Do Industries Lead Stock Markets? Evidence from the largest markets based on weekly data Master Thesis MSc Finance Stefan Bogdan Mihalache

Do Industries Lead Stock Markets?

Evidence from the largest markets based on weekly data

Master Thesis

MSc Finance

Stefan Bogdan Mihalache

Abstract: I examine the ability of industry portfolios to forecast broad market movements on a

weekly level between 2000 and 2016 for nine markets: Australia, Canada, France, Germany,

Japan, Netherlands, Switzerland, U.K. and U.S.A. I find that there are multiple significant

industries which are able to predict market movements up to one week. This supports the idea

that markets are not fully efficient and information is gradually diffused in the market, and an

investor may observe and use this phenomenon to his advantage. Even so, the pattern of

predictive industries across countries is not consistent, as different industries forecast different

markets. Moreover, weekly data contains a lot of noise due to high frequency trading and

sentiment-based transactions, and relationships may change depending on economic and

technological advancements. An investment strategy based on these findings may be viable, but

is sometimes counterbalanced by high transaction costs required by the frequency of the

rebalancing of the portfolio. Industries lead the broad market in a cyclical manner, which

determines overlaps and clustered signals for the broad market. Industries lead the market on a

weekly level sequentially, as their predictive power exhibits variation across time. This shows

that the overall market has a delayed reaction to movements of individual industries, and an

investor could integrate these sets of markers in a broader, more comprehensive strategy.

JEL: E44, G11, G14, G15, G17

Keywords: Financial Markets and Macroeconomy, Portfolio Choice, Information and Market

Efficiency, International Financial Markets, Financial Forecasting and Simulation

1

1. Introduction

One of the most important aspects of the nature of capital markets has to do with the

levels of efficiency, originally formulated by Fama (1991)

1

_{. The three forms of efficient markets}

(weak, semi-strong and strong) gradually decrease the potential of asymmetric information

happening in markets. Applying this insight nowadays, when information flows practically

instantly, capital markets should, at least theoretically, be governed by complete transparency

as information is fully available to all individuals. This renders variability and unpredictability

when it comes to stock returns, as all information is rapidly and fully reflected in stock prices,

and no causal relationships can be formulated in order to explain the behavior of certain

investment assets.

This paper focuses on the analysis of the relationship between stocks on an industrial

level and the overall market, all on a weekly frequency. Historically, certain industries have

inevitably driven the overall market as a result of the interaction between sectors, companies

and of course, investors. The following analysis is aimed at identifying and testing lagged

relationships between industries and the overall market and also incorporating the results in a

potential investment strategy. More precisely, the focus will be on the study of lagged

dependencies between industries and the overall market, in order to observe and determine

how shocks in specific sectors propagate through the economic environment, with the purpose

of integrating these markers into an investment strategy.

This view opposes the Efficient Market Hypothesis, as significant results show that there

is still predictability in the market. This phenomenon has been previously studied in the

academic literature, most notably by Hong, Torous, Valkanov (2002)

2

_{(hereafter, HTV) and Tse}

(2015)

3

_{. Both their analysis is mostly focused on the U.S.A. stock market, but HTV also expand}

towards the eight largest markets outside the U.S.A., and find very similar results.

HTV analyze monthly stock returns between 1946 and 2002 and found that 14 out of 34

U.S.A. industries can predict overall market movements. This result is also confirmed for the 8

largest markets outside the U.S.A. This supports the idea that the Efficient Market Hypothesis

does not hold perfectly and that information is not reflected instantly in stock prices, but rather

there is a gradual diffusion from the origin of a specific shock. Their interpretation of these

1 _{Fama, E., 1991. Efficient Capital Markets. The Journal of Finance 46, No. 5, 1515-1617}

2 _{Hong, H., Torous, W., Valkanov, R., 2002. Do industries lead stock markets? Journal of Financial Economics 83,}

367–396

2

results also provide a very intuitive idea: “the returns of industry portfolios that are informative

about macroeconomic fundamentals will lead the aggregate market” (p. 369).

Tse performs a reexamination on HTV’s results by expanding the time period until 2013

while using the same methodology and adding different specifications, testing different samples

and performing additional relationship tests. He analyzes only the U.S.A. market and finds less

consistent results than HTV. He presents mildly significant causalities between industries and

the overall market. Even though the data source is the same for both papers (Ken French’s data

library), recent developments have expanded the number of industries from 34 (as in HTV) to

48 (as in Tse). It may be that this disaggregation dissipates away the potential effects of

predictability, since nowadays markets are highly developed and do not depend heavily on one

single driver, but rather on a cluster of different variables. Even so, if one were to analyze this

phenomenon, the 48 industries analyzed by Tse could be grouped in increasingly larger batches

that encapsulate similarities across them, and these groups may be better predictors of market

movements as they hold effects, signals, shocks from more than one single industry.

Both papers use the same methodology, that is analyzing the effects of industries on the

market by running regressions where the dependent variable is the market return, while the

independent variables are the lagged industry return, the lagged market return and lagged

control variables: inflation, default spread, market dividend yield and market volatility. The

control variables are chosen because they are typically thought to be proxies for time varying

risk and thus, similar explanatory power of the model with and without them would mean that

the results are not generated by time varying risk. A simple form of the regression equation is

the following:

𝑅𝑀

_𝑡

= 𝛼

_𝑖

+ 𝛽

_𝑖

𝐼𝑁𝐷

_{𝑖,𝑡−1}

+ 𝛾

_𝑖

𝑅𝑀

_𝑡−1

+ 𝐴

_𝑖

𝑍

_𝑡−1

+ 𝑒

_𝑖,𝑡

(1)

where RM

t

is the monthly market return, IND

i,t-1

is the monthly return of industry i with

1-lag and Z

t-1

is a vector containing the control variables, also on a monthly frequency, with

1-lag. The specified lags are only for an easier understanding of the model, as both papers study

this relationship for more than 1 lag. The control variables are aimed at absorbing the effects of

macroeconomic factors on the market, as overall market performance is highly dependent on

these.

3

examines the explanatory power of the industries and market on the IPG (industrial production

growth rates) and found again that the market returns outperform all individual industries’

returns when it comes to the aggregation of information and signals.

The two papers have mildly contradictive results, as HTV find statistically and

economically significant results, while Tse finds less evidence that specific industries lead the

stock market, consistent with the efficient market hypothesis. Both papers analyze the

phenomenon from a descriptive perspective, and provide little information about the potential

of these findings to be implemented in a real-world environment through, most probably, an

investment strategy. Moreover, the analyzed time frame spans for a very long period of time, and

even though additional tests were made on sub-samples, they may not be very significant for an

active investor nowadays as reigning industries have changed heavily over history. Finally,

information flows nowadays at incredible speeds, and individuals have access to countless

sources from all over the world. Thus, the effects of a shock in one industry will be seen a lot

faster in stock prices than it would 20 years ago. From this point of view, analyzing monthly

returns may hide various effects and correlations about how the market reacts to various

industry movements, and may distort the relationships between different sectors.

2. Data and Methodology

4

involvements through monetary policies and fiscal rules may distort data quality and relevance.

4

Another reason for the chosen period is that industries’ power have historically shifted with the

rise of the technological industries.

5

_{Finally, a big concern nowadays are green companies which}

have a high degree of corporate social responsibility. Thus, if 80 years ago chemical and

industrial firms were the backbone of a country’s economy, nowadays people are beginning to

be increasingly concerned about ecological and social issues, shifting economic power to those

industries that have a proactive approach towards CSR issues.

6

All data was collected from Datastream for all nine countries, in order to have

comparability in the cross-section. Both HTV and Tse used, for the U.S.A., Ken French’s data

library, where he builds up industry returns from all companies from NYSE, NASDAQ and AMEX

based on their SIC code. They did so because it involved less data handling and transformations

and made calculations more straight forward. Unfortunately, this classification does not match

the one used by Datastream and a comparison analysis would have been futile, since industry

characteristics would have been mismatched between the U.S.A. and the other eight countries.

