The Causal Relationship Between the Returns of Exchange
Rates and Stock Markets in The 21st Century
Name Stan Heijnen
Student Number 10804099
Programme BSc Economics and Business Specialization Finance and Organization
Supervisor I. Sakalauskaite Date January 2018 Abstract
In this paper, we investigate the causal relationship between stock prices and exchange rates, using data on major currencies and stock markets in the period from January 2000 to December 2017. The currencies Euro, U.S. dollar, Pound Sterling, and the Japanese Yen will be compared to respectively the EuroStoxx 50, the S&P500,
the FTSE 100, and the Nikkei 225. In all regions except for the United Kingdom, evidence for unidirectional causality is found, which indicates that for the U.S., Europe, and Japan there exists a causal relationship between
Statement of Originality
This document is written by student Stan Heijnen who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
INDEX
Introduction
3
Literature Review
5
Methodology and Data
8
Results
13
Conclusions
16
Appendixes
18
Introduction
Since the introduction of stock markets, researchers and investors are trying to determine the factors that influence daily stock prices. Stock exchanges are invaluable creators of capital for businesses and together with macroeconomic factors they give a reliable representation of the economic condition of a country. The search for the establishment of connections between the stock market and macroeconomic variables, has therefore been a major topic in economics. The interest in the specific relationship between the stock market and the exchange rate has been triggered by several events, suggesting a relationship. A prominent example is the Asian crisis of 1997-1998, where there was a substantial drop of exchange rates as well as a collapse of stock markets in East-Asia. The significant economic and humanitarian effects that this event caused confirm the importance of researching the relationship between stock markets and exchange rates.
In general, the theoretical background can be divided into two camps. Classical economic theory has written about the “flow-oriented” models in which causality runs from exchange rates towards stock markets. Portfolio approaches however suggest that changes in stock prices cause changes in exchange rates. The related empirical studies are mostly conducted in developed countries and both mentioned theories are supported by empirical findings. Some studies support the portfolio approach that stock prices cause exchange rates (Ajayi and Mougoue, 1996), others the reverse relationship claimed by the classical economic theory (Abdalla and Murinde, 1997), whilst there also exists literature that reports a
bidirectional relationship (Bahmani-Oskooee and Sohrabian, 1992). The direction of the effect of this relationship is also a subject of empirical discussion. Some conclude a positive short-term relationship (Aggarwal, 1981; Roll, 1992), others postulate that it is a negative one (Soenen and Hennigar, 1988). There are also studies that find weak or no significant proof for the relationship of the two variables (Franck and Young, 1972).
Most research has examined the relationship of the stock market and exchange rate within one country. Since different methods are being used in determining this relationship, one has to be cautious when comparing results between studies. Even when the timeframes or countries of research are the same, differences in methodology could lead to different
outcomes. The safest way to compare countries and timeframes is therefore by only drawing conclusions about data that is subject to the same treatment. Having data that covers multiple countries over a long period of time provides the prospect of an international view which
evaluates the relationship over time. Since most studies do not offer this opportunity a gap in knowledge about this subject exists.
The present paper will use the following research question:
What is the relationship between exchange rates and stock markets of Europe, the U.S., the UK and Japan in the 21st century?
Trade-weighted exchange rates and stock market indices of the regions in question will be used as data. By examining the relationship between four currencies with the largest OTC foreign exchange turnover from 2000 until 2017, there will be a focus on the global
perspective. Most performed research is outdated, using data from the twentieth century. The acceleration of the globalization in the last decade makes it interesting to see whether the relationship between an equity market and a macroeconomic variable has changed. If this relationship would have changed as a result of the globalization, one could argue that this would become the most evident in regions that represent the worlds’ largest trade activity. Also, most conducted research is outdated, using data from the twentieth century. This explains the relevance of the timeframe from 2000 until 2017. The present study will provide new insights about leading economies in an interesting era, which after all includes a financial crisis. Previous crises showed that the relationship between the two variables increased during times of pressure, so this is also the expectations when the present data will be split up into the periods before, during, and after the financial crisis.
