An empirical investigation of currency exposure of
Nasdaq OMX Nordic listed firms
Master thesis MSc Finance
Corina Postolache
S3032116
Faculty of Economics and Business
University of Groningen
Supervisor: Dr. P.P.M. Smid
1
Acknowledgements
2
An empirical investigation of currency exposure of
Nasdaq OMX Nordic listed firms
Key words: foreign exchange exposure, foreign exchange regime, open economies JEL codes: F31, E42, F4
Abstract
3
Table of contents
1. Introduction ... 4
2. Literature review ... 6
2.1. Exchange rates and firm exposure to currency risk: concepts and theoretical expectations 6 2.2. Exchange rates of Nordic countries ... 8
2.3. Exchange rate exposure: empirical findings and hypotheses ... 9
2.3. Variable definition and possible measurement(s) ... 12
3. Methodology ... 14
4. Data ... 17
4.1. Sample construction... 17
4.2. Description of the variables ... 20
4.3. Descriptive statistics ... 22
5. Results ... 25
5.1. Main results ... 25
5.2. Robustness analysis ... 31
5.2.1. Exposure of stock returns to the trade weighted currency index returns ... 31
5.2.2. Equally-weighted vs. value-weighted market portfolio index ... 32
5.2.3. The effect of outliers ... 33
5.2.4. Exchange rate exposure in different time periods ... 33
5.2.5. Lagged exchange rates ... 34
5.2.6. Exchange rate exposure across industries... 35
6. Conclusion ... 36
References ... 37
Appendices ... 40
Appendix A. Control variables’ transformation ... 40
Appendix B. Explanatory variables’ definition and source ... 41
Appendix C. Descriptive statistics of the variables per periods ... 42
Appendix D. Durbin-Watson test statistic ... 43
4
1. Introduction.
Since the meltdown of the Bretton Woods fixed exchange rate system in the early 70s, there is an increase interest among academics to investigate the sensitivity of firm value to currency fluctuations.
This research aims to measure to what extent the firms listed on Nasdaq Nordic exchanges are exposed to currency risk. Specifically, it analyses the sensitivity of stock returns to exchange rate returns. In this paper, the exchange rate returns represent the monthly logarithmic returns of local currency per one unit of foreign currency, where the local currency is represented by the Danish krone (DKK), Icelandic krona (ISK) and Swedish krona (SEK) and the foreign currency is represented by the euro (EUR), British pound (GBP) and U.S. dollar (USD). The objective of the paper is twofold: firstly, to investigate if stock returns are exposed to the returns of the three major currencies: euro, pound and dollar. Secondly, to present evidence, if the foreign exchange arrangements influence the level of economic exposure of Nordic firms. This paper focuses only to the exposure of economic exchange rate risk (economic exposure). Firstly, the exposure to translation and transaction risk arises only in the presence of contracted transactions denominated in foreign currency, whereas economic exposure arises even if the firm contracted all transactions denominated in domestic currency. Secondly, since I want to capture the effects of exposure in time, economic exposure, which is long-term in nature, seems to be more appropriate. In addition, independent of the business type and foreign extensions, theory suggests that economic exposures will capture the currency risk for both multinational (MNCs) and domestic (DCs) firms (Myint and Famery, 2012).
5
exchange rate exposure under different exchange rate arrangements. Their results indicate that fixed exchange regimes boost currency exposure in stock returns.
Studies such as Jorion (1990), Bodnar and Gentry (1993), Choi and Prasad (1995), Williamson (2001), Bodnar and Wong (2003), Dominguez and Tesar (2006), Choi and Jiang (2009), investigate if the exposure differs across firms with different characteristics. The results show that the exposure is more pronounced at the industry level than at the firm level. In addition, the level of exposure is influenced by the firm determinants. The absolute values of foreign sales, foreign assets and foreign income, foreign sales ratio, foreign assets ratio, foreign income ratio, leverage, firm size, growth opportunities, liquidity levels and others, influence the sensitivity of stock returns to exchange rate risk.
Overall, this paper tries to answer the question:
How vulnerable are stock returns to currency fluctuations?
From this question, three research sub questions delimit the analysis:
Sub-question 1: Is there any evidence of exchange rate exposure in firms’ stock returns? Sub-question 2: Do exchange rate arrangements influence the level of currency exposure? Sub-question 3: Do firm characteristics influence the level of exchange rate exposure?
To answer these questions, the economic exposure of 458 firms listed on Nasdaq Nordic stock exchanges, over the period 2005-2015 is investigated with the pooled OLS and the two-way fixed effects methodologies.
The results show that firms listed on Nasdaq Nordic exchanges present evidence of exposure to the returns of the investigated currencies. In addition, the exposure betas are strongly influenced by firm size and growth opportunities and less influenced by multinationality characteristics. The hypothesis stating that firms listed on exchanges from countries with fixed exchange rate regime have less exchange rate exposure is not rejected when the pooled OLS approach is considered, and is rejected in the case of the two-way fixed effects approach.
The robustness tests show that firms have different level of exposure in different time periods. Moreover, the direction of foreign exchange exposure varies on the industry level. Also, there is weak evidence of exposure to the return on the trade weighted currency. Lastly, the robustness check of possible lag effects for exchange rate has less explanatory power.
6
Thirdly, including a macroeconomic variable, i.e. foreign exchange arrangements, the outcomes can support policy maker’s further decisions.
The remainder of the paper is structured as follows. Section 2 provides a brief review of the existing theoretical and empirical literature, which supports the further analysis of the research questions. As well, based on the earlier findings, the two hypotheses are stated. This is followed by the methodology in section 3. Section 4 presents a description of data and relevant descriptive statistics. Section 5 discloses the results from the main analysis and from the robustness tests. The last section concludes the paper and discusses the relevant practical/policy lessons learnt, as well, the limitations of the research and suggestions for further investigations.
2. Literature review
This section presents both conceptual and empirical evidence of the main determinants and measurements of currency exposure. It starts with the explanation of the theoretical concepts used in the analysis of the research. Then, the empirical findings of the articles that treat the same subject are summarized. Lastly, it presents the hypotheses supported by those findings.
2.1. Exchange rates and firm exposure to currency risk: concepts and theoretical expectations
Exchange rate risk is one of the firm’s financial risks which received much attention from the researchers over the last decades. So far, the history shows that currency exposure influence the firm’s value in different directions. Moreover, the level of currency exposure of firms is tightly linked with the country’s exchange rate policy. This shows the need to understand how a specific currency regime influences the pattern of exchange rates.
Eiteman, Stonehill and Moffett (2013, p. 275) define the exposure to currency risk (also known as exposure to foreign exchange risk / exchange rate risk) as “a measure of the potential for a firm’s profitability, net cash flow, and market value to change because of a change in exchange rates.” It is a widely-held belief by theorists that both DCs and MNCs are exposed to the
currency risk, although, the level of exposure differs across businesses. 1
Typically, firms are exposed to three broad types of currency risk: transaction, translation and economic risk. Transaction risk arises when a firm experiences changes in cash flows due to changes in the exchange rate specified in the transaction contracts. Currency variability and
1 Some determinants are: the company type - DCs tend to be less exposed to currency risk than MNCs, which are
7
currency correlations are the two factors influencing the firm’s level of transaction exposure (Madura, 2015). Translation risk (accounting risk) is associated with the changes in the balance sheet and income statement results due to changes in foreign exchange rates. Economic risk, also called strategic, operating or competitive risk is the risk associated with the sensitivity of future cash flows to unexpected currency fluctuations. It has considerable effects on the firm’s market value (Eiteman, Stonehill and Moffett, 2013). The location of production, services, the target customers and the competitive environment are among the main factors that influence the level of economic exposure. The economic exposure declines when both input costs and sales prices react similarly to currency fluctuations. Moreover, it declines when firms spread their operations in different geographic areas, so that it has more alternatives to get cheaper and faster raw inputs and easy access to a wide range of customers (Madura, 2015; Myint and Famery, 2012).
