The predictability of stock markets using business cycles and investor sentiment

(1)

The predictability of stock markets using business cycles

and investor sentiment

Eveline M. Stolk Student number: s1554182 Supervisors: Dr. R.M. Salomons Drs. B.A. Boonstra Amsterdam, February 4, 2011 University of Groningen

Faculty of Economics and Business

Master Thesis, MSc Business Administration

(2)

1

The predictability of stock markets using business cycles

and investor sentiment

Abstract

This study investigates the practical relevance of business cycles and investor sentiment for investment strategies. I analyze monthly returns for the stock markets of the US, Japan and Asia Pacific over the business cycle. The results show that long-short portfolios based on sector analyses generate on average significant positive returns for the US as well as Asia. Being able to predict stock market using the business cycle can thus be beneficial, by employing sector rotation. In addition, I find that stock returns are higher (lower) after extreme levels of negative (positive) sentiment. However, for Japan, investor sentiment is not a reliable contrarian indicator.

JEL Codes: E32, G11, G12, G14

(3)

2

Acknowledgements

(4)

3

1. Introduction

“The stock market has predicted nine of the last five recessions.”

This famous quote from Paul Samuelson (1966) suggests that the stock market is not a stable predictor of the macro economic environment. However, instead of letting the stock market forecast the state of the economy, what about the practical relevance of macroeconomics for investment strategies?

Much has been written about the predictability of stock returns. Findings in empirical literature suggest that returns can be predicted to some extent; patterns in stock returns were discovered, just as variables that can be used to forecast them (Campbell and Shiller (1987), Kothari and Shanken (1997), Pontiff and Schall (1998)). Instead of investigating the predictability of equity markets using long term cycles, I aim to examine shorter cycles, focusing on business cycles and investor sentiment. Are these factors able to explain movements in stock returns?

The role of macroeconomic forces has been discussed for quite some time. After the introduction of the Arbitrage Pricing Theory, which made it possible to include variables other than solely the market in asset pricing, Chen et al. (1986) identified several macroeconomic factors that can explain stock returns. Regarding the direction of the relation between macro economy and stock returns, the authors state that no satisfactory theory would argue that the relation between financial markets and the macro economy is entirely in one direction. However, stock prices are usually considered as responding to external forces (even though they may have a feedback on the other variables). Galbraith (1954) is more firm regarding this subject: the stock market is but a mirror which provides an image of the underlying or fundamental economic situation. Cause and effect run from the economy to the stock market, never the reverse. In this paper, I will study stock returns over the phases of the business cycle for both the United States and Asia. In addition to looking at the stock markets in general, I will study different sectors to investigate whether sector rotation over the various phases of the cycle generates higher returns by constructing long-short portfolios.

Next to studying business cycles, I will investigate even shorter cycles: those in investor sentiment. Sentiment indicates how an investor feels about the future prospects of investments. It is mostly driven by fear and greed, which are the two strongest emotions that an investor experiences. The flows of optimism and pessimism cause respectively bullish and bearish views on the market, which causes prices to overshoot fundamental values. Although sentiment does not play a role according to classical theory, market participants do believe in it nowadays. When the stock market crash in October 1987 could not be explained by fundamental factors, changes in investor sentiment, unrelated to fundamentals, became a possible explanation. Indeed, shifts in sentiment appeared to be significantly associated with changes in stock market returns (Siegel, 1992). In this paper I will study the (cyclical) features of investor sentiment and examine its relation to stock returns.

(6)

5

I find that the OECD leading indicator is the most reliable indicator of the business cycle. Consequently, I use this indicator to develop a portfolio strategy. Depending on the phase of the cycle, the strategy is to go long in the two best performing sectors and to go short in the two worst performing sectors in that particular phase. I find that a US long-short portfolio generates an average monthly return of 1.03%. The long-short portfolio for Japan has an average return of 1.58% and the Asian long-short portfolio generates 1.14% on average per month. To avoid an in-sample bias, I also look at the performance of portfolios for Japan and Asia Pacific, based on the US sector analysis. Furthermore, I included a lag in the portfolios, since the assumption of perfect forecasting of phases is not realistic. As a result, the average returns decrease but remain positive and statistically significant from zero.

Next to studying business cycles, I have analyzed investor sentiment, using two measures: the Global Risk Appetite Index (GRAI) of Credit Suisse and the weekly bull-bear survey of the American Association of Individual Investors (AAII). I find that, when looking at extreme levels of the GRAI (more than two standard deviations from the mean), positive sentiment is followed by lower returns and negative sentiment is followed by higher returns for the US and Asia Pacific. The results for the AAII show that for extreme levels of bullish sentiment and bearish sentiment, the expected relation between stock returns and extreme sentiment can also be found. However, Japan is quite unstable with respect to the use of sentiment as a contrarian indicator.

(7)

6

2. Business Cycles: Theoretical Background

2.1. The business cycle

The business cycle is defined as the recurring fluctuations in economic activity. It can be seen as periodic but irregular movements, reflecting changes in the state of the economy. The National Bureau of Economic Research (NBER) divides a business cycle into contractions and expansions. A period of expansion begins at the trough date and ends at the following peak date, whereas a period of contraction begins at the peak date and ends at the following trough date. In academic literature, however, a business cycle often consists of more than two phases. According to Taylor (1998), a business cycle can be divided into four major stages. Taylor focuses on growth cycles, with an emphasis on the output gap, measuring the difference between actual growth and potential growth. Put differently, the output gap is a measure of excess capacity.

The four phases of the growth cycle can be summarized as follows1:

1) Recession: The growth rate is below trend, falling towards a trough.

2) Early recovery: The growth rate is below trend, but rising towards the trend rate. 3) Late expansion: The growth rate is above trend, rising towards a peak.

4) Initial slowdown: The growth rate is above trend, but declining towards the trend rate. In this paper, these phase definitions will be followed.

Figure 1: A stylized business cycle

Growth and inflation are said to be two main determinants of the cycle. Figure 1 presents a graphical display. Another relevant economic indicator that can be used to describe the phases of the cycle is unemployment. To be more elaborative, the four phases can be characterized as follows:

1) The economy is in recession. Growth is below trend. Inflation slows down due to excess spare capacity and central banks cut rates in order to stimulate the economy. Low demand and falling prices result in increasing unemployment, since firms lower supply as a response to the low demand for goods and services.

2) Economy has hit the bottom at the trough. Growth expectations increase when the lower interest rates take effect. Inflation falls, because spare capacity is not used. Low prices stimulate consumption: demand starts increasing again and firms respond with a higher

1_{NB: The definition of the phases varies a lot among authors. I do not follow Taylor in defining phase 1 as the first}

(8)

7

supply. The economy starts to recover. More people get employed and earn money. Thereby the demand for goods and services will further increase. Companies have fat profit margins. 3) Economic expansion leads to growth above trend level. Eventually, inflation rise starts to rise

due to capacity constraints. When prices are increasing, people demand higher wages. However, these costs will translate into higher prices, causing inflation to keep rising. Companies are slowly facing margin pressures as input costs rise. Central banks raise rates to slow economic growth.