When analyzing two variables like the stock prices and the exchange rates we use their historic data in the form of a time-series. The conclusion about the relationship this research is pursuing is called the Granger-causality. It states that a variable Y is better predicted by the historic data of variable Y and X combined, then by the historic data of variable Y alone. In this case variable X Granger-causes variable Y. The present methodology will mainly focus on the Toda and Yamamoto (1995) approach to draw Granger-causality conclusions. To establish data that satisfies the requirements in order to perform a Granger test, the time series will be subject to certain tests. The first test is the unit root test, which ensures stationarity. Stationary data has a mean, variance and autocorrelation structure that is the same over time which provides the opportunity to make valid conclusions about the relationship of the
historic data. The second test is a lag length criteria test, which provides the optimal lag value used for the regression of the Granger test.
The paper is structured as follows. The first section provides an overview of the theoretical background and related literature. Followed by the discussion of research methodology and data in section 2, while section 3 shows the empirical results. The final section will offer a summary of the results and some concluding remarks.
Literature Review
Supporting the traditional economic theory, Dornbusch and Fischer (1980) wrote about the causality of exchange rates on stock markets, due to the international relationships of
businesses. They suggest that appreciations in exchange rates have a negative correlation with the competitiveness of a firm and therefore a negative effect on stock prices. This is due to the fact that exchange rate appreciations or depreciations lead to changes in the present value of future cash flows gained by international trade. The same goes for changes in debts to foreign creditors. These aspects influence a firm’s profit which in turn effects the stock market. For example, if we look at a listed business from the U.S. which conducts trade with Europe. They sell their goods in Europe so the revenue is valued in Euros. The exchange rate plays an important role since the profit is determined in the local currency. An appreciation of the U.S. dollar will cause that the expected revenues gained in Europe will have a lower value in U.S. dollars. In turn, this will lower the expected profit which makes the company less interesting for investors and reduces the stock prices. It is for this reason that multinationals will benefit from more accurate predictions of the exchange rate. It will help them dealing with their foreign capital accounts, decreasing the currency risk exposure and stabilizing their foreign revenues.
On the contrary, the portfolio balance approach states that it is the stock price that influences exchange rates by balancing the asset demand and supplies. Rising stock prices would attract capital flows from both local and foreign investors. This ensures a rise in the demand for the local currency, which causes an appreciation in the exchange rates. Falling stock prices will let foreign investors seek appropriate market returns elsewhere, subsequently lowering the demand for local currency and causing a depreciation in the exchange rate. Gavin (1989) contributed to this approach by showing the dynamic relationship between the exchange rate and the stock market index. He states that when stock market effects are significant enough, the real exchange rate can appreciate, rather than depreciate, as a result of a monetary expansion.
The bidirectional causality that is the result of both theorized effects working simultaneously has been the subject of empiric research throughout the world. With bidirectional causality, variable X causes variable Y, however, Y also causes X. The most prominent research has been conducted in the United States starting from 1970. After the Asian financial crisis of 1997-1998 the attention shifted to the Asian region. Many research made a distinction between a short-run and a long-run relationship. Where the short-run relationship refers to a time-frame including a few lags, calculated in days or months, the long-run takes the total time-series into account, which is often in years. Early studies only considered correlation instead of causality as a way of determining a relationship between stock prices and exchange rates.
In 1981, Aggarwal used monthly U.S. stock price data for the period 1974-1978 and stated that there existed a positive relationship that was stronger in the short-run than in the long-run. The positive correlation found indicates that when one variable decreases, the other variable also decreases and vice versa. Thus, the findings of Aggarwal showed that an
increase of the dollar exchange rate went together with an increase in the U.S. stock price. From the same reasoning, drops of the prices of the two variables also moved together.
This positive relationship was also found by Roll (1992) by using daily data from April 1988 to March 1991. The stronger dual causal relationship for the short-run was confirmed by Bahmani-Oskooee and Sohrabian (1992), who were among the first to use cointegration and Granger causality to analyse the relationship. Using monthly data over the period from 1973 to 1988, they found evidence for a short-run relationship.
On the contrary, Soenen and Hennigar (1988) found a significant negative relationship, using a different period. The results showed that a fall of the price of one variable went together with an increase of the price of the other variable, and vice versa. Thus, increasing stock prices matched with exchange rate depreciation and decreasing stock prices with exchange rate appreciation. They concluded this for U.S. stock indexes and a fifteen currency-weighted value of the dollar for the period 1980-1986. Ajayi and Mougoue (1996) investigated the long and short run relationship between the two variables for the U.S. and UK. They found that an increase in stock prices lead to a depreciation of the currencies of both countries in the long and short run.
