Are Norwegian Firms in the Risk Diversification Business?

Are Norwegian Firms in the Risk Diversification Business?

Ida Kristine Haavi

Thesis submitted for the degree of Master in Economics

30 credits

Department of Economics Faculty of Social Sciences UNIVERSITY OF OSLO

Spring 2019

Are Norwegian Firms in the Risk Diversification Business?

http://www.duo.uio.no/

Abstract

Firms make decisions on exports and investments to a destination market based on expected value. At the same time, investors with a high exposure to risk place higher value on firm shares that diversify their risk. How can the two be reconciled?

I develop a small open economy model with multiple destination markets, where I connect the export and foreign direct investment decision of the firm to that of the investor’s utility maximisation problem. I predict that firms export to, or invest more in, markets where demand has a positive covariance with the their domestic investor’s stochastic discount factor. I test the model’s predictions on the relationship between Norwegian manufacturing firms’ export and investment decisions, and changes in the oil price. I find suggestive, albeit not strong, evidence that Norwegian firms export or invest more in markets where demand has a lower covariance with the oil price.

Working on this thesis has been a roller-coaster experience, with a very steep learning-curve. I thank the Department of Economics for the opportunity to spend a semester on delving into a self-defined project.

Thank you to my supervisor, Inga Heiland, for your open door, invaluable feed- back, great discussions, and patiently listening to all my suggestions and making sure I get back on track. Thank you also to Inga Heiland and the research group for international economics at the Department for providing me with the necessary firm-level data.

A special thanks to Julie, for investing time in discussing my thesis and helpfully explaining me what I would actually like to explain myself. To Marte, for Friday breakfasts and always encouraging words. To Alexander, for useful insights at the finishing line. To Thomas, for your programming skills. To Kristoffer, for being there, and helping me understand what I wanted to achieve. To my family for always cheering me on.

And of course, to all my fellow students at the master program, for all the coffee breaks, parties, and engaging (more or less academic) discussions. This ride would not have been the same without you.

I also owe a debt of gratitude to everyone asking questions online and all those providing solutions. This has been an invaluable source of knowledge. The data preparation and estimation has been carried out in Stata. The code can be made available upon request. All remaining errors are my own.

ii

Contents

Abstract i

Preface ii

List of Figures v

List of Tables vi

1 Introduction 1

2 Literature and Theoretical Background 5

2.1 Macroeconomic volatility and openness to trade . . . 5

2.2 The gravity model . . . 6

2.3 Finance and asset pricing theory . . . 8

3 The Model 11 3.1 The Economy . . . 11

3.2 Utility maximisation of investors . . . 12

3.3 Firms’ maximisation problem . . . 14

3.4 The Oil Price Factor . . . 20

4 Data 22 5 Empirical Analysis 26 6 Results and Discussion 29 6.1 Exporting firms . . . 29

6.2 Multinational companies . . . 32

7 Conclusion 37 A Calculations 42 A.1 The risk premium . . . 42

A.2 The linear factor model . . . 42 iii

B Additional results 44

List of Figures

1.1 Norwegian consumption and the oil price . . . 2 1.2 Distribution of correlation coefficients across countries . . . 3

v

4.1 Descriptive statistics . . . 24

6.1 Trade with constant covariance terms . . . 30

6.2 Trade with ten-year covariance terms . . . 31

6.3 Cross-section analysis: years 1999 and 2004 . . . 32

6.4 FDI with constant covariance terms . . . 33

6.5 FDI with ten-year covariance terms . . . 34

6.6 Cross-section analysis: years 2000 and 2006 . . . 35

B.1 Trade with five-year covariance terms . . . 45

B.2 FDI with five-year covariance terms . . . 46

vi

Chapter 1

Introduction

A firm maximising shareholder value can contribute to diversify risk for its share- holders through exports and foreign direct investment (FDI), in effect when it discounts profits by the investor’s risk premium.

This leads to the research question of this thesis:

Do Norwegian firms diversify risk on behalf of their investors?

To analyse this question, I build a small open economy model with multiple destin- ation markets, shareholder-value maximising firms and risk-averse investors. The firm discounts profits by the investor’s stochastic discount factor. This implies that it takes export and FDI decisions based on expected profits and the risk di- versification motive of the investor. Then I test these predictions empirically. In particular, I analyse how Norwegian firms, through exports and FDI, contribute to risk diversification for Norwegian investors from oil price risk.

Norway is an oil-exporting economy, and aggregate wealth and demand seems to be positively correlated with the oil price (the wealth and income effect) (Bjørnland, 2009). Figure 1.1 shows quarterly percentage change in Norwegian consumption and the real oil price. Although not a clear positive correlation, this indicates that Norwegian consumption growth responds positively to an increase in the oil price.

Hence, Norwegians might place higher value on assets that generate higher returns when the oil price goes down, as they contribute more to diversifying his portfolio

1

Figure 1.1: Norwegian consumption and the oil price

-20246

1980q1 1990q1 2000q1 2010q1 2020q1

quarterly

Consumption percentage change Log real oil price

Norwegian consumption and the oil price

Note: This shows the real percentage change in consumption against the log real oil price. The y-axis denotes percentage points, and the x-axis the quarters included in the computation.

across states of nature.¹ In order to diversify against this aggregate risk, there must exist destination markets where a lower oil price would have a positive impact on demand. This indicates that it should be possible for a Norwegian investor to diversify his assets away from high oil price exposure. For the Norwegian investor, an obvious way to diversify risk on his portfolio would be to invest in global firms. A second option, however, is to invest in Norwegian manufacturing firms, especially if he exhibits home bias². These firms are in a good position for diversifying risk. They can export to or invest in markets where oil price reduction increases demand. An oil price production often weakens the Norwegian Krone (ter Ellen, 2016), which increases the international competitiveness of these firms (the trade channel) (Bjørnland, 2009). Hence, there are potential gains to be made from openness to international trade in terms of GDP volatility. Caselli et

1I assume that investors are risk-averse, a more thorough description of preferences is de- scribed in the model.

