Foreign exchange rate risk on industry returns:
Empirical evidence from seven major economies
Author: Bastiaan Meijer Student number: 3827542 e-mail: b.p.meijer@student.rug.nl
University of Groningen Faculty of Economics and Business
MSc. Finance
Supervisor: Prof. Dr. W. Bessler Date: 4-6-2020
Abstract
This paper examines the effects of exchange rates on industry index returns of seven major countries for the period from 2003 to 2018 using a sequential or hierarchical decomposition method. This study finds evidence of relationships between industry returns and (foreign) exchange rate risks, beyond what the market portfolio already explains. The euro and US dollar expressed in the local currency and the trade weighted local currency, significantly affect industry index returns, which are also affected by changes in interest rates. This research also provides evidence that exchange rate risk has changed over time and is especially changing during economic shocks such as the financial crisis, the unpegging of the Swiss Franc and the official Brexit announcement.
1. Introduction
2 exogenous factors, such as the state of the world economy, affect currency exchange rates. For instance, the preceding presidential elections in the United States have had a short-term negative impact on the dollar and Brexit has weakened the position of the “Great British Pound” (GBP). An exchange rate is a measure of economic health of a country. For instance, the Venezuelan Bolívar to euro rate has plummeted since the political crisis occurred in Venezuela. Other than political stability, inflation and interest ratesaffect exchange rates as well as trading deficits, government debt and the ratio of export prices to import prices (terms of trade).
Currencies adjust continuously. And therefore, exchange rates have important implications for financial decision-making, especially for internationally operating firms.
Corporations that operate in countries with different currencies are likely to run exchange rate risk. If a firm receives cash flows in a foreign currency, it has to convert the cash flows to the domestic currency. The firm runs exchange rate risk on its foreign cash flows, because the value of the cash flows in its own currency might be higher (lower) in case of a depreciation (appreciation) of the domestic currency. In order to mitigate this risk a firm can hedge by using financial derivative instruments (e.g. options, forward contracts, futures contracts or swaps). With foreign exchange rate hedges, firms are able to lock-in a certain value of the convertible cash flows.
Several studies have tried to determine the impact of exchange rate exposure on organisations. For instance, Griffin and Stulz (2001) find that exchanges rates have a significant effect on stock market indices (e.g. Canada and Sweden). In addition, Zarei et al. (2019) find significant effects of exchange rates on stock market returns in seven countries applying free-floating exchange rate regimes. Floating exchange rates are an exchange-rate regime in which the currency level is determined by supply and demand (e.g. US dollar and the euro). Conversely, in a fixed exchange rate regime the domestic currency is dependent on another country’s currency (e.g. Chinese Renminbi).
3 that the relationship between exchange rates and stock returns changes over time. To account for this time variation, a rolling regression technique is applied in this study.
Chapter 2 comprises a literature review of both theoretical and empirical research. In chapter 3 data, methodology and proposed hypotheses are discussed. The empirical findings are presented and discussed in chapter 4. This chapter also comprises robustness checks and a discussion. Finally, chapter 5 contains a conclusion that contains important implications and limitations of this research.
2. Literature review
2.1 Theoretical research
The classical asset pricing approach is the CAPM, which relates the return on an asset to an economy’s risk-free rate and the market portfolio. The CAPM, originally developed by Sharpe (1964), is a single-factor model that estimates an individual stock’s return on the j-th security or an industry index as 𝑠𝑟𝑗:
𝑠𝑟𝑗 = 𝑟𝑓 + 𝛽1𝑗(𝑟𝑚 − 𝑟𝑓) (1)
where 𝑟𝑓 is the risk-free rate, (𝑟𝑚 − 𝑟𝑓) equals the risk premium in a market, 𝛽𝑗 measures the
systematic risk of the asset(s). An extension to the CAPM is the international CAPM of Solnik (1974), which accounts for exchange rate risk. This additional factor is the security’s risk premium over its national risk-free rate. Therefore, the return on the j-th security is:
𝑠𝑟𝑗 = 𝑟𝑓 + 𝛽1𝑗(𝑟𝑚 − 𝑟𝑓) + 𝛽2𝑗(𝐹𝐶𝑅𝑃) (2)
where 𝐹𝐶𝑅𝑃 denotes the foreign currency risk premium.
For bank and insurance industries, equation two is extended to a three-factor model where the interest rate is included. Hence, the return on the j-th security for bank and insurance industries is:
𝑠𝑟𝑗 = 𝑟𝑓 + 𝛽1𝑗(𝑟𝑚 − 𝑟𝑓) + 𝛽2𝑗(𝐹𝐶𝑅𝑃)+ 𝛽3𝑗(∆𝐼𝑅) (3)
where ∆𝐼𝑅 denotes the change in monthly interest rates.
4 In their research, Fama and French (1993) emphasize: “If there are multiple common factors in stock returns, they are all in the market return”. The market return is derived from a diversified portfolio of various stocks. When a multi-factor extension model of the CAPM is used, then factors are likely to be correlated with each other.
At least theoretically, factors that determine exchange rates are explained by two models: oriented models (Dornbush and Fisher, 1980) and stock-oriented models. According to flow-oriented models, trade balances and a country’s current account are the main factors that determine exchange rates. While according to stock-oriented models, capital accounts determine exchange rates. Portfolio balance models (Branson, 1983) and monetary models (Gavin, 1989) are two subsets of stock-oriented models. According to the portfolio balance model used by Branson (1983), increases in domestic stock prices drive up domestic interest rates, leading to a decrease in the exchange rate, suggesting a negative relationship between stock prices and exchange rates. According to the monetary model used by Gavin (1989), exchange rates and stock prices are affected by common factors, but there is no link between the variables.
Exchange rate risk is most likely a systematic (unavoidable) component that investors are subject to, either indirectly or directly. For international equity and debt investments there is a direct exposure. However, exposure also occurs in domestic operating firms when their input prices (such as oil) are dependent on exchange rates. Exchange rate risk is the risk that the value of capital changes due to movements in currency exchange rates. For instance, the value of cash flows for international operating firms that receive payments in a foreign currency is dependent on exchange rate movements between the domestic and foreign currency. To lock in the value of capital (hedge), forward, futures and options contracts (Hull, 2015) can be used. Firms have not been able to completely eliminate exchange rate risk using hedges (Olugbode et al., 2014). Nevertheless, employing hedges can improve the performance of international asset portfolios (Yu et al., 2018).
2.2 Empirical research
5 On the contrary, others find a relationship between exchange rates and stock prices. For instance, De Santis and Gérard (1998) analyse four industrialized countries and find evidence that exchange rate risk is priced within several stock market indices. Kanas (2000) finds negative correlation between stock returns and exchange rate changes for several industrialized countries and Mollick and Assefa (2013) find a negative effect of the USD/EUR for US stock returns. For emerging economies, such as Brazil, India and Russia, Reboredo et al. (2015) find a positive relationship between stock prices and currency values. Using a pooled regression model, Lee et al. (2018) performed research on the pricing of exchange rate risk on individual stocks within several industries and find that exchange rate risks are priced in the US stock market for most firms. Zarei et al. (2019) find that changes in exchange rates significantly affect stock returns for several free-floating currency countries in the period from 1999 until 2016.
An extensive amount of research finds evidence of a relationship between stock returns and exchange rates. However, the sign of the relationship appears to depend on the sample, e.g. whether firms are exporters or importers.