Franck and Young (1972) were among the first to analyze the relationship between stock prices and exchange rates. They used six different exchange rates, three of European currencies and three of the U.S. dollar. Each in a time period that included a major currency realignments or event, in-between 1967 and 1971. Analysing this against the 280 largest U.S.
industrial corporations, they did not find a relationship between the two. Jorion (1990, 1991), who used the period 1971-1987 found as well as Bartov and Botnor (1994) and Ong and Izan (1999) weak or no significant evidence that suggested a relationship between equity and exchange rate markets. The latter suggested that the use of daily data instead of monthly data could have improved their empirical research, where Bartov and Botnor (1994) claimed that the use of lagged changes instead of contemporaneous changes improves the explanatory power.
When examining the empirical research for the other regions of interest for this paper, there is also no consensus to be found. Solnik (1987) employed an arbitrage pricing model to investigate the impact of several variables such as the exchange rate on stock prices for countries including Japan, U.S., UK, and the largest European countries. He found that the exchange rate can influence the U.S. stock market compared to the interest rate and
inflationary expectations, although the positive result was not significant. More recently, in a study where they examined the G-7 countries by using daily data, Nieh and Lee (2001) could not find long-term causal relations. There was evidence for short-term relationships only when using a one-day lag.
The recent Asian crisis motivated researchers to shift their attention regarding this topic towards the Asian countries, amongst which Japan. With the main question of which of the two variables caused the drop of the other, most studies found evidence that supports the traditional approach of exchange rate change leading stock prices. Qiao (1997) employed daily stock price indices and spot exchange rates for Hong Kong, Tokyo and Singapore for the period 1983 – 1994. By means of a Granger causality test he showed that for Tokyo there is a bi-directional causal relationship between stock returns and exchange rates. Pan, Fok and Liu (2007) also found a causal relationship from exchange rates to stock prices for Japan, while they could not find proof to support the causality from stock prices to exchange rates, during and before the Asian crisis.
Evaluating the empirical research, it is noticeable that the causal relationship is stronger for analysis that make use of more recent data. This suggests that the link between stock markets and exchange rates becomes more significant as economies and asset markets become more integrated. The research of Stavarek (2005) and Granger et al. (2000) are supporting this suggestion. Stavarek provides evidence for a stronger relationship between the years 1993-2003 than between the years 1978-1992, while Granger et al. used three different time series for nine Asian countries. They concluded that the more recent the period, the more significance there was detected.
Also, mixed results may be the result of flawed conclusions due to lengthy periods used in a number of studies. Bahmani-Oskooee and Sohrabian (1992) used 185 months, Granger et al. (2000) took a time-frame in consideration of more than 11 years. Although these wide time-frames add to the statistical strength of the research, the conclusions may be flawed if the causal relationship is changing within the time period. This could definitely be the case since Granger (2000) pointed out that movements in both capital markets and foreign exchange rate markets are known to be ‘intrinsically a short run occurrence’.
From this argumentation, it is therefore expected that the results of this research will vary when the time period is split-up into multiple shorter time periods. Also, the causal relationship should be stronger in later time periods. This leads to the following hypothesis for the results of the present study.
𝐻": 𝐶𝑎𝑢𝑠𝑎𝑙 𝑠𝑖𝑔𝑛𝑖𝑓𝑖𝑐𝑎𝑛𝑐𝑒 𝑤𝑖𝑙𝑙 𝑑𝑖𝑓𝑓𝑒𝑟 𝑎𝑚𝑜𝑛𝑔 𝑡𝑖𝑚𝑒 𝑝𝑒𝑟𝑖𝑜𝑑𝑠, 𝑤𝑖𝑡ℎ 𝑚𝑜𝑟𝑒 𝑠𝑖𝑔𝑛𝑖𝑓𝑖𝑐𝑎𝑛𝑐𝑒 𝑖𝑛 𝑝𝑒𝑟𝑖𝑜𝑑𝑠 𝑐𝑙𝑜𝑠𝑒𝑟 𝑡𝑜 𝑡ℎ𝑒 𝑝𝑟𝑒𝑠𝑒𝑛𝑡.
Methodology and Data
Almost all recent studies in the relationship of stock price – exchange rate employ the testing technique of Granger causality. Regression analysis could also be applied, but it is difficult to establish a causal connection by means of a statistical relationship, even when the association is strong. It is therefore that the Granger approach (1969) enjoys such popularity among studies that deal with the causal relationship between stock indices and exchange rates as well as in other applied econometric research (Gujarati, 1995). To establish short and long-run relationships, different tests are used. Only the short-run relationship can be interpreted as causality by means of the Granger test. The long run only focusses on correlation. Since this research tries to determine causality between the two variables, the short-run Granger relationship is the only one of interest.