2Home bias means investors invest more than what is optimal in domestic shares (Obstfeld and Rogoff, 2000)

Introduction 3 Figure 1.2: Distribution of correlation coefficients across countries

0.511.522.5Density

0 .2 .4 .6 .8 1

Correlation between oil price and IP

kernel = epanechnikov, bandwidth = 0.0355

Distribution across countries

Note: This shows a kernel density distribution of the correlation coefficients between IPI and the real oil price across the 39 countries in the sample.

al. (2015) have indeed found that Norway, has strongly reduced GDP volatility from openness to international trade.

To empirically analyse these diversification opportunities, I need a sample of des- tination markets for which the covariances between demand and the oil price dif- fers. Figure 1.2 shows the distribution of correlation coefficients for the Industrial Production Index (IPI) and oil prices in 39 destination markets.³ Indeed it shows that the correlation coefficient varies across countries indicating that such options to diversify exist.

The model in this analysis builds on the theoretical framework of Helpman et al. (2004) who looks at the international organisation of heterogeneous firms, stemming from productivity differences. These productivity differences lead to some firms deciding to produce only for the domestic market, some export, and some directly invest in a given destination market. Other authors such as Fillat et

3I use the industrial production index as a proxy for demand (see e.g. Heiland (2018)).

al.(2015) have shown that returns are also higher multinational firms as investor’s require a higher risk premium in equilibrium. Heiland (2018) builds a general equilibrium model showing how shareholder-maximising firms make decisions on exports when the profits of the firm is discounted by the investor’s stochastic discount factor.

The contribution of this thesis is a model for firms’ decision to export to or invest in a given destination market. I include heterogeneous firms along the lines of Help- man et al. 2004. I set it in a dynamic structure with uncertainty, assuming that firms maximise shareholder value as in Heiland (2018). This uncertainty arises from country-wide shocks due to fluctuations in the oil price. In line with Heiland 2018, I estimate covariance terms between demand in each destination market and demand in each destination market, and use these covariance terms to estimate the investor’s risk premium. Then, the firm makes a choice on whether to enter a destination market through either exports or FDI, based both on heterogeneity in productivity levels and the risk premium of the representative investor. Fur- thermore, I test these predictions empirically for Norwegian manufacturing firms, using micro-level data for both FDI and exports, and the covariance term between the IPI and the oil price in 39 destination markets. The results yield suggestive, albeit not strong, evidence that firms do discount profits by a risk premium based on the covariance between demand in destination markets and the oil price. All calculations are carried out in Stata.

This thesis proceeds as follows. In the next chapter I provide a brief overview of the theoretical background and literature, both from the international macro, finance, and international trade. Following this, I will present the theoretical model. This provides the basis for the empirical analysis that constitutes the remaining part of the thesis. In chapter 4, I present the data and descriptive statistics, before turning to the empirical analysis itself in chapter 5. In chapter 6, I discuss the results and their implications. Finally, I conclude in chapter 7 where I provide a brief summary and answer the research question.

Chapter 2

Literature and Theoretical Background

This chapter provides the theoretical background and existing literature from which I build the model and do the empirical analysis in the next chapters.

2.1 Macroeconomic volatility and openness to trade

In order to assess the potential for the firm to diversify risk on behalf of its investor, must be based on the foundation that there are gains to be made for the investor in terms of volatility from openness to trade. Caselli et al. (2015) use a macroe- conomic lense to analyse how openness to trade affects macroeconomic volatility for a country as a whole. Macroeconomic volatility is affected via two channels.

First, increased openness to trade drives the economy to increases specialisation in some sector(s). This increases volatility as the country becomes more depend- ent on demand in this sector. On the other hand, however, Caselli et al. (2015) argues, when the country is highly dependent on a specific sector, then increased openness to trade can actually reduce volatility. This reduces exposure to demand shocks in Home, as Foreign demand becomes more important. Hence, openness

5

to international trade can both increase and decrease macroeconomic volatility of Home. They find, however, that when country-specific shocks are more important, then openness is more likely to decrease volatility.

In the case of Norway, which depends on the oil price, the impact of the second channel is predicted to be more important. Indeed, Caselli et al. also find that Norway has one of the highest reductions in GDP volatility from openness to trade (Caselli et al., 2015, p. 37). Hence, the Norwegian investor can potentially make huge gains from holding share in a Norwegian exporting firm or multinational company.

2.2 The gravity model

It has been shown that international trade is determined by what can be referred to as gravity variables (see e.g. Head and Mayer 2014). Gravity equations for international trade (Head and Mayer, 2014) are based on the profit maximisation problem of the firm and uses country-specific importer-exporter-pair characterist- ics, to derive equations for the amount of trade flows between two countries. It states that due to trade costs, and in particular melting-iceberg trade costs, firms will be more likely trade with countries that are closer in geographical distance, GDP, and to which other trade barriers are lower such as being part of a free trade agreement.

The gravity equation model builds on the basic microeconomic assumption that firmj decides upon the optimal quantityq^∗they will produce in order to maximise profits π in every period t.

πjh,t =pjh,t(q_jh,t^∗ )∗q_jh,t^∗ −cjτji,q_jh,t^∗ −αj (1)

They assume that firms can freely decide upon their international organisation¹.

1The international organisation of the firm in this case is the choice it makes between export- ing, establishing an affiliate in Foreign (FDI), or producing for the Home market only

Literature and Theoretical Background 7 However, the costs varies with whether they export, do FDI, or produce for do- mestic consumers only. The main feature of the gravity equation models, however, is that the cost of production (or selling to a specific market) increases with factors such as distance, which is represented by the ”gravity” component, G.

X_ni =GS_iM_nφ_ni (2)

Equation (2) is a standard gravity equation model (Head and Mayer, 2014, p. 137).

X_ni represents bilateral trade flows between exporter i and importer n. G is the gravity component,S_irepresents the capabilities of the exporter to export,M_nthe import willingness and opportunities for the importer. φ_ni represents the market share of the exporteriin countryn, taking into account trade costs and elasticities between i and n.

