Commodity price exposure, determinants and stock returns: An analysis of Dutch ﬁrms

(1)

Master’s Thesis

Commodity price exposure, determinants and stock

returns: An analysis of Dutch firms

HIELKE DE VRIES∗

April 25, 2012

Supervisor: Dr. B.A. Boonstra

∗

(2)

(3)

Commodity price exposure, determinants and stock

returns: An analysis of Dutch firms

HIELKE DE VRIES∗†

ABSTRACT

This thesis studies the sensitivity of Dutch stock returns to seven different commodity prices and price indices using GARCH(1,1) regressions. Furthermore, six different firm specific vari-ables are regressed against all significantly estimated commodity price Betas from the first-step GARCH(1,1) regression, to see which determinants can be considered important in explaining commodity price exposure. Employing a time-series dataset with weekly stock price data from January 1998 to December 2010 and using four different subperiods, I find between 6.49% -45.45% of Dutch non-financial listed firms exposed to commodity price changes. Additionally, using an OLS regression I find that firm size, firm leverage, firm liquidity and to a lesser extent, a firm’s growth opportunities are important determinants of commodity price exposure. Foreign involvement is not found to be an important indicator.

Keywords:Commodity price exposure, determinants, two-step regression, hedging, Dutch firms. JEL classification:C22, G17, G32, F30

∗

Email: h.de.vries.30@student.rug.nl, student number: 1589776, from University of Groningen.

†

I would like to thank Dr. B.A. Boonstra for giving me valuable feedback and advice on my thesis. Dr. V. Angelini for her great support and help with all kinds of Latex problems I faced while writing this thesis. Also, she helped me writing the STATA script which automated the Beta estimation process. Furthermore I would like to thank my family and friends, and Eszter M´odos for their support throughout the writing process.

(4)

Commodity price exposure, determinants and stock returns: An analysis of Dutch firms

TABLE OF CONTENTS

I INTRODUCTION 4

II LITERATURE REVIEW 8

A Commodity price exposure definition . . . 8

A.1 Commodity price exposure . . . 10

A.2 Other economic exposures . . . 11

A.3 Dutch firms . . . 16

B Determinants of commodity price exposure . . . 18

B.1 Firm size . . . 19 B.2 Firm leverage . . . 20 B.3 Liquidity position . . . 21 B.4 Growth opportunities . . . 21 B.5 Foreign involvement . . . 22 III DATA 23 A Sample for the estimation of commodity price exposure . . . 23

A.1 Dependent variable. . . 23

A.2 Independent variables . . . 24

A.3 Control variable . . . 25

B Sample determinants of commodity price exposure. . . 25

C Descriptives . . . 26

D Non-stationarity . . . 28

E Normality . . . 29

(5)

F Multicollinearity . . . 29

IV METHODOLOGY 30 A First step regression. . . 30

A.1 Heteroskedasticity, Autocorrelation and GARCH . . . 31

B Second step regression . . . 32

V RESULTS 34 A First-step regression: Estimation of commodity price exposure . . . 34

B Second-step regression: Estimation of the determinants of commodity price exposure 38 VI ROBUSTNESS ANALYSIS: MONTHLY DATA 41 A First-step regression: Monthly data. . . 41

B Second-step regression: Monthly data . . . 42

VII CONCLUSION 42 A Conclusion . . . 42

B Limitations and implications for further research . . . 44

(6)

I. INTRODUCTION

Firm exposure to many different economic variables is tested and researched extensively. Namely interest-rate exposure and exchange-rate exposure research was conducted during the late 70’s, 80’s and 90’s for U.S. firms (Dumas, 1978), (Kaufold and Smirlock, 1986), (Jorion, 1990). Later this field of study was broadened after the 90’s with studies on non-U.S. firms and on the determinants of exposures (Choi and Prasad, 1995), (He and Ng, 1998), (Chow and Chen, 1998), (de Jong et al., 2006). To this date, relatively few literature is devoted to commodity price exposure, which can be considered odd since commodity prices exhibit even higher volatility than exchange rates for some commodities1, especially in recent years. Only several studies, and for selected firms, exam-ine the impact of certain commodity price fluctuations to firm value; for the oil industry (Faff and Brailsford, 1999), (Haushalter et al., 2002); for the gold mining industry (Tufano, 1998), (Tufano, 1996); for German firms (Bartram, 2005) and for several commercial corporations (Bilson, 1994), (Oxelheim and Wihlborg, 1995).

It is relevant to examine a firm’s financial exposure to different economic variables as one can better anticipate the magnitude of the impact of changes in these economic variables to cash flows, and thus stock returns and firm value. A classic example is how a public firm that is reliant on the oil price will see its stock price fluctuate, as oil prices go up and down in relatively short periods of time. Exposures are useful information to financial specialists and in particular risk managers and treasurers, who constantly analyze different types of risk and make important business-changing decisions based on their risk assessment. Financial agents already use interest-rate, exchange-rate and many other exposure variables in risk modeling, and due to the increase in variability in com-modity prices in the last decade (see figure 1) , they are more compelled to take comcom-modity price risk into account as well. Also, economic exposure is a popular subject in the academic world. Comparing different types of financial exposures (interest rates, commodity prices, exchange rates) gives a better economic understanding of which variables influence firm value. When knowing firm value sensitivity to several commodity prices, relevant exposures can be better identified and con-sequently better hedging strategies can be build, which can result in higher firm value (Smith and Stulz, 1985).

1_{This holds in particular for agriculture commodities such as coffee, hogs, wheat and sugar (Bartram, 2005).}

(7)

In this study I examine the financial exposure of public Dutch non-financial firms to several commodity prices and commodity index prices. Furthermore I study the determinants of commodity price exposure. Interest-rate exposure and exchange-rate exposure have been researched to a large extent in the U.S, and also to a lesser extent in Europe, but to my knowledge few literature exists that examine commodity price exposure. One study on commodity price exposure for German listed firms by Bartram (2005) indicates that corporations exhibit net exposures to several commodities, but that few cash flows being affected by commodity price movements. A paper by Tufano (1998) on the stock price exposure of gold mining firms to the gold price finds that the average mining stock moves 2 percent for each 1 percent change in gold prices, and that gold firm exposures are negatively related to the firm’s hedging and diversification activities. Studies by Haushalter (2000) and Faff and Brailsford (1999) examine the oil price exposure for firms in different countries and find opposing results. Another study on the exchange-rate exposure of Dutch firms concludes that ”firms in open economies, such as The Netherlands, exhibit significant exchange-rate exposure” (de Jong et al., 2006).

Theoretically, open economies are more affected by internationally changing prices than closed economies. It is therefore interesting to see how firms in a small and open economy, such as The Netherlands, are exposed to commodity price changes. According to the CIA fact book, The Nether-lands is ranked 9thbiggest importer and 11th biggest exporter in the world in 20102. Total exports accounted for more than 78% of total GDP and total imports were almost 71% of GDP in 20103. The country has major industries in metal working, chemical plants and agriculture but lacks its own natural resources for metals, precious metals and some chemicals. It is thus very reliant on im-port and exim-port prices of commodities since these commodities, which are used as input and output factors in production processes, have to be bought and sold for the prevailing international spot and futures prices.

The theoretical literature on financial exposures Dumas (1978), (Adler and Dumas, 1980), (Adler and Dumas, 1984) gives a fairly straightforward view on the concept of exposure. Following this theoretical view, Hodder (1982) and Jorion (1990) interpret the exposure to exchange-rates as ”the regression coefficient of the real value of the firm on the exchange rate across states of nature”.