Granger-causality
In determining a causal relation between two variables, the Granger causality test estimates a dependent variable with and without the independent variable. It uses the null hypothesis that the historical values of the independent variable make no significant contribution to the current value of the dependent variable, in excess of the contribution of the dependent variable its own past values. This will be tested by a Wald test, which tests to find out if
explanatory variables in a model are significant, thus adding value to the model. Rejecting the null hypothesis means that the alternative hypothesis that the independent variable Granger-causes the dependent variable is accepted (Granger, 1969).
The present study indicates that stock market prices Granger-cause the exchange rate, if the exchange rate can be better predicted from the past of the exchange rate and the stock market prices together, than from the past of the exchange rates alone. The same applies the other way around. When a time series X has Granger-causality on times series Y, the patterns in X are approximately repeated in Y after some time lag. This indicates that past values of X could be used for the prediction of future values of Y.
When testing two time-series for causality, it can occur that the series are integrated at different orders. One dataset can be stationary, whilst the other is nonstationary and of order I(1) or higher. In case of both series integrated of order one, it could occur that they are not cointegrated. These are all possibilities that can lead to invalid causality tests for the
traditional Granger (1969) test, the error correction model (ECM) (Engle & Granger, 1987) and the vector auto regression error-correction model (Johansen & Jesulius, 1990). As Toda and Yomamoto (1995) state, in case of the Granger causality in the VAR framework, it is clearly desirable to use a testing procedure which is robust to the integration and cointegration properties of the process in order to avoid the possible pretest biases. They propose a new method, where the integrated properties of series are not relevant for the validness of the Granger causality test. The present study will make use of this method (Toda and Yomamoto, 1995). To establish Granger-causality using the Toda and Yomamoto (1995) method, a Vector Autoregressive model (VAR) is specified. If we are testing the Granger-causality of X on Y, the VAR includes some past lag values of Y itself and some time lag values of X.
The first step in specifying the VAR is finding the optimal lag length, k. The optimal lag length stands for the number of lags that balances the minimization of the amount of estimation error in the prognosis caused by each lag with the marginal benefit of including more lags. For this, we make use of the conventional wisdom by using both the Schwarz – Bayesian (SB) criteria as the Akaike information criterion (AIC). They are criteria for model selection among a finite set of models. It is partially based on the likelihood function,
introducing a penalty term for the numbers of parameters in the model. This results in a value for every lag, where the lowest value is preferred.
According to Lütkepohl (1993), the Schwarz – Bayesian criteria is more consistent then the AIC which does not provide consistent estimators of the lag order. Therefore, we use the SB criteria for choosing the right lag. The determination of the right lag is conditional on
an autocorrelation test. Autocorrelation is a problem often encountered when using time-series and means that the error terms of old values are correlated amongst each other within the same time series. This causes the model to become misleading. We test for autocorrelation using the Breusch–Godfrey autocorrelation test. In case that the VAR with the advised lag length gives autocorrelation with the residuals, we will take the lag value given by AIC.
The maximum order of integration, d max, is the maximal order of integration that we suspect might occur in the process. This will be determined by the ADF and KPSS unit root tests and will also resolve the data requirement of stationarity. If the time-series are non-stationary at levels, but take on stationarity at first difference, we can make the assumption that our variables are integrated of the same order I(1). When performing the ADF test, the parameter of the first difference is tested on significance. 𝑌; is said to have a unit root
property if the parameter of the model is not significantly different than zero. According to Schwert (1989), who wrote an article about the unit root test, the ADF test is superior to all other unit root tests.
There are researchers that have criticized the ADF test. They say that the failure to reject the null hypothesis may be attributed to the low power against weakly stationary alternatives. Therefore, Kwiatkowski et al. (1992) state that to obtain more accurate results it is desirable to have a null hypothesis claiming a stationary time series. Thus, KPSS tests the null hypothesis of stationarity against the alternative of a unit root. The null hypothesis of stationarity is rejected if the test statistic is greater than the critical values. The unit root alternative will then be assumed.