4) A cyclical peak in growth, followed by a decline in growth rate. Inflation is still accelerating, so monetary policy remains tight. At a certain point, demand for goods will start to decrease because goods have become too expensive. As a response, companies decrease supply and lay off workers: unemployment rises. Profits of companies are under severe pressure. The economy slows down and will enter recession when conditions worsen. This first contraction phase is the worst part of the cycle.

Defining the phases of the cycle

There are several ways to define the phases of the business cycle. One of them is by using the yield curve. Since the late 1980s, a large amount of research was conducted on the yield curve as leading indicator of the business cycle. The slope of the yield curve has appeared to be a reliable predictor of future real economic activity. The slope, also called term spread, is the difference between long term interest rates and short term interest rates.

Estrella and Mishkin (1996) find that, among a set of financial variables, the slope of the yield curve has the best performance as leading indicator beyond a one-quarter horizon. The predictive power was observed for various measures of real economic activity such as changes in GDP, industrial production, consumption and recessions as defined by the NBER. Regarding recessions, every NBER recession since 1968 was predicted by an inversion of the yield curve (at least three months with negative spreads) in the twelve months preceding the start of the recession2. Thereby, an inverted yield curve is said to be an important signal with respect to the state of the economy. The phases of the business cycle can be defined based on movements in long-term and short-term interest rates:

1) In a recession, monetary policy eases in order to stimulate the economy. Long-term rates decrease less than short-term rates, resulting in a steeper curve (bull steepening).

2) Growth puts pressure on long-term rates, but short-term rates remain low because of low inflation expectations. This leads to a steeper yield curve (bear steepening).

3) Monetary policy tightens because of inflation expectations. Long-term rates increase, but less than short-term rates. As a result, the yield curve flattens (bear flattening).

4) Although inflation is rising, long-term interest rates peak when a slowing credit growth overtakes. Long-term rates decrease more than short-term rates, resulting in a flatter yield curve (bull flattening).

2.1.1. Business cycle and stock markets

According to Taylor (1998), it is the business cycle which captures and articulates the evolution of economy through time and thus holds the key to successful market timing. In order to time the market successfully, Taylor argues that it is crucial to identify which types of assets are most

2

(9)

8

favorable during the different phases of the business cycle. He summarizes the four phases with regard to the performance of asset classes as follows:

Phase 1: Equities and bonds perform exceptionally well. Phase 2: Equities and bonds outperform cash

Phase 3: Equities marginally outperform cash; bonds perform poorly Phase 4: Cash and bonds dramatically outperform equity

In this paper, I focus on stock markets. Bolten and Weigand (1998) show that the relation between the stock market and the business cycle can be expressed by the dividend discount model. In advance, they discuss the four stages of the stock market cycle, which is leading the business cycle. The direction of the stock price changes is determined by the interaction between expected future earnings and interest rates. According to the authors, this is the proposed performance of equity in the different phases of the cycle:

1) When the economy is in recession, bad economic expectations have a negative impact on expected future earnings, causing decreasing stock prices. Interest rates start falling and stock prices keep declining until rates have decreased considerably. Eventually, stock prices rebound before the economy is at the trough.

2) The economy has hit the bottom and starts to recover. Stock prices are positively affected by expectations of positive economic growth and higher future earnings. Low interest rates cause the cost of capital to be low, increasing stock prices. In addition, the low-yielding bonds are not attractive, which stimulates investors to transfer wealth to stocks. The combination of these factors leads to a relatively quick rise of stock prices.

3) The growing economy leads to inflationary pressures, causing the interest rates to rise. This implies a higher cost of capital. However, the higher expected earnings dominate this negative effect. Stock prices are still rising, only not as fast as in the previous phase.

4) Inflation keeps rising, which puts more pressure on the interest rates. Productivity slumps and earnings growth slows down. Stock prices rise slowly and reach their peak, before the economy reaches its peak. Because of the high interest rates, investors transfer wealth from stocks to bonds, which decreases stock prices.

2.1.2. Empirical literature

Various papers are suggesting that the variability of asset returns can be linked to estimated stages of the business cycle. Empirical work has found systemic movements in stock returns, which can be interpreted as a relation to macroeconomic conditions. One of the first studies that focused on the effects of the business cycle on stock returns and volatility was conducted by Schwert (1989). Schwert runs regressions on a dummy variable that indicates whether the economy is in recession or not. For the period between 1855 and 1987, he finds that stock returns are 60 percent more volatile in times of recession.

(10)

9

dynamic general equilibrium model. He argues that the presented model is useful in describing the cyclical behavior of returns.

DeStefano (2004) follows Bolten and Weigand (1998) in showing movements in economic factors and stock prices based on the dividend discount model. He concludes that “stock returns and determinants as dictated by the DDM are shown to possess clear business-cycle patterns” (DeStefano, 2004). He identifies business cycles using the NBER peak and trough dates and cuts the expansion and contraction phases in half to end up with four stages. Stock returns are positive in the early expansion phase, but decline in the later expansion phase. In the early recession phase , returns are very negative, but become positive again during the second recession phase. DeStefano (2004) emphasizes the role of expected earnings as explanation for the large increase in returns in phase 4.

2.2. Sector Rotation

Next to deciding when to shift from stocks to bonds and vice versa, it is relevant to determine where to invest within the stock market. Investors can anticipate that different sectors are strong in different phases of the cycle. With full understanding of the economic cycle, an investor can rotate into areas that are expected to do well. Stovall (1996) has studied historical data for stocks in various sectors and concludes that a pattern of sector rotation is apparent. He categorizes all industries among ten basic sectors. For each sector he determines in which phase of the business cycle it achieves the highest performance. It has to be said that Stovall divides the business cycle into five phases: three phases of expansion and two of recession. The reason is that the average duration of expansions is longer.

A sector is called cyclical when its performance is dependent on economic fluctuations. When a sector is more “immune” to the state of the economy, it is called defensive. Consequently, cyclical sectors are expected to do well during expansions, whereas defensive sectors tend to outperform in times of contraction. However, the cyclical and defensive sectors can be further broken down into four categories, by making the distinction between growth and value. Growth stocks are stocks of companies that have a substantial higher growth rate than other companies (focus on the future potential and subsequently higher valuations). Value stocks are stocks trading at a lower price relative to their fundamentals (focus on the current price). Based on this distinction, four categories emerge. Each category is expected to outperform in a certain phase of the business cycle. The four categories, linked to the phase in which they are likely to outperform, can be defined as follows:

- Phase 1 - defensive growth: companies that are not dependent on economic fluctuations, but

do show above-average growth relative to the market.

- Phase 2 - cyclical growth: companies that are dependent on macroeconomic developments,

but with a higher growth rate than the market.

- Phase 3 - cyclical value: companies that are sensitive to economic fluctuations and whose

growth is purely dependent on the business cycle.

