of observations) and period 2. Highest alpha is found during the period before the dotcom bubble crash with a monthly alpha of 3.12% and the highest T-value significance (6.75).

Period High-Low

apr-60 # n Alpha Std. Err. T-value P>|t|

jan-63 1 83 0.0056* 0.0031 1.80 0.075 dec-69 2 48 0.0006 0.0040 0.15 0.880 nov-73 3 73 0.0181*** 0.0039 4.63 0.000 jan-80 4 18 0.0023 0.0106 0.21 0.834 jul-81 5 102 0.0092*** 0.0028 3.27 0.001 jul-90 6 134 0.0312*** 0.0046 6.75 0.000 mrt-01 7 81 0.0224*** 0.0059 3.80 0.000 dec-07 8 97 0.0216*** 0.0052 4.13 0.000 dec-15 Total 636 0.014 0.005

The estimated alpha’s for the last two periods, which are the most interesting because they surround the last financial crisis, are more or less the same when comparing them to earlier periods. It seems that alpha’s in period 6-8 are remarkably larger than the periods before, but that is a subject that

(15)

14 would need further research. There is no hard evidence in the data for patterns preventing us from creating subsamples based on post and pre labels.

In order to look at the periodical differences in volatility effect we ran a rolling regression based on 36 months of returns from the High-Low portfolio during the sample period. The image reflects the higher alpha in the last few periods but also provides some further insight to their values within each period. The size of the volatility effect seem to be unstable especially at the end of the sample period.

Figure 4 Rolling regression of High-Low portfolio. Monthly alpha’s are monthly are estimated for each quarter from 1966 until 2015. All alpha’s are based on the previous 36 months of returns to ensure Central Limit Theorem and mirror the

portfolio building strategy. Indicated with red vertical lines are the break-points between periods.

When the periods are divided in post and pre-recession subsamples, the regression output shows that all High alpha’s are significantly higher than 0. Also both regressions with the complete pre and post portfolios have positive alpha’s. This is possible because the sum of the portfolios is not equal to the complete market as explained before.

(16)

15

Table 4 Regressions per subperiod. Pre and Post indicates whether the data are cumulative portfolios originated from a subperiod before or after a recession. All regressions result in positive significant alpha’s. The confidence interval provides a 95% range for the high-low portfolio alpha’s given their underlying distribution. The Pre-Post column simply substracts the pre alpha’s by the post alpha’s.

Portfolio Alpha 95% Conf. Int. Std. Err. T-value P>|t| Pre-Post

Pre Full 0.0120*** 0.0008 14.64 0.000 0.0020 Low 0.0031*** 0.0010 3.15 0.002 -0.0012 High 0.0275*** 0.0027 10.01 0.000 0.0084 High-Low 0.0244*** 0.0181 0.0307 0.0032 7.64 0.000 0.0096 Post Full 0.0100*** 0.0006 16.61 0.000 Low 0.0043*** 0.0009 4.57 0.000 High 0.0191*** 0.0025 7.74 0.000 High-Low 0.0148*** 0.0090 0.0206 0.0029 5.05 0.000

The High minus Low portfolios carry positive alpha’s in both subperiods. Pre-recession alpha is 2.44% monthly and the post-recession alpha is only 1.48% monthly. This would suggest that during the period before the recession the negative volatility effect is higher, but we cannot observe a significant difference as both alpha’s lay within each other’s confidence interval even if equal distributions are assumed. This means that the hypothesis cannot be confirmed. The volatility premium is not significantly higher after a recession.

If we zoom in on the most recent recession, a closer look can be taken on the impact of the credit crunch in the late 2000’s on the volatility effect. The negative volatility effect seems to have increased as alpha has increased, but the number of observations is not high enough to observe a significant change.

Table 5 Regressions in the pre period 7 and post period 8, surrounding the financial crisis for the high minus low portfolio.

Pre Financial Crisis

n Alpha Std. Err. T-value P>|t| 95% Conf. Int.

High-Low 40 0.0184*** 0.0041 4.45 0.000 0.0100 0.0269

Post Financial Crisis

n Alpha Std. Err. T-value P>|t| 95% Conf. Int.