For this reason, the dataset was fully extracted from Datastream, using Datastream’s in-house

classification of industries. The number of industries for each country ranges from 33

(Switzerland, Netherlands) to 41

7

_{(U.S., Japan etc.), and each industry return was built using all}

companies in the subsequent industry available on Datastream using time-varying market

weights. Thus, the time series are the most comprehensive ones available to be built from

Datastream, as the number of companies included in each industry varies from 2 to 2000,

depending on the development of each sector in each country.

An important issue at this point is survivorship bias. This plays a major role when

analyzing the effects of a particular industry on the overall market, as effects outside of those

generated by the nature of individual industries might enhance the strength of the relationship

between a particular industry and the market. Even so, by using market weights when building

an industry return series, most of the effects of this issue are taken into account. Moreover,

Datastream reports companies that have gone bankrupt with a market value of zero, up to the

4 _{Cappiello, L., Hordahl, P., Kadareja, A., Manganelli, S., 2006. The impact of the Euro on financial markets. ECB}

Working Paper 598

5 _{Jensen, M., 1993. The Modern Industrial Revolution, Exit, and the Failure of Internal Control Systems. Journal of}

Finance 48, No. 3, 831-880

6 _{Cochran, P., Wood, R., 1984. Corporate Social Responsibility and Financial Performance. Academy of Management}

Journal 27, No. 1, 42-56

7 _{Datastream has, besides the 41 mentioned, both “Other” or “Unclassified” categories which were omitted for the}

5

final date of the dataset. Thus, using market weights provides the most realistic and practical

option, as an investigation on each company present on each market between 2000-2016 would

not add significant value to the analysis.

The overall market is represented by the most popular index used for each analyzed

country, as follows: Australia – ASX 300, Canada – S&P/TSX, France – CAC40, Germany – DAX30,

Japan – Nikkei225, Netherlands – AEX, Switzerland – SMI, U.K. – FTSE100, U.S. – S&P500. The

source of the data is the same, namely Datastream.

In the standard regression, there are four control variables which were included. They

were chosen based on previous empirical research that showed that they might be well

performing forecasters, and were also included by HTV and Tse in their models. These variables

are: inflation (Fama, Schwert [1977]), default spread (Merton [1974], Fama, French [1989]),

market dividend yield (Campbell, Shiller [1998]) and market volatility (French, Schwert,

Stambaugh [1987]).

One notable fact is that inflation is a very complex unit, and is calculated only for longer

periods of time (monthly, quarterly) than the frequency used in this study, which is weekly. For

that, the series were unraveled for each country. This means that inflation was assumed constant

within a month, at a level which would render the overall monthly inflation equal to the real

inflation (takes into account compounding). By performing a unit-root test on all nine generated

inflation series, all series are stationary, as both the industry and market returns, meaning that

a regression analysis is viable.

Banks and financial institutions only seldom report analyses on average applicable rates

through the means of bond spreads. The default spread is the difference between the spreads on

AAA and BBB bonds. Thus, the only available corporate bond spreads for AAA and BBB bonds on

Datastream were for the Eurozone, U.K. and U.S.A., starting in April 2002. Even so,

unsurprisingly, the correlation between them is very high (over 70%), which shows that for

developed markets, corporate bond spreads tend to vary together. For that, a single series was

built as the average of the three markets and used for all statistical tests after taking the first

difference in order to make the series stationary.

6

to French, Schwert and Stambaugh’s methodology

8

_{, as follows. Daily market returns were}

extracted from Datastream and equation (2) was applied in order to determine weekly

volatilities.

𝜎

_𝑡2

_{= ∑ 𝑟}

𝑡,𝑖2 5 𝑖=1

+ 2 ∑ 𝑟

_𝑡,𝑖

𝑟

_𝑡,𝑖+1 4 𝑖=1 (2)

where σ

t

is the weekly volatility and r

t,i

and r

t,i+1

are daily returns from the respective

week. The mean is not taken into account due to the fact that it can be assumed to be zero, since

daily returns are theoretically zero-mean random walks. The formula expands from the regular

variance formula by adding the second term because “non-synchronous trading of securities

causes daily portfolio returns to be auto-correlated, particularly at lag one” (p. 5). The resulting

series are stationary and were used as such in the analysis.

Having collected and organized all the data, the main methodological step is to run the

regression from equation (3), which is based on HTV and Tse’s methodology.

𝑅𝑀

𝑡

= 𝛼

𝑖

+ 𝛽

𝑖

𝐼𝑁𝐷

𝑖,𝑡−1

+ 𝛾

𝑖

𝑅𝑀

𝑡−1

+ 𝜋

𝑖

𝐷𝑆𝑃𝑅

𝑡−1

+ 𝛿

𝑖

𝑀𝑉𝑂𝐿

𝑡−1

+ 𝜃

𝑖

𝑀𝐷𝑌

𝑡−1

+ 𝜆

𝑖

𝐼𝑁𝐹𝐿

𝑡−1

+ 𝑒

𝑖,𝑡 (3)

where RM

t

is the market return, IND

i,t-1

is the 1-lag return for industry i, DSPR

t-1

is the

1-lag default spread, MVOL

t-1

is the 1-lag market volatility, MDY

t-1

is the 1-lag market dividend

yield and finally, INFL

t-1

is the 1-lag inflation rate, all at a weekly frequency. The intuition of this

equation is straightforward: I want to analyze if a snapshot at a specific point in time (t-1) has

enough predictive power to determine next week’s market performance (t).

The next step is to analyze the predictive power of all industries simultaneously on the

market return in order to determine if more information provides more predictability for the

broad market. For that, the regression from equation (4) will be applied.

𝑅𝑀

𝑡

= 𝛼

𝑖

+ ∑ 𝛽

𝑖

𝐼𝑁𝐷

𝑖,𝑡−1

𝑁𝑖

𝑖=1

+ 𝛾

𝑖

𝑅𝑀

𝑡−1

+ 𝜋

𝑖

𝐷𝑆𝑃𝑅

𝑡−1

+ 𝛿

𝑖

𝑀𝑉𝑂𝐿

𝑡−1

+ 𝜃

𝑖

𝑀𝐷𝑌

𝑡−1

+ 𝜆

𝑖

𝐼𝑁𝐹𝐿

𝑡−1

+ 𝑒

𝑖,𝑡 (4)

where all variables are the same as in equation (3), the only difference being that all

industries are included simultaneously in the regression and N

i

denotes the number of industries

for each country.

Summary statistics for all variables used in this analysis are presented in Table 1.

In essence, I will test the following hypothesis:

8 _{French, K., Schwert, G., Stambaugh, R., 1987. Expected stock returns and volatility. Journal of Financial Economics}

7

H

0

: Weekly market returns cannot be forecasted by weekly individual/collective industry

portfolio returns, while controlling for lagged market returns and well-known predictors

H

1

: Weekly market returns can be forecasted by weekly individual/collective industry

8

Table 1. Summary statistics for all variables for nine countries between 2000 and 2016

The table presents summary statistics for 41 industry portfolios, the market index, default spread, market volatility, market dividend yield and inflation. All returns are on a weekly frequency between 01/01/2000 and 31/12/2016 and means and standard deviations are reported as percentages. The source of all data is Datastream. If a cell has a value “NA”, then that industry is not represented by any company in the specific country according to Datastream. Industry returns were computed using all available company data, and individual company returns within an industry were weighted through their market value at the specific point in time. The Market Return is the return of the following indexes for each country: Australia – ASX 300, Canada – S&P/TSX, France – CAC40, Germany – DAX30, Japan – Nikkei225, Netherlands – AEX, Switzerland – SMI, U.K. – FTSE100, U.S. – S&P500. The Default Spread is the first difference (for stationarity purposes) of the average spread between AAA and BBB corporate bonds of the Eurozone, U.S.A. and U.K., due to high correlation (over 70%) between them. The Market Volatility is the weekly volatility of the index computed using the French, Schwert and Stambaugh methodology. The Market Dividend Yield is the first difference (for stationarity purposes) of the actual weekly dividend yield of each corresponding index. Inflation was determined on a weekly level by unravelling the monthly series with the assumption of a constant level of inflation within a month.

Industry/Variable

Australia Canada France Germany Japan Netherlands Switzerland U.K. U.S.A.

