Macroeconomic variables and their impact on the Swedish stock market

1

Macroeconomic variables and their impact on the Swedish stock market

By: Timur Cengiz & David Holmer

Supervisor: Johanna Palmberg

Södertörn University | School of social sciences Bachelor’s essay, 15 credits

Spring Semester 2021

2

Abstract

The objective of this study is to investigate the impact of a few selected macroeconomic variables on the Swedish stock market index OMXS30. The study uses time series monthly data during the period 2000-2019. To investigate these relationships, the time series are transformed into stationary processes. Then, we construct a Vector autoregressive model (VAR) and conduct Granger causality tests. The results indicated a negative relationship between inflation and the return on stocks, interest rate and the return on stocks, as well as positive relationship between money supply and the return on stocks. The VAR-model and the Granger causality test failed to show any statistically significant relationship between exchange rate and stock prices. The same Granger Causality tests suggests a bidirectional relationship between interest rate and the return of OMXS30, as well as unidirectional relationship between inflation and the stock prices, where inflation Granger causes the return of OMXS30.

3

Table of contents

1. Introduction……….4

2. Previous Studies………...6

3. Theoretical Framework………..8

1.

2.

3.

4. Data and Methodology………...10

1. Variables ……….………..11

2. Data………....13

3. Methodology………...13

5. Empirical Results………...15

1. Unit root test and AIC………...15

2. Vector Autoregression (VAR)……….….17

3. Residuals diagnostics……….…...19

4. Granger Causality……….…21

6. Discussion………...22

7. Conclusion………..24 References

Appendix

4

1. Introduction

A stock exchange is a secondary market where current stock owners can transact securities of various companies with buyers at a given price. By the process of disinvestment and

reinvestment, it provides a mechanism that ensures that people invest in the most productive investment proposal available. This ensures an efficient allocation of capital which also spurs economic growth (Stiglitz, 1972). Furthermore, the stock market, reduces information

asymmetries, diversifies risks, mobilizes savings, reduces liquidity risks, monitors corporate control, and facilitates resource mobilizations which all spur economic growth. (Pradhan, 2018)

Investing in stocks carries risk, that is commonly divided into two parts, firm specific risk and systematic risk. The firm specific risk is the risk that is unique to a certain company, this could be a strike, unfavorable litigation or any other problem that might occur within the firm.

The latter refers to risk associated with the entire market, and this type of risk is unpredictable and impossible to avoid. This risk is composed of several factors, such as input good price shocks, like the oil crisis in the seventies, or a sudden change in other macroeconomic variables such as interest rate or inflation. Therefore, a change in these macroeconomic variables interpreted as signals to the participants of the stock market to expect higher or lower returns, and significantly impact the prices of stocks (Mankiw & Taylor, 2017).

In the midst of the Covid-19 pandemic, some of these macroeconomic variables have been subject to some erratic movement due to the implemented restrictions and sudden policy changes. For example, during 2020 central banks have via stimulus strategies increased the money supply. The Swedish money supply (M2) increased 17.6% year-over-year in January 2021, and in several countries that number is much greater. To put that number in perspective, the money supply in Sweden was only increased by about 0,2 % between 2018-2019. This means that close to a fifth of the money in circulation in Sweden was printed in 2020 (International Monetary Fund, 2021). This is one factor that has enabled the bullish markets (upward price movement) in the last year, despite the decreased demand overall. At the same time, this increase in money supply hasn't had any larger effects on the inflation rate so as of April 2021, considering the movements of the consumer price indexes (ibid).

The differences in the increase in money supply between nations is a factor that could

potentially explain some of the erratic movements in the currency markets, where the Swedish

5

Krona has appreciated by more than 13% with respect to the US Dollar in the last year, as of May 2021 (International Monetary Fund, 2021).

In the early spring of 2021, the increase in the American 10-year treasury rate had a big impact on financial markets across the globe. OMX Stockholm 30, an index containing almost only value stocks, fell by -1.42% on the 26th of February upon the news of the increased interest rate, but later increased to higher levels when the interest rate stabilized.

During the spring of 2021 the rollout of the covid-vaccines now takes place all over the developed world, this should get the economy moving back towards how it was before the pandemic. These facts have sparked an interest in investigating how these macroeconomic variables have affected the Swedish stock market before the pandemic erupted. Therefore, the main objective of this paper is to investigate the impact the macroeconomic variables interest rate, inflation rate, money supply and exchange rate have on the pricing of OMXS30 during the period of 2000 until and including 2019.

We will try to answer that question by using time series data with monthly observations to construct a vector autoregressive model (VAR) with OMXS30 as the dependent variable, and the previously mentioned macroeconomic variables as the independent variables. To examine if and how the independent variables affect the pricing of OMXS30, we will also conduct a Granger Causality test, which is used in determining whether one time series is useful in forecasting another (Hiemstra & Jones, 1994).

Previous studies on the relationship between stock market returns and one or more

macroeconomic variables have been executed in many different parts of the world, we cover a few of those in the section ‘’Previous studies’’. However we found that there is a lack of studies done in Scandinavia in particular, we therefore feel that our study fills a void and the information and relationships found in this study can be useful for anyone looking to get a better understanding of the stock market and how it might react to exogenous factors.

This thesis is structured as follows. Chapter 2 reviews and summarizes previous studies and literature related to this research question. Chapter 3 provides an overview of relevant economic theories. Chapter 4 describes the variables and the methodology used, as well as information on the collection of data. Chapter 5 focuses on the empirical results, including everything from unit-root tests, estimation of our model, regression diagnostics and Granger causality tests. In chapter 6 we will discuss our findings and conclude this paper with some

6

suggestions on further research on this topic. In chapter 7 we will state our conclusions from this research.