- Phase 4 - defensive value: companies of which the demand for products remains stable

during economic downturn, because the products are always needed, regardless of the phase of the business cycle.

(11)

10

rotation can result in maximally 2.3% annual outperformance. When transaction costs are added and assumptions are relaxed, this outperformance disappears. The authors conclude that, based on conventional wisdom, investors are not able to benefit from sector rotation.

In this paper, I use ten sectors in order to examine whether sector rotation can be beneficial: Consumer Staples (CS), which are products that are bought regardless of economic conditions (e.g. food and tobacco), Consumer Discretionary (CD), which are nonessential products (e.g. luxury goods), Information Technology (IT), Energy, Health Care, Utilities, Materials, Industrials, Financials and Telecommunication (Telecom). These sectors can be assigned to the phase of the cycle in which they are likely to achieve the highest or the lowest performance, compared to the other sectors. That is, for each phase of the cycle we can hypothesize in which sectors one should go long and go short. Following Stovall and Jacobsen et al., my hypothesis is as follows:

2.3. Business cycles in Asia

In this paper, I will focus on the United States stock market as well as on the Japanese and Asian Pacific stock markets. Most business cycle literature focuses on developed countries, whereas emerging economies receive less attention. Since the Asian economies have experienced a rapid economic development, it is interesting to compare their cyclical movements to those of the US.

(12)

11

business cycle. The authors conclude that the Asian economies have a higher volatility than the G7-countries, but the amplitude of economic fluctuations decreases over time, causing more similarities between the business cycles. There are important resemblances between the cyclical movements in Asia and the G7-countries. In addition, the authors suggest that there is an Asian-specific regional business cycle, because of the high degree of co movement between the individual country business cycles and different measures of the Asian business cycle.

Moneta and Rüffer (2006) examine whether business cycle synchronization exists in East Asia. Because of the economic expansion in the region, the question is if the increased inter and intra-regional integration (in particular trade) is the driving force behind the synchronization of business cycle. From a theoretical point of view, it is not clear in which direction the relationship between economic integration between countries and the co-movement of their business cycles runs. On the one hand, integration causes spill-over effects of demand shocks, which enforces the co- movement of business cycles. On the other hand, trade integration stimulates specialization in production. Thereby, industry-specific shocks do not affect countries to the same extent, reducing business cycle synchronization. Based on the empirical results, the authors conclude that the East Asian synchronization is mainly driven by developments outside the region, rather than by the inter- and intraregional integration. Especially the co-movement in exports accounts for a considerable part of the co-movement.

(13)

12

3. Methodology

In order to analyze the predictability of stock returns, the first step is to find a reliable indicator of the business cycle. I have looked at various indicators to use as a proxy. One can use real data, such as industrial production, which is a coincident indicator of overall economic activity. Another option using real data is the combination of inflation and the unemployment rate, which tend to lag the cycle. In addition, I can use a leading indicator (the OECD composite leading indicator) and/or market implied data (the slope of the yield curve).

In this section, I will discuss these four types of indicators and, if relevant, the accompanying definitions for the four phases of the business cycle. Subsequently, the methodology for the (sector) analyses will be addressed. Before moving on to the phase definitions, let us remind that the phases of the cycle are defined as follows:

Phase 1: Recession Phase 2: Early recovery Phase 3: Expansion Phase 4: Initial slowdown

3.1. Business Cycle Indicators

3.1.1. OECD leading indicator:

The OECD composite leading indicators (CLIs) focus on the fluctuation of the economic activity around its long term potential level3. They are developed in order to show early signals of economic expansion or economic slowdown. That is, with the turning points of the CLI consistently preceding those of the business cycle. The objective of the indicator is to have a lead time of six to nine months. The components of the OECD CLI differ per country; they are chosen based on economic significance, cyclical behavior, data quality, timeliness and availability. All components are time series which are leading the reference series (which is the index of industrial production). For example, eight components are selected for Japan, among which the ratio of imports to exports, monthly manufacturing overtime hours and the share price index (TOPIX).

I use the indicator that takes into account the OECD countries plus six major non-member countries (Brazil, China, India, Indonesia, Russia and South Africa), covering a total of 35 countries. The global OECD CLI is calculated as a weighted average of the corresponding country CLIs. Although specific leading indicators for the US and Japan are available, I choose to use this global indicator for comparison purposes. Moreover, the country specific indicators for both the US and Japan are highly correlated to the global indicator, with coefficients of 0.88 and 0.79 respectively.

In order to separate the business cycle into four phases, I need to know whether the OECD CLI is above or below its long-term trend and whether the OECD CLI is increasing or decreasing. I therefore use the following definition per phase:

Phase 1: Monthly OECD CLI < 100 (long-term trend OECD CLI); CLIt > CLI t-1

Phase 2: Monthly OECD CLI >100 (long-term trend OECD CLI); CLIt > CLI t-1

Phase 3: Monthly OECD CLI > 100 (long-term trend OECD CLI); CLIt < CLI t-1

Phase 4: Monthly OECD CLI < 100 (long-term trend OECD CLI); CLIt < CLI t-1

3

(14)

13

3.1.2. Industrial production

Next to leading data, real data can be used to identify the business cycle. I use year-over-year changes in industrial production as a measure of economic activity. This indicator represents the actual business cycle, since it usually moves simultaneously. That is, a change in the level of industrial production often reflects similar changes in economic activity. It measures the levels of output for manufacturing, mining, gas and electricity. Since its largest component, manufacturing, is a highly cyclical sector, industrial production is widely used as indicator of the business cycle.

3.1.3. Yield curve

As stated earlier, the slope of the yield curve has appeared to be a reliable predictor of real economic activity. In the literature, different interest rates have been used in order to compute the term spread. One can take various maturity combinations, such as the difference between ten-year and two-year Treasury rates. Estrella and Mishkin (1996) use the spread between ten-year and three-month Treasury rates. In a later paper it is stated that this provides a reasonable combination of accuracy and robustness in predicting US recessions over long periods (Estrella and Trubin, 2006). I follow Estrella and Mishkin (1996) by using the ten-year Treasury yield and the three-month Treasury yield for the calculation of the average yield and the spread. The phases are defined based on the state of the yield curve4.

3.1.4. Evaluation of business cycle indicators

After analysis, I decide to use the OECD leading indicator as proxy for the business cycle. Single month observations for industrial production as well as for the term spread do not result in a smooth business cycle pattern based on my phase definitions. In order to improve the outcomes, I have also looked at 3-month and 6-month simple moving averages. Although the patterns improve a little, they are still far from stable. Based on industrial production, average stock returns are higher in the two contracting phases than those in the two expansion phases. This outcome is not in line with the expectations, especially given the fact that industrial production is a coincident indicator. This indicates that the method is not really useful, because industrial production is probably too crude for determining business cycles phases.

The phase definitions based on the average yield and the term spread do not result in a reliable business cycle pattern either. For one reason, because the phases do not follow each other in logical sequence. I therefore conclude that this methodology is not appropriate for identifying business cycles5. After evaluating its predictive ability, Boulier and Stekler (2001) also conclude that the term spread is not a reliable cyclical indicator. Their identification of turning points based on the yield curve, albeit with a different methodology, leads to many false signals for both peaks and troughs (10 and 28 respectively).