Mean Std. _Dev. No. of _{obs. Mean Dev.} Std. No. of _{obs. Mean Dev.} Std. No. of _{obs. Mean Dev.} Std. No. of _{obs. Mean Dev.} Std. No. of _{obs. Mean Dev.} Std. No. of _{obs. Mean Dev.} Std. No. of _{obs. Mean Dev.} Std. No. of _{obs. Mean Dev.} Std. No. of _obs. Aerospace & Defense 0.21 8.19 845 -0.05 5.22 887 0.21 3.74 887 0.18 4.19 887 0.21 4.30 887 NA NA NA NA NA NA 0.12 3.56 887 0.22 2.86 887

9

Industry/Variable

Australia Canada France Germany Japan Netherlands Switzerland U.K. U.S.A.

Mean Std. _Dev. No. of _{obs. Mean Dev.} Std. No. of _{obs. Mean Dev.} Std. No. of _{obs. Mean Dev.} Std. No. of _{obs. Mean Dev.} Std. No. of _{obs. Mean Dev.} Std. No. of _{obs. Mean Dev.} Std. No. of _{obs. Mean Dev.} Std. No. of _{obs. Mean Dev.} Std. No. of _obs. Household Goods and Home Construction 0.25 3.09 887 0.07 3.93 887 0.20 2.77 887 0.19 3.02 887 0.17 2.98 887 0.07 4.46 887 0.16 2.86 887 0.27 2.55 887 0.13 2.67 887

Industrial Engineering 0.05 3.98 887 0.23 2.92 887 0.01 4.90 887 0.16 3.39 887 0.17 3.61 887 0.19 4.41 887 0.06 4.61 887 0.24 3.25 887 0.23 3.69 887 Industrial Metals & Mining 0.17 4.39 887 0.27 4.94 887 0.06 6.23 887 0.18 4.19 887 0.12 4.06 887 -0.03 7.23 887 -0.11 5.87 887 0.23 7.32 887 0.09 5.09 887 Industrial Transportation 0.19 3.00 887 0.29 3.01 887 0.25 2.66 887 0.17 4.69 887 0.08 3.18 887 0.02 3.75 887 0.21 3.57 887 -0.02 4.98 887 0.21 3.06 887 Leisure Goods 0.14 5.14 887 0.19 7.54 887 0.11 4.54 887 -0.41 6.47 887 0.00 3.64 887 0.30 3.76 887 NA NA NA 0.04 5.09 887 0.27 3.23 887 Life Insurance -0.04 4.01 887 0.14 3.39 887 0.08 4.16 887 -0.49 10.16 887 0.18 5.53 766 -0.21 6.43 887 -0.06 5.83 887 0.04 4.52 887 0.16 4.74 887 Media -0.05 2.86 887 0.07 2.70 887 -0.05 3.72 887 0.05 3.34 887 0.04 3.24 887 0.06 3.08 887 0.01 2.84 887 0.07 3.13 887 0.09 3.35 887 Mining 0.17 3.66 887 0.24 4.72 887 0.13 3.76 887 0.15 2.91 887 0.14 5.27 887 NA NA NA -1.34 15.00 443 0.15 4.89 887 0.32 4.40 887 Mobile Telecommunications -0.31 6.71 887 0.12 4.26 887 0.43 6.10 164 -0.17 4.30 887 -0.03 4.08 887 NA NA NA NA NA NA -0.05 4.01 887 0.07 4.03 887 Nonequity Investment Instruments 0.09 1.65 800 0.07 2.08 887 -0.01 3.03 823 -0.03 3.05 834 0.02 3.63 622 0.42 3.58 136 -0.01 3.32 503 0.25 1.76 166 0.03 2.33 887 Nonlife Insurance 0.11 3.31 887 0.18 2.93 887 -0.01 5.21 887 -0.03 4.29 887 0.09 4.72 887 0.42 3.38 29 -0.04 4.20 887 0.06 3.26 887 0.11 2.84 887 Oil & Gas Producers 0.17 3.47 887 0.20 3.67 887 0.05 3.42 887 -0.20 8.86 887 0.09 3.78 887 -0.02 3.51 887 -2.96 29.87 451 0.04 3.45 887 0.16 3.22 887 Oil Equipment & Services 0.28 4.37 887 0.21 2.20 887 0.01 5.27 887 -0.12 6.04 474 -0.05 6.46 887 0.11 5.06 887 NA NA NA 0.12 4.84 887 0.16 4.11 887 Personal Goods -0.07 5.09 887 0.35 5.50 887 0.13 3.03 887 0.22 2.85 887 0.15 2.36 887 0.09 3.15 887 0.14 4.32 887 0.16 3.02 887 0.19 2.59 887 Pharmaceuticals & Biotechnology 0.30 3.46 887 0.12 5.28 887 0.09 3.60 887 0.19 3.93 887 0.13 2.60 887 0.05 5.61 887 0.05 2.76 887 0.06 2.91 887 0.13 2.49 887 Real Estate Investment & Services 0.07 3.17 887 0.27 2.10 887 0.22 2.64 887 0.01 2.24 887 0.18 4.31 887 -0.04 4.11 887 0.05 0.98 887 0.17 2.45 887 0.32 3.65 887 Real Estate Investment Trusts 0.08 2.66 887 0.15 2.16 887 0.18 2.57 887 0.03 3.83 887 0.14 3.09 798 0.02 2.70 887 0.11 2.95 271 0.10 3.08 887 0.16 3.12 887 Software & Computer Services 0.10 3.25 887 0.23 4.09 887 0.05 4.15 887 0.09 4.75 887 0.01 4.33 887 0.01 4.78 887 0.05 5.96 887 0.07 3.99 887 0.12 3.31 887 Support Services 0.18 2.95 887 0.24 2.24 887 0.19 3.15 887 0.20 3.61 887 0.12 3.44 887 0.07 5.05 887 0.04 4.26 887 0.15 2.38 887 0.25 3.73 887 Technology, Hardware & Equipment 0.05 4.59 887 0.03 8.71 887 -0.10 4.95 887 0.08 6.34 887 -0.01 3.63 887 0.11 5.92 887 0.15 4.54 887 0.03 5.33 887 0.12 4.37 887 Tobacco NA NA NA -0.42 15.96 887 NA NA NA NA NA NA 0.18 4.35 887 NA NA NA NA NA NA 0.29 3.05 887 0.27 3.32 887 Travel & Leisure 0.12 2.37 887 0.26 3.18 887 0.07 3.37 887 -0.06 4.29 887 0.11 2.23 887 0.14 6.70 887 0.18 2.70 887 0.17 2.90 887 0.24 3.08 887

10

3. Results

3.1. Individual Industries’ Predictions

I run the regression from equation (3) for each industry from each country and analyze

the value, but mostly the significance of the β coefficients. For simplicity, since the total output

of these tests encompasses a very large data table, I will focus on the most important and

representative markets when presenting results, namely Germany, Japan, U.K. and U.S.A. These

four are arguably the most active markets of the nine, and the results for the remaining five

countries are presented in Appendix C, and are only mentioned in the following section as a

means of comparison and triangulation of the obtained results. Nevertheless, as HTV show in

their paper, this gradual diffusion of information through the market is very similar in behavior

across countries, continents and currencies. The four representative countries were chosen as

to have a good balance between differences in location and currency, and overall global presence

through the level of trading and size of companies listed.

Statistically significant results nonetheless contradict Fama’s Efficient Market

Hypothesis, and support HTV’s view that information is gradually encapsulated by the market,

depending on its point of origin and interdependencies with other sectors. The decision rule will

focus on the significance level of each β, rather than the coefficient’s level or sign. This is because

an active investment strategy that may use the fundamentals of this study has the freedom to

take advantage of a signal in both directions through both long and short positions, depending

on what the trader should expect based on individual industry movements.

The summary results for the nine countries are presented in Table 2. The individual

estimation outputs for Germany, Japan, U.K. and U.S.A. are presented in Table 3, Panels A through

D, while the full results are shown in the Appendix C.

11

The aim of this paper is to analyze the potential relationship discovered by HTV on a

higher frequency and see how results compare to their findings. By increasing the number of

lags for the independent variables, explanatory power does not improve and, moreover, it

determines results to overlap with those of HTV and Tse who studied the process on a monthly

basis. Finally, an investment strategy would be hard to implement if it relied on a trail of

numerous signals, since a favorable scenario would happen very rarely.