High-Low 48 0.0246*** 0.0084 2.93 0.005 0.0077 0.0414

A smaller scope of the rolling regression from Figure 4 does produce a remarkable finding in Figure 5. The negative volatility effect jumps right after the fall of investment bank Lehman Brothers and mortgage corporations Fanny Mae and Freddy Mac, often referred to as the peak of the financial crisis.

(17)

16

Figure 5 Rolling regression of High-Low portfolio. Monthly alpha’s are monthly are estimated for each quarter from 2003 until 2015. All alpha’s are based on the previous 36 months of returns to ensure Central Limit Theorem and mirror the

portfolio building strategy. Indicated with red vertical line is October 2008, when Lehman Brothers fell.

Robustness

To check for robustness an additional analysis is done with 3 key changes in respect to the main one described above. First of all the analysis has been done on 12/0/1 portfolios, where realised volatility is based on the 12 most recent months. Because of this the standard deviations suffer from more noise, but the shorter period could also reflect a more accurate image of the volatility of the stock as 36-month old data might be outdated.

Also added to the process are penny stocks. When penny stocks are included they primarily end up in the high volatile portfolios, giving those portfolios a small total market capitalization. This reduces the value of the general comparison as low volatile stocks need to be compared to their highly volatile peers instead of penny stocks. Appendix E, F and G provide more information on the portfolios in relation to penny stocks. Portfolio 10 for instance only contains 1.4% of the total market value, so we run the risk that this small amount of data receives disproportional attention in the upcoming analysis. Comparing portfolio 1 with portfolio 10 would be a comparison between 10.8% of

(18)

17 the market and 1.4%, which is unfair. However more stocks means a bigger dataset which provides insight to the complete market.

Lastly a different market proxy is taken. The Fama French market proxy does not completely cover the full market excess return of the CRSP database (see Appendix B). This results in an

inappropriate allocation of the market returns in the regressions. We have compiled a new CRSP Index to reflect the complete value-weighted CRSP database returns correctly. This does create an imbalance with the other Fama French factors. Because the CRSP Index has is used to compute monthly market returns as input for the regression, the SMB and HML factors are out of balance. These coefficients have been determined to optimise the joint explanatory power of the model (Fama & French, 1992). If one coefficient is altered, e.g. the market proxy, the others are automatically either under or overvalued.

The results can be checked in Appendix G. The new alpha’s are lower compared to the main alpha’s, probably because the new market proxy suits the portfolio returns better. Also the added penny stocks influence the returns negatively. General patterns are similar to the other analysis. It is remarkable though that the Pre negative volatility effect is significantly (when similar distributions are assumed) higher than in the Post subperiods.

To test the effect of the disvalued SMB and HML factors a CAPM regression is used as well (Black et al, 1972), which means that the SMB and HML factors are excluded and only the CRSP Index is taken into account. However the Capital Asset Pricing Model does not offer any new insights and confirms the other findings. Results can be viewed in Appendix H.

Conclusion

The volatility effect found in this thesis was negative in nature, meaning a positive risk premium on volatility was priced in. This contradicts the effect found by other researchers, however Bali & Cakici (2008) already found that using monthly data as a volatility proxy can result in insignificant findings, despite the higher power of future volatility predictions. This proofs once again that ex-pre volatility is a different measure than ex-post volatility. Ex-pre volatility can be estimated most precisely with a monthly volatility portfolio construction method that proofs to be worthless when trying to observe a volatility effect, but the ex-post volatility (approximated by daily volatility) works better as a volatility effect construction factor while has little to do with future return volatility. This suggests that the volatility is an anomaly rather than an additional risk factor, because the effect is not robust and has no fundament information about the future returns.

(19)

18 Besides the volatility estimation period some other factors are possible reasons for the negative effect as well. Our study has excluded penny stocks, but some other papers restrict the database even further by only including large cap firms. Also this thesis split the market up in deciles where quintiles are being used to separate the market as well regularly. This forces the results into the extremes and can have resulted in different outcomes.