2

https://www.cia.gov/library/publications/the-world-factbook/geos/nl.html

3

Centraal Bureau voor de Statistiek, Statline database http://statline.cbs.nl/StatWeb/dome/default.aspx

(8)

Commodity price exposure can be interpreted as the same regression coefficient used for exchange-rate exposure (Tufano, 1996, 1998), and should thus be defined as the sensitivity of the firm’s value related to unexpected changes in commodity prices. More specifically, if a firm uses a certain com-modity as an input for the production of a good, a price increase/decrease of this comcom-modity should affect the cash flows of the firm, and thus firm value, in a negative/positive way, net of any hedging or pass-through activities by the firm. Vice versa, if there is a price increase/decrease of a commod-ity that is an output of a production process in a firm, the firm’s cash flow and thus firm value should be affected in a positive/negative way, net of any hedging or pass-through activities by the firm.

Besides reviewing the scarce commodity price exposure literature, I use techniques developed for exchange-rate exposure and interest-rate exposure to build the methodology for commodity price exposure. Furthermore, I deduce from the complete set of exposure literature to help build the theoretical background for this study.

Additionally, this study aims to provide new evidence on the determinants of commodity price exposure. Some exposure literature examines the determinants of exchange-rate exposure, e.g. for hedging and diversification activities (Tufano, 1998), for firm size, level of parent firm’s foreign activities and dividend payout policy (He and Ng, 1998), for firm size and amount of off-balance sheet hedging (de Jong et al., 2006), and for the percentage of foreign operations (Jorion, 1990). This study estimates determinants of commodity price exposure that control for hedging incentives by firms (firm size, leverage, firm liquidity, growth opportunities), and foreign involvement of the firm (foreign sales ratio). Until now most exposure literature focuses on U.S. companies, mainly examining exchange-rate exposures and to a lesser extent interest-rate exposure and inflation-rate exposure. This study adds to the financial and economical literature by examining a non-U.S. coun-try, using a fairly not-well researched economic variable, namely, commodity prices. In addition, this study explains and examines determinants that have not been used before in combination with commodity price exposure.

Especially now this research is of interest, since developing countries have seen huge economic and output growth the last 2 decades. According to the OECD4, 40% of the economic growth was accounted for by emerging economies in 1990. This number has risen to 50% in 2010, and

4

http://www.oecd.org/

(9)

is estimated to be 60% in 20205. Consumption and demand of (luxurious) products rise in these emerging economies, which puts pressure on commodity prices and this can be perceived in the market today. Jorion (1990) argues that exchange rates are a more important source of risk due to their higher volatility compared to other prices. However, in recent years commodity prices exhibit high volatilities, even higher than most currencies and interest rates. Figure 1 shows price levels for several important internationally traded commodities. For a country such as The Netherlands, which is very reliant on international commodity prices, it is important to know how sensitive firms will react to changes in commodity prices.

0 50 100 150 200 250 300 Ja n -98 Ju n -98 No v-98 Ap r-99 Se p -99 Fe b -00 Ju l-00 Dec -00 May -01 Oct -01 Ma r-02 Au g-02 Ja n -03 Ju n -03 N ov -03 Ap r-04 Se p -04 Fe b -05 Ju l-05 Dec -05 Ma y-06 Oct -06 Ma r-07 Au g-07 Ja n -08 Ju n -08 N ov -08 Ap r-09 Se p -09 Fe b -10 Ju l-10 Dec -10 May -11

Agricultural Energy index Food Metals Non-fuel index Petroleum

Figure 1

International commodity prices, 2005=100. Source: Thomson Datastream. The lines show the commodity prices relative to the year 2005.

The remainder of this paper consists of the following sections. Firstly, the theoretical framework is presented and the relevant exposure literature is reviewed in section II. Secondly, a description of the data used in the empirical analysis is given in section III. Thirdly, the methodology is described in section IV. Fourthly, the results are presented and interpreted in section V. Thereafter, the results are tested for robustness in section VI. Lastly, a conclusion is given and limitations and potential areas of further research are discussed in section VII.

5

International Monetary Fund.

(10)

II. LITERATURE REVIEW

This section consists of the following sub-sections. Firstly, the theoretical framework is presented which explains the relation between unexpected commodity price fluctuations and firm value. Sec-ondly, discussed is why it is relevant to examine commodity price exposure for Dutch firms. Lastly, the literature on commodity price exposure and its possible determinants is discussed6_.

A. Commodity price exposure definition

Commodity price exposure explains the effect of changes in commodity prices on firm value. Simi-lar to any other economic exposure like exchange-rate exposure, interest-rate exposure and inflation-rate exposure, unexpected commodity price fluctuations affect firm value because it has an imme-diate relationship with the operative business of a firm (Bartram, 2005). The underlying assets of commodity price exposure are real goods (e.g. metals, gold, wheat, oats, gasoline, natural gas) that can be input or output factors in the production process of a firm. Industries are differently affected by commodity price fluctuations as some firms use commodities as an input in their production pro-cess, where other firms produce a commodity as an output. Assumed is that for empirical analysis, commodity price exposure can be measured at best as a result of major input/output relationships (Adler and Dumas, 1984) (Bartram, 2005) (e.g. gasoline price exposures in the chemicals industry, wheat price exposures in the agriculture industry). Table 1 presents the mayor input/output relation-ships for four important internationally traded commodities. To exemplify; when the international metal price rises, an arbitrary cutlery producer should see the import costs of his most used commod-ity (metal) appreciate, which results in higher production costs and consequently, lowered operating profits. The financial markets will value the firm, ceteris paribus, lower due to the decrease cash flows.

6

Due to the fact that the amount of commodity price exposure literature is not extensive, all the necessary exposure literature is consulted.

(11)

Commodity price exposur e , determinants and stoc k returns: An analysis of Dutc h firms Table I

Important input/output relationships between commodities and industries

This table shows four important internationally traded commodities and for wich industries these are important inputs and outputs.

Commodity Industry Output Industry Input

agricultural agriculture/forestry agriculture/forestry (oats, barley, corn, wheat); food/tobacco

rubber/plastics (rubber) (barley, corn, wheat, oils, grease, livestock, meat); apparel/

paper/wood products textile products (cotton, jute, sisal, suet ); leather/leather

publishing/printing (paper) products (skins, suet); paper/wood products/publishing/ printing

apparel/textile products (cotton, jute, sisal) (paper pulp and paper); industrial machinery and equipment

(rubber); chemicals (linseed oil, rizinus oil, suet)

petroleum mining (crude oil, natural oil) power industry (natural oil, heating oil); chemicals (crude oil,

chemicals (gasoline, heating oil, diesel) gasoline, naphtha, propane, natural oil); rubber/plastics (crude

rubber/plastics (rubber) oil, rubber); transportation (gasoline, diesel, kerosene); primary

metal industries (natural oil); ceramic, glass (natural oil)

metals mining primary metal industries primary metal industries; fabricated metal product (aluminum, plate,

copper, zinc); industrial machinery and equipment (aluminum, copper, nickel, steel scrap, titanium); transportation equipment (aluminum, copper, nickel, titanium); electrical/electronic equipment (lead,copper, mercury, silicon, tungsten)

precious metals mining primary metal industries jewelry industry; primary metal industries electrical/electronic

equipment transportation equipment (palladium)

other mining electrical/electronic equipment (selenium, silicon)

Source: Bartram (2005).

(12)

Additional indirect effects may apply for commodity price exposure when commodities are exchanged and traded in the value chain within a company. When commodity price fluctuations affect the cash flows of a firm only to a small degree, these price changes can be passed-through to other parts in the value chain, or can be passed on to other buying parties. In case a firm uses price pass-through or hedges commodity prices with futures or forwards, only weak empirical relations might be observed between unexpected commodity price changes and firm value. To illustrate, the cutlery producer may predict volatile metal prices and therefore he/she could buy metal futures or forwards to lock in the metal price. The firm is then assured to pay the locked in metal price and the financial markets will incorporate the metal futures or forward strike price when discounting the cash flows and assessing the value of the firm.