The results show a fairly well predictive power for certain industries in forecasting

weekly market returns. R-squared values vary from around 0.50% to almost 4.50%, which is

similar to those obtained by HTV, which is surprising considering that the data points have a

higher frequency in this case. Even so, one important matter is that by having weekly returns,

the amount of noise contained by the returns increases compared to the monthly data case. The

adjustments made using the Newey-West standard error corrections for heteroskedasticity and

serial correlation improve the reliability of estimates, but this does not compensate enough for

the behavioral aspects of stock trading by investors. When analyzing weekly returns, the data

are inevitably tainted by high frequency traders, sentiment trades, market anomalies like the

day of the week effect and other such phenomena. Even so, the analyzed time period, which

includes only the new millennium, is relevant for today’s trading habits and general forms of

active trading. For that, a strategy which involves taking into account this form of cross-sectional

time-lagged predictability should nevertheless take into account actions as previously

described.

12

Table 2. Industries that predict the market by 1-lag on a weekly basis at various

confidence levels between 2000 and 2016 for nine countries

Results were obtained by regressing the market return, represented by each country’s respective index, on the lagged index return, industry return and 4 control variables (default spread, volatility, dividend yield and inflation). Full results can be viewed in Table 3, Panels A through D, for Germany, Japan, U.K. and U.S.A., and in Appendix C for the other 5 countries. The significance level is based on the p-value of the β coefficient of each industry’s lagged return.

90% 95% 99% Count:

Australia

Aerospace & Defense Alternative Energy Aerospace & Defense 2

Electricity Gas, Water & _{Multiutilities} Alternative Energy 3

Equity Investment

Instruments Media Banks 1

Real Estate

Investment & Services Chemicals 1

Canada

Alternative Energy Gas, Water & _{Multiutilities} Technology, Hardware &

Equipment Electricity 2

Industrial

Transportation Equity Investment Instruments 3

France

Industrial

Transportation Food Producers Financial Services 1

Nonlife Insurance Pharmaceuticals & _{Biotechnology Telecommunications} Fixed Line 1

Germany

Food & Drug Retailers Aerospace & Defense Food & Drug Retailers 1 Healthcare Equipment

& Services Food Producers 1

Mobile

Telecommunications Forestry & Paper 1

Technology, Hardware & Equipment

Gas, Water &

Multiutilities 2

Japan

Personal Goods Equity Investment _Instruments Healthcare Equipment _{& Services} 1 Nonlife Insurance Household Goods and _{Home Construction} 1

Netherlands

Household Goods and

Home Construction Alternative Energy Transportation Industrial 2

Technology, Hardware &

Equipment Chemicals Life Insurance 1

Nonlife Insurance Media 1

Switzerland Nonlife Insurance Life Insurance _{Telecommunications} Mobile 1

UK

Electricity Nonlife Insurance 4

Fixed Line

Telecommunications Personal Goods 1

USA

13

Table 3. Predictive regressions outputs for Germany, Japan, U.K. and U.S.A.

The four panels show the results of each individual industry’s predictive power on the market, for four countries. The estimations were produced by regressing the market return, represented by indexes (Germany – DAX30, Japan – Nikkei225, U.K. – FTSE100, U.S. – S&P500) on the lagged individual industry return (IND(-1)), market return (RM(-1)) and 4 control variables (default spread – the difference between AAA and BBB corporate bond spreads [DSPR(-1)], market volatility [MVOL(-1)], market dividend yield [MDY(-1)] and inflation [INFL(-1)]). Data points have a weekly frequency and include all available observations between 01/01/2000 and 31/12/2016, all extracted from Datastream. Only coefficients are reported, and no standard deviations, for simplicity. More elaborate results are available in Appendix C. The regressions were estimated using OLS and the standard errors include a Newey-West serial correlation and heteroskedasticity correction. p-values smaller than 0.01, 0.05 and 0.10 are indicated by ***, ** and *, respectively.

Panel A: Germany

Constant IND(-1) RM(-1) DSPR(-1) MVOL(-1) MDY(-1) INFL(-1) R-squared

Aerospace & Defense 0.0012 -0.0890** 0.0901 0.4351 -0.1048 2.9432 3.1216* 2.39%

Alternative Energy 0.0011 -0.0058 0.0485 0.3395 -0.1831 3.4337 3.2082** 1.36%

Automobiles & Parts 0.0011 0.0002 0.0458 0.3208 -0.1641 3.5233 3.1960** 1.35%

Banks 0.0011 0.0275 0.0207 0.2720 -0.1291 3.8210 3.1896** 1.43%

Beverages 0.0013 -0.0842 0.0422 0.4322 -0.2619 3.2864 3.1469* 1.56%

Chemicals 0.0011 0.0232 0.0243 0.2847 -0.1614 3.5049 3.1833** 1.36%

Construction & Materials 0.0011 0.0060 0.0414 0.3029 -0.1624 3.5359 3.1831* 1.36%

Electricity 0.0011 0.0178 0.0438 0.3188 -0.1399 3.5525 3.2062** 1.37%

Electronic & Electrical Equipment 0.0010 0.0336 0.0206 0.2631 -0.1195 3.5632 3.2094** 1.41%

Equity Investment Instruments 0.0015 0.0463 -0.0804 5.4608 -0.3010 -1.9816 4.2219 4.06%

Financial Services 0.0010 0.0895 -0.0014 0.2505 -0.0901 3.6818 3.2756** 1.59%

Fixed Line Telecommunications 0.0014 -0.0076 0.0620 -0.5393 -0.2919 4.3555 3.5714* 2.21%

Food & Drug Retailers 0.0012 0.0818* -0.0076 0.1718 -0.2176 3.8393 2.9433* 1.80%

Food Producers 0.0011 -0.0080 0.0478 0.3267 -0.1592 3.4682 3.1815** 1.36%

Forestry & Paper 0.0011 -0.0085 0.0488 0.3621 -0.1769 3.4837 3.2208** 1.40%

Gas, Water & Multiutilities 0.0010 -0.0905 0.1195 0.4712 -0.2058 3.5004 3.0064* 1.94%

General Industrials 0.0011 0.0773 -0.0364 0.3661 -0.1408 3.6111 3.1433* 1.58%

General Retailers 0.0010 0.0510 0.0208 0.2492 -0.1297 3.6109 3.2761** 1.49%

Healthcare Equipment & Services 0.0014 -0.1021* 0.0898 0.2728 -0.2918 3.4621 3.1532** 1.92%

Household Goods & Home Construction 0.0011 -0.0027 0.0475 0.3257 -0.1648 3.5199 3.1952** 1.35%

Industrial Engineering 0.0010 0.0231 0.0306 0.2782 -0.1452 3.6193 3.2064** 1.38%

Industrial Metals & Mining 0.0012 -0.0386 0.0707 0.3580 -0.1977 3.2125 3.2587** 1.50%

Industrial Transportation 0.0011 -0.0299 0.0701 0.4420 -0.1872 3.4973 3.2475** 1.41% Leisure Goods 0.0011 0.0005 0.0458 0.3174 -0.1621 3.5214 3.1960** 1.35% Life Insurance 0.0010 -0.0091 0.0469 0.3461 -0.1611 3.5174 3.1458* 1.44% Media 0.0011 0.0097 0.0422 0.3017 -0.1507 3.5384 3.2008** 1.36% Mining 0.0011 0.0124 0.0454 0.3144 -0.1574 3.5345 3.1808** 1.36% Mobile Telecommunications 0.0009 -0.1335* 0.1334 -0.0810 -0.0695 3.2981 3.1278* 2.79%

Nonequity Investment Instruments 0.0009 -0.1865 0.1939 0.4477 -0.1077 3.1156 3.2566** 1.68%

Nonlife Insurance 0.0012 0.0328 0.0094 0.3175 -0.1664 3.5147 3.2381** 1.41%

Oil & Gas Producers 0.0011 0.0010 0.0460 0.3235 -0.1610 3.5246 3.1795** 1.35%

Oil Equipment & Services 0.0011 -0.0444 0.0747 -0.4099 -0.2592 4.0953 3.7666* 2.86%

Personal Goods 0.0011 -0.0030 0.0472 0.3229 -0.1632 3.5106 3.1992** 1.35%

Pharmaceuticals & Biotechnology 0.0010 0.0527 0.0122 0.3039 -0.1130 3.4784 3.3019** 1.49%