Norwegian firms should then decide on their international trade flows based on a basic profit maximisation problem. The gravity equation then determines the how much the firm should trade with a certain destination market based on production costs, its capabilities (productivity) and the gravity variables.

Helpman et al. (2004) extend this framework building a model for heterogeneous firms in terms of productivity differences. Hence, they take the perspective of the firm and look at how the firm’s choice on whether to do FDI or export to a given destination market, or not enter at all, hinges on its productivity. Hence, the most productive firms can become MNCs, less productive firms can export, while the least productive firms only will serve the Home market. This is due to the different cost structures of FDI, exports, and producing for the domestic market only. FDI incurs higher fixed costs from establishing affiliates and fixed production costs in each new destination market. Exports lead to variable costs in terms of melting-iceberg costs as defined in the gravity equation. However, there fixed costs are lower than for FDI, thought of as the cost of establishing networks.

Domestic production yields no additional fixed costs. I use their analysis to build the model in the next chapter, but extend it with shareholder value-maximising firms, which I present in the next section.

2.3 Finance and asset pricing theory

The finance and asset pricing literature takes into account the utility maximisation problem of the investor and its choice of asset portfolio. literature sheds light on how investors can trade assets, both in order to save for the future, and to spread risk across different states.

In a given period, t, the individual investor, i, can choose to spend his income stream on either consumption, C_it, risk-free assets in zero net-supply, a^f_i,t, or pur- chasing shares in Home firmj,a_ji,t. In the next period,t+ 1, the investor receives a given return on the risk-free asset,R^f_t+1, and a risky return of R_j,t+1 for holding a share in firm j (see e.g. Rogoff and Obstfeld 1996).

The investor’s portfolio consists of all assets he has purchased. The risk exposure of his portfolio depends on the share of risky assets,a_ji,t, and on thediversification of risk. A perfectly diversified portfolio implies that the expected payoff of total assets (and wealth) should be equal across the different future states.

The degree of risk diversification of an investor’s portfolio matters for the his risk exposure, and therefore the volatility of his portfolio. The value of an asset to the investor in period t, in other words, his willingness to pay for a share in firm j, is then based on two principles. First, the expected payoff, s_j,t+1, of holding this share (dividends and expected future sales price). Secondly, however, how this share contributes to risk sharing in the investor’s portfolio. Risk-sharing is then defined as how s_j,t+1 covaries with the investor’s stochastic discount factor (SDF) Heiland, 2018. The SDF for investor i is defined as

m_i,t+1 =β_iu⁰(Ci,t+1) u⁰(C_i,t) (3)

If the expected Cov(m_i,t+1, s_j,t+1) is higher, then the share in firm j contributes more to risk diversification of the investor’s portfolio. Heiland (2018) uses this fact to build a general equilibrium model for international trade flows taking into account that domestic firms maximise the domestic investor’s shareholder value,

Literature and Theoretical Background 9 and focusing on the optimal quantity choice for a firm’s exports to a given destina- tion market. The model is based on gravity equations but where the desired trade share in a given destination market is discounted by the investor’s risk premium in that market (Heiland, 2018, p. 14). The firm’s optimal quantity q_h,t^∗ to export to/produce in Foreign is determined by:

pjh,t+1(q_jh,t^∗ )q^∗_jh,t =φnjYh,t+1 (5)

The source of risk comes from the time lag between firmj’s decision on the optimal quantity, q_h,t^∗ , and the realisation of demand in Foreign, Y_h,t+1 (Heiland, 2018, p. 14). φ_nj is defined as the market share of Home firm j in Foreign market h.

φ_nj = (1−λ_ih,t)^(ε−1)(R^f_i,t+1c_iτ_ih)^(ε−1) P

iHN_it(1−λ_ih,t)^(ε−1)(R^f_i,t+1c_iτ_ih)^(ε−1) (6)

Heiland shows that the elasticity ε−1 is positive. Hence the risk premium in (6) has an effect on the market share of Home firm i in Foreign h. Since the risk premium consists of the covariance term, measuring risk diversification, she concludes that the firm actively² participates in global risk diversification through international trade. Investors might prefer to hold shares in domestic firms rather than foreign firms.³ This could be due to lower costs of investing in domestic firms. If this is the case, this implies a welfare gain for the domestic investor from investing in domestic firms participate in international trade.

Fillat et al.(2015) use the results from Helpman et al.(2004), and look at the relationship between the MNC’s (endogenous) decision to enter the Foreign market and investors’ return on shares in the MNC. They also conclude that the return to shares in MNCs is higher than for exporting firms or domestic firms. This is mostly due to the fact that the investor requires a higher risk premium for shares in the MNC. Setting up an affiliate in Foreign affects the risk premium in two opposite directions (Fillat et al., 2015, p. 37). First, it will reduce risk exposure

2”Active” here means that they make their decision on destination markets for trade/estab- lishing affiliates based on risk diversification motives of their investors

3Normally referred to as ”home bias” (see e.g. Obstfeld and Rogoff (2000)).

by increasing diversification. Second, however, it will increase risk exposure due to the sunk costs and fixed production costs. These costs are higher for MNCs than for exporting firms as they establish affiliates in foreign markets. These costs incur before knowing whether the destination market will experience positive or negative shocks. They develop a reduced form model (Fillat et al., 2015, p. 42) for the return to shares in the MNC and how this changes with entry into the Foreign market:

ret_jt =α+β₁Cov_jt+β₂F_jt+β₃X_jt+δ_k+δ_t+ε_{f t} (4)

The return to shares in firm j in period t depends on the covariance in GDP growth between Home and Foreign⁴ and F_jt represents the sunk and fixed costs required to start a business in countries where j has affiliates. X_jt are control variables that includes e.g. gravity variables.

Going back to the case of Norway, this would imply that the risk premium will be higher for firms that contribute less to diversification of returns for Norwegian investors. In particular, Norwegian investors will require higher risk premia on shares for firms that enter foreign markets that are also oil producers. Then the covariance between the investor’s SDF and the demand in that country can even be negative, and the risk premium required higher. To what extent this determines q^∗_h,t depends on the trade elasticity, ε−1.