Another interesting aspect of factors explaining stock returns, such as exchange rates, is that these are time-varying. For instance, Wetmore and Brick (1994) find that exchange rate risk varies over time for banks. Furthermore, Guo et al. (2011) find that exchange rate risk is time-varying, but also indicate that co-movement of asset prices strengthen during unstable times, such as crisis periods. Caporale et al. (2014) confirm this, as they find that correlation between bank stock returns and exchange rates is time-varying during the banking crisis in several Western countries. Furthermore, Boudt et al. (2019) find that exchange rate risk changes over time for US multinationals from May 2008 to December 2014.
Time variation can have a meaningful impact on regressions, as it implies that parameters of regression models can change over time. If not accounted for, this can lead to forecasting errors and an unreliable model. Time variation can be detected if a test for structural breaks is applied. In case of structural breaks, properties of a model exhibit a large shift. Structural breaks are caused by large and unexpected shocks, which surface new information brunches (Brooks, 2019). The new information may cause a long-term shift in parameters. Often, financial time series exhibit characteristics, such as non-normality and volatility clustering, that have several implications when using linear regressions model.
6 have the same impact on volatility. However, leverage effects are often observed in broad financial stock indices. Leverage effects imply that negative shocks have larger impact than positive shocks (Magnus and Fosu, 2006) on volatility, hence, the effect of shocks tends to be asymmetric.
3. Data, methodology and hypotheses
This section, based on Brooks (2019), outlines the data and methodology used for the empirical analysis. First, data collected in this study will be discussed. Hereafter, the methodology is reviewed and finally, the hypotheses are stated.
3.1 Data
Monthly times series data of industry indices is gathered from January 2003 to December 2018 and retrieved from Thomson Reuters Eikon. Countries within Europe with a common currency (Italy and Germany) and without a common currency (Sweden, Switzerland and UK), as well as countries from Asia with a developed economy (Japan) and developing economy (China) have been selected. At least ten industries per country are selected. The industries comprise bank, (life) insurance, utilities, pharmaceutical and biotech, healthcare, technology, chemicals, basic resources, automobile, consumer goods and construction. Moreover, monthly time series data for the full sample period on long-term government bond yields, bilateral and trade weighted exchange rates was retrieved from the Federal Reserve Bank of St. Louis (FRED) and Investing.com.
3.2 Methodology
The aim of this research is to investigate whether and how changes in foreign exchange rates and interest rates affect industry index returns. Monthly index returns are calculated according the following formula:
𝑅𝑖𝑡 = 𝐶𝑖𝑡0
𝐶𝑖𝑡−1 − 1 (4)
where 𝑅𝑖𝑡 represents the return on index i at time t and 𝐶 represent the close price of index i at time t.
7 and interest rates increases the explanatory power. If the variables are significant, this will reveal that foreign exchange rate and interest rate sensitivity affect industry stock returns, beyond the market portfolio (index). The market portfolio is used as control variable.
The relationship between industry index returns and the foreign exchange rates and interest rates is examined using a multifactor index model similar as used in Choi et al. (1992):
𝑅𝑖𝑡 = 𝛼𝑖𝑡+ 𝛽1 𝑅𝑟𝑡+ 𝛽2 𝑅𝑒𝑡+ 𝛽3 𝑅𝑚𝑡+ ∈𝑖𝑡 (5) where 𝑅𝑟𝑡 represents the change in long-term government bond yields at time t,
𝑅𝑒𝑡 represents the return on bilateral foreign or trade weighted exchange
rate at time t,
𝑅𝑚𝑡 represents the return on the market portfolio index at time t, ∈𝑖𝑡 represents the residuals for index i at time t.
Using equation 5, the relationship between industry index returns and the above stated independent variables are estimated via Ordinary Least Squares (OLS) regressions. In order for OLS to be the Best Linear Unbiased Estimator (BLUE), the data must be homoscedastic and variables must be uncorrelated with the error term. Another implicit assumption that is made when using OLS is that explanatory variables are uncorrelated with each another. If the data exhibits violations of these Gauss-Markov assumptions and violations are ignored, then OLS estimators will not have the minimum variance among unbiased estimators, i.e. be inefficient. Therefore, statistical tests will be conducted to find out whether these assumptions of OLS are violated.
3.2.1 Statistical tests for heteroscedasticity and autocorrelation
Often conditional heteroscedasticity is observed in financial time series data, consequently, it is expected in the data of this research. White’s (1980) general test for heteroscedasticity, tests whether the error variance is constant i.e., whether the data exhibits homoscedasticity. First, a regression model is estimated via OLS with the regular dependent and independent variables. Then, the squared residuals of this regression are regressed on a constant, the original dependent variables, the squares of the dependent variables and their cross-products. Next, a Lagrange Multiplier (LM) test is carried out, which follows a chi-square distribution. The LM test obtains the R² from the auxiliary regression and multiplies this by the number of observations. If the joint null hypothesis, which states that the dependent variables of the auxiliary regression are equal to zero, is rejected, then the errors are heteroscedastic. Adjusting OLS to obtain minimum variance estimators in the presence of heteroscedasticity, requires application of heteroscedastic robust standard errors.
8 autocorrelation is the Breusch-Godfrey test. This is a test for serial correlation between the error terms. The Breusch-Godfrey is an advanced test that tests for autocorrelation up to any lagged value of the error. When conducting this test, an appropriate number of lags should be specified to test for autocorrelation up to that lag. This immediately rises the potential difficulty with this test: deciding on the appropriate number of lags. Since the data in this sample are monthly, the number of lags is set to 12. This assumes that errors are expected to be related only to error values in the previous year. The null hypothesis of the Breusch-Godfrey test states no serial correlation between errors. If the null is rejected, the errors exhibit serial correlation up to 12 lags. Adjusting OLS estimators to account for serial correlation and obtain efficient estimators, standard errors that are robust to serial correlation should be applied.
When OLS estimators exhibit heteroscedasticity or serial correlation, standard errors that are robust to these forms should be applied. In several circumstances OLS estimators exhibit both heteroscedasticity and serial correlation. A convenient solution to this problem is to apply heteroscedasticity- and autocorrelation-consistent (HAC) standard errors. An example of these is the HAC standard errors proposed by Newey and West (1987). The Newey and West HAC standard errors should be applied when the data is subject to both heteroscedasticity and serial correlation.
3.2.2 Statistical test for multicollinearity
This research applies a multifactor model, including a market portfolio. As Fama and French (1993) indicate, it is likely that the market portfolio is correlated with other independent factors. When independent variables are highly correlated i.e., not orthogonal to one another, then a problem occurs called multicollinearity. Multicollinearity is often observed when a model explains much of the variation, but produces insignificant individual explanatory variable. An additional effect of multicollinearity is that small changes in the model specification, lead to large changes in coefficient values.
In order to measure multicollinearity, a correlation matrix and variance inflation factors (VIF’s) are calculated. VIF provides an explanation to what extent variance of coefficient estimators rises, due to correlation between dependent variables. When VIF gives a variable a high value, multicollinearity is present within the model. As a rule of thumb, when a VIF is higher than 5.0 it is assumed that multicollinearity is non-negligible and has to be dealt with. In addition, a correlation matrix is presented.
9 3.2.3 Statistical tests for structural breaks and time variation
Another assumption of OLS regressions is that parameters are constant for the entire sample period. However, empirical research provided evidence that exchange rate risk is time-varying. In order to test that parameters are constant the Chow test is performed. The Chow test offers the opportunity to test for structural breaks within the sample data. First a normal regression containing the complete sample data is performed. Then the Chow test can provide evidence whether there is and when there is a structural break within the sample data. Consequently, a second (before the structural break) and third (after the structural break) regression will be performed. The Chow test then provides evidence whether regressions with full sample data or with sub-sample data should be performed.