Together with the lag length, k, this will provide the order of VAR. That is, the VAR will be estimated with a (k+dmax)th-order. Finally, as Toda and Yamamoto (1995) propose, we employ a Wald test to test for Granger causality. When performing this test, the last dmax lagged vectors are ignored, since these are regarded as zeros. The following final bivariate VAR and their corresponding null hypothesis in order to test for Granger-causality are denoted:
𝑋𝑅; = 𝛼"+ DBEA𝛼AB𝑋𝑅;CB+ K LMNGEDOA𝛼FG𝑋𝑅;CG+ DBEA𝜆AB𝑆𝑃;CB+ K LMNGEDOA𝜆FB𝑆𝑃;CG+ 𝜇A; 𝐻1": 𝜆AB = 0 for all 𝑖
𝑆𝑃; = 𝛽"+ D 𝛽AB𝑆𝑃;CB
BEA + K LMNGEDOA𝛽FG𝑆𝑃;CG+ DBEA𝜙AB𝑋𝑅;CB+ K LMNGEDOA𝜙FB𝑋𝑅;CG+ 𝜇A; 𝐻2": 𝜙AB = 0 for all 𝑖
If 𝐻1" is rejected stock prices, SP, granger-cause exchange rates, XR. If 𝐻2" is rejected
exchange rates granger-cause stock prices. If both null hypothesis are rejected there is bidirectional causality.
Stationarity
According to Granger (1988), we have the critical assumption that the time series are weakly stationary when performing the Granger Causality Test. A stationary dataset reverts around a constant mean and constant variance, independent of time. It is a result of the idea that historical relationships do not change fundamentally over time, and thus can be generalized. In this way, the data provides a reliable guide to the historical relationship, by which we can make reliable conclusions (Stock & Watson, 2012). Making reliable conclusions is critical for research and testing for stationarity has therefore been one of the major topics in econometrics for the last decades. This research will use the ADF and KPSS unit root tests to ensure
stationarity.
If a time-series does not fulfill the requirements of a constant mean and a constant variance, it is nonstationary and thereby has a unit root. In econometrics, a time-series with a unit root is known to follow a random-walk. This means that is does not show any
predictability throughout time. Shocks are permanent, the variance will grow over time and its autocorrelation tends to one. This has significant impact when regressing nonstationary
datasets with each other, which our intentions are with the Granger-causality test. The estimators that emerge will not have a normal distribution asymptotically and therefore the test statistics will not be valid.
The problem of non-stationarity becomes vivid when we look at it from a mathematic perspective. Consider the following model that contains a unit root:
𝑌; = 𝛼𝑌;CA+ 𝑢; 𝑤𝑖𝑡ℎ 𝛼 = 1
𝑌; depends on its last period value and on an error term, since the unit root 𝛼 = 1. The difference between 𝑌; and 𝑌;CA is fully captured by the error term. This causes the random
walk, which makes the time-series unpredictable and the Granger test results invalid. If financial markets follow this pattern, stock picking will be useless. This is one of the reasons why this subject is one of great importance for researchers.
Ozair (2006) draws a clear picture of the empirical consequences of a unit root by adding a shock, say C, to the model of 𝑌 with time T:
𝑌[OA= 𝐶 + 𝑌[CA+ 𝑢[+ 𝑢[OA 𝑌[OD = 𝐶 + 𝑌[ODCA+ 𝑢[OD
So 𝑌[OD increases with C for all 𝑘. Thus, the effect of the shock C is permanent. Now consider the same model, only now with a unit root 𝛼 < 1, considered nonstationary.
𝑌[ = 𝛼𝑌[CA+ 𝑢[+ C 𝑤𝑖𝑡ℎ 𝛼 < 1 𝑌[OA= 𝛼(𝑌[CA+ 𝑢[+ C) + 𝑢[OA
= 𝛼𝐶 + 𝛼𝑌[CA+ 𝛼𝑢[+ 𝑢[OA
𝑌[ will change with C, but the successive values of 𝑌[ will increase by 𝛼C, 𝛼F𝐶, 𝛼a𝐶, … , 𝛼D𝐶.
Thus, the effect of the shock 𝐶 fades away over time, ensuring a constant mean and variance, that in turn secure a reliable Granger test.
The present study makes use of trade weighted exchange rates from the Bank of England. They represent the exchange rate value according to the relative amount of trade carried out with each of its trading partners. The prices of stock indices are used to represent the stock market. These are for the U.S. dollar, the Euro, the Pound Sterling and the Japanese Yen, respectively the S&P 500, the EuroStoxx 50, the FTSE 100 and the Nikkei 225. Each country will be examined individually. The sample period is from January 2000 to December 2017, taking only business days into account. The data is daily and has been obtained from Datastream, each time-series containing 4675 observations. The following table shows the descriptive statistics.