Real Estate Investment & Services 0.0010 0.0677 0.0284 0.1981 -0.1229 3.7067 3.2535** 1.48%

Real Estate Investment Trusts 0.0011 -0.0170 0.0503 0.3812 -0.1925 3.3759 3.1928** 1.38%

Software & Computer Services 0.0011 0.0758 -0.0281 0.3740 -0.1824 3.3790 3.1966** 1.78%

Support Services 0.0008 0.1227 -0.0223 0.2534 -0.0939 3.9901 3.3384** 2.11%

Technology, Hardware & Equipment 0.0010 0.0808* -0.0376 -0.0683 -0.0692 3.9046 3.1621** 2.28%

14

Panel B: Japan

Constant IND(-1) RM(-1) DSPR(-1) MVOL(-1) MDY(-1) INFL(-1) R-squared

Aerospace & Defense -0.0003 0.0006 0.0277 2.0007 0.7891 2.2039 1.3092 0.89%

Alternative Energy -0.0003 -0.0280 0.0333 2.0465 0.7925 1.6468 1.3939 1.03%

Automobiles & Parts -0.0004 0.1319 -0.0785 2.1394 0.7233 3.8507 1.3161 1.48%

Banks -0.0003 0.0151 0.0177 1.9677 0.7898 2.5109 1.3132 0.91%

Beverages -0.0003 -0.0040 0.0296 2.0009 0.7852 2.1543 1.3094 0.89%

Chemicals -0.0004 0.0988 -0.0486 1.9929 0.7260 3.2446 1.2185 1.04%

Construction & Materials -0.0004 0.0823 -0.0190 2.0321 0.7213 3.4056 1.1891 1.08%

Electricity -0.0003 0.0193 0.0292 1.9952 0.7943 2.8031 1.3621 0.93%

Electronic & Electrical Equipment -0.0003 0.0193 0.0086 1.9942 0.7831 2.1737 1.2995 0.90%

Equity Investment Instruments -0.0011 -0.0212** 0.0898 3.4429* 0.9423* 4.7088 1.4910 2.10%

Financial Services -0.0002 0.0794* -0.0497 1.8450 0.6846 3.3840 1.0517 1.31%

Fixed Line Telecommunications -0.0002 -0.0812 0.0618 2.1297 0.7325 1.5442 1.1418 1.48%

Food & Drug Retailers -0.0001 -0.0831 0.0518 1.9690 0.7516 1.3946 1.2857 1.09%

Food Producers 0.0003 -0.2586 0.1059 2.0203 0.5865 -0.7079 1.3819 2.48%

Forestry & Paper -0.0003 0.0050 0.0255 2.0042 0.7887 2.2783 1.3086 0.89%

Gas, Water & Multiutilities -0.0003 -0.0216 0.0315 2.0072 0.7894 1.8700 1.3074 0.92%

General Industrials -0.0004 0.1050 -0.0625 1.9263 0.7867 3.2693 1.3108 1.28%

General Retailers -0.0003 -0.0370 0.0544 2.0085 0.7888 2.0748 1.3303 0.93%

Healthcare Equipment & Services 0.0001 -0.1390 0.1116 1.8321 0.6876 0.6146 1.4505 1.47%

Household Goods & Home Construction -0.0004 0.0765 -0.0231 2.0475 0.6986 3.0920 1.2499 1.04%

Industrial Engineering -0.0004 0.1595 -0.1287 1.8652 0.6189 3.2042 1.2395 1.49%

Industrial Metals & Mining -0.0003 0.1457 -0.0921 2.1475 0.6743 4.6679 1.2973 2.25%

Industrial Transportation -0.0005 0.1577 -0.0830 2.2720 0.8189 4.0168 1.5050 1.63% Leisure Goods -0.0003 0.1101 -0.0721 2.0266 0.7699 2.4523 1.1580 1.35% Life Insurance -0.0003 -0.0051 0.0305 2.0151 0.7863 2.0390 1.3056 0.90% Media -0.0003 -0.0237 0.0429 1.9974 0.7824 1.9388 1.3221 0.91% Mining -0.0004 0.0316 0.0027 1.9762 0.8032 2.7845 1.2294 1.04% Mobile Telecommunications -0.0003 -0.0589 0.0661 2.0595 0.8060 2.0041 1.2243 1.10%

Nonequity Investment Instruments -0.0004 -0.0712 0.1014 3.7307* 0.8941* 2.4755 1.1650 1.62%

Nonlife Insurance -0.0003 -0.1060** 0.1074 2.2364 0.8337 -0.2509 1.5499 2.08%

Oil & Gas Producers -0.0004 0.0762 -0.0029 1.9905 0.8247 3.6866 1.1601 1.41%

Oil Equipment & Services -0.0003 -0.0004 0.0284 2.0002 0.7888 2.1926 1.3095 0.89%

Personal Goods 0.0000 -0.2061* 0.1153 1.9796 0.7261 -0.1400 1.4608 1.79%

Pharmaceuticals & Biotechnology -0.0001 -0.1206 0.0815 1.9377 0.6746 1.1545 1.1960 1.35%

Real Estate Investment & Services -0.0003 0.0510 -0.0183 1.9096 0.7247 2.7819 1.1892 1.11%

Real Estate Investment Trusts -0.0004 0.0534 0.0191 2.0155 0.8054 3.4143 1.3309 1.09%

Software & Computer Services -0.0003 0.0068 0.0230 1.9999 0.7908 2.2445 1.3202 0.90%

Support Services -0.0004 0.1669 -0.1059 1.8835 0.6501 3.5769 1.2092 2.02%

Technology, Hardware & Equipment -0.0003 0.1053 -0.0669 1.9356 0.7597 2.8701 1.1637 1.13%

15

Panel C: U.K.

Constant IND(-1) RM(-1) DSPR(-1) MVOL(-1) MDY(-1) INFL(-1) R-squared

Aerospace & Defense 0.0000 -0.0240 0.0861 -0.2939 -0.0058 3.1992 0.8538 1.21%

Alternative Energy 0.0000 -0.0035 0.0670 -0.3443 0.0124 3.1669 0.8901 1.16%

Automobiles & Parts 0.0000 0.0061 0.0553 -0.3684 0.0284 3.1818 0.9059 1.16%

Banks -0.0001 -0.0361 0.1050 -0.3999 -0.0087 3.0777 0.9591 1.27%

Beverages 0.0000 0.1047 0.0044 -0.2864 -0.0304 3.3377 0.7608 1.83%

Chemicals -0.0001 0.0164 0.0498 -0.3711 0.0427 3.2152 0.9110 1.17%

Construction & Materials 0.0000 -0.0047 0.0698 -0.3532 0.0184 3.2320 0.8973 1.15%

Electricity 0.0002 -0.1393** 0.1225 -0.2000 -0.1225 2.7285 0.8784 2.56%

Electronic & Electrical Equipment -0.0001 0.0031 0.0625 -0.3610 0.0291 3.2136 0.9086 1.15%

Equity Investment Instruments 0.0000 -0.0099 0.0668 -0.3295 -0.0177 3.1770 0.9113 1.16%

Financial Services 0.0000 -0.0040 0.0682 -0.3487 0.0190 3.2080 0.9070 1.15%

Fixed Line Telecommunications 0.0002 -0.0836** 0.1349 -0.0459 -0.1938 3.3105 0.7648 2.15%

Food & Drug Retailers -0.0001 -0.0591 0.1142 -0.3755 0.0069 3.3680 0.9583 1.43%

Food Producers -0.0001 0.0110 0.0593 -0.3567 0.0290 3.2350 0.9009 1.16%

Forestry & Paper -0.0001 0.0177 0.0481 -0.3799 0.0316 3.2371 0.8602 1.23%

Gas, Water & Multiutilities 0.0000 -0.0833 0.1206 -0.3462 0.0583 3.3021 0.8984 1.55%

General Industrials -0.0001 0.0086 0.0573 -0.3696 0.0394 3.2117 0.9065 1.16%

General Retailers 0.0000 -0.0441 0.1011 -0.2823 0.0257 3.2804 0.9356 1.32%

Healthcare Equipment & Services 0.0000 -0.0065 0.0691 -0.3536 0.0216 3.2166 0.9029 1.16%