2. Previous Studies

In this section we will take a look at previous studies that have examined the relationship between macroeconomic variables and stock prices. Our own results can then be compared to these later on to see if we found the same statistical relationships between the variables as the previous studies have.

Sirucek (2012) examines the macroeconomic variables interest rates, inflation, oil price, producer price, industrial production index, unemployment and money supply to find what relationship they have with changes in stock prices. The study looks at the United states and the indexes S&P 500 and Dow Jones industrial average (DIJA) for stock prices. The study found that the most significant factors for both were inflation and unemployment, these had a negative effect on the share prices. The study also found a relatively minor significance of the money supply regarding share prices. The most significant determinant for S&P 500 were interest rates and unemployment. DIJA was most affected by industrial manufacturing and unemployment, followed by interest rates and oil prices.

Another study by Piero (2016) also examined the effect of macroeconomic variables on stock prices, this study looked at interest rates just like Sirucek (2012) but also at industrial

production and how they affect stock prices in France, Germany and the United Kingdom.

The study concluded that changes in interest rates and industrial production clearly affects stock prices in all three countries. The study furthermore found that future changes in industrial production and current changes in interest rates account for roughly one half of stock returns. The study also found that in recent years the importance of interest rate and its effect on stock prices is lessening while the importance of industrial production is increasing.

Khan & Khan (2018) also investigated the relationship between certain macroeconomic variables and stock prices in their study. The aim of the study was to determine the effects of these macroeconomic variables on the stock prices in Pakistan by looking at monthly data between May 2000 – August 2016. The study looked at 6 macroeconomic variables, Interest rate, inflation rate, money supply, exchange rate, economic activity & exports. We can see that the variables are quite similar to the ones of previous studies however a few are new such

7

as exchange rate and exports. The study used time series secondary data with 184 monthly observations, KSE-100 index was used as the dependent variable in this case while the other macroeconomic variables were the independent variables. The KSE-100 index is a main stock market index that follows the top companies from all 34 sectors of Karachi stock exchange. A regression analysis was run using the named variables, the results showed that exchange rate and money supply have a very strong positive correlation with stock prices , economic activity and exports have a strong positive relationship with stock prices, meaning that when the exchange rate, money supply, economic activity or exports increase the stock prices will also increase. Interest rate and inflation rate were shown to be weakly associated with stock prices; the inflation rate even had a negative relationship with the stock prices.

The study concluded that there is a statistically significant long-term relationship between money supply, interest rate, exchange rate and stock prices. So, expansion in money supply or a loose monetary policy will increase the prices of shares while strict monetary policy will decrease the price of shares. A decrease in exchange rate by 1 % will lead to a 2.70 % increase in stock prices. A decrease in interest rates of 1 % will lead to an increase in stock prices of 0,26 %.

Limpanithiwat & Rungsombudpornkul (2010) also investigated the relationship between inflation similarly to Sirucek (2012) but also Khan & Khan (2018) and stock prices in Thailand. This study used time series data with monthly observations just like we will do in our study, but they used data during the period between January 2000 and March 2010, which included a devastating tsunami and the financial crisis of 2007-2008. As a proxy for stock prices they used SET50 index, stock exchange of Thailand and a CPI (consumer price index) to describe the inflation rate. Their findings implies that there is no significant relationship between inflation and stock prices in Thailand during this period. To reach this conclusion, they used a statistical method called vector autoregression (VAR) which which is a

multivariate forecasting algorithm that is used when two or more time series influence each other.

Sampath (2011) also investigated the relationship between macroeconomic variables and stock prices, this time however in India. The studied variables were real effective exchange rate similarly to Khan & Khan (2018) but also wholesale price index and index of industrial production similarly to Sirucek (2012). The study used the Bombay stock exchange (BSE) for stock prices. The study used a cointegration analysis and a unit root test is the preliminary

8

step in this analysis. The tests used in the study are The Dickey-Fuller, Augmented Dickey- fuller and Phillips-Perron. The study used monthly data from April 1993 to March 2010, total number of observations are 205. Stock prices were measured by taking monthly averages of BSE Sensex. The study found that the macroeconomic variables real effective exchange rate, wholesale price index and index of industrial production are cointegrated with the BSE stock index. The results also showed that the macroeconomic variables had a statistically significant long-run effect on stock prices. The results also showed that the macroeconomic variables had a statistically significant effect on stock prices in the short term. The study therefore

concluded that there was a significant positive relationship between economic growth and stock prices (Sampath, 2011).

Hsing & Hsieh (2012) also conducted a study where they researched the relationship between macroeconomic variables and the stock market index, this time in Poland. The study looked at quarterly data from Q1 of 2000 until Q2 of 2010. The study concluded that higher real GDP, lower government borrowing/GDP ratio, lower treasury bill rate, currency depreciation, lower expected inflation rate, higher German or US stock market index or a lower government bond yield in the euro area all led to an increase in the polish stock market index. The study also found a positive (negative) relationship between M2/GDP ratio and the stock market index.

The different previous studies have concluded different relationships between the stock market returns and macroeconomic variables. This could be because of difference in time periods, difference in econometric methodology or because the studies have been executed in different nations. Our own results will be compared to these previous studies to see how they may differ or resemble them. To summarize, interest rate was shown to be both weakly and strongly associated with stock prices in the different studies. Inflation rate seems to be either weakly and negatively associated with stock prices or not associated at all, finally there seems to be a strong association between money supply, exchange rate and stock prices.

3. Theoretical Framework

In this section we discuss what different theories we will use in this study in order to explain the relationship between the macroeconomic variables and changes in stock prices.

9

3.1 Efficient Market Hypothesis

The term efficient market was coined by Eugene Fama (1970) and he defined it as a market in which prices always fully reflect available information. This suggests that what causes prices to change in a market is new information (Fama & Malkiel, 1970). In most financial academic literature there are three different types of EMH defined; weak form, semi- strong form and strong form. The difference between these forms lies in the definition of “available

information” (Mankiw & Taylor, 2017).