Taking a closer look at the size of the negative effect shows us that the difference between the High volatility deciles and the Low volatility deciles is subject to many changes over the years. It seems that the volatility of the volatility effect has risen and so did the average effect size. A weak effect of recessions on the volatility puzzle has been found, but not enough to be make academic statements about it. Recessions in general seem to have a negative effect on the negative volatility effect, so a net positive effect on the volatility puzzle. This is in spite of the hypothesis that pre-recession volatility effect would outperform the post-recession.

The periods before and after the last financial crisis point at a different direction. The data hints that the negative volatility has increased, increasing the risk premium on volatile portfolios, however this cannot be confirmed at a scientific significance level.

Figure 5 shows the same increase as the effect spikes after the financial markets crash, suggesting that even though Ang et al (2006, p. 295) concluded that the volatility effect is robust to market sentiment, there is some relation between general market movements and the volatility effect.

The higher risk premium on risky portfolios during pre-recessions periods directly conflicts the arguments given in the introduction for the opposite statement. This means that investments after a shocking financial event are encouraged to be distributed in a riskier fashion. Low volatility based portfolios should therefore not be held during these periods as they do not offer any free lunches or other hints of market inefficiencies. Asset management should not be focused on the volatility effect as a holy grail, but carefully differentiate between stock, because risk premiums are priced in returns.

The volatility anomaly remains somewhat of a puzzle and it not still certain to which limits it extends. In relation to the general market movements the effect has been explained a bit more, but there still is enough room for further research. The correlation with (lagged) market factors are for instance interesting to take a look at. How does the market sentiment or market volatility directly affect the volatility risk premium on portfolios? Also event studies of impactful incidents should be examined further to improve understanding. The Lehman Brothers spike promises more interesting findings to be discovered.

(20)

19

Reference list

Ang, A., Hodrick, R.J., Xing, Y. & Zhang, X. (2006). The cross-section of volatility and expected returns. Journal of Finance, Vol. 61, pp. 259-299.

Bacon, C. (2010, December 20). Performance Fees – Good or Bad? Retrieved from http://www.statpro.com/blog/performance-fees-good-or-bad/

Baker, M. & Wurgler, J. (2007). Investor sentiment in the stock market. Journal of Economic

Perspectives, Vol 21, Issue 2, Spring edition, pp. 129-151.

Bali, T. G. & Cakici, N. (2008). Idiosyncratic volatility and the cross section of expected returns.

Journal of Financial and Quantitative Analyisis, Vol 43, pp. 29-58.

Black, F., Jensen, M.C. & Scholes, M.S. (1972), The capital asset pricing model: some empirical tests. in M.C. Jensen (Ed.), Studies in the theory of capital markets. New York: Praeger Publishers Inc.

Blitz, D.C. & Van Vliet, P. (2007). The volatility effect: lower risk without lower return. Journal of

Portfolio Management, Fall edition, pp. 102-113.

Clarke, R., De Silva, H., & Thorley, S. (2006). Minimum-variance portfolios in the US equity market.

Journal of Portfolio Management, Fall edition, pp. 10-24.

Dutt, T. & Humphery-Jenner, M. (2013). Stock return volatility, operating performance and stock returns: international evidence on drivers of the ‘low volatility’ anomaly. Journal of Banking

& Finance. Vol 37, pp. 999-1017.

Fama, E.F. & French, K.R. (1992), The cross-section of expected stock returns. Journal of Finance, Vol 47, pp. 427-465.

Guasoni, P. & Obłój, J. (2016). The incentives of hedge fund fees and high-water marks.

Mathematical Finance, Vol 26, Issue 2, pp. 269–295.

Haugen, R.A. & Baker, N.L. (1991), The efficient market inefficienty of capatalization-weighted stock portfolios. The Journal of Portfolio Management, Spring edition, pp. 35-40.

Haugen, R.A. & Baker, N.L. (2012). Low risk stocks outperform within all observable markets of the

world. Unpublished manuscript, Retrieved from http://dx.doi.org/10.2139/ssrn.2055431 Jegadeesh, N. & Titman, S. (1993). Returns to buying winners and selling losers: implications for stock

market effiency. Journal of Finance, Vol 48, pp. 65-92.