To sum up, all three types of risk, transaction, translation and economic risk should be considered when analysing the exposure to currency risk. This research is limited only to the exposure of economic currency risk.
The IMF (2014) distinguishes between three broad categories of currency arrangements: hard
pegs, soft pegs and floating.2 Hard and soft pegs refer to fixed exchange rate regimes or fixed
rates. Under hard pegs, countries have either a currency board arrangement, that requires to maintain a fixed exchange rate with a foreign currency, or a full dollarization (no separate legal tender) arrangement, that requires to anchor the local currency to US dollar or another strong currency (the dollar is considered the most stable currency in comparison to others). The governments of these countries tend to maintain their exchange rate fixed, even though macroeconomic factors may pressure the exchange rate to change. Hong Kong and Bulgaria are examples of countries that have such a monetary policy.
A soft peg exchange rate system is divided into conventional pegs, pegged exchange rates with horizontal bands, crawling pegs, crawl-like and stabilized arrangements. The difference between these 5 subcategories lies in the choice of the target (domestic currency can be fixed at a single foreign currency or a basket of foreign currencies) and in the conditions when the fixed exchange rate can softly change. As of 2014, about 44 % of IMF analysed countries fall under this category.
2 See the IMF’s de facto classification of exchange rate regimes of countries, as of April 30, 2014, available
8
The floating regimes / floating rates are divided into two categories: managed floating and free floating. If the Central Bank intervenes to influence the currency value, the country has a managed floating regime (characteristic for most European emerging economies such as Romania, Ukraine, Georgia). Under a free-floating regime, the Central bank almost never intervenes to change the paths of the exchange rates. EMU countries, UK, USA, Canada, Australia, Japan, and others fall in this category.
Because of the numerous advantages and disadvantages of each individual monetary policy arrangement, the theorists’ expectations about the influence of exchange rate risk on firm value are still unclear. Much more evidence concerning this issue comes from empirical studies, which will be discussed further, in the paper.
2.2. Exchange rates of Nordic countries
The Nasdaq Nordic represents the common offering from seven Nasdaq exchanges: Copenhagen (XCPH), Helsinki (XHEL), Iceland (XICE), Stockholm (XSTO), Riga (XRIG), Tallinn (XTLN) and Vilnius (XVLN) Stock Exchange. The first four exchanges refer to the Nasdaq OMX Nordic and the other three refer to the Nasdaq OMX Baltic. There is also the Nasdaq OMX First North division that describes Nasdaq’s European growth market. This market is designed for small and growing firms listed on Nordic exchanges. Moreover, it operates in parallel with the main market, where the shares are traded in a single trading system and under less extensive rules than the main market. Most of the firms that are currently listed on the main market started as being listed on First North. Further, the research focuses only on the Nordic market, including First North stocks listed on Nasdaq OMX Nordic exchanges. The current study investigates only firms with stocks listed on Nasdaq OMX Nordic exchanges, including both main market and First North divisions. Thus, the economy of Denmark, Finland, Iceland and Sweden is analysed.
The four countries have similarities in terms of economic, cultural and political aspects. These
are small open economies, characterized by a strong competitive environment,3 high
dependence on foreign trade and high vulnerability to external shocks (Kaitila and Virkola, 2014). Denmark is the only country with a fixed exchange regime. Denmark pegged its currency DKK to EUR for almost three decades. It has a conventional peg arrangement, which implies a
3 According to Global Competitiveness Index 2015-2016 Rankings, Finland, Sweden, Denmark and Iceland rank 4th,
9
fixed Danish krone to the euro rate (the central rate is 7.46038 krone per euro) and the rate is allowed for a soft float of ±2.25% around the central rate.
The rest of the target countries are characterized by a floating regime. Currently, Iceland has a managed floating arrangement which implies that in case of an economic shock, the Central Bank of Iceland will intervene to stabilize the currency. However, in the last 10 years, Iceland moved very often from a managed to a free float arrangement. Considering the same time frame, the monetary authorities of Finland and Sweden keep the decision of a free float of their exchange rates since 1992. Remarkably, the European Monetary Unit (EMU) membership of Finland (since 2002) has a positive influence on the country’s economy. With the euro, as a stabilizing factor, Liikanen (2009) argues that the economy became more long term oriented and has a less volatile business cycle.
At this point is hard to draw conclusions about which exchange regime is a better predictor of the sensitivity of stock returns to exchange rates fluctuations.
2.3. Exchange rate exposure: empirical findings and hypotheses
The first discussions and analysis about corporate currency risk emerged immediately after the
lost decade period,4 with the papers of Hodder (1982) and Adler and Dumas (1984), showing
that purely domestic companies have higher exposure to currency risk. Later, the rapid growth of trade and capital markets inspired researchers to analyse in depth this issue. In the literature, the influence of exchange rate fluctuations to the firm value is debated.
Several academic studies find strong evidence of exposure to currency risk. For instance, Parsley and Popper (2006) find that stock returns of public firms from nine Asia Pacific countries (Hong Kong, Indonesia, Japan, Korea, Malaysia, Philippines, Singapore, Taiwan & Thailand) are sensitive to exchange rate risk. Another study performed by Aggarwal, Chen and Yur-Austin, (2011) reports the same results. The stock returns of Chinese firms are significantly exposed to currency risk over 2005- 2006. Unlike Aggarwal, Chen and Yur-Austin (2011), Wu and Zhou (2011) consider only stock returns of Chinese electronics firms. Accounting for different lag periods, they find that during 2005-2011, the electronics industry in China has a significant foreign exchange exposure. An earlier study of Bodnar and Gentry (1993) based on Japanese, Canadian and US firms finds a considerable number of Japanese and Canadian firms, specifically the internationally oriented ones that are highly sensitive to currency fluctuations, while almost all US firms are less exposed to FX risk. Asaolu (2011) points toward the existence
4 The lost decade refers to the period of Latin American financial crisis (1970-1980), when several Latin American
10
of a strong relationship between exchange rate fluctuations and Nigerian stock returns. Surprisingly, the results show that both financial and non-financial firms have the same pattern of exposure. This may be explained by the fact that Nigerian financial firms have less sophisticated financial management techniques and therefore less hedging is done.
Furthermore, the same issue was addressed by Glaum, Brunner and Himmel (2000) to the Europe’s largest economy, Germany. Overall, German firms recorded high currency exposure during 1974-1997, however, there are some deviations in time. The authors divide the sample into four sub-periods and find differences in sub-samples in terms of magnitude of exposure and statistical significance.
In contrast to the above-mentioned studies several empirical papers find weak or almost no significance between firm value and the volatility of exchange rates (see, e.g., Jorion, 1990; Fraser and Pantzalis, 2004; Dominguez and Tesar, 2006; Bodnar and Gentry, 1993). A proper explanation for the contradicting results lies in the choice of the method to measure corporate exposure.