A.1. Commodity price exposure

The exposure literature is vast and comprises different fields of research. Some research is devoted to interest rate exposure and inflation rate exposure (Madura and Zarruk, 1995), (Bartram, 2002), (Faulkender, 2005), but more literature is devoted to exchange rate exposure for U.S, European and Asian firms (Hodder, 1982), (Jorion, 1990), (Bartov and Bodnar, 1994), (Allayannis and Ofek, 1997), (He and Ng, 1998), (Bartram, 2002) and one specifically for Dutch firms (de Jong et al., 2006). Not much literature examines empirically the effect of changing commodity prices on firm value; however several studies exist with mixed results. Tufano (1998) studies the financial exposure of North American gold mining firms to changes in the price of gold, and finds that the average mining stock price changes 2 percent for each 1 percent change in the gold price, but adds that exposures vary considerably over time and across firms. In total, Tufano (1998) finds that in the estimation of gold price exposures of 48 gold mining companies, more than half of the firm-quarter exposures to be statistically significant at the 5% level. In another study by Strong (1991), the oil price exposure of 25 U.S. companies is examined. 52% of the oil companies are found to have a significant exposure to oil prices at the 5% level. Other studies on commodity price exposure include a study on the oil price exposure of American Airlines (AA) (Bilson, 1994), and a study on the oil price exposure of Volvo cars (Oxelheim and Wihlborg, 1995). The former shows that AA is significantly exposed to unexpected oil price fluctuations at the 5% level, the latter that Volvo cars

(13)

is not significantly affected by price changes in oil or non-energy commodities.

In a more comprehensive analysis by Bartram (2005), 490 non-financial actively traded German firms are examined from the period 1987-1995. The results indicate that overall, the percentage of firms that exhibit significant commodity price exposure is above the 5% significance level. Due to the fact that commodity prices exhibit higher volatilities than exchange rates, a higher percentage of firms was expected. The fraction of significantly exposed firms is comparable to exchange rate exposure studies (He and Ng, 1998), (Bartram, 2004) and to interest rate exposure studies, (Bartram, 2002), (Allayannis and Ihrig, 2001), (Madura and Zarruk, 1995).

A.2. Other economic exposures

Besides interest-rate- inflation-rate- and commodity price exposure, exchange rate exposure is the most well researched area of economic exposures to firm values. Using a sample of 287 U.S. multi-national firms, Jorion (1990) identifies significant cross-sectional differences between the value of firms and the exchange rate, but only for a small fraction of the sample. Jorion (1991) uses the same methodology to examine the exchange rate exposure of a set of portfolios consisting of NYSE firms and also finds significant exposure effects. Other U.S. evidence includes a study by Amihud (1994), who uses two portfolios of 32 exporters and finds 33.3% and 50% of the firms to be significantly ex-posed at the 5% level. Choi and Prasad (1995) examine 409 multinational firms over the 1978-1989 period and find that 14.9% of the sample is significantly exposed at the 10% level. In Walsh (1994), 10.9%, 5.6% and 4.8% of a sample of 391 non-banks is found to be significantly exposed to three trade-weighted exchange rate indices. Many other U.S. exchange-rate exposure studies observe sig-nificant results, but most fail to find a high percentage of the firms to be sigsig-nificantly exposed.

More recently the focus has shifted to other countries than the U.S, mainly because the availabil-ity of data has improved in Europe and Asia. The first few non-U.S. studies were, not surprisingly, from bigger industrialized countries like Canada and Japan. Bodnar and Gentry (1993) 39 portfo-lios of NYSE and AMEX firms but also 19 portfoportfo-lios of TMX firms from Canada and TSE firms from Japan. They find significant exposures in some industries for some countries, but no strong evidence.

Using two samples of 141 and 23 Australian firms, Loudon (1993a,b) studies the exchange rate

(14)

exposure effect for a trade weighted index of currencies and finds generally weak exposure effects. Only 6% of the firms show a significant exposure effect, and the observed exposure effects are gen-erally weak. In another comprehensive study by Chow and Chen (1998) using a sample of 1110 Japanese firms, examining 5 different trade weighted exchange rate indices and consisting of 5 dif-ferent return horizons, concludes that ’Japanese firms are overwhelmingly negatively exposed, i.e. their equity returns decrease as yen depreciates’. Similar results are found by He and Ng (1998) on a sample of 171 Japanese multinational firms finding significant exposure effects for 25% of the firms in the period 1979-1993. Many other studies exist on financial exposure, Table II presents a list of the most important exposure studies with large samples. All listed studies use a similar methodology; ranging from industry portfolios, multinational portfolios, NYSE and AMEX portfo-lios using indices or commodities, with different samples ranging from mining firms to non-banks and non-financial firms, researching different countries and regions.

(15)

Commodity price exposur e , determinants and stoc k returns: An analysis of Dutc h firms Table II

Overview of the most important exposure studies

This table presents a list of the most important exposure studies. The first column shows the author(s), the second which peroid is used for the dataset, the third how many, and which type of firms are used in the dataset, the fourth column shows the results of the study, where TW is a Trade Weighted index, and in between brackets the sign of the exposure, the significance level and the amount (percentage of the sample) of firms exposed is given. The last column shows the interval of the time-series data and which control variables are used, where (M) is a Market index, (I) the Interest rate, (D) the Dividend yield, (L) the Lagged return, (S) the Spread between the short-term and the long-term yield, and (APT) APT factors.

Study Period Sample Results Characteristics

Booth and Rotenberg 1979-1983 156 firms (CAN) % USD (−/5/67.3) Monthly data

(1990) No control variables

Jorion (1990) 1971-1987 287 multinational firms (USA) % TW15 (±/5/5.2) Monthly data

40 multinational firms (USA) % TW15 (±/5/7.5) Control variables (M)

40 portfolios [287 multinational firms] % TW15 (±/5/15.0) Monthly data

(USA) Control variables (M)

Portfolios

Jorion (1991) 1971-1987 20 portfolios [NYSE firms] (USA) % TW15 (±/5/20.0) Monthly data

Control variables (M) Portfolios

% TW15 (±/5/35.0) Monthly data

Control variables (APT) Portfolios

Strong (1991) 1975-1987 25 oil firms (USA) % oil price (±/5/52) Monthly data

1982-1987 238 oil related firms (USA) % oil price (±/5/50) Control variables (M)

Bodnar and Gentry 1979-1988 39 portfolios [NYSE, AMEX firms] % TW6 (±/5/23.1) Monthly data

(1993) (USA) Control variables (M)

19 portfolios [TSE firms] (CAN) % TW6 (±/5/21.1) Portfolios

1983-1988 20 portfolios [Nikkei 500 firms] (JPN) % TW6 (±/5/25.0)

Loudon (1993a) 1984-1989 141 firms (AUS) % TW (±/5/10.6) Monthly data

(16)

Commodity price exposur e , determinants and stoc k returns: An analysis of Dutc h firms Table II–continued

Control variables (M)

Loudon (1993b) 1980-1991 23 indices (AUS) % TW (−/5/30.4) Monthly data

Control variables (M) Portfolios

Amihud (1994) 1979-1988 3 portfolios [32 exporters] (USA) % TW15 [1] (+/5/33.3) Monthly data

Control variables (M, L) Portfolios

4 portfolios [32 exporters] (USA) % TW15 [2] (+/5/50.0) Quarterly data

Control variables (M, L) Portfolios

Bartov and Bodnar 1978-1990 208 firms (USA) % TW6 [1] (+/1/100) Quarterly data

(1994) Control variables (M) Pooling

Abnormal stock returns

Walsh (1994) 1982-1993 391 nonbanks (USA) % TW (±/5/10.9), % TW [1] (±/5/5.6), Monthly data