In this chapter I have touched upon some of the relevant literature for international trade and risk diversification. In the next chapter, I will build a model based on the model of heterogeneous firms by Helpman et al. 2004, and the role of shareholder value maximising firms for determining international trade flows by Heiland 2018.

This model will then be constitute the foundation for the empirical analysis and determining whether Norwegian manufacturing firms participate in global risk sharing on behalf of their investors.

4Wherej has affiliates

Chapter 3

The Model

I build a model based on heterogeneous firms in the international trade literat- ure, as developed in Helpman et al. (2004). But rather than assuming profit- maximising firms, I assume that firms maximise shareholder value and take de- cisions based on a risk-return tradeoff, as is done in the general equilibrium model of Heiland (2018), but with heterogeneous firms. This yields predictions on firm export and direct investment decisions, when it takes the investor’s risk premium into account.

3.1 The Economy

I consider a small open economy,i, and a set of destination marketsh ∈H, with an infinite time horizon. The home country iconsists ofj ∈J firms, which produces varieties of tradable goods, both intermediate goods, qj,t and final goods Qjh,t1. These goods can be either (i) produced for the domestic market, (ii) exported to destination markets, or (iii) produced in destination markets through FDI.

From the point of view of the representative individual from the home country h, the economy is an endowment economy.² The demand in market h for product q in time t, is exogenously determined Total demand, Y_h,t(s) in each country h

1Countryhdemands Qj,tfrom firmj in eacht

2there is no labour income, but he is endowed with an initial level of wealth.

11

in future periods is uncertain and depends on the realisation of state s. These demand shocks are country-specific. Individuals maximises life-time utility, and the firm maximises shareholder value. Since choices are made for only one period ahead of time, both from the investor and the firm’s side, the problem reduces to a two-period problem.

3.2 Utility maximisation of investors

The firm maximising shareholder value must make assumptions on investor pref- erences when making its export or direct investment decisions. Here I describe the utility maximisation problem of the individual and derive the risk premium that the firm takes into account when making these decisions.

The home country, i, has a population of individual investors, i∈I, but the firm takes into account the preferences of the representative investor, i. He maximises expected lifetime utility in every period, t, as his wealth stream is uncertain and depends on the realisation of state s.

U_i = maxE_t(

∞

X

t=0

β^tu[c_it(s)]) (1)

His momentaneous utility function is given by u(c_it) = ^c

1−σ it

1−σ forσ > 1 andu(c_it) = ln(cit) forσ = 1. β^tis the discount factor of future consumption, andσ is the risk aversion parameter. The utility function is concave, u⁰⁰ < 0, hence the marginal utility of an additional unit of consumption is decreasing in consumption, and is time-separable. The investor is willing to forego some consumption today in order to secure consumption tomorrow. Furthermore, since the realisation of future states, s, is uncertain, the investor operates with the stochastic discount factor (SDF): mi,t+1 ≡ βiE[u⁰(ci,t+1(s)]

u⁰(ci,t) . In addition to maximising returns on his savings, he wishes to minimise variance on returns to his portfolio, R^W_i,t+1(s).

The Model 13 The investor has a given initial endowment, W_i,t. He maximises utility subject to his budget constraint:

W_i,t =A_i,t+1+c_i,t (2)

Hence, in each period, t, he can choose to spend his wealth either on final good consumption, c_i,t at price p_i,t, or savings, A_i,t+1. The investor’s wealth moves according to the rule:

W_i,t+1 =R^W_t+1(s)(W_i,t−c_i,t) (3)

where R^W_i,t+1 is the return on his total portfolio. There are three savings instru- ments in the economy: to purchase shares in firmj, aj,t+1 at pricevj,t with expec- ted return R_j,t+1. He can also purchase a risk-free asset,a^f_t+1, normalised at price q^f_i,t ≡1 with risk-free return R^f_i,t+1.³⁴. The investor’s total return is defined as⁵.

R^W_i,t+1(s) = X

j∈J

a_j,t+1 Ai,t+1

∗R_j,t+1(s) + a^f_t+1 Ai,t+1

∗R^f_t+1 (4)

The investor can mitigate variations in his consumption growth by diversifying investments and thereby smoothing revenue income across states (Rogoff and Ob- stfeld, 1996). Hence, he cares about the covariance of the payoffs from his savings portfolio with his SDF, Cov_t(m_i,t+1, R^W_i,t+1). An investment contributes to diversi- fying his portfolio if the expected return, E(R_j,t+1), covaries more with his SDF.

When Cov_t(m_i,t+1, R_j,t+1) is higher, the investor has a higher willingness to pay, v_j, for a share in firm j. The risk premium⁶ required by the representative in- vestor i from an investment in firm j is then λ_ij,t ≡ E_t[R_j,t+1]−R^f_t+1. However, there is a decreasing marginal value of investing in a given firm, which offsets this higher willingness to pay. This is because holding more shares in firmj, means the investor is more dependent on this firm’s profits for revenue, hence the portfolio

3The risk-free asset is available in unlimited supply

4If there are sufficient diversification benefits of a_j,t+1, we can have R^f_i,t+1 > E[R_j,t+1(s)], and the investor will still want to hold shares in firmj

5Since we operate with a representative investor from the home country h, trading Arrow- Debreu securities becomes unnecessary as they can only mitigate idiosyncratic risk.

6This is an inverse measure of the investor’s willingness to pay asRj,t+1=^sj_v^,t+1

j,t .

becomes less diversified. The higher the share of investor i’s share in firm j, the more his wealth depends on the payoff from his shares in firm j (Heiland, 2018, p. 10).

When the firm maximises shareholder value it takes into account the investor’s preferences. The set of firms J available to the investor depends on the degree of global financial market integration. If we have complete home biasObstfeld and Rogoff 2000, or financial market autarky, then the firm j will take into account the preferences of the domestic investor’s stochastic discount factor. Taking this into account, I turn to the firm’s maximisation problem and tie this to its export and FDI decisions.