Adler and Dumas (1984) is among the first to suggest measuring the economic currency exposure by regressing the stock returns on the exchange rate fluctuations. These papers try to look at the firm currency exposure from an investor perspective. The results fail to support statistical evidence for the significance of currency movements in the pattern of stock returns. Later, Jorion (1990) extended the model of Adler and Dumas (1984) by adding a new factor, the market return. The findings show that there is still weak statistical evidence that stock returns of US MNCs are affected by changes in exchange rates. Interestingly, a growing body of literature uses Jorion’s model to find how sensitive stock returns to currency fluctuations are. The mixed results can be explained by the focus of the papers. Some of the researchers focus on the industry concentration (Bodnar and Gentry 1993; Choi and Prasad, 1995; Akay and Cifter, 2014; Wu and Zhou, 2011), others focus on foreign operations (Kamil, 2006; Jorion, 2010) and other firm characteristics.
Choi and Prasad (1995) is the first paper in which Jorion’s model is adjusted. The authors suggest to orthogonalize the exchange risk factor. The findings show that the firm value of U.S. MNCs is sensitive to exchange rates and the relationship is statistically significant. However, these results vary between different industries from the sample.
11
pure exporting firms with low profit margins are more exposed to currency risk. Furthermore, MNCs register lower operating exposure.
Considering the contrasting evidence in the literature and the economic features of Nordic countries, their openness and high dependence to foreign trade, the following is hypothesized: H1: There is significant evidence of foreign exchange exposure on stock returns of Nordic listed firms.
Parsley and Pooper (2006) considers a sample of firms from South East Asia, including observations for different exchange rate arrangements among countries. The results suggest that, overall, Asia-Pacific firms have a higher exposure to currency risk. The authors conclude that a pegged regime is not a good system from the companies’ view, because, even with predictable exchange rates, the firm value is exposed to currency risk. In addition, the exposure appears to be highly statistically significant in both fixed and flexible regimes during the Asian financial crisis period.
Ye, Hutson, and Muckley, (2014) analyses the currency exposure on stock returns for a sample of firms from 20 emerging countries. The results are quite impressive: more than 50% of analysed stock returns are significantly exposed to changes in currency movements.
Chue and Cook (2008) and Lin (2011) present the results per country and firm of the emerging markets of Latin America and Africa and Asia, respectively. The results suggest that stock returns are negatively influenced by currency fluctuations. In addition, exchange rate regime matters at both macro and micro level and firms from countries with a pegged regime experience a higher exposure to currency risk.
Overall, prior studies suggest that firms are more exposed to currency risk under a fixed exchange rate regime than under a floating one, and different researchers come with different arguments.
12
2002, emerging markets will always be more exposed to currency fluctuations, independent of
the exchange rate regime.5
In conclusion, the literature suggests that firms value can be exposed to currency risk, regardless of the exchange rate arrangements. The most important is a proper selection of exchange rate variables.
Based on the findings of the most papers, the second hypothesis arises:
H2: Exposure is more pronounced in firms from countries with fixed exchange arrangements.
2.4. Variable definition and possible measurement(s)
Theoretical and empirical literature on the sensitivity of firm value to exchange rate fluctuations distinguish two models: the capital asset pricing model (CAPM) and the cash flow model. CAPM is preferred by most academics. Thus, the economic exposure is measured with the sensitivity of stock returns to the unexpected changes in exchange rate and the return of the market index.
As an alternative to “stock returns” exposure, Hodder (1982), Flood and Lessard (1986), Bodnar and Marston (2002), quantify the cash flow exposure. This means that economic exposure is defined as the sensitivity of operating cash flows of a firm to unexpected changes in exchange rate(s).
The cash flow exposure approach is less appropriate for this research. As Miller and Reuer (1998) suggests, it is a wrong approach to focus on net cash flows when firms’ level capitalization changes in time. Additionally, the cash flow exposure model is widely used to measure the exposure of unlisted firms, where the data for stock returns is not available. This study considers a sample of public firms, and hence, the market value of shareholders’ equity is used as a firm value proxy.
Dominguez and Tesar (2006) argue that the choice of the exchange rate(s) as the main explanatory variables matters. The results of the study show that using trade-weighted exchange rate understates the level of exposure. Doukas, Hall and Lang (2003) shows that even the lagged trade weighted index has no influence on firms' economic exposure.
5 The original sin problem arises when a company is unable to borrow abroad in its own currency, thus
13
Miller and Reuer (1998) suggests extending the bivariate model with multiple exchange rates. The authors show that including only one exchange rate (either bilateral or trade weighted) results in the distortion of the estimated exposure coefficients. For instance, while the estimated pound exposure coefficient might be significant under the bivariate model, it might appear to be insignificant in the multivariate model and vice versa.
Following Miller and Reuer (1998), three bilateral exchange rates are included in the model, representing the logarithmic return of the home currency per one unit of foreign currency. EUR, USD and GBP represent the foreign currencies.
In line with de Jong, Ligterink, and Macrae (2002), Asaolu (2011), Choi and Prasad (1995), among others, I expect that most of the sample firms have a significant positive relationship between exchange rate returns and stock returns.
In this study, I analyse the influence of fixed exchange regime on the marginal effect of exposure. According to Rossi (2004), Patnaik and Shah (2010), Parsley and Pooper (2006), Kamil (2006), Ye, Hutson, and Muckley (2014), I expect that the fixed exchange rate regime influence the relationship between exchange rates returns and stock returns.
Jorion (1990) shows that adding a market factor to the model considerably decreases the residual variance in the regression. Accordingly, the return on the market portfolio is considered in the model.
14
De Jong, Ligterink, and Macrae (2002) shows that MNCs experience lower levels of currency exposure. The paper of Choi and Jiang (2009) analysis the exposure U.S. MNCs and DCs. The results show that the multinationality slightly matters: there is a small and less significant currency exposure for both categories of firms. In line with the results of these three papers I expect that multinationality of the firms is a bad predictor of the relationship between exchange rate returns and stock returns.
Market share capitalization is used in the paper as a measurement of firm size. Dominguez and Tesar (2006) and Bodnar and Wong (2003) show that large-cap firms are less exposed to currency risk, because on average, the larger the firm, the more internationally oriented it is and therefore, the bigger is the probability that the exposure will be naturally hedged due to diversification. A further investigation of multicollinearity between firm multinationality and size will be conducted. I except positive and significant interaction coefficients between exchange rates and firm size (Bodnar and Wong, 2003; Dominguez and Tesar, 2006 and Choi and Jiang, 2009).
Further, growth opportunities are investigated. Market to book ratio and price-earnings ratio are the proxies for the growth opportunities commonly used in the empirical research. I use the market to book value of equity to measure the firms’ growth opportunities. Huston and Stevenson (2010) argue that firms with larger growth opportunities (measured as market to book value of equity) are more likely to engage in currency hedging and therefore, exhibit less currency exposure. Thus, I expect significant and negative beta of the interaction term between the exchange rates and growth opportunities.
All things considered, I expect Nordic listed firms to present evidence of economic exposure, and the exposure to be more pronounced for firms listed on the XCPH.