% TW [2] (±/5/4.8) Control variables (M)

Oxelheim and Wihlborg 1990-1992 1 car manufacturer (Volvo) % oil price (±/5/30) Monthly data

(1995) % non-energy commodities (±/5/20) Control variables (M)

Choi and Prasad (1995) 1978-1989 409 multinational firms (USA) % TW10 (±/10/14.9) Monthly data

% TW10 (±/10/14.9) Monthly data

Control variables (M, I)

20 portfolios [409 multinational firms % TW10 (+/10/10.0) Monthly data

(USA) Control variables (M)

Portfolios

% TW10 (+/10/10.0) Monthly data

Control variables (M, I) Portfolios

Tufano(1998) 1990-1994 48 gold mining firms (USA) % gold price (−/5/56) Weekly, daily and monthly

(17)

Commodity price exposur e , determinants and stoc k returns: An analysis of Dutc h firms Table II–continued

data

Chow and Chen (1998) 1975-1992 1110 firms (JPY) % TW14 (±/5/30.1) 1-month return Monthly data

% TW14 (±/5/23.4) 3-months return Control variables (D, S)

% TW14 (±/5/35.0) 6-months return Overlapping periods

% TW14 (±/5/47.4) 12-months return % TW14 (±/5/69.8) 24-months return

He and Ng (1998) 1978-1993 171 multinational firms (JPN) % TW9 (±/5/26.3) Monthly data

Haushalter(2002) 1992-1994 86 oil firms (INT) % oil future price (±/5/61.5) Monthly data

% oil future price (±/5/28.2) Control variables (M)

Bartram (2004) 1991-1995 373 nonfinancial corporations (DEU) % TW18(±/5/7.5) Monthly data

% USD (±/5/7.8) Control variables (M)

% TW18(±/5/7.8), % TW18(5/11.5) % USD (±/5/8.3), % USD (5/14.5)

Barttram(2005) 1987-1995 490 non-financial firms (GER) % crude oil (±/5/7.3) Monthly data

% copper (±/5/10.1) Control variables (M)

% wheat (±/5/12.5)

% agriculture index (±/5/15.6) % livestock index (±/5/15.1) % industrial metals index (±/5/13.4 % precious metals index (±/5/4.5) % energy index (±/5/6.1)

(18)

A.3. Dutch firms

With total exports accounting for more than 78% of GDP and total imports making up almost 71%7 of GDP in 2010, The Netherlands can be considered as a very open economy. In comparison, U.S. exports and imports are respectively 12.5% and 16% of GDP in 201089. The Netherlands has an important role as European transportation hub and its economy is also noted by stable industrial re-lations. Most industrial activity is in chemical, food processing, electrical machinery and petroleum refining. The total merchandise trade was 57% of GDP for exports and 63% of GDP for imports in 201010The highly mechanized agriculture sector provides large surpluses for the food-processing

0% 5% 10% 15% 20% 25% 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008

Food, beverages and tobacco

Raw materials, oils and fats mineral fuels chemicals finished products machinery Figure 2

Dutch commodity exports as a percentage of GDP. Note: due to the unavailability of more specific primary commodity data like raw metals, wheat and oil, the commodities shown in the graph are given.

industry and for exports11. The industry sector contributed 24.4% to the GDP12in 2010. Figure 2 and 3 show the amount of imported and exported commodities as a percentage of GDP, and thus illustrate the importance of commodity imports and exports for Dutch firms.

The Netherlands is well endowed with commodities including natural gases, petroleum and 7

Centraal Bureau voor de Statistiek, Statline database. http://statline.cbs.nl/StatWeb/dome/default. aspx

8_{www.census.gov/foreign-trade/}

9

www.bea.gov/newsreleases/national/gdp/2010/pdf/gdp2q11_3rd.pdf

10

Centraal Bureau voor de Statistiek, Statline database http://statline.cbs.nl/StatWeb/dome/default.aspx

11

Economist Intelligence Unit Country report The Netherlands 2010.

12_{Economist Intelligence Unit Country report The Netherlands 2010.}

(19)

agriculture. On the other hand, it lacks other resources such as metals and precious metals. This economic dependence on commodity prices in both importing and exporting can have a negative impact on the welfare and economic growth of a country (Combes and Guillaumont, 2002). Addi-tionally commodity prices can have a direct effect on the terms of trade or GDP growth. de Jong

0% 2% 4% 6% 8% 10% 12% 14% 16% 18% 20% 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008

Food, beverages and tobacco

Raw materials, oils and fats mineral fuels chemicals finished products machinery Figure 3

Dutch commodity imports as a percentage of GDP. Note: due to the unavailability of more specific primary commodity data like raw metals, wheat and oil, the commodities shown in the graph are given

et al. (2006) on exchange rate exposures for Dutch firms, point out that one of the reasons why much previous literature fails to find significant exposures, is that most studies focus on the U.S, which is one of the least open economies in the world. de Jong et al. (2006) find more than 50% of Dutch firms to be significantly exposed to exchange rate risk.

Conclusively, many U.S. studies find significant exposures but fail to find large percentages of samples exposed to exchange rate changes. This is not true for more open economies, like Japan (Chow and Chen, 1998) and The Netherlands (de Jong et al., 2006) where strong relationships are found for large fractions of the sample. Hence, based on the existing exposure literature (and eco-nomic analysis of The Netherlands), the following main hypothesis is formed:

H1₀:Dutch firms are not significantly exposed to commodity price changes

(20)

B. Determinants of commodity price exposure

This study further tries to predict whether several firm specific variables are important in deter-mining commodity price exposure for Dutch firms. Often employed determinants in exchange rate literature are foreign involvement of a firm and proxies for hedging incentives. Firms hedge expo-sures to many different economic variables, namely; interest rates, exchange rates and commodity prices. Firms use derivatives such as options, futures or forwards to lock in a price or price range that the firm has to pay for the underlying asset of the derivative, without having to worry about the actual future price13. The existing exchange-rate literature acknowledges variables that proxy for hedging incentives as important financial determinants of exposures. Bartov and Bodnar (1994) argue that systematic errors are being made when explaining exchange-rate exposures partly due to not being aware of the firm’s hedging activities.

Perfect capital markets are characterized by; (1) perfect and frictionless market conditions, where all information is available to everyone and at cost, and no investor is able to influence the market herself; (2) no taxes on capital gains and dividends; (3) firms that do not change their investment policy; (4) economic agents that can borrow and lend at the same rate. According to Modigliani and Miller (1958) when these assumptions hold, it is irrelevant how a company is fi-nanced. Modigliani and Miller’s results thus imply that hedging is irrelevant, since when a firm chooses to change its hedging policy, investors who hold claims issued by the firm can change their holdings of risky assets in order to offset any change in the firm’s hedging policy, leaving the distribution of their future income unaffected (Smith and Stulz, 1985). However in reality, market imperfections do exist14, and thus firms have incentives to employ derivative instruments to hedge against risk. Hence, in theory the use of derivatives for hedging reasons could diminish a firm’s exposure.

Because no data exists on hedging activities by Dutch firms15, several studies employ surveys to determine if specific firms use hedging instruments. Since collecting survey data on hedging

13_{This depends on the instrument as there are exotic derivatives whose strike price depends on the actual price at}

maturity of the derivative.

14

Smith and Stulz (1985) developed a positive theory for a firm’s hedging behavior. They show that it can be advan-tageous to take positions in futures, forwards or option markets if; (1) the firm faces a convex tax curve, (2) the firm experiences high costs of financial distress, (3) the firm suffers from managerial risk aversion.

15_{SFAS 105 requires U.S. firms to publish data on derivative trading activity. Allayannis and Ofek (1997) use this data}

to calculate the notional value of the contracts.