3.3 Firms’ maximisation problem

The home country has a pool of J firms. These have firms are heterogeneous in that they have different productivity levels (Helpman et al., 2004).⁷ All firms maximise shareholder value, i.e. expected profits discounted by the investor’s SDF (Heiland, 2018).

The derivation of the firm’s maximisation problem follows the framework of Help- man et al. 2004, but in a dynamic setting over an infinite time horizon with uncertainty as in Heiland 2018.⁸ I start by showing the profit function of firms. I assume that all operating firms produce for the domestic market in alltwith fixed costs, f_ji^D from labour costs. They also make two choices.

First, firms decide whether to enter a certain destination market, h by exports or FDI (the extensive margin).⁹ Secondly, exporting firms and MNCs decide on the optimal quantity, q_jh,t^∗ , to export to or produce in these markets (the intensive margin). Since demand in destination market h is determined in period t+ 1, following country-specific shocks, the firm makes it decision in period t based on

7All firms observe a productivity levelα_j, drawn from a productivity distributionG(α)

8Although the problem is analysed as a two-period problem

9I assume that each firmjwill enter a destination market heither through exportsor FDI

The Model 15 expected profits (Heiland, 2018). Whether they decide to remain in the domestic market, export to, or directly invest in destination marketh depends on expected profits to be ≥ 0 in each option. Due to differences in the cost structure arising from the three options, the productivity level of the firm is the determining factor (Helpman et al., 2004, p. 301-302). When a firm decides to start exports to destination market h additional costs incur. These are additional fixed costs in terms of establishing production networks,f_h,t^X. When exporting products, variable melting-iceberg-costs,τ_jhare also involved.¹⁰ A firm deciding to enter a destination market with FDI encounters higher fixed costs, f_jh^I , from establishing affiliates and labour costs from production, but have lower variable costs from transporting products.¹¹¹² This yields one profit function for exporting firms, π_jh,t^X , and one profit function for MNCs, π_jh,t^I with the entry condition that they should be ≥0.

π_jh,t^X =p_jh,t∗q_jh,t−c_jτ_jhq_jh,t−f_jh^X (5)

π_jh,t^I =p_jh,t∗q_jh,t−c_jq_jh,t−f_jh^I (6)

The cost, c_j reflects productivity differences between firms, and is simplified here to _α¹

j, the inverse of the productivity level. Based on the entry condition, the cutoff productivity level to enter as MNC, as compared to exports, is α^I_j = ^qjh,t_fI^(τjh−1)

jh−f_jh^X , implying that π_jh,t^I > π_jh,t^X . Firms with a lower productivity level export (or produce for the domestic market) while firms with a higher productivity do FDI in destination marketh. For a firm to enter destination markethwith FDI, profits π_jh,t^I > π^X_jh,t, implying thatc_jτ_jhq_jh,t−f_jh^X >.

The second choice the firm makes in periodt is what quantity,q_jh,t^∗ , it will produce for this market, and the price,p_jh,t+1 it will charge. I assume, however, a time lag between production and sales as in Heiland (2018). Hence, the firm decides and

10These iceberg-costs imply that only a given fraction of the product arrives at the destination market, depending on factors such as distance, tariffs, and other barriers to trade

11Also known as the proximity-concentration trade-off

12I assume that the cost of labour is the same for all countries,wi=wh∀ h

the firm will sell the quantity it produces in the next period, t+ 1. Furthermore, I assume monopolistic competition, so the firm sets the price, p_jh,t+1(q_jh,t^∗ ), based on demand and a price markup, θ, depending on its market share, ϕ_jh,t.

Since the state int+ 1 is uncertain, demand for q_jh,t^∗ from destination market his also uncertain. Hence, the firm operates with expected profits:

E_t(π_jh,t+1^X ) =E_t(p_jh,t+1∗q_jh,t^∗ −c_jτ_jhq_jh,t^∗ −f_jh^X) (7)

E_t(π_jh,t+1^I ) = E_t(p_jh,t∗q^∗_jh,t−c_jq_jh,t^∗ −f_jh^I ) (8)

The firm’s total profits is the sum of its profits in all destination markets:

E[π_j,t+1] =

I

X

i=1

E[π_jh,t+1] (9)

Since firms maximise shareholder’s value, the value of their production is discoun- ted by the investor’s SDF and reads:

max

[q_jh,s^∗ ≥0]^∞_s=0V_j,t =E[

∞

X

s=0

m_i,t+sπ_j,t+s] (10)

In t+ 1, firm j’s actual sales, and hence profits, are realised, depending on the state of the world, s. Since the firm can make a decision in every period t about the quantity it will sell to market h in the next period, t+ 1, hence we get a two-period problem:

max

q_jh,t^∗ ≥0V_jh,t =E_t[m_i,t+1π_jh,t+1] (11) Firm j’s realised profits are given by:

π_jh,t+1 =E_t(π_jh,t+1)∗Yˆ_h,t+1 (12)

The Model 17 From (8) we also have that the firm’s expecteddiscountedprofits areE_t(m_i,t+1π_jh,t+1).

Using the formula for expectations and covariances we get:

E_t(m_i,t+1π_jh,t+1) = E_t(m_i,t+1)E_t(π_jh,t+1) +Cov_t(m_i,t+1, π_jh,t+1) (13)

Using that the expected value of the investor’s SDF equals the inverse of the risk-free interest rate, ¹

R^f_i,t+1, and Cov_t(m_i,t+1, π_jh,t+1) equals Cov_t(m_i,t+1,Yˆ_i,t+1)∗ E_t(π_jh,t+1). Rewriting equation (13) we get that

E_t(m_i,t+1π_jh,t+1) = 1

R^f_i,t+1E_t(π_jh,t+1)(1 +R^f_i,t+1∗Cov_t(m_i,t+1,Yˆ_jh,t+1)) (14) From equation (11) we see that the expected discounted profits of the firm depend on expected profits as well as the covariance between the investor’s SDF and fluctuations over expected demand in market i where firmj operates.¹³