The weak form claims that the stock prices reflect all the data of historical prices, which implies that no form of technical analysis can be used to give investors an edge in the market.

The semi-strong form suggests that stock prices reflect all the data of historical prices and all public information. As a result, neither technical nor fundamental analysis can be utilized by investors, since all this information is already “priced in”. The final form, the strong form of the EMH, is the same as the semi-strong form, but it also includes that all non-public

information, for example insider information, also is completely accounted for in the pricing of the stocks (Mankiw & Taylor, 2017).

We will focus on the semi-strong form of EMH in this study. This because the semi-strong form suggests that stock prices reflect all available public information. Therefore, this form assumes that all economic factors are fully reflected in the pricing of stocks which makes it a good fit for investigating the relationship between macroeconomic variables and stock prices.

3.2 Arbitrage pricing theory (APT)

APT is an asset pricing model that predicts asset returns based on a relationship between the assets expected return and several macroeconomic variables that capture the systematic risk.

The APT model is more relevant in this study than the Capital asset pricing model (CAPM) because it can take multiple variables into account. The APT formula is the following:

𝐸(𝑟_𝑖) = 𝑟_𝑓 + 𝛽₁𝑅𝑃₁ + 𝛽₂𝑅𝑃₂ +. . . +𝛽_𝑗𝑅𝑃_𝑗 Equation 1.

10

Where 𝐸(𝑟_𝑖) is the expected return rate of an asset, 𝑟_𝑓 is the risk free rate of return, 𝛽_𝑗 is the assets price sensitivity to a certain macroeconomic variable ‘j’ and RPj is the risk premium for a certain factor ‘j’. The β can be estimated by using simple linear regression.

APT is based on three assumptions: All investors have the same investment philosophy;

investors strive for max returns at a certain risk level and investors are in a complete market.

(Ross, 1976)

Since APT suggests that an asset's price sensitivity to a certain macroeconomic variable influences its price, APT is relevant in a study that examines the relationship between macroeconomic variables and stock prices.

3.3 The expected return rate of an asset

The expected return rate of an asset should consist of expected real rate of interest and the expected rate of inflation, this according to Fisher (1930). This theory implies that the

expected real rate of interest is constant, instead the nominal rate of interest reflects all future levels of inflation. The theory suggests that the return of a stock should consist of real stock returns but also the expected inflation rate, therefore the inflation has a positive effect on stock returns. (Fisher 1930)

Fama (1981) instead showcases a negative correlation between inflation and interest rate with the ‘’Proxy effect’’. This theory suggests that the negative correlation is the result of the positive correlation between stock returns and real economic activity, but also the negative correlation between inflation rate and real economic activity. This theory therefore states that higher inflation rate will lower real economic activity which negatively affects future stock returns.

Different theories suggest different relationships between inflation and stock returns, our results will be compared to both theories in the end to determine which one best fits our results.

11

4. Data and Methodology

In this chapter we present our dependent and independent variables closer and discuss what data will be used in the study. We also discuss the methodology and what models will be used to get our results.

4.1 Variables

OMXS30 is a stock market index that contains the 30 most traded stocks on the Nasdaq Stockholm stock exchange. It’s a capitalization-weighted index that is revised semi-annually.

As of 2019-05-02, it consisted of the companies in appendix table 1 with mentioned index weight as a percentage (%). In this study, the price of OMXS30 will be our dependent variable.

We will use the Consumer price index (CPI) as a proxy for inflation. This proxy is commonly used, for example it is used by Hasan & Nasir (2008) & Limpanithiwat &

Rungsombudpornkul (2010). CPI measures the changes in prices of goods and services in the private consumption. Certain restrictions do exist with the CPI, it does not take business expenditures into account, neither does it consider all goods and services consumed by households, only the ones utilized for consumption purposes. Even though the CPI is a restricted measure of the inflation, it does give a good measure of the inflation faced by the household sector of the economy (Tsiaplias, 2008).

An exchange rate is the rate at which one nation's currency can be traded for another currency. As of Q4 2019, the US dollar makes up 60% of all (known) foreign exchange reserves, and about 90% of all trades on the foreign exchange market involve the US dollar.

Considering this, the SEK/USD exchange rate can be considered a valid proxy for the relative value of the SEK compared to the rest of the world's currencies (International Monetary Fund, 2021).

The Interest Rate Parity (IRP) claims that the ratio between the forward exchange rate and the current exchange rate equals the ratio between the interest rate in each respective country.

When comparing USD to the SEK, the equation looks like this:

12

𝐹0

𝑆₀ = ^{(1 + 𝑖𝑈𝑆)}

(1 + 𝑖_𝑠) Equation 2

Where F0 is the forward expected exchange rate SEK/USD, S0 is the current exchange rate SEK/USD, is is the interest rate in Sweden and ius is the exchange rate in the United States.

This theory implies that investors cannot earn arbitrage by borrowing in one country with lower interest rate, and invest in a country with a higher interest rate, due to gains or losses that arise when the investors want to exchange back their investments into their domestic currency later on (Mankiw & Taylor, 2017).

When the interest rate increases, stock prices seem to fall. This also means that when the interest rate decreases, stock prices increase. This negative relationship was noticed by Nasseh & Strauss (2004) among others. Further, Hassan & Nasir (2008) found that an

increase in interest rate leads to an increased discount rate, which means a decrease in present value of future cash flows, which negatively affects stock prices.

We will use the Stockholm interbank offered rate (Stibor) as a proxy for interest rate. Stibor is designed to reflect the average rate at which the Swedish banks are willing to lend to one another without collaterals. Stibor is classified as a critical interest rate benchmark and is also being regulated by the Swedish Finansinspektionen (FSA), the stibor rate is therefore an appropriate proxy to use for interest rate (SFBF, 2021).