Lastly, the unemployment rate and inflation are both important economic indicators and are therefore expected to be suitable for defining business cycle phases. However, the phase definitions based on a combination of the unemployment rate and inflation do not provide stable results. I therefore do not use this proxy for identifying the business cycle.

4

See page 7, bull steepening/flattening, bear steepening/flattening.

5_{This does not mean that the yield curve cannot be a useful indicator: when it comes to recessions, the term spread does}

(15)

14 3.2. Analysis

Once the phases of the business cycle are determined, monthly returns of the relevant MSCI indices are calculated and linked to the corresponding phases. These stock returns will be analyzed in order to examine their cyclical features. For each phase, I will look at the mean return, standard deviation and higher moments of the distribution. The United States will serve as base case and the analysis will be extended by including stock markets of Japan and Asia Pacific. I use data on:

- MSCI USA (MXUS): available since December 31, 1969 - MSCI Japan (MXJP): available since December 31, 1969

- MSCI Asia Pacific ex Japan (MXAPJ): available since December 31, 1987

(16)

15

4. Data analysis

4.1. Economic cycles

We now turn to the analysis of the business cycle phases, as defined by the OECD leading indicator. In the period between December 31, 1969 and July 30, 2010, without including the first phase and the last phase of the series (since we do not know when the first one started/the current one will end), 57 phases were identified based on the global OECD leading indicator. The length of the phases varies between 2 months and 16 months. Even when the OECD CLI data is smoothed, phases of two months remain present. The average length of phases in times of contraction is shorter than the average length in times of expansion: the duration of phase 1 and 4 is 7.7 months and 8.2 months respectively. The duration of both phases 2 and 3 is 8.7 months. It is therefore confirmed that the average duration of expansions is longer than that of contractions.

In the data series, several smooth business cycles can be observed: cycles that run from phase 1 to phase 2, 3 and 4 without deviations (or with a different starting phase, but still followed by three successive phases). Regardless of the starting point, those cycles have an average duration of 36 months6. However, not every cycle is stable. At a few points in time the cycle moves back and forth between two successive stages. Furthermore, sometimes one phase is “skipped” (e.g. the second phase was missing in 2005).

We can compare the turning points of the business cycles based on the OECD global indicator with the actual NBER business cycle reference dates, in order to check similarity. The months that are defined as recession by the NBER are indeed observable in phase 1 (recession). The trough dates of the NBER cycles are all observable in one of the first months of phase 2. That is, the phases of early recovery indeed fall at the same time as the start of the NBER expansions. Furthermore, the peak dates of the NBER are observable either in the end of phase 4 or the beginning of phase 1 (which can be explained by the use of a leading indicator).

It has to be said that the NBER has defined only six business cycles since 1969, whereas my methodology identifies more business cycles in the same period. However, the NBER follows US cycles, while I focus on global cycles.

Before turning to the analysis of stock returns over the cycle, let us first look at the macroeconomic data. Figure 1 displays the OECD leading indicator and the year-on-year changes in industrial production (the actual cycle). The graph shows that the OECD leading indicator is indeed leading the economic cycle. The correlation between these two indicators is 0.79. Looking at figure 1, we can identify the major recessions since 1969, as defined by the NBER.

December 1969 - November 1970

The first large dip in industrial production can be explained by a relatively mild recession in the United States, which took place after a long period of expansion (the second longest period of growth in the US history). The recession was characterized by high unemployment and high inflation. The government tightened monetary policies to fight inflation, as well as fiscal policies in order to reduce budget deficits caused by the Vietnam War.

6_{Several types of business cycles with different lengths have been identified: the Kitchin inventory cycle (3-5 years), the}

(17)

16

November 1973 - March 1975

A few years later, a severe recession hit the western economies. The OPEC placed an embargo on oil shipments to the West as a punishment for the support to Israel during the Yom Kippur War (specifically targeted at the US and The Netherlands). As an immediate response, the oil price quadrupled. The oil crisis, combined with the high costs of the Vietnam War for the US, were the main causes of the recession. An important event was the stock market downturn in 1973-1974 that followed the fall of the Bretton Woods system. Many countries experienced a period of stagflation: rising unemployment together with rising inflation.

January 1980 - July 1980

In the beginning of the eighties a short recession occurred, followed by a deep recession. In order to fight inflation, which reached 13.5% in 1980, the Fed reduced the growth rate of the money supply and raised interest rates (the federal funds rate rose from 11% in 1979 to 20% in 1980), resulting in an economic slowdown.

July 1981 - November 1982

After the short recession of 1980, one can observe a period of recovery in the graph, followed by another recession. In fact, this “double dip” was caused by the Iranian revolution, which led to the energy crisis/second oil crisis. The new regime in Iran exported oil on an irregular basis and at lower volumes, resulting in a large increase in the price of oil. High inflation led to tight monetary policies in the US: the prime interest rate peaked at 21.5% in 1982. This is said to be the main reason why the economy fell back into recession in the early eighties.

July 1990 - March 1991

Since the peak of the economic cycle in 1984, the next recession defined by the NBER took place in the early nineties. On Black Monday in October 1987 the stock market in the US collapsed: de DJIA dropped 22.6% in one day. Although the stock market recovered relatively fast, another problem arose: the savings and loans industry in the US collapsed (the savings and loans crisis). Moreover, the start of the first Gulf War caused international stock markets to decline. This combination of causes led to the recession of the early nineties.

March 2001 - November 2001

After a decade of growth in the US, the next, relatively mild, recession was caused by the collapse of the dot-com bubble. As a response, the NASDAQ crashed. After the attacks of September 11, the Dow Jones suffered a major loss as well. Although September 11 did not cause the recession, “the attacks clearly deepened the contraction and may have been an important factor in turning the episode into a recession7.” Furthermore, the corporate scandals at for example Enron and Worldcom might have made things worse. The recession hit mostly developed countries. It was predicted by many economists, because the period of expansion in the nineties was expected to come to an end.

December 2007 - June 2009

The most recent recession was the global financial crisis, initially caused by the subprime mortgage crisis in the US, which led to the collapse of the US housing market. The problems spread to other countries and turned into a global financial crisis: bank collapses in US and Europe, declines in global stock markets, etc. Governments and central banks responded with bailouts and all kinds of policies to prevent further damage. The NBER declared that the recession in the US ended in June 2009.

7

(18)

17

Figure 2: Global OECD Composite Leading Indicator and Industrial Production (year-on-year) index.

4.2. Analysis of phases

In the following sections, I discuss the results for the different MSCI indices and the corresponding sector analyses.

4.2.1. Analysis of returns MSCI United States, based on the OECD CLI

Table 1 displays the average monthly returns of the MXUS for each phase of the business cycle, based on the global OECD leading indicator for the period between December 31, 1969 and July 30, 2010. The average returns per phase of the cycle are in line with the expectations. In phase 1 (recession), the average return is negative and is phase 4, the average return is only slightly positive. The returns are clearly the highest in the two expansion phases.