The Chow test only tests for one structural break. However, there could be multiple structural breaks within data. Therefore, as an extension to the Chow test, a rolling regression technique is applied to research the time variability of the regression coefficients. An estimation period of 30 months is applied, with a 1-month shift for every regression. Rolling regression results will be graphed to find out whether parameters change over time more than once. Consequently, several of these structural breaks will be tested as sub-periods.
3.2.4 Statistical test for normality and ARCH Effects
Often, financial time series data are not normally distributed. For hypothesis testing about model parameters, it is required that disturbances are normally distributed, according a bell-shaped curve. To test for normality, a skewness (symmetry around mean) and kurtosis (‘fat tails’) test is conducted on regression residuals. Originally, the test was designed by D’Agostini et al. (1990), but empirically corrected by Royston (1991c). One source for non-normality may be that the residuals contain outliers, which are observations that deviate from an overall pattern on a sample. Outliers may arise due to extreme events such as an outbreak of a crisis. Another source of non-normality in financial data, could be a certain type of heteroscedasticity, namely autoregressive conditional heteroscedasticity (ARCH) or volatility clustering. Therefore, a test for ARCH effects is performed to determine whether the errors exhibit autocorrelation. Engle’s LM test will be applied, with a lag of 12 months, to test for ARCH effects. In this research, a minority of industries are subject to ARCH effects. Hence, (E)GARCH models are used to account for ARCH effects.
3.2.5 Statistical test for stationarity
10 result of Akaike’s information criterion less one. The null hypothesis states dependent variable contains a unit-root. When the null hypothesis is rejected, the variable follows a stationary process.
3.3 Hypotheses
The objective of this research is to determine whether there is a significant effect of different exchange rates on industry index returns for each of the seven economies during the period from January 2003 to December 2018. Moreover, during 2008, there was a worldwide financial credit crisis. The financial credit crisis may have caused a structural break in the data. A structural break causes a change in the parameters of a model. Therefore, this research aims to find a difference between the effect of exchange and interest rate risk on industry index returns before and after the financial credit crisis. This requires a test of differences between these two periods. In addition to the financial crisis, the scrapped peg of the Swiss Franc to the euro in January 2015 and the official Brexit announcement in June 2016 are analysed similarly. The main research questions in this paper is: Do changes in foreign exchange rates and interest rates significantly affect industry index returns?
In order to be able to answer the research question, the following hypotheses are being tested: Hypothesis 1: There is significant effect of exchange and interest rates on industry index returns.
Hypothesis 2a: There is significant change in effect of exchange and interest rates on industry index returns during the financial crisis.
Hypothesis 2b: There is significant change in effect of exchange and interest rates on industry index returns since the unpegging of the Swiss Franc to the euro.
Hypothesis 2c: There is significant change in effect of exchange and interest rates on industry index returns since the Brexit announcement.
4 Empirical results
This chapter presents the empirical results of the orthogonal time series regressions. Section 4.1 comprises descriptive statistics of the sample. Section 4.2 discusses the full sample period results per risk factor. In section 4.3 specific periods regression results are discussed. Additionally, section 4.4 discusses robustness checks of the empirical results. Finally, section 4.5 contains a discussion on this research.
11 distinguishes itself by focussing on the effect of the bilateral euro. Therefore, the order of the variables is as follows:
1. The first factor is the euro expressed in the local currency.
2. The second factor is the US dollar expressed in the local currency. 3. The third factor is the trade weighted local currency.
4. The final (control) factor is the market portfolio, which is a country’s national index. As mentioned in chapter 3, for bank and insurance industries the factor model comprises an additional input factor; the interest rate factor. Empirical research finds that bank and insurance stock returns are significantly affected by interest rate changes (Elyasiani & Mansur, 1998). Therefore, the interest rate becomes the first factor for bank and insurance industries, followed by the abovementioned order of factors.
For Germany and Italy, the euro is the local currency, which implies that there is no bilateral euro. Consequently, for these countries the model has dropped the bilateral euro factor. The trade weighted euro is used to account for euro effects. The order of factors for Germany and Italy is (interest rate,) US dollar, trade weighted euro and market portfolio, respectively.
4.1 Descriptive statistics
Table 1 provides descriptive statistics for all industries across the seven countries. Most industries comprise the full sample period from January 2003 to December 2018. However, for some industries data was only available from 2005 or 2009. On average, most industry returns are positive, except for banks in the UK and Germany, technology in the UK and utilities in Sweden.
Appendix 1 presents a correlation matrix between regressors by country, and figure 1 and 2 show the exchange rate factors for Japan and the United Kingdom over time. Consistently with what Fama and French (1993) suggest, the factors are correlated. At least two of the exchange rate factors are highly correlated (>0.45) for every country. This indicates multicollinearity between the regressors. To resolve the multicollinearity problem, regressors are made orthogonal using a sequential or hierarchical (variance) decomposition method.
Table 1. Descriptive statistics by country and industry.
13 Automobile 192 1.1215% 0.0790 -0.3063 0.2300
14 Figure 1. Foreign exchange rates over time for the United Kingdom.
4.2 Full period results
This section describes the empirical results for the full estimation period. Via sequential decomposition, the regression results are made orthogonal. The model factors are discussed separately. First, the interest rate factor (4.2.1) on bank and insurance industries is reviewed. Hereafter, the effects of the bilateral euro (4.2.2) and US dollar (4.2.3) exchange rates are discussed. Finally, the effects of the trade weighted local currencies (4.2.4) are reviewed. The intercept and market portfolio factors of the orthogonal regressions are presented in appendix 3.
4.2.1 Interest rate
Table 2 presents the time series regression results for the interest rate factor by country and industry with both robust and Newey West standard errors. The interest rate factor is significant for banks in all countries. Japanese banks returns are most sensitive to interest rate changes with a positive coefficient of 0.2583. Banks from the UK are least sensitive to interest rate changes with a positive coefficient of 0.0680. Where bank returns in Sweden and Switzerland are similarly sensitive to interest rate changes, the coefficients of Germany and Italy are divergent. Where Germany has a positive coefficient of 0.1543, Italian banks have a negative coefficient of 0.1035. As figure 3 graphs, there is a noticeable difference between German and Italian interest rates around 2012. During the end of the financial crisis, Italian banks were subject to a much larger default risk than German banks, caused by the high debt burdens of Italy (Alter and Schüler, 2012). The higher default risk between the countries, might explain the difference between interest rate changes and banks’ exposure to interest rate risk.
16 In most countries, the interest rate factor has a significant effect on the insurance industry. Only in the UK and China, the interest rate factor remains insignificant. Again, Japan has the highest exposure to the interest rate factor, with a positive coefficient of 0.3036. Sweden and Switzerland have similar coefficients. Whereas the Italian insurance industry reacts negatively to interest rate changes, with a coefficient of -0.0774, Germany has a positive coefficient of 0.0904. Again, the high default risk of Italian firms may explain the difference between the coefficients of the German and Italian industries. Overall, the findings that interest rates affect industry index returns are consistent with other empirical results (Bessler and Murtagh, 2004 and Olugbode et al., 2014).
The null hypothesis that interest rates have no significant effect on industry returns is, in favour of the alternative, rejected for all bank industries and all insurance industries, except for China, Sweden and the United Kingdom.