The alternative to using interbank loan rates as a proxy for the interest rate would be using T- bill rate. However, this is mostly used in regions with a not as sophisticated banking sector as in Sweden. Considering this, we believe that the Stibor is a better proxy for the true cost of borrowing money in Sweden, where the companies of OMXS30 operate, than the treasury bill rate.

The money supply can be explained as all currency and other liquid instruments in a country’s economy at a certain time. In theory, an increase in money supply leads to a surplus in money balance that can be used to purchase stocks. This increased demand for stocks eventually leads to an increase in the prices of stocks. According to Khan & Khan (2018), a rising money supply seem to have a positive effect on stock prices.

13

We will use M3 as a proxy for money supply. Hasan & Nasir (2008) used M1 as a proxy for money supply, which is defined as the non-bank sector's holdings of notes and coins, demand deposits and other checkable deposits. Considering that M1 doesn’t include bank reserves, we believe that M3 is a better proxy for money supply, since it doesn’t neglect the banking sectors holdings on the stock market and other financial instruments (Mankiw & Taylor, 2017). M3 includes cash, checking deposits, easily convertible near money and long-term bank deposits. M3 is not the most used proxy in the previous studies but we believe that it gives the best and broadest representation of the money supply.

4.2 Data

The main objective of this paper is to determine the impact of some macroeconomic variables on the pricing of OMXS30. This stock index is our dependent variable, and our four

independent variables are inflation, exchange rate, interest rate and money supply. We will use time series data with monthly observations, covering the period of 2000 until and including 2019.

We will collect the data from numerous sources. Data on the price of OMXS30 will be collected from the Database FinBas. The CPI-data is gathered from Federal Reserve Economic Data. We’ll use the International Monetary Fund’s database “International

Financial Statistics” to retrieve data on the SEK/USD-exchange rate. We will collect data on day-to-day Stibor-rate from Sveriges Riksbank. Lastly, we’ll use Swedish Statistics (SCB) to collect data on the money supply.

4.3 Methodology

We are going to use a vector autoregression (VAR) in order to capture the relationship between our dependent and independent variables. The primary difference between VAR and other autoregression models is that the latter are unidirectional, i.e. that the predictors

influence the dependent variable. VAR on the other hand is bidirectional, i.e. variables influence each other. Considering the macroeconomic variables we examine, this could be very advantageous. VAR-models have proven themselves to be very useful for describing the dynamic behavior of financial and economic timer series (Zivot & Wang, 2006).

14

Before using VAR however, we must check the validity of the data. A key assumption when dealing with time series data is stationarity. This means that the joint probability of the series is independent of time, so that mean and variance remain constant. Non-stationary processes can cause spurious relationships between unrelated variables, causing the coefficient of determination to inflate. In some of the time series there are also obvious trends, meaning that for example the mean change over time. Before we use the data to determine the Granger causality, we must make sure that all our series are stationary, otherwise detrend the data to make them stationary. To determine which series that are stationary, we conduct a unit root test for each series. If a process contains one or more unit roots, it requires differencing to be made stationary. One of the most popular unit root tests is the so-called Augmented Dickey- Fuller (ADF). This test will tell us which series that already are stationary, and which we must detrend or take first difference, or even second difference, in order for them to become

stationary. (Cheung & Lai, 1995).

In the previous research we have covered, for example Limpanithiwat & Rungsombudpornkul (2010), there are mainly two models that are commonly used. One is the already mentioned VAR, and the other is called an autoregressive distributed lag (ARDL) model. The common denominator is that both are based on ordinary least square (OLS) estimation. A standard assumption when dealing with regression analysis is exogeneity. This refers to that the independent variables are exogenous, and exogenous variables are able to influence the system without being influenced by it. However, as we’ve already mentioned, VAR is bidirectional, allowing the independent variable to be influenced by other independent variables or the dependent variable. ARDL models require the explanatory variables to be at least weakly exogenous, unlike VAR models that have no exogeneity requirements. This is very useful, considering that for example the interest rate can be decreased as a proactive stimulus strategy, and increased reactively with the purpose of stabilizing an overheated economy and vice versa (Shrestha & Bhatta, 2018).

ARDL models don't require all variables to be stationary, but the interpretation of the models differs vastly depending on the stationarity or non-stationarity of the variables. In VAR models all variables must be stationary, which allows for a straightforward interpretation of the results (Shrestha & Bhatta, 2018).

15

Before we compute our estimates, we also must decide which lag length to use for our model.

There are primarily three different criterias that have widespread use in determining lag length; Akaike information criterion (AIC), Bayesian information criterion (BIC) and Hannan–Quinn information criterion (HQ) (Gredenhoff & Karlsson, 1999). We’re going to use the AIC, because it under-estimates the lag-length less frequently than BIC and HQ and may thus be the preferred criteria when performing inference (Gredenhoff & Karlsson, 1999).

When we have computed our estimates from the regression, we’re going to run some diagnostics of our model to check if our estimates are reliable and robust. To check for autocorrelation in our data, we will conduct a Breusch- Godfrey test, with the null hypothesis that there is no autocorrelation in the data. The potential presence of heteroscedasticity is examined by a Breusch-Pagan test. Finally, we’ll perform a Jarque-Bera test to see if the normality assumption holds.

When we have completed our diagnostics, we want to determine the Granger casualties, which we of course do by conducting Granger causality tests. It is used to determine whether one time series is significant in forecasting another. So, Y is Granger caused by variable X if variable X assists in predicting the value of variable Y (Hiemstra & Jones, 1994).

All tests, estimations and computations will be done in the software R. All tests will be conducted with the significance level α = 0.05. All variables that are not already in percentages will be log-transformed.