Household Goods & Home Construction -0.0001 0.0561 0.0288 -0.4230 0.0158 3.1630 0.8884 1.34%

Industrial Engineering 0.0001 -0.0487 0.1027 -0.2749 -0.0825 3.0490 0.9347 1.33%

Industrial Metals & Mining -0.0001 0.0110 0.0550 -0.3833 0.0664 3.3180 0.8857 1.23%

Industrial Transportation 0.0000 -0.0067 0.0692 -0.3447 0.0046 3.1625 0.9105 1.17% Leisure Goods 0.0000 -0.0271 0.0771 -0.2569 -0.0014 3.2827 0.8904 1.31% Life Insurance -0.0001 0.0523 -0.0105 -0.5097 0.0083 3.2356 0.9368 1.51% Media 0.0000 -0.0321 0.0915 -0.3225 0.0116 3.2042 0.8839 1.19% Mining 0.0000 -0.0005 0.0655 -0.3523 0.0216 3.2108 0.9036 1.15% Mobile Telecommunications 0.0000 -0.0777 0.1332 -0.5294 0.0645 3.3294 0.7572 1.89%

Nonequity Investment Instruments -0.0022 -0.0068 -0.2044 -1.7174 6.3539*** -4.0163 -0.8462 4.46%

Nonlife Insurance 0.0000 -0.0524 0.1076 -0.1920 0.0587 3.3837 0.8096 1.39%

Oil & Gas Producers 0.0000 0.0570 -0.0005 -0.4209 -0.0518 3.1065 0.9069 1.39%

Oil Equipment & Services 0.0000 -0.0085 0.0740 -0.3584 0.0098 3.1881 0.9283 1.16%

Personal Goods -0.0001 0.0149 0.0558 -0.3751 0.0303 3.2328 0.8971 1.17%

Pharmaceuticals & Biotechnology -0.0001 0.0164 0.0523 -0.3340 0.0163 3.2002 0.9280 1.17%

Real Estate Investment & Services -0.0001 0.0225 0.0561 -0.3666 0.0883 3.3199 0.9165 1.18%

Real Estate Investment Trusts 0.0001 -0.0510 0.1090 -0.2853 -0.0664 3.2369 0.8405 1.39%

Software & Computer Services 0.0000 -0.0017 0.0662 -0.3516 0.0193 3.2146 0.9015 1.15%

Support Services 0.0000 -0.0318 0.0919 -0.3244 -0.0036 3.2571 0.8773 1.18%

Technology, Hardware & Equipment 0.0000 -0.0072 0.0709 -0.3293 -0.0003 3.2141 0.8926 1.16%

Tobacco 0.0001 -0.0451 0.0891 -0.3837 -0.0149 3.0878 0.8938 1.32%

16

Panel D: U.S.A.

Constant IND(-1) RM(-1) DSPR(-1) MVOL(-1) MDY(-1) INFL(-1) R-squared

Aerospace & Defense 0.0002 0.0739 -0.0244 -0.5388 -0.1665 3.7336 1.5815 0.71%

Alternative Energy 0.0004 -0.0079 0.0516 -0.4339 -0.2585 3.5618 1.5408 0.51%

Automobiles & Parts 0.0004 0.0707 -0.0314 -0.5018 -0.2707 4.4330 1.3112 0.95%

Banks 0.0005 -0.1037* 0.0874 -0.4680 -0.2250 -0.5497 1.3192 1.48%

Beverages 0.0003 0.0351 0.0171 -0.4322 -0.1974 3.4801 1.5572 0.54%

Chemicals 0.0003 0.1108 -0.0808 -0.3958 -0.2256 3.9017 1.3654 1.06%

Construction & Materials 0.0004 -0.0041 0.0478 -0.4582 -0.2394 3.7185 1.5057 0.49%

Electricity 0.0004 -0.0323 0.0624 -0.4171 -0.2649 3.7095 1.4894 0.54%

Electronic & Electrical Equipment 0.0005 0.0707 -0.0517 -0.6131 -0.3194 3.3860 1.3347 0.69%

Equity Investment Instruments 0.0007 -0.2037*** 0.0132 -0.2365 -0.6760 -3.2027 1.4655 1.78%

Financial Services 0.0004 -0.0585*** 0.0873 -0.4326 -0.1147 2.3955 1.7236 1.12%

Fixed Line Telecommunications 0.0004 -0.0257 0.0611 -0.4519 -0.2369 3.6589 1.4268 0.54%

Food & Drug Retailers 0.0004 0.0350 0.0089 -0.4368 -0.2729 3.1141 1.4990 0.54%

Food Producers 0.0005 -0.0876 0.1034 -0.5331 -0.1420 3.5576 1.4839 0.77%

Forestry & Paper 0.0003 0.0912*** 0.0066 -0.7236 -0.0735 7.1518 1.2606 1.59%

Gas, Water & Multiutilities 0.0003 0.0397 0.0194 -0.4986 -0.2175 3.8585 1.5637 0.56%

General Industrials 0.0006 0.1270 -0.0927 -0.5498 -0.4597 3.9115 1.3151 1.25%

General Retailers 0.0003 0.0460 -0.0064 -0.3906 -0.2285 3.4977 1.6010 0.55%

Healthcare Equipment & Services 0.0008 -0.1564 0.1440 -0.4793 -0.4566 2.5630 1.5410 1.31%

Household Goods and Home Construction 0.0004 -0.0138 0.0532 -0.4814 -0.2515 3.7550 1.4823 0.49%

Industrial Engineering 0.0004 0.0170 0.0193 -0.4751 -0.2611 3.7079 1.4565 0.50%

Industrial Metals & Mining 0.0004 0.0486 -0.0200 -0.4887 -0.2231 4.4699 1.3506 0.90%

Industrial Transportation 0.0004 0.0313 0.0159 -0.4606 -0.2563 3.9664 1.5129 0.53% Leisure Goods 0.0004 0.0690 -0.0319 -0.4912 -0.2495 3.3807 1.2957 0.82% Life Insurance 0.0005 -0.0349 0.0598 -0.3523 -0.3007 2.0068 1.4883 0.63% Media 0.0003 0.0654 -0.0187 -0.5206 -0.2012 4.2641 1.5578 0.62% Mining 0.0004 -0.0014 0.0436 -0.4531 -0.2406 3.7079 1.5129 0.49% Mobile Telecommunications 0.0004 -0.0511 0.0683 -0.4082 -0.2606 2.4830 1.5408 0.84%

Nonequity Investment Instruments 0.0005 0.2355 -0.1712 -0.5253 -0.2525 3.0377 1.3491 0.79%

Nonlife Insurance 0.0005 -0.0518 0.0855 -0.4285 -0.2500 3.4376 1.3851 0.57%

Oil & Gas Producers 0.0004 0.0004 0.0427 -0.4598 -0.2421 3.7395 1.5048 0.49%

Oil Equipment & Services 0.0004 -0.0209 0.0551 -0.4307 -0.2729 3.2093 1.5969 0.54%

Personal Goods 0.0004 0.0112 0.0365 -0.4441 -0.2371 3.8559 1.5113 0.49%

Pharmaceuticals & Biotechnology 0.0004 -0.0185 0.0611 -0.4925 -0.2391 3.9203 1.4874 0.50%

Real Estate Investment & Services 0.0003 0.0352 0.0081 -0.4822 -0.2028 3.9798 1.4560 0.63%

Real Estate Investment Trusts 0.0004 -0.0044 0.0458 -0.4600 -0.2465 3.6603 1.5045 0.49%

Software & Computer Services 0.0004 0.0139 0.0279 -0.4618 -0.2537 3.6816 1.4850 0.49%

Support Services 0.0004 0.0096 0.0306 -0.4639 -0.2402 3.6188 1.5084 0.50%

Technology, Hardware & Equipment 0.0005 0.0683 -0.0576 -0.5158 -0.3922 2.8288 1.3291 0.83%

Tobacco 0.0004 -0.0030 0.0457 -0.4609 -0.2448 3.7792 1.5065 0.49%

17

In Germany’s case, there are four industries significant at a 90% level and one at a 95%

level: Food & Drug Retailers has a coefficient of 0.08, Healthcare Equipment & Services -0.10,

Mobile Telecommunications -0.13, Technology, Hardware & Equipment 0.08 and Aerospace &

Defense -0.09, respectively. It is not immediately clear why the signs of these coefficients are

positive or negative. A behavioral or intuitive explanation might be, for example, that if

healthcare companies rise in value, then a country is more dependent on it. As a consequence,

the overall population feels a decrease in overall life quality and, for that, using the market index

as a proxy for the economic development of a country, it has a negative correlation determined

by the opposing forces. The same interpretation can be attributed to the Aerospace & Defense

industry, as when value goes up, individuals might become unsettled and feel less safe, and the

market’s actors’ reaction is to pull out, in order to preserve their wealth and/or sense of security.