3. Methodology
This section describes the models and the estimation method used to perform the analysis. Similar to the previous empirical research in this area, I follow the procedure in Jorion (1990) to find the level of currency exposure in firm’s stock returns. The following four-factor regression model is applied:
𝑅𝑖𝑡 = 𝛽0𝑖+ 𝛽1𝑖𝐸𝑋𝑡𝐸𝑈𝑅+ 𝛽2𝑖𝐸𝑋𝑡𝐺𝐵𝑃+ 𝛽3𝑖𝐸𝑋𝑡𝑈𝑆𝐷 + 𝛽4𝑖𝑅𝑚𝑡 +𝜖𝑖𝑡 (1)
15
where 𝑅𝑖𝑡 is the logarithmic return of i-th stock at time t; 𝛽0𝑖 is the intercept; 𝐸𝑋𝑡𝐸𝑈𝑅, 𝐸𝑋𝑡𝐺𝐵𝑃
and 𝐸𝑋𝑡𝑈𝑆𝐷are the logarithmic returns of the exchange rates representing the home currency per
one unit of foreign currency, which are euro, U.S. dollar and British pound at time t; 𝛽1𝑖, 𝛽2𝑖 and 𝛽3𝑖 are the sensitivities of stock returns to the respective exchange rate returns of
i-th stock; 𝑅𝑚𝑡 is the logarithmic return of the Nordic, equally weighed market portfolio index
at time t; 𝛽4𝑖 is the sensitivity of i-th stock return to market risk at time t; and 𝜖𝑖𝑡 is the idiosyncratic error term for i-th stock at time t .
If β1i, β2i and β3i from equation (1) are significantly different from zero, then the first
hypothesis, specifying that there is significant evidence of currency exposure on stock returns is not rejected.
The second regression equation includes several control variables and the interaction variables between control variables and exchange rates. This model aims to explain if the specified predictor variables influence the level of stock return exposure to currency risk.
𝑅𝑖𝑡 = 𝛽0𝑖+ 𝛽1𝑖𝐸𝑋𝑡𝐸𝑈𝑅+ 𝛽2𝑖𝐸𝑋𝑡𝐺𝐵𝑃+ 𝛽3𝑖𝐸𝑋𝑡𝑈𝑆𝐷 + 𝛽4𝑖,𝑛(𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑛∗ 𝐸𝑋𝐸𝑈𝑅)𝑡 +
𝛽5𝑖,𝑛(𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑛∗ 𝐸𝑋𝐺𝐵𝑃)𝑡+ 𝛽6𝑖,𝑛(𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑛∗ 𝐸𝑋𝑈𝑆𝐷)𝑡+ 𝛽7𝑖𝑅𝑚𝑡+ 𝛽8𝑖,𝑛𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑡,𝑛+ ϵit
(2) for i = 1, …, N; t = 1,…,T and n = 1,…,N
where, the new variables denote: 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑡,𝑛 is the n-th control variable at time t; the
(𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑛∗ 𝐸𝑋𝐸𝑈𝑅)𝑡 , (𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑛∗ 𝐸𝑋𝐺𝐵𝑃)𝑡 and (𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑛 ∗ 𝐸𝑋𝑈𝑆𝐷)𝑡 are the interaction variables of the exchange rates with the n-th control variable. The explanation for the remaining variables is the same as for equation (1). The set of control variables is further explained in the Data section.
To test the second hypothesis, that firms listed on the stock exchange of countries with a fixed exchange regime are more exposed to currency risk, the following model is applied:
Rit= β0i+β1iEXtEUR + β2iEXtGBP + β3iEXtUSD + β4iFIXDUMMYt + β5i(EXtEUR*FIXDUMMY)t+
β6i(EXtGBP*FIXDUMMY)t + β7i(EXtUSD*FIXDUMMY)t + β8iRmt + 𝛽9𝑖,𝑛𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑡,𝑛+ ϵit
(3) for i = 1, …, N; t = 1,…,T and n = 1,…,N
where the new variables denote: 𝐹𝐼𝑋𝐷𝑈𝑀𝑀𝑌𝑡 is a dummy variable which takes the value of 1
if the stock exchange where the firm’s stock is listed belongs to a country with a fixed exchange
rate regime and 0 otherwise; (𝐸𝑋𝑡𝐸𝑈𝑅*FIXDUMMY)
16
(𝐸𝑋𝑡𝑈𝑆𝐷*FIXDUMMY)
t are the interaction variables of the exchange rates with the fixed exchange rate regime dummy at time t. The explanations for the remaining variables are the same as for equation (1) and (2).
In other words, it is tested if, on average, an increase in exchange rate return will mean higher stock returns for firms listed on XCPH than for firms listed on other exchanges. In case of either positive or negative statistically significant interaction effects, the second hypothesis is not rejected.
Most empirical papers apply cross-sectional regression estimation. This study analyses the sensitivity of stock returns to exchange rate fluctuations using panel techniques. As Brooks (2014) explains, there is a lot of advantages to use panel data approach. A combination of time-series and cross-sectional data increases the power of the test by increasing the number of degrees of freedom. Furthermore, Brooks explains the limitation of separate time-series or cross-sectional data analysis. Estimating time-series regressions for each entity do not account for any common structure in the series and the problem of multicollinearity may arise. Estimating cross-sectional regressions at each point in time is not a good approach when there is a common variation in the series over time. Alternatively, estimating a pooled regression for the whole data have also some limitations. A pooled OLS regression assumes homogeneity in the sample which implies a constant relationship between the average values of the variables over time and across entities. These issues are tackled by the panel estimation methods. Moreover, in the last case, the problem of multicollinearity is less likely to arise. Another advantage of panel estimation is that entity effects account for unobserved or unmeasured variables which vary across entities but not over time. Thus, the impact of omitted variables bias is removed.
17
4. Data
This section presents a detailed description of the data. Firstly, it specifies the sample and it explains the steps used in the sample selection process. It includes also a brief description of the variables: the measurement, data sources and data transformation. Lastly, it presents the main descriptive statistics of the variables included in the analysis.
4.1. Sample construction
An unbalanced panel data set is used to investigate the sensitivity of stock returns to exchange rate fluctuations. The data set covers monthly observations over 2005-2015 period and includes only non-financial firms that are listed on Nasdaq OMX Nordic exchanges during this period. With the time frame of 11 years, the sample size is sufficiently large to conduct such a research. Moreover, monthly instead of yearly observations decrease the mean variance of the sample and hence increase the precision of the estimates.
As mentioned in the Literature Review section, Nasdaq Nordic describes the common offering from Nordic and Baltic stock exchanges. During January 2005-December 2015, there are 884 stocks listed on seven Nasdaq Nordic exchanges, from which 607 stocks belong to the main Nordic market, 66 stocks belong to the Baltic market and 211 stocks belong to the First North market.
18
listed on Nordic exchanges. Since the number of firms listed on Baltic exchanges is quite small as compared to the number of firms listed on Nordic exchanges, firms listed on Baltic exchanges are excluded from the sample and a further investigation is suggested.
In the second step, firms with stocks listed on XHEL are excluded, as Finland entered the Eurozone in 2002 and euro is a currency of great interest for about 80% of firms. Further, in the paper I refer to firms listed on NASDAQ OMX Nordic as those with stocks listed on XSTO, XCPH and XICE.