Table 2. Interest rate factor coefficients for bank and insurance industries in all countries. Interest rate factor Industry Robust standard errors Newey West standard errors United Kingdom Bank 0.0680** 0.0680**
Insurance 0.0159 0.0159 Sweden Bank 0.0922** 0.0922*** Insurance 0.0558* 0.0558** Switzerland Bank 0.1645*** 0.1645*** Insurance 0.0654** 0.0654** Germany Bank 0.1543*** 0.1543*** Insurance 0.0904** 0.0904* Italy Bank -0.1094*** -0.1094*** Insurance -0.0774*** -0.0774*** Japan Bank 0.2583*** 0.2583*** Insurance 0.3036*** 0.3036*** China Bank -0.1035** -0.1035** Life insurance -0.1283 -0.1283 *** p<0.01, ** p<0.05, * p<0.1 4.2.2 Euro
17 The null hypothesis that the bilateral euro exchange rate has no significant effect on industry returns is rejected for 38 industries across all countries. As can be observed from table 3, the bilateral euro exchange rate affects most industries. Significant coefficients lie within the range of 1.0710 (technology, Switzerland) and -1.0878 (chemicals, Sweden). As expected, the euro has an impact on industries in countries that are close to the Eurozone. Both Sweden and Switzerland are exposed to changes in the euro exchange rate. Specifically, the effect of the euro on Swiss industries is highly significant and positive. One explanation is that the Swiss Franc was pegged to the euro between September 2011 and January 2015. For Sweden, the coefficients are significant and negative. Sweden’s economy relies heavily on its trades with the European Union. Import from EU countries to Sweden is larger than Sweden’s export to EU countries, which might explain why an appreciation of the euro has a negative effect on Swedish industries. Interestingly, the effect of the euro on UK’s industries is limited, although the EU is the UK’s largest trading partner. However, not all industries are affected by the euro. This might be explained by the low adjusted R², which is derived through a sequential variance decomposition and presented in appendix 4. For industries where the euro has a significant impact, the bilateral euro explains a substantial part of the variation. For these cases, the size of the effect is relatively small compared to other countries.
Another interesting finding is that the euro has an effect on all Japanese industries and most Chinese industries. Both countries’ industries reflect positive return increase for every return unit increase in the bilateral exchange rate. The effect of the euro on these countries is expected, because both countries are extremely export oriented.
Table 3. Bilateral euro factor coefficients by country and industry.
Euro € Industry Robust standard errors Newey West standard errors
United Kingdom Bank 0.2885 0.2885
Insurance 0.2690 0.2690
Utilities 0.2991** 0.2991**
Pharmaceuticals & Biotech 0.3371** 0.3371**
18 Automobile -0.7756*** -0.7756** Consumer goods -0.4839** -0.4839* Construction -0.3184 -0.3184 Switzerland Bank 0.8119*** 0.8119*** Insurance 0.8169*** 0.8169** Utilities 0.4328** 0.4328**
Pharmaceuticals & Biotech 0.3322** 0.3322**
Healthcare 0.4191*** 0.4191*** Technology 1.0710*** 1.0710*** Chemicals 0.5536*** 0.5536*** Basic resources 0.5698*** 0.5698*** Automobile 0.3020 0.3020 Consumer goods 0.4236*** 0.4236** Construction 0.4879* 0.4879*** Japan Bank 0.8503*** 0.8503*** Insurance 0.9666*** 0.9666*** Utilities 0.9891*** 0.9891***
Pharmaceuticals & Biotech 0.3689*** 0.3689***
Healthcare 0.4516*** 0.4516*** Technology 0.9501*** 0.9501*** Chemicals 0.7282*** 0.7282*** Basic resources 0.9332*** 0.9332*** Automobile 1.0087*** 1.0087*** Consumer goods 0.8991*** 0.8991*** Construction 0.6215*** 0.6215*** China Bank 0.7362*** 0.7362*** Life insurance 0.9554*** 0.9554*** Utilities 0.5399** 0.5399***
Pharmaceuticals & Biotech 0.2390 0.2390
Technology 0.4185 0.4185 Chemicals 0.6475** 0.6475** Basic resources 0.8840*** 0.8840*** Automobile 0.7380** 0.7380*** Consumer goods 0.6138** 0.6138*** Construction 0.6010** 0.6010*** *** p<0.01, ** p<0.05, * p<0.1 4.2.3 US dollar
Table 4 provides the results of the US dollar on industry index returns, after the effects of the euro have been taken care of. The dollar is expressed in the local currency.
19 currency has a negative impact on industry returns. For Japan, however, an increase in the USD/JPY causes an increase in Japanese industries’ returns.
As mentioned previously, the Swiss Franc was pegged to the euro and, which might explain why Swiss industries were highly affected by the bilateral euro exchange rate. The Chinese Renminbi was pegged to the US dollar. In addition, the United States is China’s largest trading partner. Therefore, one could expect the dollar to have an impact on the Chinese industries. However, there is no evidence found in this sample that the US dollar impacts Chinese industries.
Compared to the euro, the dollar has relatively little impact. This seems unexpected as the US dollar is the dominant currency in the world. For instance, several important products such as oil, are solely traded in US dollars. For all Chinese industries, the US dollar is insignificant. One source that the US dollar is insignificant, might be due to the order of orthogonalization. Therefore, at section 4.4 empirical analysis is presented, where the effects of the bilateral US dollar are being taken into account, before the effects of the trade weighted local currency and bilateral euro exchange rate.
Table 4. Bilateral US dollar factor coefficients by country and industry.
Dollar $ Industry Robust standard errors Newey West standard errors United Kingdom Bank -0.5660** -0.5660**
Insurance -0.5465** -0.5465**
Utilities -0.0752 -0.0752
Pharmaceuticals & Biotech 0.1258 0.1258
Technology -1.1018** -1.1018** Chemicals -0.6763*** -0.6763*** Basic resources -0.5703 -0.5703 Automobile -0.9373** -0.9373** Consumer goods -0.0281 -0.0281 Construction -0.4833 -0.4833 Sweden Bank -0.4255** -0.4255* Insurance -0.1290 -0.1290 Utilities -0.4120 -0.4120* Healthcare 0.1498 0.1498 Technology 0.1707 0.1707 Chemicals -0.1334 -0.1334 Basic resources -0.3521** -0.3521 Automobile -0.0693 -0.0693 Consumer goods -0.1392 -0.1392 Construction -0.1133 -0.1133 Switzerland Bank -0.0030 -0.0030 Insurance -0.1769 -0.1769 Utilities -0.2228* -0.2228**
20 Healthcare 0.0941 0.0941 Technology -0.2724 -0.2724 Chemicals 0.0304 0.0304 Basic resources 0.0176 0.0176 Automobile -0.5396 -0.5396 Consumer goods 0.0345 0.0345 Construction -0.2450 -0.2450 Germany Bank -0.9083*** -0.9083** Insurance -0.4610** -0.4610* Utilities -0.4375*** -0.4375***
Pharmaceuticals & healthcare 0.0833 0.0833
Technology -0.3298* -0.3298 Chemicals -0.3945** -0.3945 Basic resources -0.6097*** -0.6097*** Automobile -0.0309 -0.0309 Consumer goods -0.1867 -0.1867 Construction -0.4252* -0.4252 Italy Bank -0.7804*** -0.7804*** Insurance -0.5959*** -0.5959** Utilities -0.4321*** -0.4321**
Pharmaceuticals & Biotech -0.1600 -0.1600
Technology 0.0107 0.0107 Chemicals -0.4197 -0.4197 Basic resources -0.1907 -0.1907 Automobile -0.7505** -0.7505*** Consumer goods -0.4647** -0.4647* Construction -0.4954*** -0.4954* Japan Bank 0.2280 0.2280 Insurance 0.1641 0.1641 Utilities 0.1253 0.1253
Pharmaceuticals & Biotech 0.1923 0.1923
Healthcare 0.1911 0.1911 Technology 0.4345*** 0.4345** Chemicals 0.3261** 0.3261** Basic resources 0.3563** 0.3563* Automobile 0.7470*** 0.7470*** Consumer goods 0.5581*** 0.5581*** Construction 0.1829 0.1829 China Bank -0.7552 -0.7552 Life insurance -0.0958 -0.0958 Utilities -0.5921 -0.5921
Pharmaceuticals & Biotech -1.3389 -1.3389
Technology -1.6581 -1.6581
Chemicals -1.0278 -1.0278
Basic resources -1.0761 -1.0761
21
Consumer goods -1.1878 -1.1878
Construction -0.4061 -0.4061
*** p<0.01, ** p<0.05, * p<0.1 4.2.4 Trade weighted local currency
Table 5 presents the effects of local trade weighted currencies, after the effects of the euro and US dollar have been taken care of. The null hypothesis that the trade weighted local currency has no significant effect on industry returns is rejected for 11 industries in Germany, Italy, Japan and Switzerland. Coefficients are ranging between -.2547 (construction, Japan) and -0.9712 (basic resources, Germany). Even after the effects of the bilateral euro and US dollar have been taken care of, the domestic currency still has an effect for several industries. This indicates that these industries are especially exposed to changes in the local trade weighted currency.