The exporting firm will then only decide to enter a destination market if its ex- pected value will be positive, i.e. if:

V_jh,t^X = 1 +R^f_i,t+1∗Cov_t(m_i,t+1,Yˆ_jh,t+1)

R^f_i,t+1 ∗E_t(π_jh,t+1^X )≥0 (15)

Inserting for Et(π^X_jh,t+1) from (x) yields:

V_jh,t^X = 1 +R_i,t+1^f ∗Cov_t(m_i,t+1,Yˆ_jh,t+1)

R_i,t+1^f ∗E_t(p_jh,t+1∗q_jh,t^∗ −c_jτ_jhq_jh,t^∗ −f_jh^X)≥0 (16) A multinational company will similarly only enter a destination market h if:

V_jh,t^I = 1 +R^f_i,t+1∗Cov_t(m_i,t+1,Yˆ_jh,t+1)

R^f_i,t+1 ∗E_t(π_jh,t+1^I )≥0 (17)

Inserting for E_t(π^I_jh,t+1) from (x) we get:

V_jh,t^I = 1 +R^f_i,t+1∗Cov_t(m_i,t+1,Yˆ_jh,t+1)

R^f_i,t+1 ∗E_t(p_jh,t∗q_jh,t^∗ −c_jq^∗_jh,t−f_jh^I )≥0 (18)

13I use hat notation for realised value as a fraction of expected value, i.e. ˆX =_E(X)^X

I assume that the firm’s value function is ”market-separable”, i.e. we can express the total value of firm j as the sum of its values in all destination markets i:¹⁴

V_j,t =

I

X

i=1

V_jh,t (19)

The firm’s total value is given by the value for its shareholders V_j,t. The share price,v_j,t of a share in firmj,s_j,t+1, is determined by investori’s willingness to pay.

The investor’s willingness to pay depends on the return he requires in equilibrium:

E_t[R_j,t+1] =E_t[sj,t+1

v_j,t ] =R^f_i,t+1−Cov_t(m_i,t+1, s_j,t+1) (20)

Rearranging, I get the equilibrium price, v_j,t, as a function of the covariance term, using the relationship between expectations and covariances:

v_j,t =E_t[m_i,t+1s_j,t+1] = E_t[s_j,t+1]

R_i,t+1^f +Cov_t(m_i,t+1, s_j,t+1) (21)

The value its shareholders places on the firm, Vj,t, is then the number of shares times the price of these shares, v_j,t∗s_j,t+1.

In addition, since total demand in country h in t+ 1, Y_h,t+1, determines sales of firm j, sjh,t+1 = pjh,t+1(q_jh,t^∗ )q_jh,t^∗ = ϕjh,t+1Yh,t+1, where 0 < ϕjh,t+1 < 1 is the share of demand for firm j products in country h. Since sales determine payoffs to the investors, the return from firm j to investori becomes:

E(R_j,t+1) = E(s_j,t+1

v_j,t ) = E(ϕ_jh,t+1Y_h,t+1(s)

v_j,t ) (22) Then, equation (15) can be rewritten to:

v_j,t =E_t[m_i,t+1s_j,t+1] =ϕ_jh,t+1E_t[Y_h,t+1(s)]

R_i,t+1^f +ϕ_jh,t+1Cov_t(m_i,t+1, Y_h,t+1) (23) The investor’s willingness to pay for a share in firmj, in other words, also depends on the covariance between his SDF and demand in countryhwhere firmjoperates.

14This implies that the value of entering one marketj= 1 does not effect the value of entering marketj= 2.

The Model 19 This fact provides the basis for the link between the investor’s risk premium and the firm’s international organisation decision.

I have already defined the risk premium required by the investor asλij,t≡Et[Rj,t+1]−

R_t+1^f . This can be rearranged to yield:¹⁵

λij,t =−R^f_t+1∗Covt(mi,t+1, Rj,t+1) (24)

In effect, when the Cov_t(m_i,t+1, R_j,t+1) positive and higher in absolute value, the risk premium required by the investor is lower (discounted by the risk-free interest rate).

Here, I have shown that we can extend the framework of Helpman et al. with heterogeneous firms make different decisions on whether to export to or invest in a destination market, to shareholder-maximising firms who discount profits by the investor’s SDF within the framework of Heiland. Taking heterogeneous firms into account in equations (15) and (17), yields separate value functions for exporting firms and multinational companies. This provides a framework for analysing the effect of the investor’s risk premium on the firm’s decision, while taking into ac- count different profit maximisation problems, and separates this model from that of Helpman et al.. This also changes the cutoff levels for entering as an exporter or with FDI into a given destination market h, as a higher covariance term for the investor of entering the market will increase the value of the firm of entering.

Then, this reduces the cutoff productivity level to enter the market as an exporter or investor compared to Helpman et al.’s prediction Helpman et al., 2004, p. 302.

In order for the exporting firm or MNC to maximise shareholder value, it must make some assumptions about the investor’s stochastic discount factor as this is not directly observable. This is what I turn to next.

15See Appendix A for calculations.

3.4 The Oil Price Factor

The linear factor model that we can use factors that impact the return on invest- ments to estimate the SDF of the investor (see e.g. Cochrane, 2009, p.106-7). This gives the firm a way to determine the risk premium, λ_ij,t+1, its shareholders will require from a decision to enter a given destination market.

The starting point is the covariance term of interest for the investor, namely between his SDF and expected returns to his portfolio: Cov_t(mb_i,t+1, R^W_i,t+1(s)).