Regarding the former, the same behavior can be seen within the other three countries from Table

3, which is that Healthcare Equipment & Services is a negative lagged predictor for the market,

although insignificant. Aerospace & Defense has more variability, being actually positive for the

U.S.A. From this perspective, a cultural, behavioral interpretation might be better suited in order

to explain the relationship between companies from this sector and the overall market. The

U.S.A., being one of the top superpowers and arms manufacturers in the world, relies heavily on

the export of weapons and defense spending. Thus, inevitably, a major driver of the U.S.A.

economy is the Aerospace & Defense sector, which, if it is well performing, boosts the overall

economy.

18

commodities) like Oil & Gas Producers, Oil Equipment & Services and Mining will have an

expected negative lagged correlation with the market. This also emphasizes the dismissal of the

Efficient Market Hypothesis by supporting the idea that markets lack efficiency, or how HTV

present in their paper: “The fact that the signs of these predictive relations are consistent with

conventional wisdom on the relation of these industries to the macro-economy reassures us that

these predictive regressions are indeed capturing the slow diffusion of sector information into

the broad market index as opposed to being the result of chance” (p. 14).

For the U.S.A., there are four driving industries, one significant at a 90%, and 3 at a 99%

level: Banks with a coefficient of -0.10, Equity Investment Instruments with -0.20, Financial

Services with -0.06, and Forestry & Paper with 0.09, respectively. The negative relationships are

surprising, as they deviate from HTV’s and Tse’s results, which are positive in both instances.

This is because the financial sector is a big part of the American economy, and even though the

market, which in this case is represented by the S&P500 index, includes roughly 50 companies

from this sector, it may not be well diversified enough to encapsulate the full effects of this sector.

Other countries have mixed reactions to these sectors, since, for the U.K., they are all negative as

in the U.S.A. case, and for Germany they are all positive. Another potential reason for this

relationship is the inclusion of the Financial Crisis of 2008 in the dataset, when the biggest hit

was taken by the financial sector, among the real-estate sector. In the build-up to this downturn,

the financial sectors around the world were booming, as laissez-faire activities, NINJA loans,

CDOs and other products were filling up financial institutions’ balance sheets, unknowingly

rupturing the economy from the inside. Even neglecting the financial crisis, the financial sector

heavily outperformed the market. For example, €1 invested on 01/01/2000 would have a value

on 31/12/2016 of: €1.52 if it were invested in the market index (S&P500), €2.18 for Banks,

€0.92 for Equity Investment Instruments and €10.50 for Financial Services. Equity Investment

Instruments was actually outperforming the market, but it took the hardest hit of the 4 in

2007-2008.

Overall, the results are less significant than what previous research has showed.

Nevertheless, this was expected, as by increasing the frequency of data points from monthly to

weekly, a lot of noise is captured by the underlying market-weighted prices of each industry. For

that, alternative specifications will be tested in the following sections, by performing robustness

tests through the split of the dataset in multiple periods.

19

markets on a weekly level. Both HTV and Tse found significant results for these activities. As

previously stated, this might be because the technological cycle is longer for such industries than

other ones, and information might have a slower flow to the economy. Moreover, previous

research had datasets spanning for a very long period of time, well before the year 2000 (HTV

had 56 years of data and Tse had 69 years), and in the second half of the 20

th

_{century all}

commodity prices were rising: oil, copper, natural gas and gold. Inevitably, industries which

extracted, processed or used these for manufacturing were leading the market as more and more

investors were free-riding on this trend. Moreover, the 1979 energy crisis also had a big impact

on the prices of commodities, as energy supplies became scarcer. Considering that both HTV’s

and Tse’s datasets mostly consist of pre-21

st

_{century data, and given that, in that period,}

economies were heavily industrialized and depended fundamentally on this type of “hard”

industries, markets were inevitably driven by sectors such as Industrial Metals, Mining, Oil & Gas

Producers etc. In the new millennium, increasingly more companies started to develop more

efficient ways of using and, sometimes, producing energy. Corporate Social Responsibility plays

a major role nowadays, as people become less dependent on fossil fuels. For that, investors have

shifted away from these so-called “sin stocks” and individuals have started to value increasingly

more companies which take advantage of clean energy.

Regarding individual leading industries presented in Table 2, there is no clear pattern as

to what industries lead a market. Results differ between countries, although there are a few

overlaps, like: Nonlife Insurance, Technology, Hardware & Equipment, Alternative Energy and

Equity Investment Instruments. This goes in line with the idea that advancements in newer,

high-tech sectors have become market drivers. Investors started valuing companies differently

than a few decades ago, and human capital and R&D developments have become increasingly

valuable for a company.

Although there is a lot of variability about what industries do lead market by one week,

there is a possible explanation for this heterogeneity. Lo and MacKinlay (1990)

9

_{argue that large}

stocks lead small stocks by studying lead-lag relationships on the U.S.A. market on a weekly basis

between 1962 and 1987. This effect can also be seen in the results from Tables 2 and 3, as, for

each country, the industries that lead the overall market are some of the most developed ones in

the intracommunity area. Moreover, they might be the most popular or liquid ones, if one were

to analyze the trading volumes compared to other industries. For that, there is a mild time-lagged

9 _{Lo, A., MacKinlay, C., 1990. When are contrarian profits due to stock market overreaction? The Review of Financial}

20

size effect in the cross-section, between the reigning industries of a country and the overall

economic performance, using the broad market index as proxy.

3.2. Collective Industries’ Predictions

A natural question is of course the following: what happens if we include all industries

simultaneously in the regression? For that, the regression equation that was used is equation

(4). Individual industries which do not include all observations were left out, in order for the

data to be as representative as possible for the period between 01/01/2000 and 31/12/2016.

The industries which have significant coefficients are reported in Table 4. I do not report

individual coefficients for brevity, but they are presented in Appendix E. I run several Wald tests

in order to determine the joint significance of the independent variables, and the F-test values

(p-values) are for Australia, Canada, France, Germany, Japan, Netherlands, Switzerland, U.K. and

U.S.A.: 1.369 (0.057), 1.593 (0.009), 1.614 (0.009), 1.356 (0.068), 1.422 (0.039), 1.008 (0.456),

1.395 (0.071), 1.599 (0.009) and 4.486 (0.000), respectively. Except for Netherlands, the Wald

test shows at least one statistically significant coefficient at a 90% confidence level for each

country, so the null hypothesis is rejected. The test includes also the control variables which

have less explanatory power than in previous research, due to the longer timeframe that

macroeconomic variables need to fully encapsulate market information.

The results from Table 4 deviate only slightly from those of Table 2. Thus, industries

which were predictors before, are most probably predictors now. The R-squared values also

increased, meaning that there is a joint cross-predictability in the market, and the market does

not react to individual signals, but rather clusters of different drivers. This result is useful, as it

shows, especially for countries with high adjusted R-squared values (i.e. Canada, with 10.41%),

that a snapshot of this form at a specific moment in time can predict a good proportion of next

week’s broad market movements.

21

higher predictive power than the rest.

Table 4. Predictive Regressions for Market Returns using All Industry Returns

Simultaneously

Results were obtained by regressing the broad market return (index) on all industry returns from each respective country with 1-lag and 4 control variables with 1-lag (default spread, market volatility, market dividend yield and inflation). Observations are on a weekly frequency between 01/01/2000 and 31/12/2016. Younger industries, which do not include all observations, were excluded from the regressions. The table shows the industries with statistically significant β coefficients [see equation (4)] and

the adjusted R-squared values of the equations. The 3 significance levels are based on the p-value of each coefficient. The

regressions were estimated using OLS and the standard errors include a Newey-West serial correlation and heteroskedasticity correction. Individual industry returns were built based on market weights of individual companies and the source is Datastream.