This study makes use of the opportunity to analyze the relationship between stock markets and exchange rates before, during and after the financial crisis. In order to distinguish
the possible differences in these periods, the sample period is split up into the periods January 2000 – June 2007, July 2007 – December 2010, and January 2011 – November 2017. June 2007 was chosen as the start date of the crisis, since in this month Standard and Poor’s and Moody’s Investor services started to excessively downgrade bonds, which indicated the start of the collapse of the financial market. Around the end of 2010, the economy had returned to operating at normal capacity, indicating the end of the financial crisis.
Results
The results of the performed Wald test to find Granger-causality are depicted in Table 2.
The varying empirical conclusions about the causal relationship between stock prices and exchange rates drawn throughout history, are in conformity with our results. Our tests provide evidence that supports conclusions denying the causal relationship as well as evidence that supports a unidirectional relation and a bidirectional relationship, varying per region. The direction of causality differs from country to country, and also from period to period. This arouses the implication that whether stock price movements Granger cause exchange rate or vice versa is country and time dependent. This is in line with the part of the drawn hypothesis that stated an expected difference in causal significance among time periods, due to the changing causal relationship over time.
Appendix 1 shows the results of the ADF and KPSS tests. The ADF null hypothesis that there is a unit root cannot be rejected for all data where the KPSS null hypothesis that the data is stationary is rejected at 1% for all data, excluding a few exceptions. Thus, we can conclude that almost all data is integrated of order I(1). This result is as expected, as it is known since Nelson and Plosser (1982) that macro-economic data normally contain a unit root. With Eurostoxx50 for the ‘complete period’ and GDP for the time-series ‘before crisis’, the tests show contradictory results. Since we follow the Toda and Yomamoto (1995)
approach this will not cause any statistical complications for the Granger causality test. If an ADF test-statistic can cause a rejection of the null hypothesis, the same test in first difference is performed. A dataset integrated of order I(1), is made stationary by taking the first
difference of the series. If this test-statistic has a higher significance than the test in levels, there is reason to assume that the dataset is integrated of order higher then I(1). If the KPSS test-statistic gives a test-statistic that is less significant than 1%, the same differenced test is performed. Likewise, this is to ensure that the data is not integrated of order higher then I(1). Comparing the test-statistics of the ADF and KPSS level data against the differenced data shows that the significance increases and decreases respectively when the test is performed on the differenced data. Thus, we can safely assume that all datasets are at the most of order I(1) when determining the maximal order of integration for the VAR model.
Appendix 2 represents the outcome of the Schwarz – Bayesian (SB) criteria and the Akaike information criterion (AIC). For every advised lag, a VAR model is created to test for autocorrelation. Using the Breusch–Godfrey auto/serial correlation test, we accept the null hypothesis for autocorrelation for all significance levels 1%, 5%, and 10%. This is to ensure the absence of autocorrelation. The SB criteria are taken into consideration before the AIC. In case of autocorrelation among the error terms of the VAR models with both the advised lag of
SB and AIC, the nearest value to SC without serial correlation is chosen. The results of the accurate Schwarz – Bayesian criteria are conforming expectations. That is, the acceleration of the globalization in the last decade caused that the information flow in the financial world to be near perfect. This is reflected by a respectively short time lag.
Looking at the Granger-causality results for the United States we see that for the complete sample period there is evidence at 1% significance level for SP → XR, that is the S&P500 Granger-causes the U.S. dollar exchange rate. For the reverse relationship, shown in the table as XR → SP, we found little power of explanation. These results support the portfolio balance approach, which claims that stock prices cause exchange rates by means of the supply and demand of assets. Also, the series with data from before the crisis and during the crisis confirm this one way relationship. The estimated coefficients of the VAR models of the periods with Granger causality indicate whether the sign of causality is negative or positive. We find that the most significant coefficients of all three periods state a negative effect of the stock price on the exchange rate. This is in line with the findings of Ajayi and Mougoue (1996), who stated that an increase in stock prices leads to a depreciation of the currency of the U.S.