Table 5. Trade weighted local currency factor by country and industry.
Local currency Industry Robust standard errors Newey West standard errors United Kingdom Bank -0.0731 -0.0731
Insurance -0.0849 -0.0849
Utilities 0.2613 0.2613
Pharmaceuticals & Biotech 0.1782 0.1782
Technology 0.4930 0.4930 Chemicals -0.1906 -0.1906 Basic resources 0.6007 0.6007 Automobile -0.0070 -0.0070 Consumer goods 0.0063 0.0063 Construction 0.4479 0.4479 Sweden Bank -0.2290 -0.2290 Insurance -0.0480 -0.0480 Utilities 0.4280 0.4280 Healthcare -0.0372 -0.0372 Technology 0.0201 0.0201 Chemicals -0.0490 -0.0490 Basic resources -0.4652 -0.4652 Automobile 0.1336 0.1336 Consumer goods -0.0683 -0.0683 Construction -0.1845 -0.1845 Switzerland Bank -0.3935 -0.3935 Insurance -0.8565* -0.8565** Utilities -0.6497* -0.6497*
Pharmaceuticals & Biotech -0.4127* -0.4127**
Healthcare -0.4598** -0.4598***
Technology -0.5899 -0.5899
Chemicals -0.2952 -0.2952
Basic resources -0.2759 -0.2759
22 Consumer goods -0.4260* -0.4260** Construction -0.9243** -0.9243*** Germany Bank -0.6408 -0.6408 Insurance -0.4270 -0.4270 Utilities -0.1528 -0.1528
Pharmaceuticals & healthcare -0.3263 -0.3263
Technology -0.3039 -0.3039 Chemicals -0.5898* -0.5898 Basic resources -0.9712*** -0.9712** Automobile -0.6647* -0.6647 Consumer goods -0.3936 -0.3936 Construction -0.5752 -0.5752 Italy Bank -0.5961* -0.5961** Insurance -0.5705* -0.5705* Utilities -0.3376 -0.3376
Pharmaceuticals & Biotech -0.8030** -0.8030*
Technology -1.0040* -1.0040* Chemicals -0.0519 -0.0519 Basic resources -0.9464** -0.9464** Automobile -0.9269* -0.9269* Consumer goods -0.7221* -0.7221* Construction -0.4562 -0.4562 Japan Bank -0.2937 -0.2937** Insurance -0.2828 -0.2828* Utilities -0.5190** -0.5190**
Pharmaceuticals & Biotech -0.1989 -0.1989
Healthcare -0.2439 -0.2439** Technology -0.3758*** -0.3758*** Chemicals -0.3673*** -0.3673*** Basic resources -0.4315*** -0.4315*** Automobile -0.4452*** -0.4452*** Consumer goods -0.3826*** -0.3826*** Construction -0.2574* -0.2574** China Bank 0.0089 0.0089 Life insurance -0.5433 -0.5433 Utilities 0.4327 0.4327
Pharmaceuticals & Biotech -0.0740 -0.0740
23
4.3 Financial crisis and currency shocks
In this section, parameter changes during unstable periods are highlighted and discussed. First, the 2008 credit crisis is discussed. Hereafter, the January 2015 scrapped peg of the Swiss Franc (CHF) to the euro is reviewed. Finally, focus is put on the official Brexit announcement in June 2016.
4.3.1 Financial credit crisis
On the 15th of September 2008, Lehman Brothers collapsed and had to declare bankruptcy.
Many other large corporations followed. The collapse of Lehman Brothers is marked as the outbreak of the financial credit crisis. It was the largest impact a crisis would have on the world economy, since the great depression in the 1930s. The financial credit crisis had a large impact on the world economy and it triggered unexpected changes for corporations around the world. Moreover, the crisis had such an impact, that parameters drastically changed over time. Therefore, table 6 reports the coefficients with robust standard errors of before and since September 2008. Only industries with at least both 30 observations prior and since the financial crisis are analysed. The null hypothesis that the financial crisis has not led to a significant shift exchange and interest rate risk is rejected for 43 industries in all seven countries. Figures 4 and 5 are examples of rolling regression estimates, with an estimation window of 30 months (2.5 years), which show that estimates are changing over time for the bank and technology industry in the UK and Switzerland, respectively.
The results provide evidence that interest rate risk has significantly changed for bank industries for the UK, Switzerland, China and Italy. This indicates that the effect of the financial crisis caused the interest rate risk for banks to shift for these countries. Since the outbreak of the financial crisis, central banks lowered interest rates. Especially during crises, interest rates are important as they represent the price of liquidity (Altavilla et al., 2017). The largest positive shift in interest rate parameter is for banks in China (0.1888). Italy has the largest negative shift in the interest parameter (-0.1178). The change in the interest rate factor for the UK (0.1320) and Switzerland (0.1694) appears to be similar. However, for Sweden the shift is insignificant, indicating that the financial crisis didn’t have a drastic impact on interest rate risk. Hansen and Welz (2011) explain this by referring to a smaller increase in the risk premia for Swedish banks compared to the rest of the euro area, United Kingdom and United States.
For Japan and Germany, the shift in interest rate risk is insignificant. Interestingly, banks in Italy are affected by the (negative) shift in interest rate risk, which is probably caused by the high default risk of Italian banks.
24 Switzerland). For the UK changes are minor, as effects of the bilateral euro did only change for its pharmaceuticals and biotech industry. For Swiss industries, the effects of the bilateral euro diminished sharply, as the Swiss Franc appreciated steadily. On the contrary, the effect of the bilateral euro increased for Japanese industries. During the financial crisis, the euro depreciated greatly vis-à-vis the Japanese Yen. Moreover, as table 7 presents, the correlation between Japanese industry returns and the bilateral euro exchange rate intensified. Especially during the financial crisis, the correlation increased sharply.
Figure 4. Rolling regression estimates bank industry UK.
Significant changes in the bilateral US dollar are found in all countries except for China and Japan. The Yen and Renminbi both appreciated vis-à-vis the US dollar. Parameter changes are all negative ranging between -0.5412 (healthcare, Sweden) and -2.0475 (automobile, UK). Almost all significant changes cause US dollar parameters to switch sign. Coefficients change from positive, before the financial crisis, to negative sign, since the financial crisis. During the financial crisis, the US dollar appreciated sharply against a lot of other currencies. The ECB (Fratzscher, 2009) explains that there were three factors that have caused the US dollar to appreciate. First, countries’ financial liabilities with regards to the United States. Second, cash reserves of domestic Central Banks were insufficient to intervene of a sharp depreciation of domestic currencies. Finally, countries with a large trading deficit experienced a sharp depreciation of the local currency.