Table 1: Returns of the US stock index over the cycle

This table displays the average monthly returns of the MXUS over the business cycle for the period December 31, 1969 – July 30, 2010. The phases of the cycle are based on the OECD leading indicator.

Full cycle Phase 1 Phase 2 Phase 3 Phase 4

Recession Recovery Expansion Slowdown

Average return 0.59% -0.83% 1.89% 1.09% 0.04%

It seems remarkable that the highest performance is achieved in phase 2 and 3. Since the OECD provides us with a leading indicator, which aims at a lead time of six to nine months, one would expect that the performance would be leading the true cycle as well. However, since financial markets are also leading the economic cycle, one can actually think of the OECD leading indicator as a coincident indicator of the stock market cycle. That is, it is not unexpected that the two expansion phases are outperforming over the cycle.

(19)

18

For the sector analysis, I analyze the period 1995-2010. Table 2 displays the average monthly returns of the sectors per phase of the cycle for the MXUS8. I perform t-tests to check whether the returns are significantly different from the return of the index in each phase. As can be seen in the table, there are large differences between the performances of the various sectors. In phase 1 all sectors have an average negative performance, but several do outperform the index (although not significantly). In phase 2 only the two worst performing sectors are significantly different from the index, whereas this is the case for the two best performing sectors in phase 4. In phase 3,bonly IT and Utilities are different from the index.

Table 2: Sector analysis for the US stock index over the cycle

This table shows the results of the sector analysis for the MXUS. Presented is an overview of the average monthly returns of the stock index per sector and per phase of the business cycle for the period 1995 – 2010. The phases of the cycle are based on the OECD leading indicator. When the average return of a sector is significant difference from the index in a particular phase, this is indicated with an asterisk (* =10% significance level, ** =5% significance level).

Sector Returns 1 Sector Returns 2 Sector Returns 3 Sector Returns 4

Health Care -0.1% IT 3.7% IT* 2.7% Utilities** 1.5%

CS -0.8% Financials 3.3% Industrials 1.7% Energy* 1.4%

Telecom -1.2% Materials 3.1% Energy 1.7% Health Care 0.9%

IT -1.4% CD 3.0% Index 1.5% CS 0.9%

CD -1.4% Industrials 2.5% CD 1.5% Industrials 0.4%

Energy -1.5% Index 2.2% Materials 1.3% Telecom 0.0%

Index -1.5% Energy 1.7% Financials 1.3% Materials -0.1%

Utilities -1.7% CS 1.5% Health Care 1.0% Index -0.2%

Financials -2.1% Health Care 1.0% CS 0.8% Financials -0.2%

Materials -2.1% Telecom* 0.6% Telecom 0.8% CD -0.9%

Industrials -2.2% Utilities** 0.4% Utilities* 0.7% IT -1.1%

Looking back at the hypothesis, we can observe the following9_:

Phase 1: In line with the hypothesis, Health Care is the best performing sector and Industrials is the worst performing sector in this phase. However, IT performs better than expected, whereas Financials unexpectedly appears at the bottom.

Phase 2: In line with the hypothesis, IT and Materials belong to the best performing sectors in this phase. Financials does better than expected. As was hypothesized, Utilities and Telecommunication are the worst performing sectors.

Phase 3: In line with the hypothesis, IT, Industrials and Energy are the best performing sectors. As expected, Consumer Staples is one of the bad performing sectors.

Phase 4: In line with the hypothesis, Utilities is the best performing sector. Contrary to expectations, Consumer Staples and Telecommunication do not belong to the best performers. As expected, IT has the worst performance in this phase.

8_{An overview of the ranking of sectors per phase based on the Sharpe Ratio can be found in the Appendix.} 9

(20)

19

4.2.2. Analysis of returns MSCI Japan, based on the OECD leading indicator

Table 3 displays the average monthly returns of the MXJP per phase of the business cycle. Both phases of contraction have negative returns; both phases of expansion have positive returns.

Table 3: Returns of the Japanese stock index over the cycle

This table displays the average monthly returns of the MXJP over the business cycle for the period December 31, 1969 – July 30, 2010. The phases of the cycle are based on the OECD leading indicator.

Average return 0.49% -1.69% 2.09% 1.62% -0.33%

Table 4 presents the overview of the sector analysis for the MXJP. As was the case for the MXUS, IT is the best performing sector in both expansion phases. It is also the worst performing sector in both contraction phases. For Utilities, the opposite is true. For example, in phase 2 it is the only sector with a negative return, whereas in phase 1 it is the only sector with a positive average return.

Table 4: Sector analysis for the Japanese stock index over the cycle

This table shows the results of the sector analysis for the MXJP. Presented is an overview of the average monthly returns of the stock index per sector and per phase of the business cycle for the period 1995 – 2010. The phases of the cycle are based on the OECD leading indicator. When the average return of a sector is significant difference from the index in a particular phase, this is indicated with an asterisk (* =10% significance level, ** =5% significance level).

Utilities** 0.9% IT 4.2% IT** 3.5% Energy 0.2%

Health Care* -1.5% Materials 3.9% CD 1.9% Utilities** 0.0%

CS -1.5% CD 3.5% Health Care 1.6% Health Care -0.4%

Energy -2.3% Financials 3.0% Index 1.5% CS -0.8%

Materials -3.2% Industrials 2.9% Industrials 1.3% Industrials -1.2%

Index -3.2% Energy 2.8% Materials 1.2% CD -1.7%

CD -3.3% Index 2.8% Financials 1.0% Materials -1.8%

Industrials -3.3% CS 1.7% CS 0.5% Index -1.8%

Financials -4.1% Health Care 1.6% Utilities** -0.1% Financials -2.0%

IT -4.3% Utilities** -0.6% Energy* -0.2% IT -2.8%

Looking back at the hypothesis, we can observe the following:

Phase 1: In line with the hypothesis, Health Care is one of the best performing sectors and Industrials and IT belong to the worst performing sectors in this phase. Financials appears at the bottom, contrary to the hypothesis.

Phase 2: In line with the hypothesis, IT, Materials and Consumer Discretionary are the best performers in this phase. Moreover, Utilities is the worst sector and significantly underperforms the index.

Phase 3: In line with the hypothesis, IT has the highest average return in this phase. However, Industrials and Energy do not belong to the best performing sectors, with Energy even having the lowest average return.

(21)

20

4.2.3. Analysis of returns MSCI Asia Pacific ex Japan, based on the OECD leading indicator

Table 5 displays the average monthly returns of the MXAPJ per phase of the business cycle. Again, both phases of contraction have on average negative returns and both phases of expansion have positive returns, with phase 2 having a very high return (compared to the MXUS and MXJP). Table 6 displays the overview of the sector analysis for the MXAPJ. Once again, IT has the highest and lowest ranks in the expansion phases and contraction phases, respectively.