5. Empirical Results

In this chapter, we will present our findings from the tests we have conducted on our

empirical data. We will also present all relevant values from our estimated model. This will be done based on the discussion in chapter 4.

5.1 Unit Root test and AIC

As discussed in chapter 4.3, we must ensure that all our time series are stationary, to avoid spurious regression results. To examine which time series that already are stationary, and which that must be differenced, we will conduct Augmented Dickey- Fuller tests (ADF), with the following hypotheses:

16

H0: Process contains at least one unit root (Variable is not stationary) H1: Process does not contain a unit root (Variable is stationary)

When we run the test in the software R, we will retrieve a p-value for each variable. Since we’re conducting this analysis with the significance level α = 0.05, we require a p-value of 0.05 or smaller to be able to reject the null hypothesis. It is also worth mentioning that the function we use in R takes whether there is a trend in the process into account, allowing use to use the same test function for all variables. The results from these tests can be seen in table 1.

Table 1. ADF test results at level.

Variable Null hypothesis P-value Results Conclusion

LOMXS30 Non-stationary 0.1107 Do not reject H0

Non-stationary

LCPI Non-stationary 0.4376 Do not reject H0

Non-stationary

LM3 Non-stationary 0.7015 Do not reject H0

Non-stationary

LER Non-stationary 0.06194 Do not reject H0

Non-stationary

IR Non-stationary 0.5319 Do not reject

H0

Non-stationary

According to the results from these tests, no variable is stationary. This means that we will take the first difference from all variables. When we have done this, we will do the same test over again for our now differenced variables. To separate the variables that have been

differenced, we will add a ‘d’ to the variable name. The results from these tests can be seen in table 2.

Table 2. ADF test at 1^st difference

Variable Null hypothesis P-value Results Conclusion

LOMXS30d Non-stationary <0.01 Reject H0 Stationary

LCPId Non-stationary <0.01 Reject H0 Stationary

LM3d Non-stationary <0.01 Reject H0 Stationary

17

LERd Non-stationary <0.01 Reject H0 Stationary

IRd Non-stationary <0.01 Reject H0 Stationary

We can conclude from this test that all variables are stationary after taking first difference.

This means that we have successfully transformed all our time series into stationary processes. We must decide on the number of lags to include in our model, before we can construct our model. As previously discussed in chapter 4.3, we will use Akaike information criterion (AIC) to determine this. We use a built-in function in R, that will generate the recommended lag length according to several information criterias. In this case, the AIC recommended using six (6) lags for our model.

5.2 Vector Autoregression (VAR)

As mentioned in the methodology chapter (4.3), we’re going to construct a VAR-model. We will use six (6) lags in our model, based on the AIC, as mentioned in chapter 5.1.1. This model will later be used to examine Granger causality. As also mentioned in chapter 4.3, VAR is bidirectional model, meaning that we will create several regression models, one for every variable to be exact. However, in this chapter we will focus on our primary model, the model with OMXS30 as the dependent variable. This model will have the following form:

𝐿𝑂𝑀𝑋𝑆𝑑_𝑡= ∅₁₁𝐿𝑂𝑀𝑋𝑆𝑑_𝑡−1 + ∅₂₁𝐿𝐶𝑃𝐼𝑑_𝑡−1+ ∅₃₁𝐿𝑀3𝑑_𝑡−1+ ∅₄₁𝐿𝐸𝑅𝑑_𝑡−1 + ∅₅₁𝐼𝑅𝑑_𝑡−1 + ∅₁₂𝐿𝑂𝑀𝑋𝑆𝑑_𝑡−2 + ∅₂₂𝐿𝐶𝑃𝐼𝑑_𝑡−2+ ∅₃₂𝐿𝑀3𝑑_𝑡−2 + ∅₄₂𝐿𝐸𝑅𝑑_𝑡−2 + ∅₅₂𝐼𝑅𝑑_𝑡−2 + ∅₁₃𝐿𝑂𝑀𝑋𝑆𝑑_𝑡−3 + ∅₂₃𝐿𝐶𝑃𝐼𝑑_𝑡−3 + ∅₃₃𝐿𝑀3𝑑_𝑡−3+ ∅₄₃𝐿𝐸𝑅𝑑_𝑡−3 + ∅₅₃𝐼𝑅𝑑_𝑡−3 + ∅₁₄𝐿𝑂𝑀𝑋𝑆𝑑_𝑡−4 + ∅₂₄𝐿𝐶𝑃𝐼𝑑_𝑡−4+ ∅₃₄𝐿𝑀3𝑑_𝑡−4+ ∅₄₄𝐿𝐸𝑅𝑑_𝑡−4 + ∅₅₄𝐼𝑅𝑑_𝑡−4 + ∅₁₅𝐿𝑂𝑀𝑋𝑆𝑑_𝑡−5 + ∅₂₅𝐿𝐶𝑃𝐼𝑑_𝑡−5+ ∅₃₅𝐿𝑀3𝑑_𝑡−5+ ∅₄₅𝐿𝐸𝑅𝑑_𝑡−5 + ∅₅₅𝐼𝑅𝑑_𝑡−5 + ∅₁₆𝐿𝑂𝑀𝑋𝑆𝑑_𝑡−6 + ∅₂₆𝐿𝐶𝑃𝐼𝑑_𝑡−6+ ∅₃₆𝐿𝑀3𝑑_𝑡−6 + ∅₄₆𝐿𝐸𝑅𝑑_𝑡−6 + ∅₅₆𝐼𝑅𝑑_𝑡−6

Equation 3.

Table 3 presents the output from R of our primary equation of our VAR-model. L1 denotes that it’s the first lag of the said variable, L2 denotes the second lag of said variable and so on.