National Bureau of Economic Research. (2010). US Business Cycle Expansions and Contractions. Retrieved from: http://www.nber.org/cycles.html

Van Doesburg, P. & De Kleer, M. (2011). Are performance fees beneficial to mutual fund investors?

(21)

20

Appendix A: Stata commands for creating decile portfolio’s

import delimited E:\CRSP.csv replace prc=abs(prc)

destring ret, generate(ret2) force drop ret

rename ret2 ret

replace ret=. if ret==-66.0 | ret==-77.0 | ret==-88.0 | ret==-99.0 drop if missing(ret)

drop if prc<1

merge m:1 date using "E:\FFfactors.dta" drop _merge

label variable hml "High minus Low" label variable smb "Small minus Big" label variable rf "Risk-free rate (monthly)" label variable mktrf "Market excess return" label variable umd "Momentum"

generate marketcap = shrout*prc generate daterank = date(date, "MDY") tsset permno daterank, monthly compress

generate rmin1 = ret[_n-1] generate rmin2 = ret[_n-2] generate rmin3 = ret[_n-3] generate rmin4 = ret[_n-4] generate rmin5 = ret[_n-5] generate rmin6 = ret[_n-6] generate rmin7 = ret[_n-7] generate rmin8 = ret[_n-8] generate rmin9 = ret[_n-9] generate rmin10 = ret[_n-10] generate rmin11 = ret[_n-11] generate rmin12 = ret[_n-12] generate rmin13 = ret[_n-13] generate rmin14 = ret[_n-14] generate rmin15 = ret[_n-15] generate rmin16 = ret[_n-16] generate rmin17 = ret[_n-17] generate rmin18 = ret[_n-18] generate rmin19 = ret[_n-19] generate rmin20 = ret[_n-20] generate rmin21 = ret[_n-21]

(22)

21 generate rmin22 = ret[_n-22]

generate rmin23 = ret[_n-23] generate rmin24 = ret[_n-24] generate rmin25 = ret[_n-25] generate rmin26 = ret[_n-26] generate rmin27 = ret[_n-27] generate rmin28 = ret[_n-28] generate rmin29 = ret[_n-29] generate rmin30 = ret[_n-30] generate rmin31 = ret[_n-31] generate rmin32 = ret[_n-32] generate rmin33 = ret[_n-33] generate rmin34 = ret[_n-34] generate rmin35 = ret[_n-35] generate rmin36 = ret[_n-36]

egen vola = rowsd(rmin1 rmin2 rmin3 rmin4 rmin5 rmin6 rmin7 rmin8 rmin9 rmin10 rmin11 rmin12 rmin13 rmin14 rmin15 rmin16 rmin17 rmin18 rmin19 rmin20 rmin21 rmin22 rmin23 rmin24 rmin25 rmin26 rmin27 rmin28 rmin29 rmin30 rmin31 rmin32 rmin33 rmin34 rmin35 rmin36)

label variable vola "1-year historic volatility" generate exret = ret - rf

label variable exret "Excess return" generate valexret = marketcap*exret label variable valexret "Value excess return" drop if missing(rmin1) drop if missing(rmin2) drop if missing(rmin3) drop if missing(rmin4) drop if missing(rmin5) drop if missing(rmin6) drop if missing(rmin7) drop if missing(rmin8) drop if missing(rmin9) drop if missing(rmin10) drop if missing(rmin11) drop if missing(rmin12) drop if missing(rmin13) drop if missing(rmin14) drop if missing(rmin15) drop if missing(rmin16) drop if missing(rmin17) drop if missing(rmin18) drop if missing(rmin19) drop if missing(rmin20) drop if missing(rmin21) drop if missing(rmin22) drop if missing(rmin23) drop if missing(rmin24) drop if missing(rmin25) drop if missing(rmin26)

(23)