(21)

activities for all listed Dutch non-financial firms is beyond the scope of this thesis, I employ indirect measures of hedging activities similar to He and Ng (1998), Chow and Chen (1998), Tufano (1998) and Choi and Kim (2003). These measures are explained in the next subsections.

B.1. Firm size

Chow and Chen (1998), He and Ng (1998) and Choi and Kim (2003) employ in their studies on exchange-rate exposure several proxies for hedging incentives by firms; one being firm size. Chow and Chen (1998) base their theory about hedging costs and whether a firm should engage in hedg-ing activities partly on a paper by Nance et al. (1993). In their paper Nance et al. (1993) contend that because managing risks and building hedging strategies can be costly, firms use hedging instru-ments if benefits are greater than costs. Chow and Chen (1998) presume two factors to determine jointly the relationship between firm size and the incentive to hedge; (1) economies of scale and (2) bankruptcy costs. The first factor relates back to Nance et al. (1993); since larger firms have better access to risk expertise and have economies of scale in hedging costs (Block and Gallagher, 1986), (Booth et al., 1984), larger firms are more likely to hedge than smaller firms. The other view which is also put forward by Warner (1977), is that smaller firms are more likely to hedge than bigger firms as they face higher bankruptcy costs.

The empirical literature does not give one straightforward answer to the question whether smaller or bigger firms are more exposed. He and Ng (1998) find that exposure increases in firm size, and Chow and Chen (1998) find smaller firms to have smaller exposures when measuring one month horizons, while for longer-return horizons larger firms have larger exposures. No previous empirical research was conducted on relation between firm size as a hedging incentive and unex-pected commodity price changes. However commodity price risk can be hedged in the same way as exchange rate risk, and are at least as volatile, and in recent year even more volatile (see figure 1), than exchange rates. Hence, the same theoretical arguments hold for empirical testing. Following the theory of economies of scale in hedging costs, the hypothesis is formed:

H2:The size of a Dutch non-financial firm is negatively related to its commodity price exposure

(22)

B.2. Firm leverage

Mayers and Smith (1982) and Smith and Stulz (1985) argue that by using hedging instruments to reduce the variance of the cash flows of the firm, the probability that the firm encounters financial distress is reduced. The probability of the firm encountering financial distress is directly related to a firm’s leverage as the costs of issuing debt can be reduced (Chow and Chen, 1998). Ceteris paribus, firms that have higher leverage face larger expected costs of financial distress and are more likely to use hedging instruments, hence are less exposed to commodity price risk. Myers (1977) and Froot et al. (1993) explain the relation of leverage and hedging incentives from the agency costs point of view. They state that the underinvestment problem is more pronounced with more debt in a firm’s capital structure due to the conflict of interest between bondholders and equity holders. Because of the high leverage of the firm, the propensity of excessive risk taking beviour by equity holders in-creases. Hedging can restrict the states in which the firm would default on debt so that bondholders are less hurt by equity holders’ excessive risk taking. Hence, firms with higher leverage are more likely to hedge.

The empirical literature to be divided on the subject of leverage and hedging incentives, but to my knowledge more researchers find a positive relationship between firm leverage and hedging practices by firms than a negative relationship, or no relationship at all. Amongst others, Gay and Nam (1998) and Haushalter (2000) study whether more leveraged firms hedge more and find posi-tive significant results for a relation between leverage and hedging where Nance et al. (1993) fail to find a relation. Block and Gallagher (1986) find a positive but statistically insignificant relationship between the debt-to-equity ratio and hedging. The empirical literature on the relation between lever-age as a proxy for hedging incentives and financial exposure is less ambiguous. Chow and Chen (1998), He and Ng (1998) and Choi and Kim (2003) find leverage significantly negatively related to exposure, using different proxies. Based on the theories of expected bankruptcy costs and the underinvestment problem, the following hypothesis is formed:

H₃:The leverage of a Dutch non-financial firm is negatively related to its commodity price expo-sure

(23)

B.3. Liquidity position

Another way to reduce the probability of default and thereby the expected costs of financial distress, is by reserving more liquid assets and restricting dividend payouts (Nance et al., 1993). Without hedging, firms could be forced to underinvestment in some states of the world because it is costly or impossible to raise external finance (Froot et al., 1993). Both Nance et al. (1993) and Froot et al. (1993) contend that liquidity is negatively related to hedging activities.

Amongst empirical studies, Chow and Chen (1998) use the current ratio and dividend payout ra-tio as proxies for the liquidity posira-tion of a firm. They find in their study, examining Japanese firms, that ”firms with high current ratios or low cash dividend payouts tend to have low exposures”, which is in contrast with hedging theories. He and Ng (1998) find results in favour of hedging theories and conclude that firms that have a low dividend payout ratio, or a high quick ratio, have less of an incentive to hedge and thus are more exposed. I deduce from the theoretical arguments given by Nance et al. (1993) and Froot et al. (1993) and the empirical results by He and Ng (1998) the following hypothesis:

H4:The liquidity position of a Dutch non-financial firm is positively related to its commodity price exposure

B.4. Growth opportunities

The underinvestment problem is also more marked in firms that have high growth opportunities. Equity holders engage more in risk taking activities when they see more potential growth oppor-tunities. The underinvestment hypothesis (Mayers and Smith, 1987) suggests a negative relation between growth opportunities and costly external financing. Because hedging reduces the under-investment problem, the higher the growth opportunities, the greater a firm’s incentive to employ hedging instruments to reduce underinvestment costs.

In empirical research, Geczy et al. (1997) find that firms with more growth opportunities, prox-ied by book-to-market ratio, are more likely to use currency derivatives. Similar results are demon-strated by Gay and Nam (1998), who emphasize that a firm’s derivative use is partly driven by

(24)

reducing underinvestment costs. He and Ng (1998) and Choi and Kim (2003) estimate the rela-tion between a firm’s growth opportunities and exchange-rate exposures and find negative results. However, the result by He and Ng (1998) are inconsistent over time periods. They find significant results in some periods, while in other periods they find no significant results. Following Geczy et al. (1997) and based on the underinvestment theory, the hypothesis is stated as follows:

H5:The growth opportunities of a Dutch non-financial firm are negatively related to its commod-ity price exposure

B.5. Foreign involvement

In general it can be hypothesized that the more a firm is involved in doing business abroad, the more foreign currencies it should hold, and thus should be more exposed to exchange-rates. U.S. studies find that a firm’s exchange-rate exposure is significantly related to its foreign involvement as prox-ied by export ratio or foreign sales ratio Jorion (1990) Choi and Prasad (1995). Allayannis (1995) shows that the level of exports and imports of U.S. manufacturing firms is an important determinant of exchange-rate exposure. Jorion (1990) finds that the depreciation of a dollar is positively related to the firms foreign sales ratio. Another study by He and Ng (1998) on Japanese multinationals concludes that the export ratio is a significant determinant of exchange-rate exposure. Yet, there is no such obvious link between commodity price exposure and foreign involvement as exchange-rate exposure and foreign involvement. Commodity price fluctuations might influence firm value through input/output relations but it should not matter if these firms do business domestically or abroad. However, the very reason that firms are exposed to commodity prices could mean that they are involved in trading these commodities with foreign parties. Hence, I investigate the relation between commodity price exposure and foreign involvement hypothesizing a positive relationship:

H6:The foreign involvement of a Dutch non-financial firm is positively related to its commodity price exposure

(25)

III. DATA

In this section firstly a description of the data sample is given. Secondly, the data series are tested on non-stationarity, normality and multicollinearity. Thirdly, descriptive statistics are presented for the variables used in the regression.