All variables are integrated of order 1 (I(1)) and log-transformed, except for IRd, our interest rate variable, that already is in percentage. Constructing the model based on the number of

18

lags recommended by the AIC provides us with a model containing 30 regressors, and only a few of them are statistically significant under the various significance levels. The first lag of our interest rate variable, IRd, is shown to be significant at the 10% significance level. The coefficient for this variable is negative, suggesting that an increase in the interest rate is associated with a decrease in the returns of OMXS30. The third lag of LCPId and the fourth lag of our money- supply variable, LM3d, are significant under the 5% significance level.

This implies that inflation have a negative impact on the returns of OMXS30, while an increase in money supply have a positive impact on the returns of OMXS30. The variable representing exchange rate, LERd, didn’t show any significance, indicating that OMXS30 moves independently regardless of this variable. According to this model, OMXS30 is also dependent of its own second and third lag. Although these lags are significant at the 5%

significance level, as well as the fourth lag being significant at the 10% significance level, it’s difficult to conduct any inference about these findings considering that the second lag have a negative coefficient while the third have a positive coefficient.

Table 3. Regression output from VAR(1)-model with LOMXS30d as the dependent variable.

Variable Coefficient Std. Error p-value

LOMXS30d L1 0.007 0.072 0.927

LCPId L1 -0.537 1.032 0.604

LM3d L1 0.175 0.247 0.479

IRd L1 -0.031* 0.017 0.076

LERd L1 0.012 0.114 0.914

LOMXS30d L2 -0.154** 0.071 0.031

LCPId L2 -0.096 1.048 0.927

LM3d L2 0.314 0.243 0.198

IRd L2 -0.024 0.018 0.193

LERd L2 -0.006 0.110 0.958

LOMXS30d L3 0.173** 0.070 0.015

LCPId L3 -2.445** 0.991 0.014

LM3d L3 0.208 0.236 0.379

IRd L3 0.004 0.019 0.838

LERd L3 0.002 0.110 0.988

LOMXS30d L4 -0.017 0.068 0.797

LCPId L4 -1.466 0.994 0.142

19

LM3d L4 0.465** 0.232 0.046

IRd L4 -0.025 0.019 0.189

LERd L4 -0.087 0.111 0.434

LOMXS30d L5 0.129* 0.069 0.064

LCPId L5 -1.136 1.055 0.283

LM3d L5 0.359 0.233 0.126

IRd L5 -0.010 0.018 0.557

LERd L5 -0.138 0.113 0.222

LOMXS30d L6 0.064 0.069 0.353

LCPId L6 -1.099 1.054 0.299

LM3d L6 0.090 0.229 0.693

IRd L6 -0.019 0.018 0.266

LERd L6 -0.009 0.113 0.933

R2 0.222

Adj. R2 0.106

Sample size N = 239

Note: * = significant under 10%

significance level

** = significant under 5% significance level

*** = significant under 1%

significance level

5.3 Residuals diagnostics

To verify that the results from our VAR-model are robust and reliable, we conduct a series of test to make sure that the residuals of the model are white noise, as discussed in chapter 4.3.

There should not be any systematic information in the residuals, since this would imply that there is information not captured by the VAR-model, affecting our estimates.

The first test we conduct is the Breusch- Godfrey test. It is used to assess whether there is any serial correlation in the residuals or not. The test is conducted with the following hypothesis:

H0: No autocorrelation H1: Autocorrelation

Then we proceed to test for the presence of heteroscedasticity. To examine this, we use the Breusch- Pagan test. It tests whether the variance of the residuals is dependent on the values of the independent variables, using the following hypothesis:

20 H0: Homoscedastic (no heteroscedasticity) H1: Heteroscedastic

The final test we conduct is the Jarque-Bera test, which is used to examine if the error terms follows the normal distribution or not. The hypothesis for this test is stated as follows:

H0: Residuals are normally distributed H1: Residuals are not normally distributed

These tests are conducted with a 5% significance level as stated in chapter 4.3. the results from these tests can be seen in table 4.

Table 4. Results from Breusch- Godrey, Breusch- Pagan and Jarqu- Bera tests

Test Null hypothesis p-value Results Conclusion

Breusch- Godfrey No autocorrelation 0.183 Do not reject H0

No autocorrelation in the residuals Breusch- Pagan Homoscedastic 0.139 Do not reject

H0

Residuals are homoscedastic Jarque- Bera Residuals are

normally distributed

< 2.2e-16 Reject H0 Residuals are not normally distributed

Considering that we received p-values less than our critical value of 0.05 for both the Breusch- Godfrey and the Breusch- Pagan tests, we can conclude the absence of both autocorrelation and heteroscedasticity in or model. However, this is not true for the Jarque- Bera test, we’re we must reject the null hypothesis in favor of the alternative hypothesis that the residuals do not follow the normal distribution. The non-normal behavior of the residuals can possibly be derived from the from the cyclical behavior of the CPI (appendix figure 2), and the relatively big changes in interest rate associated with the financial crisis 2007-2008, but the impact of this event is not formally tested. Although the residuals don’t follow the normal distribution according to the Jarque- Bera test, we find our p-values from the regression output (table 3) to be reliable, since we have a large sample size in our VAR- model.

It’s difficult to construct a model where the residuals are perfect white noise, and this is confirmed by the rejection of the null hypothesis in the Jarque- Bera test. However, since we

21

successfully concluded the absence of both heteroscedasticity and autocorrelation, we will proceed with our analysis as if the residuals do not contain any systematic information.

5.4 Granger Causality

As the final part of this analysis, and as discussed in chapter 4.3, we want to investigate whether the time series of our dependent variables are significant in forecasting the return of OMXS30. To investigate this, we’re conducting Granger causality tests, which aim at determining whether lags of a variable is useful in predicting changes in another variable.

Also, the statement “Y Granger causes X” is true if Y is useful in predicting the values of X (Hiemstra & Jones, 1994).