22 drop if missing(rmin27) drop if missing(rmin28) drop if missing(rmin29) drop if missing(rmin30) drop if missing(rmin31) drop if missing(rmin32) drop if missing(rmin33) drop if missing(rmin34) drop if missing(rmin35) drop if missing(rmin36) drop if missing(exret) sort daterank ssc install egenmore

egen rankvola = xtile(vola), by(daterank) p(10(10)90) label variable rankvola "Portfolio member"

compress

sort daterank rankvola

by daterank rankvola: egen portmarketcap = total(marketcap) label variable portmarketcap "Portfolio Value"

by daterank rankvola: egen portvalexret = total(valexret) label variable portvalexret "Portfolio Value Return" generate portret = portvalexret / portmarketcap label variable portret "Portfolio Excess Return"

collapse (mean) portret vola mktrf smb hml umd rf, by (daterank rankvola) merge m:m date using "E:\Thesis backup.dta", nogenerate

drop rmin1 rmin2 rmin3 rmin4 rmin5 rmin6 rmin7 rmin8 rmin9 rmin10 rmin11 rmin12 rmin13 rmin14 rmin15 rmin16 rmin17 rmin18 rmin19 rmin20 rmin21 rmin22 rmin23 rmin24 rmin25 rmin26 rmin27 rmin28 rmin29 rmin30 rmin31 rmin32 rmin33 rmin34 rmin35 rmin36 exret valexret

portmarketcap portvalexret marketcap shrout ret prc permco comnam ticker permno keep in 1/6360

Appendix B: CRSP Index returns

Rate (Monthly, Annualized) Mean Std. Dev. Min Max Fama French market excess returns 6.29% 15.35% -95.82% 499.77%

CRSP Index excess returns 14.64% 15.30% -94.72% 572.20%

(24)

23

(25)

24

Appendix C: Low and High portfolio returns

Appendix D: Subsample regressions of Low and High portfolio

Period Low High

apr-60 # n Alpha Std. Err. T-value P>|t| Alpha Std. Err. T-value P>|t| jan-63 1 83 0.0021** 0.0009 2.38 0.020 0.0077*** 0.0028 2.77 0.007 dec-69 2 48 0.0349*** 0.0010 3.54 0.001 0.0041 0.0041 1.00 0.320 nov-73 3 73 0.0030** 0.0013 2.28 0.026 0.0211*** 0.0033 6.37 0.000 jan-80 4 18 0.0064** 0.0023 2.72 0.017 0.0086 0.0097 0.89 0.389 jul-81 5 102 0.0042*** 0.0011 3.72 0.000 0.0134*** 0.0024 5.54 0.000 jul-90 6 134 0.0040** 0.0016 2.53 0.013 0.0353*** 0.0412 8.56 0.000 mrt-01 7 81 0.0044** 0.0017 2.58 0.012 0.0268*** 0.0052 5.15 0.000 dec-07 8 97 0.0016 0.0017 0.95 0.345 0.0232*** 0.0042 5.56 0.000 dec-15 Total 636 0.0076 0.0015 0.0175 0.0091

(26)

25

Appendix E: Penny stocks

Observations Mean MarketCap Mean Return (Annualized) Return S.D. (Annualized) Penny Stocks 191,413 $ 12,321.26 -40.90% 89.20%

Total observations were 3,690,003

Which portfolio contains how many penny stock observations? (Based on 1-year realised volatility)

Portfolio member Percent Cum. Average Price ($)

1 0.88% 0.88% 41.71 2 0.87% 1.75% 63.78 3 1.34% 3.09% 42.45 4 2.12% 5.21% 27.49 5 3.50% 8.71% 22.69 6 5.63% 14.34% 18.31 7 9.14% 23.49% 15.45 8 14.49% 37.98% 12.65 9 22.65% 60.63% 10.07 10 39.37% 100% 7.67

Appendix F: Large companies

Mean Return (Annualized) Return S.D. (Annualized)

Top 10% High value 22.80% 30.95%

CRSP Market 19.52% 15.30%

Which portfolio contains how much of the top 10% largest firm observations? (Based on 1-year realised volatility)

Portfolio member Percent Cum. % of total Market Cap

1 10.45% 10.45% 10.8% 2 16.63% 27.07% 19.0% 3 17.14% 44.21% 19.2% 4 15.53% 59.74% 16.4% 5 12.86% 72.61% 12.2% 6 9.88% 82.49% 8.6% 7 7.25% 89.74% 6.0% 8 5.24% 94.99% 3.9% 9 3.36% 98.35% 2.6% 10 1.65% 100% 1.4%

(27)

26

Appendix G1: Portfolio characteristics, robust

Portfolio member Mean Excess Return (Annualized)

Std. Dev.