A. Sample for the estimation of commodity price exposure

A.1. Dependent variable

This study aims to explain how Dutch firms’ value changes in reaction to commodity price changes, and which determinants can be considered important in explaining this reaction. Stock price data is collected for Dutch listed firms for the period January 1998 to December 2010 from Datastream International, chosen due to completeness and accuracy. Another reason to use this period is the fact that it includes three important periods for commodity prices as can be seen in Figure A.1. The period January 1998 to December 2001 shows fairly stable commodity prices and not much volatility. The second period, January 2002 to August 2008, is characterized by a sharp increase in commodity prices which has not been seen in the last 25 years before this period16. The last pe-riod (September 2008 to December 2010) shows the aftermath of the 2008 financial crisis17, where commodity prices dropped substantially, recovered in 2009 and quickly rose to even higher levels than pre-financial crisis. The three periods will be used for empirical analysis separately and as one whole period.

The initial sample consisted out of 123 Dutch firms listed on AEX, AMX and AScX. How-ever, after evaluation of the data 78 firms were kept for the final sample. All financially affiliated firms18(21) are left out of the sample as there is no clear link with internationally traded commodi-ties19. Another eight firms were listed twice as holdings, these were removed from the sample as

16_{Domanski and Heath (2007) wrote a paper on the growing appeal of commodities in the period 2002-2006 and}

conclude that due to the increased investor activity the commodity markets have become more like financial markets in some respects.

17

Also commonly referred to as ’the credit crunch’ or ’the credit crisis’.

18

This includes banks, financial service firms and insurance firms.

19_{Banks and other financially affiliated firms provide services instead of producing products, where commodities are}

needed as input factors.

(26)

well. One firm showed incorrect stock prices and 10 firms did not have stock prices for the whole period from January 1998 to December 2010, again these firms were excluded from the sample. The last four firms that were removed are three firms that did not have financial statement data and two firms which were virtually bankrupt.

The final sample consists of weekly stock return20data for 78 Dutch listed non-financial firms. Many studies discuss the use of different return horizons. Some argue the use of bi-weekly or monthly data because daily data suffers from problems with non-synchronous trading (de Jong et al., 2006). Tufano (1998) and Scholes and Williams (1977) show that using daily data to calcu-late exposures causes biased results. I choose weekly data to maintain power, by means of avoiding loss of information due to large time-intervals in between data points, and because of the relatively small sub-periods with the smallest being two years.

A.2. Independent variables

The independent variables consist out of the most traded commodities according to the Dutch bu-reau of statistics, CBS. As shown in Figure 1 and Figure 2, machinery, mineral fuels, chemicals, raw materials and food and beverages are the most traded goods in The Netherlands and thus represent the importance and dependency on these products for the Dutch economy. Weekly price data was collected for seven closely related internationally traded commodities and commodity indices, used for determining commodity price exposure. These are; aluminum, which is the most extensively used metal in the world and important in machinery production, and crude oil, important in ma-chinery, raw materials and chemicals production. Furthermore, five different S&P Goldman Sachs Commodity Indices (GSCI)21are applied. These include an industrial metals index, agriculture in-dex, energy inin-dex, non-energy index and precious metal index. The components of the indices are weighted by the world production of these commodities.

20

Returns are calculated using Rit= ln_PPit

it−1

21_{The Standard and Poor’s Goldman Sachs Commodity Prices Index is calculated on a production-weighted basis and}

is comprised of the commodities that are the subject of active, liquid futures markets.

(27)

Table III

Overview of all variables used for the estimation of commodity price exposure

This table shows a qualitative description of the dependent, control variable and all independent variables used to estimate commodity price exposure.

Variable Type Composition Notation Measure

Firm stock price Dependent Shares of the firm R Weekly return

Aluminum Independent Metric tons of aluminium RC(ALU) Weekly return

Oil Independent Barrels of Dated Brent oil RC(OIL) Weekly return

S&P GSCI Industrial Metals Independent Aluminum, copper, lead, zinc,

nickel

RC(MET) Weekly return

S&P GSCI Agriculture Independent Wheat, corn, soybeans, cotton,

sugar, coffee, cocoa

RC(AGR) Weekly return

S&P GSCI Energy Independent Crude oil, brent crude, unleaded

gasoline, heating oil, gas oil, nat-ural gas

RC(ENRG) Weekly return

S&P GSCI Non-Energy Independent All commodities excluding

GSCI energy index

RC(NENRG) Weekly return

S&P GSCI Precious Metal Independent Gold, silver RC(PREMET) Weekly return

AEX Control All stocks on AEX index Rm(AEX) Weekly return

A.3. Control variable

To correct for market movements and to reduce noise, AEX weekly market price data was also col-lected. According to Jorion (1990) the influence of market reactions should be included in the ex-posure model to overcome exaggerated measurements in foreign currency movements. Essentially, the market index is a way to ’soak up’ the remaining correlation between other macroeconomic vari-ables and the error term. Table III presents an overview of all varivari-ables used to empirically estimate commodity price exposure.

B. Sample determinants of commodity price exposure

The second stage of this study consists of estimating the possible determinants of commodity price exposure as described in Section II. For this estimation, a second dataset with yearly data for the period 1998 to 2010 for six firm specific variables was downloaded from Datastream international. Bloomberg data was consulted for firm specific foreign sales data due to the unavailability thereof

(28)

Table IV

Proxies for hedging incentives

This table shows the different determinants for heding incentives by firms and which firm specific variables are used as proxies.

Determinant Proxy Reference Notation Measure

Firm size Logarithm of sales Warner (1977), He and Ng

(1998)

SIZE Yearly average

Leverage Debt-to-equity ratio∗ He and Ng (1998), Chow and

Chen (1998),Tufano (1998)

DE Yearly average

Liquidity position Quick ratio, Dividend payout

ratio

Nance et al. (1993), Froot et al. (1993)

QR,DIV Yearly average

Growth opportunities Book-to-marktet ratio Geczy et al. (1997), He and

Ng (1998)

BM Yearly average

Foreign involvement Foreign sales ratio† Chow and Chen (1998),

Jo-rion (1990), de Jong et al. (2006)

FOR Yearly average

∗ Debt-to-equity ratio is calculated usinglong−term debt_{total equity}

† Foreign sales ratio is calculated usingnon−domestic sales

domestic sales

by Datastream. The variables chosen are proxies for hedging incentives by firms, as described in section II. Firms size is proxied by the logarithm of sales (Warner, 1977), firm leverage is proxied by debt-to-equity ratio (He and Ng, 1998) and (Chow and Chen, 1998), the liquidity position of the firm is proxied by two different measures, quick ratio and dividend payout ratio (Nance et al., 1993) and (Froot et al., 1993), the growth opportunities of the firm is proxied by the book-to-market ratio (Geczy et al., 1997) and foreign involvement by the firm is proxied by the foreign sales ratio (Jorion, 1990),(Chow and Chen, 1998) and (de Jong et al., 2006). Table IV presents the list of firm specific variables used.

C. Descriptives

Table V presents the descriptive statistics for the sample used for estimating commodity price ex-posure. All variables listed consist of 679 calculated weekly returns from financial price data, from the period January 1998 to December 2010. Table VI shows descriptive statistics for the sample used to estimate the determinants of commodity price exposure.

(29)

Table V

Descriptive statistics of the first sample used to estimate commodity price exposure

This table shows the descriptive statistics for the independent and control variables used for estimating commodity price exposure for the period January 1998 - December 2010 using weekly return data. AEX is the AEX market index, AGR is the S&P GSCI Agriculture index, ALU is Aluminum, ENRG is the S&P GSCI Energy index, MET is the S&P GSCI Industrial metals index, NENRG is the S&P GSCI Non-energy index, OIL is Dated Brent oil, and PREMET is the S&P GSCI Precious metal index.