The next step is the exclusion of 162 stocks which belong to firms from the financial industry. This firms operate with foreign currency in larger amounts as compared to firms from other industries. Such firms are regulated by Financial Regulatory Authorities, which set the limits of currency risk and provide firms with the models to calculate the risk associated with expected and unexpected exchange rate changes. Subsequently, their management strategies for such risk might differ significantly in terms of volume, frequency and type, as compared to other firms from the sample. This can influence the “total sample” exposure results.
In addition, there are two firms with a multi-listed stock each. The data for these firms is kept based on the number of trades the stock has on a specific stock exchange. That stock listed on the stock exchange with the higher number of trades is further investigated.
Finally, to keep the sample size as homogeneous as possible, 19 stocks which belong to firms whose local currency is other than DKK, ISK or SEK are excluded from the sample.
The final sample includes monthly observations of 484 stocks (458 firms) during 2005-2015 period. The unbalanced panel data set is described by 484 cross-sections and 132 time-series observations.
19
Table 1. Distribution of firms per stock exchange
Note: This table reports the distribution of number of firms and stocks per stock exchange, according with the final
sample of firms. The number of stocks is reported in parentheses. The Main market and First North denote the Nasdaq Nordic market divisions. The L, M and S are abbreviations for large-cap, mid-cap and small-cap firms. MNCs and DCs denote multinational and domestic firms. XCPH, XICE and XSTO denote Copenhagen, Iceland and Stockholm stock exchange, respectively.
XCPH XICE XSTO Total sample
No. of firms % of total No. of firms % of total No. of firms % of total No. of firms % of total Main market 88 (93) 95 (95) 9 (9) 82 (82) 213 (231) 60 (62) 310 (333) 68 (69) First North 5 (5) 5 (5) 2 (2) 18 (18) 141 (144) 40 (38) 148 (151) 32 (31) L 23 (26) 25 (27) 0 (0) 0 (0) 51 (65) 14 (17) 74 (91) 16 (19) M 17 (18) 18 (18) 6 (6) 55 (55) 70 (71) 20 (19) 93 (95) 20 (20) S 53 (54) 57 (55) 5 (5) 45 (45) 233 (239) 66 (64) 291 (298) 64 (61) MNCs 71 (75) 76 (77) 10 (10) 91 (91) 263 (283) 74 (75) 344 (368) 75 (76) DCs 22 (23) 24 (23) 1 (1) 9 (9) 91 (92) 26 (25) 114 (116) 25 (24)
20
Table 2. Distribution of firms per industry
Note: This table presents the distribution of number of firms and stocks per industry. The number of stocks is
reported in parentheses. The classification of industries is in line with the first level Industry Classification Benchmark (ICB) adopted by Nasdaq OMX to classify all share classes. The calculations are based on the final data sample. The L, M and S are abbreviations for large-cap, mid-cap and small-cap stocks. MNCs and DCs denote multinational and domestic firms respectively.
MNCs DCs
Code Industry L M S L M S Total % of
total
0001 Oil & Gas 2 1 10 0 1 2 16
(17) 4 (4) 1000 Basic Materials 7 0 9 0 0 12 28 (31) 6 (6) 2000 Industrials 25 36 53 1 1 23 139 (147) 30 (30) 3000 Consumer Goods 10 16 22 1 0 8 57 (62) 12 (13) 4000 Health Care 12 14 31 0 3 20 80 (82) 17 (17) 5000 Consumer Services 6 13 19 1 1 18 58 (62) 13 (13) 6000 Telecommunications 4 1 2 1 0 1 9 (10) 2 (2) 7000 Utilities 0 0 1 0 0 2 3 (3) 1 (1) 9000 Technology 4 6 41 0 0 17 68 (70) 15 (14) Total 70 (87) 87 (89) 188 (192) 4 (4) 6 (6) 103 (106) 458 (484)
4.2. Description of the variables
Dependent variable. In this paper, the dependent variable is represented by the firm value,
which is further measured with the logarith of stock returns, in percentage. Monthly closing prices, in local currency, for each individual stock are obtained from Datastream database. The following equation is used to transform the prices in returns:
𝑅𝑖𝑡 = 100 ∗ 𝑙𝑜𝑔 ( 𝑝𝑖𝑡
𝑝𝑖(𝑡−1)) (4) for i = 1, …, N and t = 1,…,T
where 𝑅𝑖𝑡 is the return of i-th stock at time t, 𝑝𝑖𝑡 is the price of i-th stock at time t, and 𝑝𝑖(𝑡−1) is
the price of i-th stock at time t-1.
Independent variables: exchange rates, exchange rate regime and market portfolio. This
21
experience higher currency exposure. Therefore, exchange rates and fixed exchange regime represent the main explanatory variables.
The impact of exchange rate fluctuations is measured with the logarithmic return on three bilateral exchange rates: local currency/EUR, local currency/GBP and local currency/USD. The local currency refers to DKK, ISK and SEK. The average monthly (middle) exchange rates are obtained from the Central Bank of Denmark, Iceland and Sweden websites. Equation (4) is used to transform exchange rate prices into percentage returns.
The fixed exchange rate regime variable is a dummy variable. It takes the value of 1 if the firm is listed on a stock exchange from a country with fixed exchange rate arrangements and 0 otherwise.
In line with the CAPM model, a market portfolio index is included to show the reaction of stocks to market changes. Nordic all share market index is used as an appropriate measure of the market portfolio. The index includes all shares listed on XCPH, XICE, XHEL and XSTO main market. The index prices in DKK, ISK and SEK are obtained from the Nasdaq OMX Nordic official website and average monthly prices are computed. Then, equation (4) is used to calculate logarithmic returns, in percentage.
Control variables: size, growth opportunities and multinationality.
A list of control variables is used in the paper to explain if firm characteristics influence the level of currency exposure in stock returns. Besides, the paper investigates whether the marginal effect of exchange rates return depends on these control variables.
Firm size is measured by the logarithm of market value of equity and is expressed in millions of local currencies.
The logarithm of market to book value of equity ratio, expressed in local currencies measures the firms’ growth opportunities. This series contains observations with negative values. To apply the logarithmic transformation and at the same time to keep all the information about the ratio, a constant that allows all the data to be greater than zero is added.
Logarithmic transformation is applied to increase normality. Appendix A shows the distribution of the variables. Even though Jarque Bera test for normality is significant for all series (p-values equals 0.000), logarithmic transformations considerably increase normality.
22
multinational. If the firm has no subsidiaries abroad, but it has observations for FSR greater than 25% it is also considered multinational. For those firms that were not recognized by Orbis and have no FSR data, the information about the organization of the firm, the recorded location of subsidiaries, employees, business areas, and location of target customers is analysed from the companies’ websites. In total, 344 MNCs and 114 DCs firms are recorded. The multinationality variable is a dummy variable; it is set up one if the firm has at least one subsidiary abroad and the foreign sales ratio is equal or higher than 25% and zero otherwise. Monthly data for market value of equity; market to book value of equity and FSR is obtained from Datastream. Further, refer to Appendix B for an extended description of variables, together with the direct link to the data source (s).
4.3. Descriptive statistics
Before proceeding with the descriptive statistics analysis, the data is checked for stationarity. A potential problem with non-stationary data is that the output from OLS regressions can lead to incorrect conclusions (Brooks,2014). A panel unit-root test, namely the Maddala and Wu test
(MW)6 is performed for each individual time-variant variable specified in equations (1) – (3).