The test results of the exchange rate of the Euro and the European stock indices Eurostoxx 50 do not provide enough evidence to support a significant relationship, although the data indicate that the variables do have some explanatory power on each other, with the exception of the period before the crisis, where a p-value of 0.02 is denoted. Here, the exchange rate has significant causality on the stock prices. The overall low evidence for a causal relationship for the Euro could be due to the fact that most European countries conduct most trading activity with European neighbors, limiting the international activity and
therefore reducing the effect that the traditional economic theory and the portfolio approach have on the causal relationship between exchange rates and stock prices.
The Granger-causality test gives contrary results for the UK, as indicated by the data in Table 2. The strong dual causality found for the complete period, is not reflected when looking at the sub-periods. This could be due to the difference in lags used for the
determination of the VAR models. The sub-periods used lags of respectively 1, 3, and 1, where the complete period uses a lag length of 7. Adding more lags to the model increases the coefficients the model contains, which subsequently increases the amount of estimation error entering the forecast. This could cause the test result for the complete period to be biased.
The Japanese market gives high significance to assume that exchange rates Granger-cause stock prices. This perspective applies for all periods. The VAR model coefficients imply that this is a negative effect. For the most recent period, after the crisis, there is
evidence to support a bidirectional causality. Pan, Fok and Liu (2007) stated similar findings. They also found a causal relationship from exchange rates to stock prices for Japan, while they could not find proof to support the causality from stock prices to exchange rates, during and before the Asian crisis. Especially when comparing the sub-periods of the present test with these findings, a significant match is found. The data also fails to find proof to support stock price to exchange rate causality, during and before the time of a crisis.
By reviewing the related literature, the present research drew up the expectations of its results in a hypothesis. The expectations were that there would be different causal
significance among time periods, with more significance in periods closer to the present. Overall, the results are in consensus with the first part of the hypothesis relating to the differences in causal significance among time periods. For all regions, the conclusion can be drawn that the significance of causality differs among the different time periods. Thus, the conclusion that movements in both capital markets and foreign exchange rate markets are known to be ‘intrinsically a short run occurrence’ made by Granger (2000) is supported by the current research. The results show lack of evidence that support the expectation that periods closer to the present would show higher significance due to the increasing integration over time. The time period with the lowest amount of observations, which was the time period during the crisis, made use of 912 observations. Using the argumentation of the short run movements cycle in both markets, the wide time-frames used in this research could have had a negative influence on the precision of the results. This could explain the fact there was no significant evidence found that the periods closer to the present showed stronger causality.
Conclusions
This paper examined the causality between stock prices and exchange rates for the U.S., the UK, Japan and Europe. We tested for Granger causality using the Toda and Yamamoto (1995) approach. In order to do this, we performed the Wald test for four different time frames, the periods before, during, and after the 2008 financial crisis as well as the complete period that covers the years of the last decade.
The test gave inconclusive results, where some regions showed a causal relationship and other regions did not. The portfolio balance approach was confirmed by the test results of the U.S., where a causal relationship from stock prices to exchange rate became evident. The highest explanatory power was found for Japan. Here, the exchange rate Granger-caused stock prices in all periods with the period after the crisis showing bidirectional causality. The findings of this paper regarding the stock markets and exchange rates for Europe and the UK showed that for most datasets there is little argumentation to support a relationship. For Europe, only one period out of four supported a unidirectional causality from exchange rates to stock prices. The UK showed bidirectional causality only for the time-series including all years. The optimal lags used for this period were significantly higher than the ones used for the other periods. Since adding more lags cause an increase in the estimation error, this could give the explanation why the results for the complete period gave contradictory results in comparison with the other periods. Relatively high lags are also used in the time period ‘during crisis’. This could also be the reason that against expectancy, the results showed that the causal relationship was not stronger in times of pressure.
The empirical literature showed that previous studies as well as the present study still find contradictory evidence in this field of research. It is therefore suggested to continue the investigation of the relationship between stock prices and exchange rate. A suggestion to improve the research results is to increase about the sign of the causality by using the Hatemi-J (2012) asymmetric causality test, where the positive and negative shocks are split. In addition, there is a chance that a third variable z, effects both stock prices and the exchange rate, which causes that a possible Granger causality is unjustified since both variables are controlled by z (Stock and Watson, 2012). This could be resolved by adding more variables to the models.
Appendixes
References
Abdalla, I. S. A., and Murinde, V. (1997). Exchange rate and stock price interaction in emerging financial markets: Evidence on India, Korea, Pakistan, and Philippines. Applied Financial Economics 7, 25-35.