Variable N Min Median Max Mean Std. Dev.

AEX 679 -0.1598 0.0038 0.1303 -0.0002 0.0322 AGR 679 -0.1761 0.0010 0.0998 0.0011 0.0287 ALU 679 -0.1299 0.0017 0.1064 0.0006 0.0295 ENRG 679 -0.1975 0.0052 0.1725 0.0022 0.0458 MET 679 -0.1349 0.0023 0.1073 0.0016 0.0310 NENRG 679 -0.1248 0.0023 0.0681 0.0012 0.0206 OIL 679 -0.2367 0.0071 0.1938 0.0025 0.0550 PREMET 679 -0.1094 0.0028 0.1850 0.0023 0.0267 Table VI

Descriptive statistics of the second sample used for estimating the determinants of commodity price exposure

This table presents descriptive statistics for all firm specific variables for estimating the determinants of commodity price exposure for the period January 1998 - December 2010 using yearly data. DE is the debt-to-equity ratio, DIV is the dividend payout ratio, QR is the quick ratio, SIZE is the firm size and FOR is the foreign sales ratio.

Variable N Min Median Max Mean Std. Dev.

(30)

The minimum of DE, DIV and FOR is 0, which can be explained by the fact that firm(s) in the sample exist that only have equity holders or do not pay out dividend or have no foreign sales. The maximum of 1 for FOR indicates there are firm(s) in the sample that exclusively do business abroad.

D. Non-stationarity

All independent variables and the control variable are tested for non-stationarity by applying a Dickey-Fuller Unit Root test (Dickey and Fuller, 1979).22The Dickey-Fuller test checks if there is evidence for a unit root. If the test fails to reject the null-hypothesis of a unit root in the dataset, it is not appropriate to examine the coefficient standard errors and t-values in the regression. Table VII shows the results of the Dickey-Fuller Unit Root test for all independent variables. All test statistics are smaller in absolute terms than their critical values, hence I find no evidence of non-stationarity in the datasets of the tested variables.

Table VII

Dickey-Fuller test statistics

This table shows the Dickey-Fuller Unit Root test statistics for the independent and control variables used for estimating commodity price exposure for the period January 1998 - December 2010 using weekly return data. AEX is the AEX market index, AGR is the S&P GSCI Agriculture index, ALU is Aluminum, ENRG is the S&P GSCI Energy index, MET is the S&P GSCI Industrial metals index, NENRG is the S&P GSCI Non-energy index, OIL is Dated Brent oil, and PREMET is the S&P GSCI Precious metal index

Z(t) 1% Critical 5% Critical 10% Critical

AEX -25.753* -3.43 -2.86 -2.57 AGR -25.838* -3.43 -2.86 -2.57 ALU -25.378* -3.43 -2.86 -2.57 ENRG -27.348* -3.43 -2.86 -2.57 MET -25.556* -3.43 -2.86 -2.57 NENRG -25.887* -3.43 -2.86 -2.57 OIL -25.525* -3.43 -2.86 -2.57 PREMET -26.261* -3.43 -2.86 -2.57

* Significant at the 1% level.

22

Before the influential work by Granger and Newbold (1974) on non-stationarity in time-series data, it was common practice to conduct OLS regressions assuming time-series data to be stationary. However, many stochastic processes follow a non-stationary path. Because OLS regression relies on a stationary process, results would be spurious.

(31)

E. Normality

For the estimation of commodity price exposure weekly time-series data is used, and returns are calculated from these prices. A goodness-of-fit measure by the means of a Jarque-Bera test is con-ducted to see if the independent variables and control variable depart from normality (see figures B.1-B.8). None of the variables tested show a sufficiently high enough Jarque-Bera value and prob-ability, hence all tests fail to reject to the alternative hypothesis23 of a non-normal distribution. However, as can be seen in the histograms in figures B.1-B.8, the independent variables are close to normal. The non-normality is caused by the fat tails in the distribution, especially due to the 2008 financial crisis. I scale back the extreme outliers that show more than 5 times the standard deviation using the Winsorization method. These outliers are unrealistic, and are probably errors of the Datastream database, in which data input is done manually. The total number of scaled back outliers is 12. The other somewhat less extreme outliers are maintained because these data points can be important in explaining effects due to the financial crisis.

F. Multicollinearity

Correlations are calculated between independent variables to test for multicollinearity (See table C.I for the first dataset). Several variables show high correlations with each other. NENRG is 70% correlated with MET and 87% with AGR. This is easily explained as NENRG consists of all commodities excluding the commodities in the GSCI Energy index and thus has a high correlation with all variables excluding OIL (0.31), ENRG (0.39) and AEX (0.05). Furthermore OIL is highly correlated with ENRG because ENRG consists of OIL and other fuels. ALU is highly correlated with MET for the same reason, MET is composed of ALU and other metals. Care should be taken with correlated variables as the use of these variables may lead to biased estimates24. However, the independent variables will not be used in the same regression and thus cannot bias the estimation itself. I anticipate the correlation problem by expecting the variables that show high correlations to produce similar exposures to the variables they are correlated with. Yet, some of the indices may show high correlations with each other but since they comprise several commodities, some of

23_{The null hypothesis assumes that the variable is normally distributed.}

24_{The threshold correlation coefficient used in this thesis is 0.8.}

(32)

these commodities can be important in explaining exposures. For example, because ENRG also consists of natural gas, which is very abundant in The Netherlands, this variable can be important in determining commodity price exposure. MET is important since it consists of other metals which are important in machinery industry besides aluminum.

Correlations for the second dataset (firm specific variables) are shown in table C.II. Except for DIV and SIZE, that show a correlation coefficient of 0.43, no high correlation coefficients are calculated. Since a correlation of 0.43 is well below the threshold value of 0.8, no adjustments have to be made.

IV. METHODOLOGY

This section describes the methodology used for estimating the relation between the stock returns of Dutch listed non-financial firms and six different commodity price returns. The model employed is a well-known two-step regression formula, widely utilized in the exposure literature (Jorion, 1990), (Choi and Kim, 2003), (Bartram, 2005). Firstly, the first-step regression is described which consists of estimating the separate Betas for the six commodity prices (indices) using all firms in the sample separately25. Secondly, the second-step regression estimates the relation between the six different firm specific variables (SIZE, DE, QR, DIV, BM and FOR) and the estimated Betas in the first step of the regression as the dependent variable.

A. First step regression

To measure commodity price exposure I use a model first theoretically defined by Dumas (1978), Hodder (1982) and Adler and Dumas (1984). I use this model because it is the most commonly used to estimate exposure effects and is (successfully) applied by Jorion (1990), Bartov and Bodnar (1994), He and Ng (1998), Tufano (1998), Bartram (2005) and de Jong et al. (2006) amongst oth-ers. Dumas (1978), Adler and Dumas (1980), Hodder (1982) and Adler and Dumas (1984) define

25

Two statistical packages are used. STATA is used to estimate the first step regression. Due to the fact that many Betas have to be estimated using GARCH(1,1) a STATA script was written to automate the process running the model and produce manageable output data. Eviews is used for estimating the second-step regressions as well as for calculating the descriptives, unit root, normality and correlations.

(33)

economic exposure to exchange rate movements as the regression coefficient of the stock returns of the firm on exchange rates across states of nature. Ceteris paribus, commodity price exposure can be measured in the same way as any other exposure in the sense that one measures the relation-ship between the firm value and any other arbitrarily exposure variable (Adler and Dumas, 1984), (Jorion, 1990). I apply the same methodology used by Bartram (2005), who estimates commodity price exposure.