The number of lags is determined automatically based on Schwarz Information Criterion. Only the intercept is included in test equation. The results show that all variables, except SIZE are stationary in level and first difference at the 1% significance level. The unit root test for the SIZE series fails to reject the null hypothesis of the presence of a unit root at the 10% significance in level and at the 1% significance in first difference.7 In sum, all variables are stationary and the bias of spurious regressions is eliminated.
The summary statistics of all the variables (after transformation) included in the main analysis are presented in Table 3. The analysis is performed based on a common sample of 43,725 observations. The other 21,163 observations are excluded due to missing data.
On average, the percentage of stock returns is close to zero (common sample mean equals -0.021). Moreover, stocks listed on the Nordic market register positive returns (mean equals 0.296%), while those listed on First North market register negative returns (mean equals -1.132 %). A possible explanation can be that First North stocks belong to small and growing firms and these categories of firms might have less diversification opportunities. As per total sample, the
6 The Maddala and Wu unit root test tests the null hypothesis of non-stationarity in both balanced and unbalanced
panels.
7 The SIZE variable is stationary in level at 10% significance when only the intercept is included in the equation and
23
euro, pound and dollar rates, as well the market portfolio have, on average, positive returns. The highest return on euro, pound and dollar rates is 14.723%, 16.282 % and 22.371% respectively. In contrast, the lowest return on these currencies is -11.957%, -11.929% and -8.744%. On average, the highest market portfolio return is about 25% and the lowest market return is about -25%. As is presented in Appendix C, the 2008-2009 crisis period had an important impact on stock returns and market portfolio returns, as well as exchange rate returns. Moreover, the crisis effect is still persistent in returns. In addition, during the after-crisis period, 2010-2015, stock returns, returns on the market portfolio and return on euro rate are, on average, lower as compared to the pre-crisis period, 2005-2007. After the crisis period, there is a slight increase in the average pound and dollar rate returns, with 0.798% and 0.022%, respectively, and a decrease of 0.429% in the euro rate returns.
The descriptive statistics of the firm size variable show that on average, large firms are concentrated on the Nordic main market and smaller firms on the First North market with a mean of 7.348 and 4.368 million of local currency. In addition, firms listed on First North market have higher growth opportunities with a maximum value of logarithm of market to book value of equity being 6.231 units of local currency and minimum value of -4.170 units of local currency as compared to the 5.744 and -5.662 maximum and minimum values of the same ratio for the firms with shares listed on the main market. As Appendix C shows, after the financial crisis, both size and growth opportunities indicators increase.
24
Table 3. Descriptive statistics
Note: This table presents the descriptive statistics of variables used in the main analysis (see the explanation of the variables in Appendix B) for monthly observations of 484
cross-sections and 2005-2015 period. Panel A and Panel B shows the common sample descriptive statistics. The statistics from Panel A are calculated based on the observations for the entire sample of firms, and the statistics from Panel B are based on the two subsamples: Nordic, which includes observations for firms with shares listed on the Nasdaq OMX Nordic main market and First North, which includes observations for firms with shares listed on First North division.
Full sample
Panel A Mean Median Max. Min. St.Dev. Skewness Kurtosis Obs.
Ri -0.021 0.000 688.073 -230.259 14.827 2.746 130.189 43,725 EXEUR 0.006 0.001 14.723 -11.957 1.310 -0.011 11.913 43,725 EXGBP 0.010 0.125 16.282 -11.929 2.054 -0.174 4.572 43,725 EXUSD 0.195 0.234 22.371 -8.744 2.664 0.461 4.977 43,725 Rm 0.544 1.409 24.840 -25.149 4.568 -1.517 8.698 43,725 FIXDUMMY 0.266 0.000 1.000 0.000 0.442 1.058 2.119 43,725 SIZE 6.764 6.537 13.626 -1.139 2.337 0.278 2.524 43,725 GRO 1.217 1.122 6.231 -5.662 0.760 0.153 12.429 43,725 MNDUMMY 0.828 1.000 1.000 0.000 0.377 -1.742 4.034 43,725
Nordic First North
Panel B Mean Media n
25
The coefficients of Kurtosis show that fix regime dummy and firm size variables have a platykurtic distribution relative to the normal with a Kurtosis less than three. All other variables have a leptokurtic distribution relative to the normal with a Kurtosis greater than three.
Table 4. Correlation matrix
Note: This table presents the correlation coefficients of all the variables used in the main analysis (see the
explanation of the variables in Appendix B). It is based on a balanced sample (list-wise missing value deletion) of monthly observations, for the 2005-2015 period. ***, ** and * denote statistical significance at the 1%, 5% and 10% level, respectively. The value expressed in bold shows a high correlation between the variables.
Ri EXEUR EXGBP EXUSD Rm FIXDUMMY SIZE GRO MNDUMMMY
Ri 1.000 EXEUR -0.125*** 1.000 EXGBP -0.020*** 0.477*** 1.000 EXUSD -0.093*** 0.441*** 0.602*** 1.000 Rm 0.315*** -0.202*** 0.028*** -0.203*** 1.000 FIXDUMMY 0.002 -0.002 -0.009* -0.008* -0.002 1.000 SIZE 0.086*** -0.010** 0.001 -0.015** 0.016*** 0.017*** 1.000 GRO 0.096*** -0.015*** 0.001 -0.029*** 0.027*** -0.102*** 0.158*** 1.000 MNDUMMY 0.033*** -0.001 -0.009* -0.010** 0.002 -0.088*** 0.366*** -0.001 1.000
Table 4 reports the correlation matrix of the variables, after transformation. It shows that stock returns are significant and negatively correlated with the euro, dollar and pound exchange rate returns. In contrast, there is a significant, but positive correlation of returns on the market portfolio, size, growth opportunity, and multinationality dummy with the dependent variable. In general, correlation coefficients show that there is no multicoliniarity between the explained and explanatory variables. A possible multicollinearity is observed between the pound and dollar exchange rate returns, with a correlation coefficient of 0.602, but since the analysis is performed based on panel estimation approach, with both cross-sections and time series observations, the multicollinearity is not an issue (Brooks, 2014).
5. Results
This section presents the empirical results. It starts with the investigation of firms’ exchange rate exposure using the models, methods and data specified in Methodology and Data sections. The empirical investigation is then extended by performing several robustness tests.
5.1. Main results
26
estimates for the regression models specified in equations (1) – (3). The results are analysed using different estimation methods. Firstly, a common constant method is applied. As the name suggests, this method estimates a common constant for all stocks. The motivation for this approach is that all firms in the analysis activate in highly developed countries, from the same region - Nordic. Secondly, the same regressions are estimated with a two-way fixed effects method. This method aims to account for the possible effects of the variables which are left unobserved cross-sectionaly, like cultural differences, differences in corporate governance, differences in business practices, etc. Moreover, the method controls also for unobserved variables that change over time, like national monetary policies. In such a way, the results of the two-way fixed effects are expected to incorporate the effects of omitted variables.