The linear one-factor model can be set up as:

mb_i,t+1 =α_i+β_1ix_1i+β_2iZb_t+1 (25)

Inserting equation (4) for R^W_i,t+1 and then equation (22) forRj,t+1 into the covari- ance term of interest, and rearranging, we get:¹⁶

Cov_t(mb_i,t+1, R^W_i,t+1)'Cov_t(mb_i,t+1,Yb_jh,t+1) (26)

Inserting (25) into the covariance term that we care about,

Cov_t(mb_i,t+1,Yb_jh,t+1) =Cov_t(α_i+β_1i−→x_1i+β_2iZb_t+1,Yb_jh,t+1) (27)

Following rules for covariances we get that:

Covt(mbi,t+1,Ybjh,t+1) = (...) +β2iCovt(Zbt+1,Ybj,t+1) (28)

Hence, we have established a relationship between the covariance term of interest for the investor,Cov_t(mb_i,t+1, R^W_i,t+1), and a factor that determines the relationship between this covariance term and demand in destination markets.

16Here I use the relationship between returns from firm j from operating in country h. See Appendix A for calculations.

The Model 21 In the empirical analysis, I will use the oil price as a factor. From what I have argued in the introduction, it seems reasonable to assume that a Norwegian man- ufacturing firm assumes that the representative Norwegian investor looks to the oil price as a factor to be considered in his SDF.

Inserting this factor into the linear factor model yields:

mb_i,t+1 =α_i+β_1ix_1i+β₂OPd_t+1 (29) Inserting (29) into the covariance term that we care about,

Cov_t(mb_i,t+1,Yb_jh,t+1) =Cov_t(α_i+β_1ix_1i+β₂OPd_t+1,Yb_jh,t+1) (30)

Following rules for covariances we get that:

Covt(mbi,t+1,Ybjh,t+1) = (...) +β2iCovt(dOPt+1,Ybjh,t+1) (31)

For the Norwegian investor, the β_2i is negative. Defining this factor provides the firm with a measure of the risk premium for investing in marketh, that affects the decision of the firm to enter this market as an exporter or investor.

From equations (15) and (17) we have that the value of the firm depends posit- ively on Cov_t(m_i,t+1,Yˆ_jh,t+1). Then, the above analysis states that when we have Cov_t(Yb_jh,t+1,OPd_t+1)< 0, we have that Cov_t(m_i,t+1,Yˆ_jh,t+1)>0, i.e. the value of the firm increases. Hence, the model predicts that for both exporting firms and multinational companies, Cov_t(Yb_jh,t+1,OPd_t+1) will have a negative effect on the value of sales or operating income, respectively, in destination market h.

In the remaining part of this thesis, I will test these predictions. I start by de- scribing the data used and the empirical set up. Then, I turn to the results of the empirical analysis and discuss whether these confirm the predictions of the model.

Data

For the empirical analysis, I use micro-level data on Norwegian firms and produc- tion, data on gravity variables for Norway, and industrial production growth in Norway and international partners, and oil price movements. The data and their sources are described in closer detailed below.

FDI of Norwegian manufacturing firms

The panel data set, FDI9606 is generated by Statistics Norway, and adapted by the research group for international economics at UiO. The data set contains in- formation on Norwegian and non-Norwegian firms with affiliates in Norway and their foreign affiliates. It includes data on an annual basis from 1996-2006 on des- tination markets, operating income, institutional sector, and ownership structure of investments ¹

Exports of Norwegian manufacturing firms

The panel data set, TRADE9604 is generated by Statistics Norway, and adapted by the research group for international economics at UiO. This contains informa- tion on all Norwegian manufacturing firms who were exporters in the years 1996- 2004. It includes data on an annual basis on destination markets for each firm, the export value, quantity, and the HS8 sector specification of the products.

1These data are also described in Irarrazabal et al. 2013.

22

Data 23 International Gravity DataIn order to control for gravity variables, I use grav- ity data collected fromCEPII Research Center.² This includes exporter-importer gravity variables such as distance to destination markets, GDP, population, parti- cipation in trade agreements, and the cost of starting a business in the destination markets. Since the only exporter in this analysis is Norway, I exclude all other exporting countries. In the final data set, I only include the 39 destination markets for which I have the data to estimate the covariance term between the oil price and the Industrial Production Index.

Industrial Production Index (IPI)As a proxy for demand in destination mar- kets, I use monthly data on the Industrial Production Index (IPI), collected from the OECD Main Economic Indicators. There is data available for 39 destination markets, though for varying lengths of time. I use data from 1985 to 2006 to estimate the covariance with the oil price.

Norwegian Consumption Growth To find the SDF of the Norwegian repres- entative investor and estimate his marginal utility growth the IPI is not a good proxy for dependence on the oil price, since Norwegian manufacturing compan- ies are likely also dependent on oil as a production input. Rather, I use data quarterly Norwegian consumption growth. This is collected fromStatistics Norway Database: 09173: Final consumption expenditure of households. I use seasonally adjusted data and real prices to estimate the consumption growth.

Oil Price DataFinally, I use monthly data on real oil price movements, collected from Macrotrends ³. This contains monthly data on the world oil price, in both real and nominal terms, and I use the WTI Brent Crude oil price index from 1985 to 2006 to estimate covariance terms with the IPI of the destination markets in the OECD country sample.

2http://www.cepii.fr/CEPII/en/

3https://www.macrotrends.net/

Table4.1:Descriptivestatistics ExportsFDI TotalFranceRussiaTotalFranceRussia MeanSDMeanSDMeanSDMeanSDMeanSDMeanSD Logsales/income10.715672.18253511.106092.38293111.085982.33826110.093322.47786310.137082.40358710.126712.716906 Constantcorr.346209.1925941.27657180.88458780.3127647.1858659.27657180.8845878- Constantcov1.098145.9384846.447728402.755370.9967348.8715439.447728402.75537- 10ycov-2.01091921.907935.10313712.8931913.733474.51058910.9019933.2510310.882616.3723733.7191234.79958 5ycov6.29260414.529926.20021612.6629513.733474.51058911.5751619.873035.64594112.1448230.4074432.22451 Observations11629014052018872408921449344 Distinctfirms4263354102171215116657 Countries39--39-- Thistableshowsdescriptivestatisticswithmeansandstandarddeviation(SD)forthepaneldatasetsforexportingfirmsandmultinationalcompanies Ireportvaluesonlogsalesandlogoperatingincomeforthetotalsample,aswellasforonenetoilimporter,France,andonenetoilexporter,Russia.Thisistoillustratethespreadbetween differentdestinationmarketsinthecovarianceterms,whichisnecessarytoanalysetheeffectsintheempiricalanalysis.