25 affected before and after the financial crisis. Changes of the parameters range between 2.5035 (utilities, Sweden) and 1.0959 (utilities, Switzerland). Interestingly, all countries are within the Eurozone. Where Sweden has only its utilities industry affected by a significant change, Switzerland has a significant change for its utilities, technology and construction industries. Germany has a significant change for its basic resources and construction industries. Finally, the construction industry in Italy is significantly affected.
Figure 5. Rolling regression estimates technology industry Switzerland.
4.3.2 Currency peg Switzerland
On the 15th of January 2015, Switzerland announced that it was going to scrap its peg of the
Swiss Franc to the euro. This information came at a surprise, causing the Swiss Franc to appreciate by 20%. The effect of the currency shock on Swiss industries that followed after the announcement of the Swiss National bank is analysed. Consequently, the data of Switzerland is split into the period before the announcement and after the announcement. Table 8 reports the regression results of the Swiss currency shock. Only industries with at least both 30 observations prior and since the Swiss currency shock were analysed. The null hypothesis that the unpegging has not led to a significant shift in at least one parameter is rejected for five industries in Switzerland.
26 bilateral US dollar on the automobile industry changed drastically from -1.0318 to 0.7036. A similar change happened for the trade weighted Swiss Franc for the insurance and construction industries. The parameter switched sign for both industries.
After the unexpected announcement of its unpegging, the Central Bank of Switzerland created uncertainty. Speculators accelerated the purchase of the Swiss Franc, which caused it to appreciate by over 20% (Zhu, 2016). Obviously, the unpegging resulted in a change of interest and exchange rate effects for several Swiss industries. Interestingly, the industries of which the sensitivity with regards to exchange rates has significantly changed all became less sensitive to exchange rate risk.
4.3.3 Brexit announcement
The results of the Brexit referendum held on the 23rd of June 2016, led to a currency shock of
the Pound Sterling. After the announcement, the currency instantly decreased by more than 10% in value. Consequently, the sample of the UK is split into two periods. The first period is before the Brexit announcement, whereas the second period comprises the period since the Brexit announcement in June 2016. Table 9 reports the regression results before and after the Brexit announcement. Only industries with at least both 30 observations prior and since the Brexit announcement were analysed. The null hypothesis that the Brexit announcement has led to no significant shift in at least one parameter is rejected for four industries in the United Kingdom.
The parameter of the Euro expressed in Pound Sterling significantly changed for the insurance, utilities and consumer goods industries. This parameter switched sign for the insurance industry by changing from 0.4253 to -0.4863. For the utilities industry the coefficient increased from 0.1982 to 0.9174. Finally, for the consumer goods the parameter increased from 0.1597 to 0.5154. The US dollar parameter switched sign for the utilities industry by changing from -0.2075 to 0.5548. On the contrary, the coefficient of the trade weighted Pound Sterling decreased from 1.0194 to -0.8984 for the technology industry.
Overall, industries which experienced a significant change in sensitivity to exchange rate risk, show greater sensitivity to the bilateral euro and US dollar exchange rates after the Brexit announcement. However, sensitivity to the Pound Sterling has decreased.
4.4 Robustness checks
27 Table 6. Factor coefficients and changes for sub-samples by country and industry.
United
Kingdom Before crisis Since crisis Difference
Interest
rate EUR USD GBP
Interest
rate EUR USD GBP
Interest
rate EUR USD GBP Bank -0.0221 0.2572 0.4058** 0.2221 0.1099*** 0.2950 -0.9680*** -0.1524 0.1320** 0.0378 -1.3738*** -0.3746 Insurance -0.0755 0.2789 0.4634** -0.1314 0.0455 0.2704 -0.8243*** -0.0841 0.1210* -0.0085 -1.2876*** 0.0474 Utilities -0.2685 0.1421 0.6334 0.3482*** -0.1214 0.2124 0.6167** -0.2635 -0.421 Pharmaceuticals & Biotech 0.1307 0.3731* 0.4395 0.3796** 0.0206 0.1095 0.2489 -0.3525 -0.3301 Technology -0.9956 -0.9634** 0.8551 -0.2271 -1.1615** 0.4061 0.7685 -0.198 -0.449 Chemicals -0.0548 -0.093 0.4293 0.5715*** -0.9110*** -0.3462 0.6263* -0.8181** -0.7755 Basic resources -0.926 -0.1472 1.2705 -0.5271 -0.6544 0.5046 0.3989 -0.5072 -0.7659 Automobile -0.2911 0.6662 -0.2587 0.1847 -1.3812*** 0.0224 0.4758 -2.0475*** 0.2811 Consumer goods -0.2617 0.2876 0.8694 0.3126*** -0.1440 -0.204 0.5743 -0.4316* -1.0734* Construction -0.6167 -0.4364 0.8276 -0.4963 -0.4977 0.3939 0.1204 -0.0612 -0.4337
Sweden Before crisis Since crisis Difference
Interest
rate EUR USD SEK
Interest
rate EUR USD SEK
Interest
28 Consumer
goods -0.3738 0.3273** -0.5671 -0.5052** -0.3199** 0.1051 -0.1314 -0.6472*** 0.6722 Construction 0.2557 0.5245** -1.0671* -0.4283 -0.3580** 0.1171 -0.684 -0.8825*** 1.1843
Switzerland Before crisis Since crisis Difference
Interest
rate EUR USD CHF
Interest
rate EUR USD CHF
Interest
rate EUR USD CHF Bank 0.0546 3.1531*** 0.5984* -1.3188* 0.2240*** 0.5656*** -0.2248 -0.1357 0.1694** -2.5876*** -0.8232** 1.1831 Insurance 0.0572 2.3006** 0.2878 -1.7268** 0.0719** 0.6772** -0.3536** -0.6419 0.0147 -1.6233 -0.6415** 1.0849 Utilities 1.0972** 0.0237 -1.4375*** 0.2986 -0.2937** -0.3416 -0.7985 -0.3174 1.0959** Pharmaceuticals & Biotech 1.3645*** 0.3579** -0.7700** 0.211 0.0508 -0.2975 -1.1535** -0.3071 0.4725 Healthcare 1.4238*** 0.3399** -0.8699*** 0.3106** 0.0046 -0.3438* -1.1132** -0.3353* 0.5261 Technology 3.9569*** 0.8169** -2.4086*** 0.7565** -0.6704** -0.0632 -3.2004*** -1.4872*** 2.3454** Chemicals 1.3803*** 0.4326** -1.0162** 0.4632** -0.1162 -0.0859 -0.9171* -0.5488** 0.9303 Basic resources 1.4809*** 0.4264** -0.9185** 0.4667** -0.1304 -0.0842 -1.0143* -0.5568** 0.8343 Consumer goods 1.3688*** 0.5493*** -0.9278** 0.3161** -0.152 -0.2759 -1.0527** -0.7014*** 0.6518 Construction 1.4577* 0.7204*** -2.3598*** 0.3681 -0.5940*** -0.4966 -1.0896 -1.3144*** 1.8632***
Japan Before crisis Since crisis Difference
Interest
rate EUR USD JPY
Interest
rate EUR USD JPY
Interest
29 Pharmaceuticals & Biotech 0.3862** 0.2499 -0.0912 0.3645** 0.1741 -0.2304 -0.0218 -0.0758 -0.1392 Healthcare 0.3448** 0.1840 -0.0379 0.4725*** 0.1936 -0.3070* 0.1277 0.0096 -0.2691 Technology 0.0318 0.3657 -0.1200 1.1308*** 0.4568*** -0.4560*** 1.0990*** 0.0911 -0.336 Chemicals 0.2466 0.2431 -0.3717 0.8273*** 0.3510** -0.3747*** 0.5807** 0.1079 -0.003 Basic resources 0.1325 0.1215 -0.4391 1.0833*** 0.4370** -0.4212*** 0.9508** 0.3155 0.0179 Automobile 0.5288** 0.5028* -0.6662** 1.1050*** 0.8265*** -0.3827** 0.5762** 0.3237 0.2836 Consumer goods 0.3887 0.3410 -0.4140 1.0012*** 0.6286*** -0.3773*** 0.6124** 0.2877 0.0367 Construction -0.0510 0.0332 -0.2472 0.7592*** 0.2289 -0.2709* 0.8103** 0.1957 -0.0237
China Before crisis Since crisis Difference