Baltagi (2005) demonstrates that fixed effects estimation fulfils all conditions of Gauss-Markov Theorem, and provides best linear unbiased estimates (BLUE) only under the assumption that investigated explanatory variables are non-stochastic. For all estimates reported in Table 5, I use White standard errors to correct for any possible heteroskedasticity in the residuals and to provide reliable inferences about the coefficient estimates. Moreover, I analyse the output of Durbin-Watson (DW) statistic for each regression in part to see if there is presence of autocorrelation in the residuals. The DW statistic fails to reject the null hypothesis that the residuals are uncorrelated for all six regressions at the 1% significance level (refer to Appendix D for details about the test). Now, providing evidence that the estimates are BLUE, the results from each model is further analysed.
The standard approach specified in Jorion (1990) is first tested by pooled OLS. The results show that the euro exchange rate is negative and statistically significant at the 1% significance level. This implies that a 1% increase in the returns of local currency/euro rate causes a 0.703% decrease in the value of stock returns, leaving all other variables constant. There is also a significant and positive relationship between returns on the market portfolio and stock returns, indicating that a 1% increase in the returns on market index causes a 0.986% increase in the value of stock returns, holding all other variables constant.
The estimates of the second equation reported in Panel A show that the adjusted R2 slightly
27
Table 5. Stock return exposure to exchange rate risk: evidence from traditional panel data estimation
Note: This table presents the coefficient estimates for the regression models specified in equations (1) – (3). The
results are based on a panel data set with monthly observations for 484 cross-sections over 2005-2015 period. The dependent variable in each regression is represented by the logarithm of stock returns. Refer to Appendix B for the explanation of the independent variables. White diagonal standard errors are reported in parentheses in columns 1,2 and 3 of Panel A and in columns 2 and 3 of Panel B. White cross-section standard errors are reported in parentheses in column 1 of Panel B. ***, ** and* indicate statistical significance at the 1%, 5% and 10% levels, respectively. Empty cells occur when the variable is not included in the regression or a specific statistical test is not performed. RE and FE denote regression with cross-section random effects and regression with cross-section and period fixed effects, respectively.
Panel A: Common constant Panel B: Entity and time effects
1 2 3 1 2 3 c -0.558*** (0.068) -5.353*** (0.297) -5.577*** (0.304) -0.570* (0.299) -21.569*** (1.297) -21.340*** (1.293) EXEUR -0.703*** (0.068) -1.442*** (0.297) -0.678*** (0.071) -0.706*** (0.197) 2.314 (1.981) 4.888** (2.001) EXGBP 0.045 (0.051) 0.610** (0.243) -0.093 (0.058) 0.045 (0.214) -2.476 (1.694) -3.975** (1.625) EXUSD -0.046 (0.036) -0.414*** (0.156) 0.029 (0.041) -0.046 (0.158) 0.720 (2.134) 0.696 (2.124) Rm 0.986*** (0.017) 0.967*** (0.017) 0.964*** (0.017) 0.985*** (0.049) -0.816 (0.981) -1.451 (0.994) SIZE 0.409*** (0.037) 0.410*** (0.037) 3.000*** (0.174) 3.019*** (0.174) GRO 1.451*** (0.112) 1.512*** (0.112) 1.263*** (0.186) 1.291*** (0.185) MNDUMMY 0.365 (0.258) 0.394 (0.261) SIZE*EXEUR 0.068** (0.033) 0.090*** (0.033) SIZE*EXGBP -0.087*** (0.026) -0.094*** (0.026) SIZE*EXUSD 0.037** (0.019) 0.030 (0.018) GRO*EXEUR 0.133 (0.105) 0.128 (0.102) GRO*EXGBP -0.240*** (0.084) -0.216*** (0.083) GRO*EXUSD 0.211*** (0.063) 0.144** (0.062) MNDUMMY*EXEUR 0.156 (0.238) 0.175 (0.232) MNDUMMY*EXGBP 0.307* (0.177) 0.295* (0.175) MNDUMMY*EXUSD -0.110 (0.134) -0.135 (0.135) FIXDUMMY 0.374** (0.151) FIXDUMMY*EXEUR 0.160 (0.376) 0.143 (0.440) FIXDUMMY*EXGBP 0.447*** (0.116) 0.591*** (0.111) FIXDUMMY*EXUSD -0.173** (0.083) -0.228*** (0.080)
Hausman test statistic 7.705 480.108*** 406.445***
Estimation specification RE FE FE
Durbin-Watson statistic 2.184 2.033 2.034 2.196 2.037 2.039
Adjusted R2 0.100 0.117 0.116 0.101 0.158 0.158
F-statistic 1,264.909*** 361.671*** 522.096*** 1,270.312*** 14.745*** 14.861***
28
returns and in the same direction with the returns on pound rate. There is also evidence that returns move in line with the Nordic all-share market portfolio index. The market return variable has a positive and significant at the 1% level relationship with stock returns. Controlling for firm size, firm growth opportunities and firm implication in foreign operations with MNDUMMY, it appears that stock returns are sensitive to the changes in size and growth opportunities but not statistically significant across MNCs. Holding all other variables constant, as the market value of equity increases with one million, the returns will increase by 0.409%. Likewise, independent of other variables, a one unit increase in the value of market to book equity ratio will increase the value of stock returns with 1.451%. The MNDUMMY is positive and statistically insignificant, meaning that, on average, stock returns follow a pattern which is unaffected by the fact that the firm is multinational or not. The interactions between control variables and exchange rate variables aim to investigate if the marginal effect of exchange rate returns is also affected by the firm characteristics. Of the three interaction variables, only SIZE has a significant and positive effect on the marginal effect of euro rate return on stock returns. This implies that stock returns are sensitive to euro exchange rate returns, and the level of exposure differs among firms with different size levels. The larger the firms, the less sensitive are stock returns to euro rate returns: the marginal effect becomes -1.374 (-1.442 + 0.068), which is less than -1.442. A possible explanation is that large firms have more opportunities to diversify currency risk.
The exposure of stock returns to pound rate is highly influenced by the size and growth opportunities. The larger the size and growth opportunities for the firm, the smaller is the exposure of stock returns to pound and dollar exchange rate returns: the calculated exposure to pound rate is 0.523 and 0.370 for size and growth opportunities, respectively and the calculated exposure to dollar rate is -0.377 and -0.203, respectively. In addition, the exposure to pound rate is influenced by the MNDUMMY. The stock returns of MNCs are higher exposed to pound rate: the exposure increases by 0.307 (from 0.610 to 0.917).
Considering the above findings and the fact that the coefficients are jointly different from zero (the F-statistic for regressions is rejected at the 1% level), the pooled OLS results show evidence of a significant relationship between stock returns and exchange rate returns. The results support the findings of the pooled OLS estimation of Ye, Hutson and Muckley (2014).
29
exchange rate returns show that the exchange rate regime matters in the analysis of currency exposure. The results show that under a fixed exchange regime, firms have significant exposure to dollar and pound rate and no exposure to the fixed currency – euro. The exposure beta increases in the pound rate from 0.093 to 0.354 and decreases in the dollar rate from 0.029 to -0.144.
Overall, the results of pooled OLS estimations support the hypothesis that stock returns are sensitive to exchange rate returns and that firms which are listed on stock exchanges from countries with fixed exchange arrangements are more exposed to fluctuated currencies and not significantly exposed to the pegged currency.