Data 25 In the empirical analysis, I separately merge the export and FDI data with the gravity data, and the constructed covariance terms for each available destination market between the IPI and the oil price. Since the empirical analysis hinges on having an available covariance term, all destination markets for which I do not have this information are dropped from the export and FDI. We are left with two panel data sets, one for exporting firms and one for multinational companies.

Table 4.1 shows descriptive descriptive statistics of the final panel data used in the empirical analysis for both FDI and exports, respectively. It shows the mean and standard deviations for sales and operational income, which are the main dependent variables in the empirical analysis. I also include the the constant covariance and correlation terms, as well as the time-variant five-year and ten- year covariance terms. As a reference point in addition to the total value, I have included the mean and standard deviation for France, an oil importer, and Russia, an oil exporter, to show the spread in their covariance terms for and value of sales and FDI to these two markets. Even though trade and FDI to these countries also depends on gravity variables and other fixed effects, it still indicates that Norwegian manufacturing firms take the covariance term into account when they determine FDI or exports to a destination market. We also note that there are many more exporting firms than multinational companies, which is in line with the prediction that only the most productive firms will choose to enter a destination market with FDI as in the model in chapter 3.

The data is limited by the number of destination markets (39) in the final sample, compared to the full sample of the trade and FDI data. This weakens the validity of the results. It is also below the rule-of-thumb value of 50 to be able to use cluster-robust standard errors on the country level, in addition to the firm level Cameron and Miller, 2015.

Here, I have provided an overview of the data used and the relevant variables and what these can and can not explain. In the next chapter, I will develop the model for which to analyse the research question, before I turn to a description of the empirical approach for the analysis.

Empirical Analysis

In this chapter I describe the empirical approach.

I use a fixed effects model to control for effects that vary across countries, firms, products, and years to estimate the coefficient on the covariance term (Baltagi et al., 2003). This yields consistent estimates ofβ_h1. The estimates ofβ_h1 are also efficient when we treat the fixed effects, α_i and λ_t, as nuisance parameters to be ignored (Cameron and Trivedi, 2005, p. 726-727). I include country-fixed effects when the covariance term is allowed to vary over time. I run separate regressions for the exporting firms and multinational companies. The model does assumes no product differentiation or productivity differences across countries, so I include fixed effects on the firm level, and also firm-product fixed effects, will yield the most reliable results following the model. When we also control for gravity variables, in effect country size and cost of entering the market related to e.g. being part of a free-trade agreement (FTA), we get the same cutoff levels in productivity for entering a destination market as an exporter or with FDI, so that we can isolate the effect of the covariance term Helpman et al., 2004, p. 303.

The first step is to estimate the covariance term which is the main independent variable of interest. From the linear factor model defined in equation (31), I estim- ate a constant covariance term between destination markets’ industrial production and the oil price, based on monthly data from 1985 to 2006. To account for the

26

Empirical Analysis 27 fact investors may change their portfolio composition as preferences change over time (Merton, 1973), I also construct a ten-year and five-year backward-looking covariance term between all destination markets and the industrial production (see e.g. Heiland (2018) for a similar approach).

The main regression equations are a log-linear version derived from equations (16) and (18) after solving for the optimal quantity, q_j^∗h, t for exporters and multina- tional companies, respectively, from the model:¹

logsales_hjp,t =β_h1Cov(Yb_hj,t+1,OPd_t+1) +β_h2x_hj,t+α_i+λ_t+ε_it (x)

logincome_hjp,t=β_h1Cov(Yb_hj,t+1,OPd_t+1) +β_h2x_hj,t+α_i +λ_t+ε_it (x)

The independent variable of interest is the covariance between the destination market’s industrial production and the oil price, which is related to the Norwegian representative investor’s covariance of his SDF and his returns through the linear factor model (see equation (31)). The coefficient on the covariance term should then be negative for the predictions of the model to hold. I include a vector of gravity variables as controls x_hj,t including log distance to the destination market and log GDP of the destination market. I also include entity-fixed and time-fixed effects (α_i andλ_t). To assess the goodness of fit, it is essential to look at thewithin R², which excludes the effect of the between variation.

The fixed effects are on both the country-level and the firm-level. For the export data I include interactions with firm-product fixed effects and firm-product-year fixed effects. For the FDI data I use also include fixed effects along the firm-sector and firm-sector-year dimensions.² Hence I use an estimator taking into account this multilevel fixed-effects model as developed by Correia (2016). The estimates

1I use value of sales and operating income rather than quantity as the dependent variable due to data availability, but as we get higher expected sales or operating income with higher quantities this should be not be of huge importance.

2As the sector also determines whether the multinational firm is foreign-owned, so that in- cluding fixed effects along the firm-sector dimension will be more reliable as the foreign-owned firms might not take into account the SDF of theNorwegianinvestor.

using firm-product-year fixed effects are should yield the most reliable results as is discussed by Baltagi et al.(2003). I use cluster-robust standard errors on the firm level to avoid that t-statistics are too high, and p-values and standard errors are too low. I do not to cluster on the country-level as the small number of countries in the final sample (39), is below the ”rule of thumb” of 50 clusters (see e.g. Cameron and Miller (2015)).

The model predicts a negative coefficient on the covariance independent variable.

To assess whether the coefficient on the covariance term yields a small or large effect on firm choices, it is crucial to also look at the standard deviation. Table 4.1 shows that the standard deviation on the constant covariance term is equal to 0.94.

Since we are interested in the spread of covariance terms for the opportunity to diversify risk, a higher standard deviation means that, and we can infer that firms might still export more to some markets where the covariance of demand with the oil price is lower, which is not captured in the coefficient (see e.g. Helpman et al., 2004, 311-312).

Next, I describe the results of the empirical analysis, and discuss whether these show that Norwegian firms indeed maximise shareholder-value for the Norwegian representative investor.