This test, just like the previous ones, are done with a significance level of 5%, and with a lag of 6, based on the Aikake Information Criteria from chapter 5.1. It’s done after we have taken the first difference of all variables, because we want all variables to be stationary. In other words, the variables are in the same form as in the VAR-model. The results can be seen in table 6, and the following hypothesis have been used for the test:

H0: Y does not Granger cause X H1: Y Granger causes X.

Table 5. Granger Causality tests between stock index and macroeconomic variables

Null hypothesis p-value Result Conclusion

LCPId does not Granger cause LOMXS30d 0.002 Reject H0 Unidirectional LOMXS30d does not Granger cause LCPId 0.158 Do no reject H0 relationship

LM3d does not Granger cause LOMXS30d 0.447 Do no reject H0 Unidirectional LOMXS30d does not Granger cause LM3d 0.017 Reject H0 relationship

LERd does not Granger cause LOMXS30 0.798 Do no reject H0 No relationship LOMXS30 does not Granger cause LERd 0.288 Do no reject H0

22

IRd does not Granger cause LOMXS30d 0.002 Reject H0 Bidirectional LOMXS30d does not Granger cause IRd 0.025 Reject H0 Relationship

The results from the Granger Causality tests suggests that inflation Granger causes the pricing of OMXS30, which is aligned with our findings from the VAR-model. We can also conclude that LOMXS30d Granger causes the money supply. Both these relationships are

unidirectional, but whether the pricing of OMXS30 affects the money supply is not a part of our research question. Like our VAR-model, the Granger causality test leaves us with no indication of exchange rate having any relationship with the returns of Swedish stocks.

Finally, the test implies that there is a bidirectional relationship between the interest rate and the OMXS30. This is aligned with our discussion in chapter 4.3 where we briefly spoke about the dynamics of the stock market.

6. Discussion

As we mentioned in our introduction, a stock exchange is a secondary market where current stock owners can transact securities of various companies with buyers at a given price. We have in this study performed analyses to determine what impact a few selected

macroeconomic variables have on the Swedish stock market, this by using VAR and Granger causality tests. Our results from the VAR-model showed that the inflation rate and interest rate are negatively correlated with the return of OMXS30.The results also indicates that the money supply have a positive relationship with stock returns, while the exchange rate showed no significant correlation with stock prices. The results from the Granger causality test

suggest that both interest rate and inflation rate have predictive powers, i.e. that according to the test, both variables can be useful in predicting the return of OMXS30.

The negative relationship between inflation rate and stock returns can be explained by looking at the theory presented by Fama (1981) where he argues that there is a positive correlation between stock prices and real economic activity but also a negative relationship between real economic activity and inflation rate. Real economic activity concerns the production and purchase of goods and services within the economy. Higher inflation leads to lower real economic activity which affects stock prices negatively. This negative correlation between inflation rate and stock prices can also be found in previous studies such as Sirucek (2012),

23

Khan & Khan (2018) and Hsing & Hsieh (2012). The opposite result can be found in

Limpanithiwat & Rungsombudpornkul (2010) that found no significant relationship between inflation rate and stock prices.

We had one significant lag that showed a positive relationship between the money supply and stock returns. However, the Granger causality test showed unidirectional relationship between the variables, where OMXS30 Granger causes the money supply. This positive correlation between these two variables can also be seen in Khan & Khan (2018) while other studies such as Sirucek (2012) found a very minor relationship between the variables.

We found a significant negative relationship between the interest rate and stock returns in the VAR-model at the 10% significance level. These variables were also shown to have a

bidirectional relationship in the Granger causality test, meaning they affect each other. This is aligned with the results from previous studies such as Nasseh & Strauss (2004), Hassan &

Nasir (2008) and Piero (2016). Hassan & Nasir (2008) found that an increase in interest rate leads to an increased discount rate which means a decrease in present value of future cash flows which negatively affects stock prices. This theory could explain our findings.

Finally, no significant relationship was found between the exchange rate and stock prices. Our results differ from previous studies such as Khan & Khan (2018) and Sampath (2011).

However, we believe that it’s possible to further improve this study, getting more significant and robust results. For example, having weekly data instead of monthly data would of course imply a greater sample size. This would also increase the probability of finding more

significant lags. Sadly, there is no weekly data of money supply available in Sweden, which is why we were stuck with monthly observations.

This study also sheds little or no light on the drastic events of the dot com bubble in the early 2000s or the financial crisis of 2007-2008. This caused some erratic movements among our variables, but the impact of these events on the outcome of our regression or the Granger causality tests we’re never formally tested.

To summarize we found the relationships between inflation & money supply and the return of OMXS30 to be significant at the 5% significance level, as well as the interest rate variable being statistically significant on the 10% significance level. This can be tied together with

24

APT that we described in our theories, one variable that affects stock prices in the APT model is the asset price sensitivity to a certain macroeconomic variable, which is confirmed by our results. However, we would recommend any further studies on this subject to be conducted with weekly data. It would also be interesting to see whether the usage of another model would affect the results, considering the various types of models used in the previous studies and the differences in their results. A more complete study should also further investigate the impacts of any type of economic or financial crisis.

7. Conclusion

The aim of this study was to investigate the impact certain macroeconomic variables have on the Swedish stock market index Stockholm OMX30. Our study concluded that the inflation rate and money supply have a statistically significant relationship with stock prices at the 5%

significance level, and interest rate at the 10% significance level. The Granger causality test suggests a bidirectional relationship between interest rate and OMXS30. The test also implies that inflation can be useful in predicting the return of OMXS30. As for money supply, the Granger test implies that there also is a unidirectional relationship between money supply and the return on stocks, but in the opposite direction, i.e. that the returns of OMXS30 is useful in predicting the money supply. Furthermore, our study concluded that exchange rate has no statistically significant relationship with stock prices.