Panel B presents the results with fixed effects estimations. Notably, the null hypothesis that cross-section and time fixed effects are jointly equal to zero is rejected at the 1% significance level, meaning that there is cross-sectional and time variation in exposure. This supports the need to include fixed effects as a specification. In addition, MNDUMMY and FIXDUMMY are time-invariant variables and are not included as individual variables in the model. This is done
to avoid any possible multicollinearity issue. 8
Column 1 of Panel B presents the coefficients estimated with cross-section random effects. The Hausman test fails to reject the null hypothesis that coefficients estimated with random effects are the same as the ones estimated with fixed effects at the 10% significance level. The results of the four-factor model with cross-section random effects estimates are very similar to the ones of pooled OLS: there is evidence that stock returns are exposed to euro exchange rate returns. A 1% increase in the euro rate returns decreases stock returns by 0.706%. Also, stock returns move in the same direction with the market portfolio returns. Keeping all other variables constant, a 1% increase in the market portfolio returns increases stock returns by 0.985%. A contrasting evidence is presented in column 2, Panel B. By adding additional independent variables as controls, the model fails to support evidence of euro, pound and dollar exchange rate exposure on stock returns: the relationship of exchange rates as individual variables is statistically insignificant. The results indicate also that stock returns are not statistically sensitive to the market portfolio returns, but they are highly, positively related to the size and growth opportunities of the firms, meaning that stock returns are sensitive to the changes in market value of equity and market to book value of equity. Moreover, the results of two-way fixed
8 Estimations with cross-section fixed effects allow to include only time-varying independent variables in the model.
30
effects contradict the ones of pooled OLS. The interaction variables show evidence that large firms with high growth opportunities are more exposed to exchange rate risk. The larger the firm, the higher is the exposure to the euro and pound rate. The exposure betas increase to 2.404 and -2.570 respectively. The higher the firm growth opportunities, the higher is the exposure to pound and dollar rate. A 1% increase in the pound rate return will decrease the stock returns of firms with high growth opportunities by 2.692%, as compared to 2.476%. Likewise, a 1% increase in the dollar rate will increase stock returns of firms with high growth opportunities by 0.864%, as compared to 0.720%. MNCs have lower exposure to pound rate. A 1% increase in the pound exchange rate returns will decrease stock returns of MNCs by 2.181% as compared to 2.476%.
Thus, I conclude that exposure to exchange rates varies in line with the firm’ characteristics. Column 3 reports the results with the inclusion of control variables and interactions between exchange rate returns and FIXDUMMY. The results indicate different sign for the relationship between euro exchange rate returns and stock returns. There is a positive and significant relationship between euro rate returns and stock returns, a negative and significant relationship between pound rate returns and stock returns and no significant relationship at all between dollar rate returns and stock returns. This denotes that, on average, stock returns increase by 4.888% with the returns on euro rate and decrease by 3.975% with the returns on pound rate, keeping all other variables constant. The interactions of exchange rates with FIXDUMMY coefficients show that the exchange regime influence the exposure of stock returns to exchange rate returns, but the exposure can vary among different currencies. There is evidence that the exposure of the pound rate decreases from -3.975 to -3.384. As well, the exposure to dollar rate decreases from 0.696 to 0.468. Hence, the results fail to support the hypothesis that the exposure is more pronounced in firms from countries with fixed exchange arrangements.
As a robustness test, equations (2) and (3) are estimated with the inclusion of growth opportunities measured by the log of price to earnings ratio. Overall, the results show similar evidence in terms of statistical significance of the coefficients. The results are not included in the paper, but are available upon request.
To sum up, the results indicate evidence that firms have significant economic exposure and supports the findings of Choi and Prasad (1995), de Jong, Ligterink and Macrae (2002), among others. This implies that the first hypothesis is not rejected.
31
compared to the exposure of firms from countries with floating regime. Hence, the empirical results fail to support the findings of Rossi (2011), Ye, Hutson and Muckley (2014), Patnaik and Shah (2010), Parsley and Pooper (2006) and Kamil (2006).
5.2. Robustness analysis
This subsection reports the findings of several robustness tests. An analysis is performed to see if the choice of the independent variables was appropriate. Firstly, the return on the bilateral exchange rates is replaced with the return on the trade weighted currency index specific for each local currency in part. Secondly, the return on the all share Nordic portfolio index is replaced with the return on the equally-weighted market index. The same approach is undertaken by replacing the equally-weighted index with the value-weighted portfolio index. Further, a robustness test is conducted to check if the outliers detected in data influenced the exposure results. The next robustness approach splits the sample in three periods: pre-crisis, crisis and after-crisis period to see the reaction of stock returns to currency fluctuations in time. In addition, the models are adjusted to investigate if the results are robust among the lagged effects of exchange rate returns. The last robustness check aims to present evidence if the exposure results hold across different industries. Notably, all the robustness tests are performed with fixed effects estimates. Only in case when the Redundant Fixed Effects test in rejected on the 10% significance level, then the inferences are based on the pooled OLS estimates.
5.2.1. Exposure of stock returns to the trade weighted currency index returns.
Following Jorion (1990), de Jong, Ligterink and Macrae (2002), Doukas, Hall and Lang (2003), Fraser and Pantzalis (2004), Akay and Cifter (2014), Ye, Hutson and Muckley (2014) among others, I replace the bilateral exchange rates with a trade weighted currency index. The monthly average prices for the nominal effective exchange rates are obtained from the Denmark, Iceland and Sweden Central Banks for the DKK, ISK and SEK currencies, respectively. Then, the prices are converted in percentage of logarithmic returns using the formula specified in equation (4). The variable is checked for stationarity. The Maddala and Wu test fails to reject the null hypothesis of the presence of a unit root at the 1% significance level.
32
The results show that the model specified in Jorion (1990) fails to present evidence of significant exposure of stock returns to the returns of the trade weighted currency index. The evidence of significant exposure appears when the control variables are added to the model. Notably, the
adjusted R2increased for the second model, which implies that the model has a better fit in terms
of explained variance. The results now present evidence of highly statistically significant (positive) relationship between stock returns and trade weighted currency returns at the 1% level. This implies that a 1% increase in the trade weighted currency returns will increase stock returns by 0.106%. Moreover, the results show that firm size, level of growth opportunities and multinationality are not among the best predictors of economic exposure, when the exchange rate is measured with a trade weighted index. This supports the choice of the bilateral exchange rates. In addition, there is evidence that firms listed in countries with the fixed exchange rate regime are more exposed to exchange rate risk. The exposure for this firms increase to 0.355 and is characterized by an inverse significant relationship between stock returns and exchange rate returns.
To sum up, the findings are in line with the ones of de Jong, Ligterink and Macrae (2002). There is still evidence of exposure to the trade weighted exchange rate index, and this research confirms the fact that, in open economies firms are more exposed to currency risk. Moreover, the results are robust with the evidence that fixed exchange rate arrangements increase the economic exposure of firms.
5.2.2. Equally-weighted vs. value-weighted market portfolio index
Dominguez and Tesar (2006) and de Jong, Ligterink, and Macrae (2002) argue that the choice
of the market index can influence the results. In this section I replace the common equally-weighted market portfolio index with the local equally-equally-weighted market portfolio indices and then the local value-weighted market portfolio indices to investigate if the choice of the market index was appropriate and if this choice affects the results in terms of significance. The estimation results are reported in Table E2 from Appendix E.
Both series, the return on equally-weighted and the return on value-weighted market index are stationary. The regressions (six in total) present no evidence of autocorrelation (see Appendix D) and the estimates are corrected for heteroskedasticity.