Aggarwal, R. (1981). Exchange rates and stock prices: a study of the US capital markets under floating exchange rates. Akron Business and Economic Review Fall 12, 7-12. Ajayi, R.A., and Mougoue, M. (1996). On the dynamic relation between stock prices and exchange rates. Journal of Financial Research 19, 193- 207.
Bahmani-Oskooee, M., Sohrabian, A. (1992). Stock prices and the effective exchange rate of the dollar. Applied Economics 24 (4), 459– 464.
Bartov, E., and Bodnar, G. M. (1994). Firm valuation, earning expectations, and the exchange rate exposure effect. Journal of Finance 49, 1755-85.
Dickey, D. A., and Fuller, W. A. (1979). Distribution of the estimators for autoregressive time series with a unit root. Journal of the American Statistical Association 74, 427–431.
Dornbusch, R., and Fischer, S. (1980). Exchange Rates and Current Account. American Economic Review 70, 960-971.
Engle, R.F., and C.W.J. Granger. (1987). Cointegration and error correction: representation, estimation and testing. Econometrica 55, 251-276.
Franck, P., and Young, A. (1972) Stock price reaction of multinational firms to exchange realignments. Financial Management 1: 66-73.
Gavin, M. (1989). The stock market and exchange rate dynamics. Journal of International Money and Finance 8, 181-200.
Granger, C.W.J. (1969). Investigating causal relations by econometric models and cross spectral methods. Econometrica 37, 428-438.
Granger, C.W.J., Huangb, B.N., and Yang, C.W. (2000). A bivariate causality between stock prices and exchange rates: evidence from recent Asianflu, Quarterly Review of Economics
and Finance 40 (3), 337–354.
Gujarati, D. (1995). Basic Econometrics, third edition, McGraw-Hill, New York. Johansen, S., and K. Juselius. (1990). Maximum likelihood estimation and inference on cointegration-with application to the demand for money. Oxford Bulletin of Economics and Statistics 52, 169-210.
Jorion, P. (1990). The Exchange Rate Exposure of U.S. Multinationals. Journal of Business 63, 331-345.
Jorion, P. (1991). The pricing of exchange rate risk in the stock market. Journal of Financial and Quantitative Analysis 26, 361-376.
Kwiatkowski, D., Phillips, P. C. B., Schmidt, P., & Shin, Y. (1992). Testing the null hypothesis of stationary against the alternative of a unit root. Journal of Econometrics 54, 159–178.
Lütkepohl, H. (1993). Introduction to Multiple Time Series Analysis. Berlin: Springer-Verlag. Nelson, C.R., and Plosser, C.R. (1982). Trends and random walks in macroeconomic time series: evidence and implications. Journal of Monetary Economics 10, 139-162.
Nieh, C.C., Lee, C.F. (2001). Dynamic relationship between stock prices and exchange rates for G-7 countries. The Quarterly Review of Economics and Finance 41, 477-490.
Ong, L.L., and Izan, H.Y. (1999). Stocks and currencies: Are they related? Applied Economics 9, 523-532.
Ozair, A. (2006). Causality Between Stock prices and Exchange Rates: A Case of The United States. Florida Atlantic University, Master of Science Thesis.
Pan, M.S., Fok, R.C.W., Liu, Y.A. (2001). Dynamic Linkages between Exchange Rates and Stock Prices: Evidence from Pacific Rim Countries. Working Paper at College of Business Shippensburg University
Schwert, G.W. (1989). Tests for unit roots: a Monte Carlo investigation. Journal of Business and Economic Statistics, 7, 147–159.
Soenen, L., Hennigar, E. (1988). An analysis of exchange rates and stock prices: the US experience between 1980 and 1986. Akron Business Economic Review 19, 7–16.
Solnik, B. (1987). Using financial prices to test exchange rate models: a note. Journal of Finance 42, 141-149.
Stavarek, D. (2005). Stock prices and exchange rates in the EU and the USA: Evidence of their mutual interactions. Journal of Economics and Finance 55, 141-161.
Stock, J.H., Watson, M.W. (2012) Introduction to econometrics. Pearson Education Limited. Toda, H.Y., and Yamamoto, T. (1995). Statistical inference in vector autoregression with possibly integrated processes. Journal of Econometrics 66, 225–250.
Roll, R. (1992). Industrial structure and the comparative behaviour of international stock market indices. Journal of Finance 47, 3-41.
Qiao, Y. (1997). Stock Prices and Exchange Rates: Experience in Leading East Asian