Table 5: Returns of the Asia Pacific stock index over the cycle

This table displays the average monthly returns of the MXAPJ over the business cycle for the period December 31, 1987 – July 30, 2010. The phases of the cycle are based on the OECD leading indicator.

Average return 0.72% -1.87% 3.93% 1.59% -0.90%

Table 6: Sector analysis for the Asia Pacific stock index over the cycle

This table shows the results of the sector analysis for the MXAPJ. Presented is an overview of the average monthly returns of the stock index per sector and per phase of the business cycle for the period 1995 – 2010. The phases of the cycle are based on the OECD leading indicator. When the average return of a sector is significant difference from the index in a particular phase, this is indicated with an asterisk (* =10% significance level, ** =5% significance level).

Utilities -1.0% IT 6.4% IT* 3.3% Energy 0.1%

Health Care -1.3% CD 5.1% Health Care 2.5% Health Care -0.3%

CS -1.4% Financials 4.7% Energy 2.3% Utilities -0.4%

Financials -2.0% Materials 4.6% CD 1.9% CS -0.6%

Energy -2.2% Index 4.3% Materials 1.8% Materials -0.7%

CD -2.4% Energy 3.6% Index 1.7% Telecom -1.1%

Materials -2.5% Industrials 3.6% Financials 1.5% Industrials -1.3%

Index -2.5% CS 3.3% CS 1.5% Financials -1.4%

Telecom -2.6% Telecom 2.7% Industrials 1.4% Index -1.5%

Industrials -3.1% Health Care* 2.3% Telecom 1.3% CD -1.6%

IT -3.3% Utilities** 2.1% Utilities 0.9% IT -2.8%

Looking back at the hypothesis, we can observe the following:

Phase 1: In line with the hypotheses, Health Care is one of the best performing sectors and Industrials and IT are the worst performing sectors in this phase. Financials does not belong to the best sectors.

Phase 2: In line with the hypothesis, IT and Consumer Discretionary are the best performers in this phase. However, Materials does not belong to the best sectors. As expected, Utilities is the worst performing sector and significantly underperforms the index. Telecom appears at the bottom as well. Health Care significantly underperforms the index, which was not expected.

Phase 3: In line with the hypothesis, IT is the best performing sector. As expected, Energy also performs well in this phase. However, Industrials and Consumer staples were expected to do better and worse, respectively.

(22)

21 4.3. Long-short portfolios

4.3.1. Portfolio Strategies

Based on the sector analyses, I formulate portfolio strategies. Depending on the phase of the cycle, the strategy is to go long in the two best performing sectors and to go short in the two worst performing sectors for that particular phase. Thereby, an equally weighted long-short portfolio is constructed. The composition of the portfolio depends on the OECD leading indicator, which is used to define the phase of the business cycle on a monthly basis and thereby determines the sector holdings in the portfolio10. The portfolio will thus be altered every time the cycle moves into another phase. This may happen on a frequent basis.

When back testing the portfolio strategies, it is assumed that sector rotation can be executed immediately once the business cycle moves into another phase. However, the OECD data is not readily available each month. Since there is a two month delay in the release of the OECD CLI, the current method imposes a look-ahead bias. For a realistic portfolio strategy, the composition should be based on lagged OECD data. I therefore create long-short portfolios with a two month lag as well. That is, it is possible that the business cycle has already moved to the next phase, while the portfolio composition is still based on the previous phase.

In order to check robustness, I also formulate portfolio strategies for Japan and Asia Pacific based on the sector analysis of my base case (United States). This might lead to a lower performance, since the US portfolio strategy probably involves different sectors than a strategy based on a Japanese or Asian sector analysis. Rapach and Wohar (2006) observe that existing literature on stock return predictability mainly relies on in-sample tests, whereas it is typically believed that out-of-sample tests provide a measure of protection against data mining, as statistical models are tested using out-of-sample observations that are not used in the estimation of the statistical model itself.

Once the long-short portfolios are constructed, I analyse the average and cumulative returns. Subsequently, I perform t-tests in order to check whether the long-short portfolio returns are significantly different from zero. Finally, we need to take transaction costs into account. Every time the business cycle moves into a new phase, we rotate between sectors: we shift the assets to the well-performing sectors for that particular phase. For a realistic case, the average return will be reduced with 25 basis points (50 basis points for Asia) per phase change of the business cycle.

4.3.2. Analysis long-short portfolios

Table 7 displays the results for the MXUS-based long-short portfolio. The average monthly return of the portfolio over the period 1995 – 2010 is 1.03%. Moreover, the standard deviation of the portfolio is substantially lower. As we can see in the table, the two month lag lowers the average return for the long-short portfolio to 0.59%, although it remains significantly different from zero. We can also observe that the portfolio has the highest average performance in phase 2 and phase 4.

Table 8 displays the results for the long-short portfolios for Japan. Over the full cycle, the average returns of all short portfolios are significantly different from zero. The regular long-short portfolio results in an average monthly return of 1.58%. While the MXJP has a significant negative performance in the two contraction phases, the portfolio has a significant positive return in each phase of the cycle. Using a two month lag, more than half of the average full cycle return disappears and the hit ratio decreases. The out-of sample portfolio (based on the US sector analysis) with two month lag still provides a positive return of 0.46% with a relatively low standard deviation.

10_{An overview of the portfolio compositions per phase of the cycle for the US, Japan and Asia Pacific can be found in the}

(23)

22

Once transaction costs are taken into account, the out-of-sample long-short portfolio with two month lag generates a positive average return of 0.44%, which is significantly different from zero.

Table 7: Long-short portfolio results for the United States

This table shows the average returns of the long-short (LS) portfolios for the US for the full business cycle and per phase, as well as the average returns of the MXUS for comparison purposes. The long-short portfolios are constructed following a sector analysis: the strategy is to go long in the two best performing sectors and to go short in the two worst performing sectors for each phase of the cycle. The portfolios are equally weighted and one includes a two month (2M) lag. The average returns, standard deviations and hit ratios are for the time period 1995 - 2010.

United States Full cycle Phase 1 Phase 2 Phase 3 Phase 4

Average return Index 0.57%* -1.53% 2.23%** 1.51%** -0.21%

Standard Deviation 4.63 6.26 4.1 3.34 3.87

Average return LS Portfolio 1.03%** 0.87%* 1.52%** 0.72%* 1.22%**

Hit ratio 65% 67% 68% 62% 66%

Average return LS 2M lag 0.59%** 0.52% 0.85%* 0.45% 0.63%

Hit ratio 62% 63% 70% 59% 57%

Table 8: Long-short portfolio results for Japan

This table shows the average returns of the long-short (LS) portfolios for Japan for the full business cycle and per phase, as well as the average returns of the MXJP for comparison purposes. The long-short portfolios are constructed following a sector analysis: the strategy is to go long in the two best performing sectors and to go short in the two worst performing sectors for each phase of the cycle. The portfolios are equally weighted. One includes a two month (2M) lag and another one (out-of-sample) is based on the US sector analysis, instead of the Japanese sector analysis (also with two month lag). The average returns, standard deviations and hit ratios are for the time period 1995 - 2010.