Partial equilibrium is assumed, implying that commodity prices are exogenous to the value of a firm (Johnston and DiNardo, 1997). Therefore, exposure can be measured by the following regression model:

Rit= βi0+ βimRmt+ βiCRCt+ it, it_eN (0, σt2) (1) Where Ritis the return of the ith stock (R) in period t, βimis the market beta, Rmtis the return on the capital market (AEX) in period t, RCt is the percentage change in the commodity (ALU, OIL, MET, AGR, ENRG, NENRG, PREMET) price (index) in period t, and itis the error term. βiC is the slope coefficient in the regression. The estimate of βiC signifies the commodity price exposure, because it implies the sensitivity of stock returns to unexpected changes in commodity prices. The return on the market index, Rmt is used to control for all other systematic impacts on the stock price. The constant is expressed by βi0.

An increase in the price of a commodity (index) raises the input factor price for production firms and this may lead to higher costs for these firms, if the extra costs are not transferred to the customer or the firm is not fully hedged against commodity price risk. Thus, in general, a price increase of a commodity used as an input factor for a specific firm should induce a negative commodity price exposure. Conversely, a price increase of a commodity should display a positive commodity price exposure when the firm in question produces the commodity as an output.

A.1. Heteroskedasticity, Autocorrelation and GARCH

According to the classical linear regression model residuals should be uncorrelated with each other and the variance of the residuals should be constant (Brooks, 2008). Since the data used is time-series data, the presence of autocorrelation and non-constant variance in the residuals is suspected.

(34)

A Durbin-Watson test (Durbin and Watson, 1951) is used to test the relationship between an er-ror and its immediately previous value, which is the autocorrelation. The Durban-Watson statistic shows values between 2.27 and 2.39 for all regressions, where a value of 2 would suggest little or no evidence of autocorrelation. Hence, there is evidence that autocorrelation is existent in the residuals. A further graphical interpretation is presented in figures D.1-D.8.

The classical linear regression model also assumes that the variance of the residuals is constant. If the variance of the residuals is not ought to be constant, an implication would be that standard error estimates could be wrong (Brooks, 2008). Looking again at figures D.1-D.8, which present a graphical presentation of the variance of all independent variables and the control variable, it can be concluded that the variance of the residuals is not constant. A GARCH(1,1)26model is used to correct for heteroskedasticity, following Bollerslev (1986) and Engle (2001). The GARCH model is preferred over an ARCH model because the former is more parsimonious, and avoids over fitting (Brooks, 2008).

Accordingly, the first step regression model with the GARCH(1,1) (Hull, 2009) specification becomes:

Rit= βi0+ βimRmt+ βiCRCt+ it, it_eN (0, σt2) (2)

it = w + aiu2it−1+ β2it−1 (3)

where w consists of the product of the long-term variance and a constant, 2_it−1 signifies the variance of the last period for the ith stock and u2_it−1expresses the squared return of the last period.

B. Second step regression

The relationship between the estimated Betas in the first step of the regression and the factors that should affect them is examined using an OLS-regression. The significantly exposed Betas from the first sample are used as the dependent variable, regressed against the five firm specific variables that

26_{Generalized autoregressive conditionally heteroskedastic.}

(35)

proxy for hedging incentives by firms, and one variable that proxies for foreign involvement by a firm, as discussed in Section II,

ˆ

βiC = a0+ a1SIZEi+ a2DEi+ a3QRi+ a4DIVi+ a5BMi+ a6F ORi+ vi, viteN (0, σ 2 t) (4)

where ˆβiC is estimated from equation (2), F ORiis defined as the foreign sales ratio, SIZEiis the logarithm of total sales for the ith stock, DEi signifies the debt-to-equity ratio, QRi is the quick ratio, DIVi is defined as the dividend payout ratio, BMiexplains the book-to-market ratio and vi signifies the error term. All firm specific variables are yearly averages (see table IV).

The second step regression is conducted using the estimated Betas from the two commodity indices that reflect the relative importance for Dutch firms (see figure 2 and 3). These indices are ENRG and MET since machinery, chemicals, mineral fuels and finished products are the most imported and exported goods by Dutch firms. Hence, equation (4) is redefined as two separate regression models, one for MET and one for ENRG,

ˆ βiM ET = a0+ a1SIZEi+ a2DEi+ a3QRi+ a4DIVi+ a5BMi+ a6F ORi+ vi , viteN (0, σ 2 t) (5) and, ˆ

βiEN RG= a0+ a1SIZEi+ a2DEi+ a3QRi+ a4DIVi+ a5BMi+ a6F ORi+ wi , witeN (0, σ

2 t)

(6)

where ˆβiM ET embodies the significantly exposed Betas to MET and ˆβiEN RG comprises the sig-nificantly exposed Betas to ENRG. The estimated coefficients a1, a2, a3, a4, a5 and a6 imply the sensitivity of the estimated Betas from equation (2) to SIZE, DE, QR, DIV, BM and FOR re-spectively. To illustrate, a positive and significant a1 implies a positive relationship between the estimated Betas and firm size, which tells us that large firms tend to be more exposed to commodity prices. Table VIII presents a summary of the regression analyses and their expected signs.

(36)

Table VIII

Expected relationships for regression analysis

This table shows a summary of the first step regression as modelled by Rit= βi0+ βimRmt+ βiCRCt+ it(2), and the

second step regression as modelled by ˆβiC= a0+ a1SIZEi+ a2DEi+ a3QRi+ a4DIVi+ a5BMi+ a6F ORi+ vi

(5), (6). A priori expected sign describes the hypothesized sign of the relationship.

Hypothesis Indicator Measure Notation A priori

exp. sign

Source

First-step regression (2)

1 Commodity price exposure Commodity price returns RC +/- Datastream

Second-step regression (4)

2 Firm size Log sales SIZE - Datastream

3 Firm leverage Debt-to-equity ratio DE - Datastream

4a Liquidity position Quick ratio QR + Datastream

4b Liquidity position Dividend payout ratio DIV - Datastream

5 Growth opportunities Book-to-market ratio BM - Datastream

6 Foreign involvement Foreign sales ratio FOR + Bloomberg

V. RESULTS

This section is divided into two main sections. Firstly, the results of the first-step regressions are presented and interpreted. Secondly, the second-step regression results are presented and discussed.

A. First-step regression: Estimation of commodity price exposure

Table IX shows the estimated commodity price exposures for the sample of 78 Dutch non-financial firms. Between 6.49% - 45.45% of the firms shows significant commodity price exposure at the 5% level in different time periods. In every period more firms are positively than negatively and the mean exposure coefficient is primarily positive for all commodities (indices). From this can be concluded that the majority of Dutch non-financial firms benefit in terms of stock returns from a commodity price increase, which is a function of the relevant input/output factors by firms. Evi-dence thus shows that apparently, more firms in the sample produce commodities as an output than using commodities as an input. Comparing the different periods gives interesting results. For period A, between 23.38% - 45.45% of the firms are significantly exposed to commodity prices (indices). The three different sub-periods, B, C and D show a very different distribution of the exposure Betas.

(37)

Table IX

Distribution of commodity price exposure coefficients ˆβiCof Dutch non-financial

firms

This table shows the distribution of the estimated Betas ( ˆβiC) obtained from the following model: Rit= βi0+βimRmt+

βiCRCt+it(1). The sample period used is 1998-2010. Where Ritis the return on the ith stock, Rmtis the return on the

AEX index and RCtis the return on the commodity (index). MET is the GSCI Metal index, AGR the GSCI Agriculture

index, ENRG the GSCI Energy index, NENRG the GSCI Non-energy index, PREMET the GSCI Precious metal index, OIL signifies Dated Brent oil and ALU is Aluminum. ’−’ reports the percentage of negatively exposed firms at the 5% level, ’+’ reports the percentage of positively exposed firms at the 5% level and ’±’ reports the total percentage of firms exposed at the 5% level.

Variable Min Median Max Mean − + ±