25

References

Cheung, Y.W. and Lai, K.S. (1995). Lag order and critical values of the augmented Dickey–

Fuller test. Journal of Business & Economic Statistics, 13(3), pp.277-280.

Fama, E. F. (1981). Stock returns, real activity, inflation, and money. The American economic review, 71(4), 545-565.

Fisher, I. (1930). The Theory of Interest. New York: M.

Gredenhoff, M. and Karlsson, S. (1999). Lag-length selection in VAR-models using equal and unequal lag-length procedures. Computational Statistics, 14(2), pp.171-187.

Hasan, A. and Nasir, Z. M. (2008). Macroeconomic factors and equity prices: An empirical investigation by using ARDL approach. The Pakistan Development Review, 501-513.

Hiemstra, C. and Jones, J.D. (1994). Testing for linear and nonlinear Granger causality in the stock price‐volume relation. The Journal of Finance, 49(5), pp.1639-1664.

Hillier, D., 2013. Corporate finance. London [etc.]: McGraw-Hill Education.

Hsing, Y. and Hsieh, W. (2012). Impacts of macroeconomic variables on the stock market index in Poland: new evidence. Journal of business economics and management, 13(2), pp.334–343.

International Monetary Fund. (2021). Currency Composition of Official Foreign Exchange Reserves (COEFR). Available: https://data.imf.org/?sk=E6A5F467-C14B-4AA8-9F6D- 5A09EC4E62A4

Khan, J. and Khan, I. (2018). The impact of macroeconomic variables on stock prices: A case study Of Karachi Stock Exchange. Journal of Economics and Sustainable Development, 9(13), pp.15-25.

26

Limpanithiwat, K. and Rungsombudpornkul, L. (2010). Relationship between inflation and stock prices in Thailand.

Malkiel, B.G. and Fama, E.F. (1970). Efficient capital markets: A review of theory and empirical work. The journal of Finance, 25(2), pp.383-417.

Mankiw, N.G. and Taylor M.P. (2017). Economics. Fourth edition. Hampshire: Annabel Ainscow.

Nasseh, A. and Strauss, J. (2004). Stock prices and the dividend discount model: did their relation break down in the 1990s?. The quarterly review of economics and finance, 44(2), 191-207.

Persons, W. (1911). Fisher's "The Purchasing Power of Money". Publications of the American Statistical Association, 12(96), pp.818-829.

Peiró, Amado, 2016. Stock prices and macroeconomic factors: Some European evidence.

International review of economics & finance, 41, pp.287–294.

Pradhan, Rudra P, 2018. Development of stock market and economic growth: the G-20 evidence. Eurasian economic review, 8(2), pp.161–181.

Sampath, T. (2011). Macroeconomic Variables and Stock Prices in India: An Empirical Analysis. IUP Journal of Monetary Economics 9.4 : pp.43-55.

Shrestha, M.B. and Bhatta, G.R. (2018). Selecting appropriate methodological framework for time series data analysis. The Journal of Finance and Data Science, 4(2), pp.71-89.

Sirucek, M. (2012). Macroeconomic variables and stock market: US review. IJCSMS International Journal of Computer Science and Management Studies, Vol. **, Issue **, Month Year ISSN (Online): 2231-5268

27

Statistics Sweden. (2021). Money Supply. Available: https://www.scb.se/en/finding- statistics/statistics-by-subject-area/financial-markets/financial-market-statistics/financial- market-statistics/pong/tables-and-graphs/money-supply/

Stiglitz, J.E. (1972). On the optimality of the stock market allocation of investment. The Quarterly Journal of Economics, 86(1), pp.25-60.

Sveriges Riksbank. (2021). Sök räntor & valutakurser. Available:

https://www.riksbank.se/sv/statistik/sok-rantor--valutakurser/

Tobin, J. (1969). A General Equilibrium Approach To Monetary Theory. Journal of Money, Credit and Banking, 1(1), pp.15-29.

Tsiaplias, Sarantis, 2008. The CPI and Other Measures of Australian Inflation. Australian economic review, 41(1), pp.105–113.

Zivot, E. and Wang, J. (2006). Vector autoregressive models for multivariate time series.

Modeling Financial Time Series with S-Plus®, pp.385-429.

28

Appendix

Table 1. The constituents of OMXS30, with index weight as a percentage of the total index.

Company Weight (%)

ABB Ltd 2.80

Alfa Laval 2.26

Assa Abloy B 5,28 Astra Zenica 2,30 Atlas Copoo A 5,99 Atlas Copoo B 2,56 Autoliv Inc. SDB 1,15

Boliden 1,84

Electrolux B 1,75

Ericsson B 7,02

Essity B 4,44

Getinge B 0,84

Hennes & Mauritz B 6,07

Hexagon AB 4,35

Investor B 5,06

Kinnevik B 1,63

Nordea Bank 4,26

Sandvik 5,33

Securitas B 1,41

SEB A 4,90

Skanska B 1,64

SKF B 1,80

SSAB A 0,26

SCA A 1,31

Svenska

Handelsbanken A

4,94

Swedbank A 4,29

Swedish Match 2,02

Tele2 B 2,08

Telia Company 4,32

Volvo B 6,11

29

Figure 1. Data graph before and after first difference for LOMXS30

Figure 2. Data graph before and after first difference for LCPI

Figure 3. Data graph before and after first difference for LM3

30

Figure 4. Data graph before and after first difference for IR

Figure 5. Data graph before and after first difference for LER

31

Figure 6. Akaike information criterion (AIC) used for lag selection.

Figure 7. VAR-output with LCPId as the independent variable

32

Figure 8. VAR-output with LM3d as the independent variable

Figure 9 VAR-output with IRd as the independent variable

33

Figure 6 VAR-output with LERd as the independent variable