(Annualized) Freq.

Sharpe

Ratio Sharpe # Market Cap %

1 9.4% 11.6% 636 0.81 9 10.8% 2 11.4% 13.3% 636 0.86 8 19.0% 3 11.6% 15.3% 636 0.76 10 19.2% 4 15.8% 17.3% 636 0.91 6 16.4% 5 16.5% 18.5% 636 0.89 7 12.2% 6 19.6% 21.4% 636 0.92 5 8.6% 7 23.3% 23.6% 636 0.99 4 6.0% 8 27.1% 26.8% 636 1.01 3 3.9% 9 35.6% 30.2% 636 1.18 2 2.6% 10 44.0% 33.9% 636 1.30 1 1.4% Total 21.0% 22.4% 6360 0.96 100%

(28)

27

Appendix G3: Portfolio regressions, robust

Portfolio member Alpha Std. Err. T-value P>|t|

Low -0.0008 0.0007 -1.15 0.252 2 -0.0009 0.0006 -1.63 0.103 3 -0.0025*** 0.0006 -4.44 0.000 4 -0.0007 0.0005 -1.27 0.205 5 -0.0004 0.0007 -0.65 0.515 6 0.0008 0.0008 1.09 0.276 7 0.0032*** 0.0009 3.41 0.001 8 0.0043*** 0.0012 3.73 0.000 9 0.0090*** 0.0014 6.52 0.000 High 0.0135*** 0.0019 7.15 0.000

High minus Low 0.0143*** 0.0022 6.42 0.000

Total 0.0026 0.0009

Appendix G4: High-Low regressions per period, robust

Period High-Low

apr-60 # n Alpha Std. Err. T-value P>|t|

jan-63 1 83 0.0087** 0.0038 2.30 0.024 dec-69 2 48 0.0050 0.0057 0.86 0.393 nov-73 3 73 0.0157*** 0.0039 4.03 0.000 jan-80 4 18 -0.0151 0.0119 -1.27 0.225 jul-81 5 102 0.0098*** 0.0035 2.80 0.006 jul-90 6 134 0.0272*** 0.0052 5.21 0.000 mrt-01 7 81 0.0141** 0.0061 2.31 0.023 dec-07 8 97 0.0133** 0.0062 2.15 0.035 dec-15 Total 636 0.010 0.006

Appendix G5: High and Low portfolio regressions per period, robust

Period Low High

apr-60 # n Alpha Std. Err. T-value P>|t| Alpha Std. Err. T-value P>|t| jan-63 1 83 -0.0015 0.0009 -1.62 0.108 0.0072** 0.0033 2.16 0.034 dec-69 2 48 -0.0007 0.0014 -0.47 0.642 0.0043 0.0059 0.73 0.469 nov-73 3 73 -0.0034** 0.0017 -2.00 0.050 0.0123*** 0.0035 3.57 0.001 jan-80 4 18 0.0035 0.0044 0.79 0.440 -0.0116 0.0091 -1.27 0.225 jul-81 5 102 -0.0020 0.0014 -1.44 0.154 0.0078** 0.0032 2.48 0.015 jul-90 6 134 -0.0023 0.0016 -1.48 0.141 0.0249*** 0.0046 5.36 0.000 mrt-01 7 81 0.0002 0.0016 0.11 0.912 0.0143** 0.0055 2.60 0.011 dec-07 8 97 0.0017 0.0018 0.96 0.342 0.0150*** 0.0051 2.93 0.004 dec-15 Total 636 -0.0006 0.0019 0.0093 0.0050

(29)

28

Appendix G6: Regressions per subperiod, robust

Portfolio Alpha Std. Err. T-value P>|t| Pre-Post 95% Conf. Int.