Interest
rate EUR USD RMB
Interest
rate EUR USD RMB
Interest
rate EUR USD RMB
30
Germany Before crisis Since crisis Difference
Interest
rate USD EUR Interest rate USD EUR
Interest
rate USD EUR
Bank 0.0567 0.1200 -0.3741 0.2059*** -1.1971*** -0.7827 0.1491* -1.3170*** -0.4086 Insurance 0.1285 0.0700 -0.2679 0.0686 -0.6334*** -0.4734 -0.0599 -0.7033 -0.2055 Utilities 0.0252 -0.2286 -0.5197*** -0.2119 -0.5449 0.0166 Pharmaceuticals & healthcare 0.0291 -0.1822 0.1244 -0.4258 0.0953 -0.2437 Technology 0.5354** -0.6854 -0.5816*** -0.1655 -1.1170*** 0.5199 Chemicals 0.2751 -0.6428 -0.5731*** -0.6082 -0.8482** 0.0347 Basic resources 0.6968* -2.3264*** -0.9239*** -0.5594 -1.6207*** 1.7670** Automobile 0.6887** -0.3068 -0.2303 -0.8386* -0.9191** -0.5318 Consumer goods 0.3671* -0.4120 -0.3569* -0.3828 -0.7239** 0.0292 Construction 1.1789*** -2.1708*** -0.8659*** -0.0051 -2.0448*** 2.1657**
Italy Before crisis Since crisis Difference
Interest
rate US dollar $ Euro € Interest rate US dollar $ Euro €
Interest
31 Table 7. Correlation coefficients between EUR/JPY and Japanese industries.
Full period Pre crisis Since crisis During crisis 2008m9 - 2011m12
EUR/JPY EUR/JPY EUR/JPY EUR/JPY Bank 0.4071 0.0137 0.4071 0.4207 Insurance 0.4712 0.0776 0.4712 0.4757 Utilities 0.4761 0.0603 0.4761 0.6766 Pharma 0.2757 0.2339 0.2757 0.3070 Healthcare 0.3520 0.2438 0.3520 0.4193 Technology 0.5256 0.0128 0.5256 0.6959 Chemicals 0.4935 0.1338 0.4935 0.6090 Basic resources 0.5039 0.0527 0.5039 0.6333 Automobile 0.5387 0.2408 0.5387 0.5243 Consumer goods 0.5625 0.2028 0.5625 0.5869 Construction 0.3733 -0.0190 0.3733 0.4840
Table 8. Factor coefficients and changes for sub-samples in Switzerland.
32 Chemicals 0.6142** -0.031 -0.4771 0.4567** 0.3575 0.1905 -0.1575 0.3886 0.6676 Basic resources 0.6333** -0.045 -0.4472 0.4683** 0.3516 0.1817 -0.1649 0.3966 0.6289 Automobile 0.2198 -1.0318** -1.1171 0.4432 0.7036 -1.6824** 0.2233 1.7354*** -0.5653 Consumer goods 0.3229* 0.0021 -0.5003 0.5854*** 0.2236 -0.2265 0.2625 0.2215 0.2738 Construction 0.484 -0.3112* -1.2926** 0.4950*** 0.1281 0.0610 0.0110 0.4394 1.3536** *** p<0.01, ** p<0.05, * p<0.1 Table 9. Factor coefficients and changes for sub-samples in the United Kingdom.
*** p<0.01, ** p<0.05, * p<0.1
United Kingdom Before Brexit announcement Since Brexit announcement Difference
Interest
rate EUR USD GBP
Interest
rate EUR USD GBP
Interest
33 4.4.1 Changing the order of regressors
Appendix 5 presents the empirical results of time series regression with a different order of variables. Here, the industry returns are regressed on the bilateral US dollar and trade weighted local currency before the bilateral euro. Since this order was originally applied for Germany and Italy, these countries are left out of this discussion.
Biancolillo and Naes (2019) emphasize that coefficients and standard errors might change due to the order of variables when sequential or hierarchical decomposition is applied. Consistently, the coefficients and their significance have drastically changed with the alternative order. This has ruled out the dominant effect of the bilateral euro on most industries. Here, nine industries are affected by the euro when the alternative order of variables is applied. This implies that, even after the effects of the bilateral US dollar and trade weighted local currency have been taken into account, nine industries are still sensitive to changes in the bilateral euro. In the United Kingdom, the euro has an effect on the insurance, utilities and chemicals industries. In Japan, the insurance and utilities industries are sensitive to changes in the bilateral euro. The bank, utilities, basic resources and automobile industries in China are significantly affected by the bilateral euro. Except for the insurance industry in the UK, all these industries are sensitive to the bilateral euro, where the order of the factors is irrelevant. This implies that these eight industries are highly sensitive to the euro expressed in the local currency.
The US dollar expressed in the local currency now has an effect on 23 industries in Japan, Sweden, Switzerland and the United Kingdom. Consistently with previous findings, the US dollar expressed in Renminbi has no significant effect on Chinese industries. This is an unexpected finding, since the US dollar is the most traded currency in the world. Moreover, the United States is China’s largest trading partner. However, a possible explanation for the insignificant effect is that the Renminbi is considered undervalued (Dai, 2013). The economy of China is not likewise open as economies of the other countries in this sample. The currency policy of China is that is aims to affect the Renminbi to benefit its exports. This might explain why the currency is undervalued and the link between the US dollar expressed in Renminbi is insignificant.
For Japan and Switzerland, the trade weighted local currency remains to affect several industries. For China, one industry is affected by the local currency with the alternative order. Now a total of 17 industries are affected by local currency movements, whereas with the original order 11 industries showed sensitivity to local exchange rate risk.
4.4.2 Newey and West heteroscedasticity- and autocorrelation-consistent standard errors
34 precise than they truly are. This research applies monthly data, therefore, a test for serial correlation with a lag of 12 months is performed. This would indicate that a variable has a relationship with its lagged version of a year ago. Six industries showed significant effects of serial correlation. Hence, Newey and West HAC standard errors are presented for the full period in section 4.1.
For the interest rate factor only one, out of 14 coefficients, becomes significant and one becomes insignificant, when comparing robust with HAC standard errors. This indicates that the empirical findings are robust for the interest rate factor.
Consistent with the robustness of the interest rate factor, the euro factor is also robust. Only two coefficients out of 52 become significant and two become insignificant. This indicates that there is little difference between the application of robust or HAC standard errors for this regressor. For the US dollar factor, one coefficient becomes significant and seven coefficients become insignificant out of 52 coefficients. However, none of the industry coefficients that changed significance, have significant serial correlation with a lag of twelve months. This would indicate that the robust standard errors are more efficient than the HAC standard errors. Therefore, the findings on the US dollar factor, as presented in section 4.1.2, are robust to serial correlation. Finally, eight out of 72 coefficients of the local trade weighted currency are significant and one coefficient is insignificant, when changing from robust to HAC standard errors. However, of these changed industry coefficients only banks in Japan are subject to serial correlation with a lag of 12 months. This coefficient has become significant when applying the HAC standard errors. This implies that the trade weighted Japanese Yen has a significant effect on returns of its bank industry.