Japan Full cycle Phase 1 Phase 2 Phase 3 Phase 4

Average return Index -0.12% -3.20%** 2.82%** 1.46%** -1.83%*

Average return LS Portfolio 1.58%** 1.93%** 1.77%** 1.45%** 1.24%**

Hit ratio 66% 74% 68% 57% 70%

Average return LS 2M lag 0.71%** 0.51% 0.56% 0.79% 0.94%*

Hit ratio 58% 63% 57% 54% 62%

Average return LS out-of-sample 0.46%* -0.07% 0.50% 0.57% 0.83%*

Hit ratio 58% 51% 62% 56% 64%

The results for the long-short portfolio based on the MXAPJ can be found in table 9. The average return of the regular long-short portfolio over the full cycle is 1.14%. Once again, the two month lag strongly reduces the average returns. The out-of-sample portfolio with two month lag has an average return of 0.38%, whereas the hit ratio remains 56%.When transaction costs are included, the average return reduces to 0.34%, which is still significantly different from zero.

(24)

Table 9:

This table shows the average returns of the long

as well as the average returns of the MXAPJ for comparison purposes. The long

sector analysis: the strategy is to go long in the two best performing sectors and to go short in the two worst performing sectors for each phase of the cycle. The portfolios are equally weighted. One includes a two month (2M) lag a

sample) is based on the US sector analysis, instead of the

returns, standard deviations and hit ratios are for the time period 1995

Asia

Average return Index Standard Deviation Average return LS Portfolio Standard Deviation Hit ratio

Average return LS 2M lag Standard Deviation Hit ratio

Average return LS out-of-sample Standard Deviation Hit ratio

Figures 3, 4 and 5 show the cumulative returns of the three long

cumulative returns of the relevant stock market indices since 1995. Assumed is scenario: the long-short portfolios include a

are based on the US sector analysis

after the 15-year period than the stock market indices.

portfolio return and the MXJP return is very large. When looking at the US and Asia Pacific, observe that the cumulative returns

indices. In general, the returns show an upward trend

Figur

This figure presents the cumulative returns for the US stock index and which is constructed following a sector analysis . T

short in the two worst performing sectors for each phase of the business cycle. The portfolio is equally weighted and two month lag.

23

: Long-short portfolio results for Asia Pacific

f the long-short (LS) portfolios for Asia Pacific, for the full business cycle and per phase as well as the average returns of the MXAPJ for comparison purposes. The long-short portfolios are constructed following a

long in the two best performing sectors and to go short in the two worst performing sectors The portfolios are equally weighted. One includes a two month (2M) lag and another one (out

is, instead of the Asia Pacific sector analysis (also with two month lag). hit ratios are for the time period 1995 - 2010.

0.53% -2.49%* 4.32%** 1.70%** -1.49 6.7 8.3 5.5 4.2 6.8 1.14%** 1.02%* 1.77%** 0.91%** 1.06 2.84 3.58 2.84 2.28 2.75 68% 72% 73% 61% 68% 0.44%* 0.08% 0.66% 0.33% 0.77 3.02 3.60 2.36 2.59 3.49 56% 53% 57% 52% 64% sample 0.38%* 0.23% 0.61%* 0.33% 0.40 2.53 2.51 1.97 2.31 3.29 56% 56% 59% 48% 64%

Figures 3, 4 and 5 show the cumulative returns of the three long-short portfolios compared to the cumulative returns of the relevant stock market indices since 1995. Assumed is the most realistic short portfolios include a two month lag and, relevant for Japan and Asia Pacific, are based on the US sector analysis. As can be seen, all portfolios achieve a higher cumulative return year period than the stock market indices. For Japan, the difference between the io return and the MXJP return is very large. When looking at the US and Asia Pacific,

observe that the cumulative returns of the long-short portfolios are less volatile than those of the turns show an upward trend.

Figure 3: Cumulative returns for the US

This figure presents the cumulative returns for the US stock index and the cumulative returns of a long-short portfolio a sector analysis . The portfolio strategy is to go long in the two best performing sectors and to orst performing sectors for each phase of the business cycle. The portfolio is equally weighted and

business cycle and per phase, short portfolios are constructed following a long in the two best performing sectors and to go short in the two worst performing sectors nd another one (out-of-sector analysis (also with two month lag). The average

Phase 4 1.49% 6.8 1.06%** 2.75 68% 0.77% 3.49 64% 0.40% 3.29 64% compared to the the most realistic month lag and, relevant for Japan and Asia Pacific, higher cumulative return For Japan, the difference between the io return and the MXJP return is very large. When looking at the US and Asia Pacific, we can than those of the

(25)

Figure 4

This figure presents the cumulative returns for the Japanese stock index and the cumulative returns of a long Japan, which is constructed following a sector analysis.

and to go short in the two worst performing sectors for each phase of the business cycle. The portfolio is equally weighted and includes a two month lag and is based on the US sector analysis.

Figure 5

This figure presents the cumulative returns for the Asia Pacific stock index and the cumulative returns of a long for Asia Pacific, which is constructed following

sectors and to go short in the two worst performing sectors for each phase of the business weighted and includes a two month lag and is based on the US sector analysis.

24

Figure 4: Cumulative returns for Japan

cumulative returns for the Japanese stock index and the cumulative returns of a long

a sector analysis. The portfolio strategy is to go long in the two best performing sectors two worst performing sectors for each phase of the business cycle. The portfolio is equally weighted and

and is based on the US sector analysis.

Figure 5: Cumulative returns for Asia Pacific

returns for the Asia Pacific stock index and the cumulative returns of a long

which is constructed following a sector analysis . The portfolio strategy is to go long in the two best performing e two worst performing sectors for each phase of the business cycle. The portfolio is equally

and is based on the US sector analysis.

cumulative returns for the Japanese stock index and the cumulative returns of a long-short portfolio for he portfolio strategy is to go long in the two best performing sectors two worst performing sectors for each phase of the business cycle. The portfolio is equally weighted and

(26)

25 4.4. Fama and French thee-factor model

4.4.1. Factors over the cycle

As was mentioned earlier in this paper, value stocks are expected to outperform in phase 3 and phase 411_{. To examine whether value investing is indeed beneficial in those phases of the cycle, I can}

use the value factor of the Fama and French three-factor model. In their landmark paper, Fama and French (1992) find a weak relation between a portfolios beta and its average return. Aiming to explain the cross-sectional variation in stock returns, the authors expand the Capital Asset Pricing Model by developing the three-factor model. They observe that small cap stocks and stocks with a high book-to-market ratio (value stocks) tend to outperform the market. Therefore, they add factors for size and value in addition to beta, the single factor included in the CAPM. The expected stock return is thereby related to three factors: the excess return of the market portfolio, excess returns of small caps over big caps and excess returns of value stocks over growth stocks. In addition to analysing the returns of the value factor over the cycle, I will also look at the size factor and momentum. Carhart (1997) extended the three-factor model by adding momentum, to explain the tendency of stocks with high performance over the past period to continue rising.