Pre Full 0.0030*** 0.0007 4.30 0.000 0.0006 Low -0.0019* 0.0011 -1.77 0.078 -0.0020 High 0.0172*** 0.0027 6.46 0.000 0.0062 High-Low 0.0191*** 0.0032 6.00 0.000 0.0083 Post Full 0.0024*** 0.0006 4.19 0.000 0.001252 0.003455 Low 0.0001 0.0009 0.16 0.869 -0.00159 0.00188 High 0.0110*** 0.0025 4.37 0.000 0.00604 0.015928 High-Low 0.0108*** 0.0029 3.75 0.000 0.005148 0.016529

Appendix H1: CAPM Portfolio regressions

Portfolio member Alpha Std. Err. T-value P>|t|

1 0.0003 0.0007 0.41 0.682 2 0.0000 0.0006 0.00 0.998 3 -0.0016*** 0.0005 -2.90 0.004 4 -0.0002 0.0005 -0.42 0.673 5 -0.0005 0.0007 -0.72 0.470 6 0.0000 0.0009 0.02 0.982 7 0.0015 0.0011 1.38 0.167 8 0.0025* 0.0014 1.80 0.073 9 0.0065*** 0.0018 3.67 0.000 10 0.0107*** 0.0022 4.82 0.000 10 minus 1 0.0104*** 0.0029 3.58 0.000 Total 0.003 0.0012

Appendix H2: High and Low portfolio regressions per period, CAPM robust

Period Low High

apr-60 # n Alpha Std. Err. T-value P>|t| Alpha Std. Err. T-value P>|t|

jan-63 1 83 -0.002 0.0011 -1.64 0.104 0.012** 0.0051 2.46 0.016 dec-69 2 48 0.001 0.0014 0.56 0.581 -0.005 0.0056 -0.89 0.378 nov-73 3 73 -0.004** 0.0017 -2.40 0.019 0.022*** 0.0052 4.18 0.000 jan-80 4 18 0.005 0.0050 1.03 0.319 -0.008 0.0106 -0.74 0.470 jul-81 5 102 0.002 0.0016 1.17 0.246 0.003 0.0034 0.79 0.432 jul-90 6 134 0.021 0.0020 1.04 0.302 0.011 0.0065 1.64 0.104 mrt-01 7 81 0.002 0.0017 0.93 0.357 0.007 0.0063 1.19 0.239 dec-07 8 97 0.002 0.0020 0.91 0.363 0.014*** 0.0052 2.75 0.007 dec-15 Total 636 0.003 0.002 0.007 0.006

(30)

29

Appendix H3: Regressions per subperiod, CAPM robust

Portfolio Alpha Std. Err. T-value P>|t| Pre-Post

Pre Full 0.0014** 0.0006 2.19 0.029 -0.001 Low 0.0006 0.0011 0.53 0.599 0.000 High 0.0092*** 0.0033 2.78 0.006 -0.003 High-Low 0.0086 -0.003 Post Full 0.0025*** 0.0006 4.41 0.000 Low 0.0001 0.0009 0.15 0.881 High 0.0122*** 0.0030 4.14 0.000 High-Low 0.0121

Referenties

GERELATEERDE DOCUMENTEN

First and third order peripheral equipment energy demands are split into value adding and non-value adding shares because they are considered to be directly energetically

Keywords: Corporate Social Responsibility (CSR), Corporate Political Activity (CPA), complementarity, CSR-CPA complementarity, Government Dependence, Corporate Financial

The  Swedish  International  Development  Agency  (Sida)  has  been  supporting  the  University  Eduardo  Mondlane  (UEM)  since  1978.  Currently  Sida  is 

To analyze the multilayer structure we combined the Grazing Incidence X-ray Reflectivity (GIXRR) technique with the analysis of the X-rays fluorescence from the La atoms excited

Despite the overt effects of vaccination on lung virus titers several observations indicate that vaccination did not result in sterilizing immunity: (i) the breathing frequency

The high reaction order of 2 in hydrogen as well as the negative order in nitrite at low hydrogen pressures (0.05 bar) have never reported before to the best of our knowledge, which

With respect to the first question (What work process can characterize creation and production in the creative industries?), we derived a six-phase process, based on analysis

Responsible innovation; liminal innovation; emerging technologies; anticipation; clinical practice; postanoxic coma; practice-based