4.4.3 EGARCH and GARCH models
Appendix 6 lists the regression results using (E)GARCH models. The coefficients using these models are presented in maximum likelihood, which are estimates that are difficult to interpret. Therefore, to circumvent this difficulty, the size and sign of the parameters will not be interpreted. There will solely be discussed whether parameters are significant or not.
35 When an EGARCH (1,1) model did not provide estimates, a GARCH (1,1) model was applied. In some cases, both models weren’t able to get estimates for an industry, consequently, these industries are left out of appendix 6.
According appendix 6, only for the bank industry in Switzerland and insurance industries in China and Switzerland, the interest rate has a significant effect. In comparison with OLS, less industries are sensitive to interest rate changes. This contradicts the empirical findings in this and other researches (Bessler and Murtagh, 2004 and Olugbode et al., 2014) discussed previously.
Consistently with the findings on the interest rate, the bilateral euro has an effect on fewer industries in comparison with OLS estimations. Only seven industries are affected by this factor. These results are inconsistent with previous findings in section 4.2.2.
The US dollar expressed in the local currency is significant for 15 industries, somewhat less industries in comparison with OLS estimations. However, only six industries are affected by the bilateral US dollar in both OLS and (E)GARCH estimations. For the six industries the findings are therefore robust to both models.
For ten industries, the local trade weighted currency has a significant effect. Findings are robust for only five industries applying (E)GARCH models.
The advantage of using EGARCH over GARCH, is the possibility to test for leverage effects. Evidence of leverage effects is provided for six industries. However, twelve industries show significant effects that positive shocks have a larger impact on volatility. Hence, the majority of industries show opposite leverage effects. This is a contradiction with findings of Magnus and Fosu (2006).
However, problems that arise when applying (E)GARCH models for small sample sizes (N<500), are that estimates are biased and that in many cases estimates are not possible with non-negative conditions (Hwang and Valls Pereira, 2006). Hwang and Valls Pereira find evidence that GARCH estimates are negatively biased for small sample sizes. The maximum number of observations per industry analysed is 192.
4.5 Discussion
36 of. Finally, the study shows that changes in long-term government bond yields, have a significant effect on banking industries in all countries and on insurance industries in all countries, except for China, Sweden and the United Kingdom.
The findings are robust to Newey West HAC standard errors. In addition, the relationship is tested using (E)GARCH (1,1) models, that account for volatility clustering. The findings are robust for a minority of industries to the (E)GARCH models. However, findings using these models are considered to be inaccurate and unreliable, as Hwang and Valls Pereira (2006) find evidence that GARCH estimates are negatively biased for small sample sizes. They suggest that this also accounts for variations of GARCH models, such as EGARCH. According to their study, a sample size of at least 500 observations is advised to circumvent this problem. However, the maximum number of industry observations in this study is 192, consequently, (E)GARCH results are neglected in this study. In order to retrieve reliable estimations using these models, the sample size should have been increased.
To account for multicollinearity, a sequential decomposition is applied. A drawback of this decomposition is that the order of regressors influences the coefficients and standard errors of estimates. Moreover, the adjusted R² changes with the order of regressors. A robustness check shows that when the effects of the US dollar expressed in the local currency are first accounted for, more industries are sensitive to changes in this bilateral exchange rate. With this alternative order, 34 out of 72 industries are sensitive to changes in the US dollar expressed in the local currency. This is a substantial number of industries and it confirms that the US dollar affects industries worldwide. When comparing the goodness of fit of factors with each other, the US dollar expressed in the local currency explains on average more variation for industries in Sweden and the UK, whereas the bilateral euro exchange rate explains on average more variation for industries in China, Japan and Switzerland. However, the differences between the goodness of fit are minor, except for China and Switzerland. Furthermore, the adjusted R² is industry specific, which makes it risky to compare the goodness of fit across countries as a whole. Moreover, a cross-sectional analysis of countries by industry should be performed, to analyse and compare differences across countries by industry.
With the alternative order of regressors, nine industries in China, Japan and the UK are affected by the bilateral euro exchange rate. This indicates that these industries are highly sensitive to changes in this exchange rate, as the effect of the bilateral US dollar and local currency have already been taken into account. The effect of the bilateral euro exchange rate is expected, because most countries in this sample are within Europe.
37 explaining exchange rates during the financial crisis, as Central Banks uniformly lowered the interest rates. The US dollar unexpectedly appreciated sharply against most other currencies, caused by high financial liabilities, low FX reserves and large trading deficits of other countries. This changed the effects of the bilateral US dollar exchange rate on industry returns in most countries. Moreover, the sensitivity of several industries to the bilateral US dollar exchange rate intensified in Germany, Italy, Japan and the UK. A drawback of this sub-sample, however, is that the sub-sample is split in before and since the financial crisis. For most (European) countries, the peak of the crisis was between 2010 and 2012. By 2012 most European economies, with the exception of Italy, started to grow again. Whereas the effects before and since the financial crisis are analysed, the analysis could have been more accurate. For future research, it is recommended to perform a more accurate analysis on financial crises, by splitting data into three samples: before, during and after a financial crisis.
In addition to the recent financial crisis, evidence is provided that the unpegging of the Swiss Franc to the euro has caused a change in the sensitivity of several Swiss industries to exchange rate risk. Overall, the sensitivity, of both the bilateral euro exchange rate and the trade weighted Swiss Franc, decreased for Swiss industries. Sensitivity to the bilateral US dollar has only decreased significantly for the Swiss automobile industry.
Finally, the analysis on the official Brexit announcement shows that sensitivity of UK’s industries to exchange rate risk has changed significantly. The sensitivity to the bilateral euro and US dollar exchange rates has increased for the insurance, utilities and consumer goods industries. The sensitivity to the trade weighted Pound Sterling has decreased for the technology industry.
The findings on the sub-samples show that economic time series relationships are time dependent for exchange and interest rate risk. However, the order of regressors as a consequence of sequential or hierarchical decomposition, may have changed the size of the coefficients and standard errors. Instead an egalitarian decomposition as suggested by Chow (2013) could have been applied. In the egalitarian model, the problem of order-dependence, which arises when applying a sequential decomposition, is evaded. A drawback of the egalitarian model, however, is that it is extremely complex to derive the orthogonalized factors.
5 Conclusion
Over the past years there have been large fluctuations in exchange and interest rates across countries worldwide, especially during unstable periods. The relationship between stock returns and exchange and interest rate risk has been widely researched. Nevertheless, empirical research provides mixed results.
38 Especially the bilateral euro and US dollar exchange rates appear to affect industry returns, whereas local trade weighted currencies appear to have an effect on a smaller amount of industries in Italy, Japan and Switzerland. Additionally, the effect of exchange and interest rate risk on sub-sample periods is examined. The analyses indicate that the effect of exchange and interest rate risk has changed during the financial crisis, and after the unpegging of the Swiss Franc and the official Brexit announcement. Moreover, the results indicate that the relationship between industry returns and exchange and interest rate risk is time dependent. The findings show that exchange and interest rate risks are important for financial-decision making.
This study has applied a sequential decomposition method to take care of multicollinearity between the regressors. A drawback of this method is the order-dependence of the regressors. The size of coefficients and standard errors change as different orders of regressors are applied. A decomposition method that avoids this bias is the egalitarian decomposition method. For future research it is, therefore, suggested that the egalitarian decomposition method should be applied in case of multicollinearity.