Table 10, 11 and 12 display the results of the analysis. Although the results for the value factor for the US are not significantly different from zero, the value factor does have a significant positive performance over the full cycle in Japan and Asia Pacific. Furthermore, I find significant positive returns for both indices in phase 1 and phase 2. In phase 3 and phase 4, for which the highest returns were expected, the outcomes are in general insignificant and in some cases do not have the right sign.

With respect to the size factor, I expect to find lower returns during periods of contraction. Returns of small firms may be more sensitive to credit market conditions. Small firms tend to be firms with high financial leverage and are therefore less likely to survive economic downturns (Chan and Chen, 1991). Table 12 shows that the results for the size factor are not unambiguous. The negative average return in phase 1 is only significant for Japan. In phase 4, Asia Pacific even has a significant positive return.

When looking at the results for momentum, we see significant negative returns in phase 2, while the average returns are positive and significant in phase 3. Over the full cycle, average monthly returns are not significant.

Table 11: The value factor over the business cycle

This table displays the average monthly returns of the size factor of the Fama and French three-factor model for Japan over the different phases of the business cycle. When the average return of the factor is significantly different from zero, this is indicated with an asterisk (* =5% significance level, ** =1% significance level)

Value Full cycle Phase 1 Phase 2 Phase 3 Phase 4

US 0.19% -0.05% 0.85% -0.54% 0.84%

Japan 1.11%** 1.92%** 1.76%* -0.76% 2.50%*

Asia Pacific 1.02%** 1.91%* 2.38%** 0.63% -0.55%

11

(27)

26

Table 10: The size factor over the business cycle

This table displays the average monthly returns of the size factor of the Fama and French three-factor model for the US over the different phases of the business cycle. When the average return of the factor is significantly different from zero, this is indicated with an asterisk (* =5% significance level, ** =1% significance level)

Size Full cycle Phase 1 Phase 2 Phase 3 Phase 4

US 0.31% -0.04% 1.41%** 0.73% -0.82%

Japan -0.38% -1.26%* -1.94%** 1.53%** -1.04%

Asia Pacific -0.42% -1.16% -1.50%* -0.52% 1.40%**

Table 12: The momentum factor over the business cycle

This table displays the average monthly returns of the size factor of the Fama and French three-factor model for Asia Pacific over the different phases of the business cycle. When the average return of the factor is significantly difference from zero, this is indicated with an asterisk (* =5% significance level, ** =1% significance level)

Momentum Full cycle Phase 1 Phase 2 Phase 3 Phase 4

US 0.50% 1.46% -2.11% 1.39%** 0.55%

Japan -0.64% 0.44% -4.80%** 2.09%** -2.02%

Asia Pacific 0.07% -0.52% -2.07%* 1.28%** 0.93%*

4.4.2. Fama and French regressions

I will now examine whether the returns of the long-short portfolios can be explained with risk factors. To analyse the risk characteristics of the portfolios, the returns are regressed on the factors of the Fama and French three-factor model. The equation of the three-factor model is as follows:

ܴ݉ െ ܴ݂ ൌ ߙ ൅ ߚሺܴ݉ െ ܴ݂ሻ ൅ ݏܵܯܤ ൅ ݄ܪܯܮ ൅ ߝ (1)

Following equation 1, expected returns are determined by three factors: - Beta: the market risk factor

- SMB: Small Minus Big (market capitalization), the size factor - HML: High Minus Low (book to market ratio), the value factor

I perform regressions on the long-short portfolios with two month lag of the US, Japan and Asia Pacific. Data on the factors for the US is extracted from Kenneth R. French’s website. For Japan and Asia Pacific, data on the factors comes from JP Morgan. In addition, I perform regressions using a fourth factor: momentum. Both the three-factor model and the four-factor model are often used to explain the performance of portfolios. After including momentum (MOM) as fourth factor, the regression equation becomes:

ܴ݉ െ ܴ݂ ൌ ߙ ൅ ߚሺܴ݉ െ ܴ݂ሻ ൅ ݏܵܯܤ ൅ ݄ܪܯܮ ൅ ݉ܯܱܯ ൅ ߝ (2)

(28)

27

For Japan, only the intercept and the HML factor are significant (adjusted R-squared 0.11). The negative HML factor loading indicates that the long-short portfolio of Japan can be defined as a growth portfolio. The regression on the Asian long-short portfolio does not provide significant results (adjusted R-squared 0.02).

Table 13: Regression results three-factor model (US)

This table shows the results of an OLS regression of the US long-short portfolio returns on the Fama and French three factors: Rm-Rf (the market risk factor), SMB (the size factor and HML (the value factor). Statistical significance of the coefficients is indicated with asterisks (*** = 1% level, ** = 5% level and * = 10% level).

Variable Coefficient t-Statistic

C 0.436 2.190**

Rm-Rf -0.083 -1.922*

SMB -0.202 -3.437***

HML -0.231 -5.119***

Table 14: Regression results three-factor model (Japan)

This table shows the results of an OLS regression of the long-short portfolio returns for Japan on the Fama and French three factors: Rm-Rf (the market risk factor), SMB (the size factor and HML (the value factor). Statistical significance of the coefficients is indicated with asterisks (*** = 1% level, ** = 5% level and * = 10% level).

C 0.658 2.572**

Rm-Rf -0.074 -1.566

SMB -0.092 -1.145

HML -0.261 -3.397***

Table 15: Regression results three-factor model (Asia Pacific)

This table shows the results of an OLS regression of the long-short portfolio returns for Asia Pacific on the Fama and French three factors: Rm-Rf (the market risk factor), SMB (the size factor and HML (the value factor). Statistical significance of the coefficients is indicated with asterisks (*** = 1% level, ** = 5% level and * = 10% level).

C 0.197 0.860

Rm-Rf -0.060 -1.650

SMB 0.064 0.643

HML 0.006 0.068

Table 16, 17 and 18 display the regression results after adding the fourth factor: momentum. For all three portfolios, the momentum factor has a positive coefficient and turns out to be highly significant. The adjusted r-squared improves for every model, although the explanatory power of all models remains low (US: 0.25, Japan: 0.22, Asia Pacific: 0.16).

The US long-short portfolio is still defined as a large cap portfolio. However, the beta coefficient becomes insignificant. For Japan, the HML factor is not significant anymore, whereas the SML factor becomes negative and significant, implying a small cap portfolio12. For the Asian long short portfolio, the beta and HML factor are now significant on a 10 percent level. The portfolio can be characterized as a value portfolio, given the positive coefficient.

12

The predictability of stock markets using business cycles and investor sentiment

The predictability of stock markets using business cycles

and investor sentiment

The predictability of stock markets using business cycles

and investor sentiment

Abstract

Acknowledgements

Table of Contents

1.

Introduction

2.

Business Cycles: Theoretical Background

3.

Methodology

4.

Data analysis