90% 95% 99% _squared Adj. R- Count:

Australia

Aerospace & Defense Electricity Media

3.11%

Aerospace & Defense 2

Equity Investment Instruments

Alternative Energy 2

Alternative Energy Beverages 1

Gas, Water &

Multiutilities Chemicals Electricity 2 3

Real Estate Investment &

Services Electrical Equipment Electronic & 1

Canada

General Industrials Alternative Energy

10.41%

Equity Investment

Instruments 3

Life Insurance Industrial

Transportation Telecommunications Fixed Line 3

Technology, Hardware & Equipment

Food & Drug

Retailers 1

Food Producers 1

France

Beverages Fixed Line

Telecommunications

Food Producers

3.33%

Gas, Water &

Multiutilities 2

Chemicals Pharmaceuticals &

Biotechnology

General Industrials 1

Industrial Metals & Mining

Leisure Goods Healthcare

Equipment & Services 1

Germany

Food & Drug Retailers Mobile

Telecommunications

Aerospace & Defense 4.04%

Industrial

Engineering 1

Support Services Industrial Metals &

Mining 2

Technology, Hardware &

Equipment Industrial

Transportation 1

Japan

Equity Investment

Instruments Telecommunications Fixed Line 4.43%

Leisure Goods 1

Life Insurance 3

Nonlife Insurance Media 1

Netherlands Electronic & Electrical _Equipment Chemicals 1.77%

Mobile

Telecommunications 2

Nonlife Insurance 2

Switzerland

Life Insurance Nonlife Insurance

5.30%

Personal Goods 1

Personal Goods Pharmaceuticals &

Biotechnology 1

Real Estate Investment &

Services Real Estate

Investment &

Services 2

UK Life Insurance Electricity _{Telecommunications} Fixed Line _3.42%

Mobile Telecommunications Support Services 1 Technology, Hardware & Equipment 2 USA

Healthcare Equipment & Services Electricity 3.50% Equity Investment Instruments Industrial Engineering

Industrial Metals &

22

4. Robustness Tests

In order to analyze the reliability of previously presented results, I arbitrarily split the

dataset into two equal parts: 2000-2007 and 2008-2016. I run the same regression from

equation (3) and present the results in Table 5. For brevity, I only show what industries had

predictive power at confidence levels of 90%, 95% and 99%, and do not report individual

coefficients. Even so, for previous statistically significant coefficients there is only little

variability and the sign of the relationship (positive/negative) almost never changes.

An important matter at this point is the fact that the overall dataset contains the period

of the global financial crisis of 2008. For that reason, the dataset was split in order to observe

how relationships changed from the pre-crisis period to the post-crisis and moreover, if they are

consistent with previous results for the full sample, presented in Table 2.

Unsurprisingly, the pre-crisis period resembles an increase in the number of industries

from the financial sector as predictors for the broad market. Industries such as Banks, Financial

Services, Life/Nonlife Insurance, (Non-)Equity Investment Instruments, but also Real Estate are

significant and in a greater number than the case of the full sample. At least in the years leading

up to the crisis both the financial sector and the housing market were booming, especially in the

U.S.A. For that, inevitably they became trend-setters for other businesses and consequently, the

economy, as more and more investors started to free-ride this bubble.

23

Table 5. Robustness tests results of industries that lead the overall market by 1-lag in a

two period analysis

Results were obtained by regressing the broad market return (index) on all industry returns from each respective country with 1-lag and 4 control variables with 1-lag (default spread, market volatility, market dividend yield and inflation). Observations are on a weekly frequency between 01/01/2000 and 31/12/2016. Panel A shows the results for the first half of the dataset, between 01/01/2000 and 31/12/2007, while Panel B shows the results for the second half, namely between 01/01/2008 and 31/12/2016. Industries which do not include sufficient observations for a regression analysis are excluded from the estimations. The table shows the industries with statistically significant β coefficients [see equation (3)]. The 3 significance levels are based

on the p-value of each coefficient. The regressions were estimated using OLS and the standard errors include a Newey-West serial

correlation and heteroskedasticity correction. Individual industry returns were built based on market weights of individual companies and the source is Datastream.

Panel A: Time frame: 2000-2007

90% 95% 99%

Australia

Banks Aerospace & Defense

Forestry & Paper Technology, Hardware & _Equipment Pharmaceuticals & Biotechnology Travel & Leisure

Canada Fixed Line Telecommunications Banks

Industrial Engineering

France _Chemicals Banks Equity Investment Instruments

Germany

Financial Services Fixed Line Telecommunications Pharmaceuticals & _{Biotechnology}

Nonlife Insurance Food Producers

Software & Computer Services

Support Services

Japan Support Services

Netherlands

Switzerland

Construction & Materials

Nonequity Investment Instruments

UK

Electricity Automobiles & Parts Alternative Energy

Industrial Engineering General Industrials Forestry & Paper

Life Insurance Healthcare Equipment & Services Mining

Media Travel & Leisure

Nonlife Insurance

Pharmaceuticals & Biotechnology Real Estate Investment Trusts

Support Services USA

Real Estate Investments &

Services Alternative Energy Beverages

Construction & Materials Financial Services Oil Equipment & Services Industrial Engineering

24

Panel B: Time frame: 2008-2016

90% 95% 99%

Australia

Electricity Equity Investment Instruments

Gas, Water & Multiutilities Media

General Retailers Real Estate Investments & Services

Canada

Banks Alternative Energy Technology, Hardware & _Equipment

Chemicals Industrial Transportation

Gas, Water & Multiutilities Software & Computer Services General Industrials

France

Fixed Line Telecommunications Food Producers

Pharmaceuticals &

Biotechnology

Germany

Forestry & Paper Aerospace & Defense

Healthcare Equipment &

Services Life Insurance

Technology, Hardware &

Equipment Mobile Telecommunications

Nonequity Investment Instruments

Japan Nonlife Insurance Equity Investment Instruments

Netherlands

Chemicals Alternative Energy

Household Goods & Home Construction Nonlife Insurance

Technology, Hardware & _Equipment

Switzerland Life Insurance

Personal Goods

UK

Electricity Fixed Line Telecommunications

Forestry & Paper

Mobile Telecommunications

USA

Aerospace & Defense Electronic & Electrical Equipment Equity Investment Instruments

Banks Forestry & Paper

Financial Services Media Nonlife Insurance Real Estate Investments &

Services

One potential reason for this might be, as previously mentioned, the size effect or the

popularity/notoriety effect. The more trades a market has, the more information can be

extracted from market data. The only “black swan” in this case is Japan, which shows very little

predictability. This was also shown by HTV, as industries present on the Japanese market had

the least predictive power from all analyzed countries.

25

contains a lot of noise which comes from trading based on sentiment or news, rather than

thorough valuation analyses. For that, caution is needed when interpreting the results.

Moreover, a significant coefficient means little to the overall explanatory power of the model.

Analyses on weekly returns for stocks will undoubtedly render small correlations and R-squared

values, as the “random walk” behavior of stock prices is increasingly emphasized when data

point frequency rises.

Nevertheless, considering that results are repeatedly consistent, and although vary

across countries, it cannot be denied that there is some predictive power in certain industries

for some countries. This is also backed up by the fact that individual regression coefficients

deviate very little from the full sample results, while their signs (positive/negative) almost never

change. This phenomenon also has a dose of intuition behind it, because of the inter-relation

between industries and also the market. A shock at a specific level cannot instantly be captured

by the entire economy. This is clearer if we think about the fact that shocks or certain events

affect the real market price of a stock, but this price is determined on the open market, through

basic supply and demand. Thus, individual investors will nevertheless have asymmetric

information between each other, which still is a fundamental strategy for some, although

frowned upon. There will always be one trader who finds and takes advantage of information

before another, and this can only mean that effects of shocks do not disperse simultaneously in

all directions, but rather the information is gradually diffused within each outskirt of the market.

For that, a causality waterfall exists, be it between industries and the overall market or other

variants, on a daily, weekly or monthly level.